Category: Inpatient Practice

Artificial intelligence (AI) is causing a paradigm shift in the education, business, and legal professions. But AI is also poised to irreversibly change the way medicine is practiced. Jobs that traditionally relied on extensive training by memorization may be able to be performed as well (or better) by AI than by humans. Jobs that have relied on image analysis or sound pattern analysis are also at risk of being displaced by AI.

Let me give you an example from my hobby of birdwatching. In the past, bird species identification relied on comparing a bird that you saw in a tree to a drawing or photograph in a bird identification book. To become an expert birder, you needed thousands of hours of birding to identify birds by their calls and by their seasonal plumage. But now, we have the Merlin app. When birding, I can turn on my phone’s microphone and Merlin will identify bird species by bird calls. I can take a photograph of a bird, upload it to the app, and Merlin will tell me what bird I saw. With Merlin, even a novice birder like me can identify birds like a seasoned expert.

So, what if we have the Merlin app equivalent to identify heart sounds by auscultation? Or diagnose a rash by a photograph? Or interpret CT scan images? Or read cytology specimens on microscopic slides? Like it or not, artificial intelligence is coming to medicine and it will make many skills traditionally requiring hundreds of hours of training, obsolete.

Artificial intelligence and advance practice providers

To become a primary care physician (general internist, general pediatrician, or family physician) requires 11 years of education and training after high school. To become a nurse practitioner (NP) or physician assistant (PA) only requires 6 years of training after high school. As a result, it is far less expensive to become an NP or PA than to become a primary care physician but the trade-off is that NPs and PAs generally have a  lower annual income than physicians. However, if the salary of an NP and a primary care physician was the same, most hospitals would prefer to hire the physician under the presumption that additional 5 years of training to be a primary care physician would translate to greater skill and knowledge than the NP or PA. On the other hand, if the skillset and knowledge of an NP or PA was the same as that of a primary care physician, most hospitals would prefer to hire the NP or PA because they are cheaper.

Artificial intelligence now offers the possibility of eliminating the need to acquire many of the skills and much of the knowledge currently required to become a physician. This offers a future where an NP armed with a few AI apps may be able to perform many of the tasks currently relegated to physicians.

Need to diagnose a child with a fever and a rash? There’s going to be an app for that. Need to decide the best blood pressure medication to prescribe for a patient with newly diagnosed hypertension? There’s going to be an app for that. Need to recommend follow-up of a pulmonary nodule given a patient’s age and smoking history? There’s going to be an app for that.

The FDA and medical devices

At first glance, it would seem logical to embed artificial intelligence into electronic medical record (EMR) software programs. After all, the EMR is the database of all information about a patient – their blood pressure, their family history, their medication list, etc. However, a barrier to incorporating AI into the electronic medical record is that the U.S. Food and Drug Administration considers AI to be a medical device whereas the electronic medical record is just considered a documentation tool. Medical devices are regulated differently than documentation tools. Medical devices require extensive clinical trials and then FDA approval; documentation tools do not. Clinical trials and FDA regulation are very expensive and can pose a barrier to regular EMR software upgrades. For these reasons, the major electronic medical record companies have been reluctant to incorporate artificial intelligence algorithms into their EMR programs.

For the most part, this makes sense. You don’t want to have an artificial intelligence program to recommend a chemotherapy regimen for advanced lung cancer unless it has been shown in clinical trials to be accurate and has been approved by the FDA. The fear of the electronic medical record companies is that if their EMRs become classified as medical devices, then they will have to get FDA approval every time they want to change the font size in their blood chemistry test results in the EMR. So, at least for now, the electronic medical record and artificial intelligence programs will need to be separated, and that means that there will have to be a human to do a history and physical examination and then to interface between the EMR and the AI. But in many situations, that human can be an NP or a PA, rather than a physician.

Artificial intelligence and primary care

Much of primary care is based on clinical practice guidelines. The U.S. Preventative Services Taskforce has guidelines for everything from colon cancer screening to pre-exposure prophylaxis to prevent HIV. The American College of Cardiology has a hypertension diagnosis and management guideline. The Advisory Committee on Immunization Practices has guidelines for childhood and adult vaccination schedules. And the American Diabetes Association has a guideline for the prevention, diagnosis, and treatment of diabetes. If you roll all of these clinical practice guidelines into one artificial intelligence program, then you have the majority of primary care medicine routine visits covered.

As a medical student, I spent hours memorizing vaccination schedules, hypertension treatment algorithms, diabetes medication drug interactions, and the staging systems for various cancers. And guess what? An artificial intelligence program can do all of these things better than my memory allows me to do. In other words, AI eliminates the need for much of the education and training that we currently require in medical school and residency. Artificial intelligence will allow a practitioner with lesser training (such as an NP or PA) to be just as good as a physician when it comes to preventative care medicine and algorithm-based management of most common medical conditions.

However, artificial intelligence is not infallible

Artificial intelligence is actually not new in medicine. I’ve been using simple forms of AI for decades. Every EKG and pulmonary function test that I have ordered in the past 30 years that comes with a computer interpretation has incorporated rudimentary AI into those interpretations. These interpretations programs are fairly good at identifying normal but invariably come up with an incorrect diagnosis in a substantial percentage of those tests that are abnormal. So, before I am willing to allow an AI program to diagnosis breast cancer from a histopathology slide and before I am willing to allow an AI program to diagnose idiopathic pulmonary fibrosis from a chest CT scan, these programs are going to have to get very, very good. Until then, the use of artificial intelligence for more complex pathologic and radiologic diagnoses will supplement rather than replace a physician.

And then there is legal liability…

If a radiologist misses a lung cancer on a chest X-ray, the radiologist is named in a medical malpractice lawsuit. If a patient dies of sepsis when a hospitalist made an incorrect antibiotic choice for the patient’s pneumonia, the hospitalist is named in the malpractice suit. But if an artificial intelligence program misses the lung cancer or selects the wrong antibiotic, who gets named in the malpractice case? The company that created the AI program? The hospital that purchased the AI program? The FDA that approved the program? The physician who entered the patient’s clinical data into the program? All four of them?

Currently, a physician in primary care practice will pay about $12,000 per year in malpractice insurance premiums whereas a primary care nurse practitioner pays about $1,200. The reason for the 10-fold difference is that in most situations, a nurse practitioner is understood to be working under the supervision of a physician and that physician is ultimately responsible or at least shares responsibility for the management of patients seen by the nurse practitioner.  Artificial intelligence is likely to be similar – if it is considered to be a medical device then that device will need to be used by a licensed medical practitioner who will have the greater burden of malpractice liability. Clearly, laws will need to be written to clarify liability before artificial intelligence can be autonomously implemented in clinical practice.

Who will AI benefit the most – nurse practitioners or physicians?

A recent study from MIT researchers found that artificial intelligence has the greatest impact on the least skilled workers. Workers who were new or had low skills were helped more by AI than highly skilled workers. In other words, AI allows those with less training to be “upskilled” much more than those with advanced training.

Extrapolating from this study, it is likely that nurse practitioners and physician assistants will derive greater benefit from artificial intelligence than physicians. Artificial intelligence can make up for the fewer years of training that it takes to become an NP or PA.

Which physicians are most vulnerable to being displaced by artificial intelligence?

Although artificial intelligence has received a lot of press about its potential in radiology, I would argue that primary care physicians are most vulnerable to being displaced by artificial intelligence. Notice that I used the word “displaced” rather than “replaced”. That is because artificial intelligence is likely to be used to supplement a practitioner rather than become a practitioner, at least in the foreseeable future. In this regard, an NP or PA using an artificial intelligence program can replicate much of the skillset of a primary care practitioner. Thus the combination of an NP or PA plus an artificial intelligence program will together displace the primary care physician.

Physicians who are the least vulnerable are those who perform procedures such as surgeons and interventional cardiologists. Although this could change in the future, for now, no AI program or nurse practitioner is capable of independently performing a hip replacement surgery or a coronary artery stent placement. In primary care practice, the office procedures are far less complex – cerumen removal, IUD placement, and laceration suturing can be performed by an NP or PA and do not require a physician.

Also less vulnerable are physicians who are highly specialized. For example, an artificial intelligence program for brain MRI imaging will need to be used under the supervision of a practitioner who can confirm or contradict the AI’s findings. This will require a practitioner who is already an expert in brain MRI image interpretation, in other words, a physician specializing in neuroradiology. Artificial intelligence can still benefit the neuroradiologist, however, by serving in a capacity similar to that of a radiology resident who performs a preliminary read of the MRI that is then over-read and confirmed by the attending neuroradiogist.

“I’m a medical student, should artificial intelligence affect my career choice?”

The answer is… maybe. Fully implemented artificial intelligence in medicine is still a long way off. There will have to be significant improvements in software, significant legal liability questions resolved, and supervision requirements defined. However, if AI can replace certain medical specialists at a lower cost, then economic theory indicates that it eventually will. General internists, general pediatricians, and family physicians may be more vulnerable to displacement than other specialties, especially if the field of medical artificial intelligence matures coincident with an increase in the number of nurse practitioners and physician assistants. However, when it comes to cajoling a cardiologist to add in a patient with chest pain to their already full Friday afternoon schedule, an AI program simply cannot replace a persuasive family physician. The primary care physician may become more of a manager: coordinating care and overseeing a group of nurse practitioners who each have access to the artificial intelligence program.

Things are about to get interesting…

Change in medicine is inevitable but initial resistance to change is also inevitable. When electronic medical records were initially implemented, physicians universally hated them and many refused to use them. Now, no physician in his or her right mind would want to return to an era of paper records kept in manila folders. Ten years ago, the idea of driverless vehicles was met with skepticism but today, you can order a driverless Waymo taxi in San Francisco and you can buy a driverless John Deere tractor to plow your farm.

Artificial intelligence is coming in medicine and its widespread implementation is unavoidable. The question is whether it will augment physicians or displace physicians. I believe that it will do both, depending on the specialty.  From my vantage point, primary care physicians may be the most vulnerable to displacement. And employment opportunities for NPs and PAs are looking bright.

August 30, 2023

In March 2023, the U.S. Drug Enforcement Agency (DEA) announced new education requirements for all physicians applying for new or renewal DEA licenses. This was the result of provisions in the Consolidated Appropriations Act of 2023 that enacted a one-time requirement of 8 hours of continuing medical education (CME) on the treatment and management of patients with opioid or other substance use disorders. The requirement went into effect on June 27, 2023. Because DEA licenses are renewed on a rolling 3-year basis, all physicians with DEA licenses must meet this requirement sometime in the next 3 years.

Another provision of the Consolidated Appropriations Act of 2023 was to eliminate the DATA-Waiver (X-Waiver) Program that was previously required for physicians to prescribe buprenorphine. In the past, hospitalists, emergency medicine physicians, and other practitioners needed to obtain an X-Waiver to initiate buprenorphine when patients with opioid use disorder were admitted to the hospital or seen in the emergency department. Because only a small number of physicians took the time and effort to obtain an X-Waiver, the requirement was seen as a barrier to getting patients started on treatment. Now, any practitioner with a current Drug Enforcement Administration (DEA) registration may prescribe buprenorphine for opioid use disorder (if permitted by state law). The trade-off for elimination of the X-Waiver was the requirement that all practitioners with a DEA license be trained in the treatment of opioid use disorder, including the use of buprenorphine.

Who does this affect?

Any practitioner with a DEA registration must meet this requirement. This includes physicians, dentists, nurse practitioners, and physician assistants. However, only practitioners who prescribe controlled substances need to register with the DEA and obtain a DEA number. Although the majority of U.S. physicians have DEA numbers, some physicians do not, either by nature of their practice (for example, pathologists and researchers) or by choice (for example, general practitioners who do not want the hassle of prescribing opioids and other controlled substances).

To obtain a DEA number, a physician must apply to the DEA and pay an $888 fee. DEA numbers are valid for 3 years at which time the physician must re-apply. The DEA waives the fee for certain physicians including those who work in the military, for U.S. government hospitals or institutions, and for state government hospitals or institutions. As an employee of the Ohio State University (a state government institution), my DEA fees were waived. However, even if the fee is waived, the practitioner must still apply for and obtain a DEA number and the practitioner must still meet the new education requirements.

Certain practitioners are exempt from the new educational requirement including veterinarians, physicians board-certified in addiction medicine, and practitioners who have graduated from their professional school within the past 5 years. The latter means that most residents in training are exempt.

What are the specifics of the requirement?

When applying for a new or renewal DEA number, physicians (and other practitioners) must check a box attesting to having completed 8 hours of training on treatment and management of patients with opioid or other substance use disorders. This is a fairly broad topic area and it is up to physicians to maintain their own documentation of completion of education in the event of an audit. In addition, if the physician faces legal action (such as a medical malpractice lawsuit), documentation of completion may be necessary to establish physician competency. The details of the requirement are as follows:

• The 8 hours of education do not need to occur in one session and (for example) can be 8 individual 1-hour CME events.
• This is a one-time requirement and will not need to be repeated every three years when re-applying for a DEA number.
• Education can take the form of grand rounds, classroom sessions, on-line materials, or professional society meetings.
• Education hours obtained prior to the new requirement also count. For example, attending a grand rounds on buprenorphine in past years can count; just be sure that you have documentation of participation or attendance. Physicians with an X-Waiver can count the training hours from their original X-Waiver application.
• The education can come from any organization accredited to provide CME credits by the Accreditation Council for Continuing Medical Education.

What do hospitals need to do?

Although the DEA requirement is left to the responsibility of the individual practitioner applying for a DEA number, hospitals do have an obligation to facilitate education. First, if practitioners fail to get the required 8 hours of training and are unable to obtain a DEA number, the hospital’s ability to dispense controlled substances or manage patients requiring controlled substances will be compromised. Second, in the event of a medical malpractice lawsuit involving a practitioner on the medical staff who lacks documentation of completion of the educational requirements, the hospital could be accused of being complicit by not confirming that their practitioners were appropriately trained. Specific steps that hospitals should take now include:

• Make sure that all members of the medical staff are aware of the new DEA requirements.
• Inventory practitioners’ DEA license expiration dates and remind practitioners at least 6 months in advance of that date that they must fulfill the educational requirements prior to the renewing their DEA number.
• Require practitioners with DEA numbers to submit documentation of completion of the educational requirements and then maintain that documentation in each practitioner’s employment record.
• Require any new practitioners to include documentation of completion of substance abuse treatment CME as part of their application to the medical staff. Those lacking documentation should be required to complete training during their provisional/probational appointment period.
• Schedule grand rounds or other CME events covering treating and managing patients with opioid or other substance use disorders.
• Provide practitioners with links to on-line CME resources. For nearly a quarter of a century, I moderated the CME webcast, OSU MedNet-21. We produced many CME webcasts on substance abuse disorders and these webcasts are available to anyone. A recent example is:
  • Addiction, Stigmata, & First-Person Language
• Many professional societies have included sessions on substance abuse disorders as part of their annual meetings or have prepared on-line CME sessions to help fulfill the requirements. Examples of on-line education programs include:
  • The American College of Physicians
  • The American Society of Addiction Medicine
  • The American Medical Association
  • The American Society of Anesthesiologists
  • The American Academy of Pediatrics
• The Centers for Disease Control offers a free on-line 1-hour CME activity about substance abuse disorders
• Journal subscription materials can count. Practitioners with subscriptions to resources such as UpToDate, JAMA, and the New England Journal of Medicine can obtain CME credit by reading relevant articles and then applying for CME hours.

Why has Congress required this?

The primary impetus for the new requirement is a directive of the U.S. Congress to address the opioid epidemic. Eliminating the X-Waiver program was seen as a way of improving access to treatment for patients with opioid use disorder. But to justify elimination of the X-Waivers, Congress needed a mechanism to ensure that all practitioners were knowledgable in initiating treatment for opioid use disorder.

In 2021, a total of 106,699 Americans died of a drug overdose. Although street-purchased fentanyl was the most common drug implicated, prescription opioids accounted for 16,706 of the overdose deaths in 2021. In fact, the number of deaths from prescription opioids exceeded the number of deaths from heroin (9,173).

Drug overdose deaths are particularly high in Appalachian states. West Virginia has the highest overdose death rate at 90.9 per 100,000 population, followed by Kentucky and Tennessee (each 56.6 per 100,000 population) and Louisiana (55.9 per 100,000 population). My state of Ohio ranks 7th highest at 48.1 per 100,000 population. Nebraska comes in lowest at 11.4 per 100,000 population.

To put these numbers in perspective, last year, the U.S. COVID death rate was 61.3 per 100,000 population. Opioids are abused by more than 10 million Americans each year (3.8% of Americans) and 2.7 million Americans have an opioid use disorder. About half of those who become addicted to opioids first use opioids in the form of prescription pain medications. An estimated 3% – 19% of people who take prescription opioid pain medications will become addicted to opioids. Addiction can occur with only 3-5 days of prescription opioid use.

The good news is that there are effective treatments for opioid use disorder including buprenorphine (often combined with naloxone), methadone, and naltrexone. In addition, the FDA has now approved naloxone to be sold over-the-counter to treat opioid overdose. The goal of the DEA education requirements is that any practitioner in the U.S. who is licensed to prescribe opioids is also trained in identifying and treating opioid abuse.

A quarter of a century of change

In the 25 years since the American Pain Society advocated that physicians adopt “pain as the 5th vital sign” and since Purdue Pharmaceuticals falsely promoted OxyContin as a non-addictive opioid, physicians have become much more aware of the role that we have played in catalyzing the current opioid epidemic. The new DEA education requirements were created as one step in remedying the epidemic. By helping our physicians meet these new requirements, hospitals can help reduce the number of Americans who become addicted and help increase the number of Americans who get their addiction treated.

August 28, 2023

A recent study in JAMA Internal Medicine showed that the number of hours physicians work per week has fallen significantly over the past 20 years. The data was derived from the U.S. Census Bureau’s Current Population Survey that included 87,297 monthly surveys of physicians between 2001 and 2021. During this 20-year period, the average number of hours worked per week has steadily fallen.

In the study, respondents were asked how many total hours they worked at all jobs during the previous week. The average weekly work hours from 2001 – 2003 were compared to average weekly work hours from 2019 – 2021. Overall, the average physician worked 52.6 hours per week in 2001 – 2003 and this number dropped to 48.6 hours per week in 2019 – 2021. When only physicians working full-time were included, the work hours decreased from 55.6 hours per week in 2001 – 2003 to 51.1 hours per week in 2019 – 2021.

There has been a change in work hours by physician age. In the time period 2001 – 2003, the youngest physicians (age 35 – 44) worked the most hours per week, followed by middle aged physicians (age 45 – 54), and then older physicians (age 55 – 64). In the more recent time period of 2019 – 2021, the opposite was true – older physicians (age 55 – 64) worked the more hours per week than younger physicians.

On average, male physicians currently work more hours per week (49.7 hours) than female physicians (46.8 hours). However, this gap has been narrowing with a gender difference of 5.3 hours per week in 2001 – 2003 versus a gender difference of 2.9 hours per week in 2019 – 2021. Differences in the percentage of women in different specialties may be responsible for some of the gender difference in hours worked per week. Other physician demographic variables did not differ significantly in the current number of hours worked per week including race, country of origin, urban vs. rural, and dual household earners versus single household earners.

The 2023 Medscape Physician Compensation Report surveyed 10,011 physicians between October 7, 2022 and January 17, 2023. The report found that there is substantial variation in physician work hours among different specialties. Hospital-based specialties that tend to involve patient care at night and on weekends work the most hours per week including critical care, general surgery, cardiology, and nephrology. On the other hand, outpatient specialties that generally do not require seeing patients at night or on weekends had the lowest work hours per week including allergy, dermatology, and ophthalmology. Emergency medicine also had a low number of hours worked per week, owing to the shift work nature of the specialty.

A problem with the Medscape survey is that all of the information is self-reported and thus susceptible to either over-estimation or under-estimation. Furthermore, the survey is voluntary and the physicians who choose to report data may not be truly representative of the population of physicians as a whole. Nevertheless, the recent Medscape data for work hours by specialty are remarkably similar to a 2011 study published in the Archives of Internal Medicine.

At the onset of the COVID-19 pandemic (during the 2nd quarter of 2020), the average number of hours physicians worked per week fell as elective procedures were canceled. However, weekly work hours quickly rebounded by the 3rd quarter of 2020 as shown in the graph below derived from data from the recent study in JAMA Internal Medicine. Notably, over the the 24 quarters from 2016 through 2021, the overall trend has been for physician work weeks to become shorter.

Do work week hours correlate with income?

Intuitively, one might assume that specialties with longer work weeks also have higher incomes. However, the 2023 Medscape Physician Compensation Survey indicates that there is little relationship between the number of hours worked per week and the annual compensation by specialty. The graph below shows average number of hours worked per week in red and annual compensation in blue for 29 specialties.

By combining the data from the two graphs above, we can calculate the average income per work-week hour. Note that this is not same as hourly compensation – that would require dividing the income per work-week hour by the number of weeks worked per year. The Medscape Physician Compensation Survey did not report the number of work vs. vacation weeks per year by specialty so true hourly compensation cannot be determined. The graph below shows that plastic surgeons, orthopedic surgeons, dermatologists, and radiologists have the highest income per work-week hour. Infectious disease specialists, family practitioners, pediatricians, and general internists have the lowest compensation per work-week hour. Notably, family medicine, pediatrics, and general internal medicine have shorter residencies (3 years) than the other specialties.

As stated in a previous post, the data for infectious disease is particularly alarming. Physicians specializing in infectious disease have the lowest income per work week hour of all 29 specialties in the Medscape Physician Compensation Survey. In order to become board-certified in infectious disease, a physician must first complete a 3-year internal medicine residency and then complete a 2-year infectious disease fellowship. However, infectious disease specialists have both a lower total annual income and a lower income per work week hour than general internists. The implication is that an infectious disease specialist is financially penalized for doing a 2-year fellowship after internal medicine residency. As a result, many infectious disease specialists are either supplementing their income by working part-time as hospitalists or are leaving the specialty of infectious disease altogether to work as general internists or hospitalists. Because of the nature of physician billing and RVU determination, it is not possible for infectious disease specialists to increase their income by professional billing alone. There is an urgent need for U.S. hospitals to financially supplement their infectious disease specialists in order to preserve the infectious disease physician workforce.

Physicians are working fewer hours but are they happier?

The Medscape Physician Compensation Report found that 73% of physicians would choose medicine again if they were just now starting their career. However, the Medscape survey five years ago found that 77% of physicians would choose medicine again, indicating that physicians are less satisfied with their careers now than five years ago.

Undoubtably, the COVID pandemic has had an impact on physician job satisfaction. During the pandemic, many physicians retired early or left the workforce for other jobs. In 2019, there were 989,684 clinically active physicians practicing in the United States. In 2021 that number fell to 923,419, a 6.7% decrease. Hopefully, as the pandemic winds down and the practice of medicine gets back to normal, the exodus of physicians from the profession will slow.

The continued creep in paperwork and administrative tasks is a dissatisfying factor for many doctors – physicians reported spending an average of 15.5 hours per week on these chores, of which 9 hours are for electronic medical record documentation. Advances in artificial intelligence technology offers hope that the use of electronic medical records will be streamlined in the near future, giving physicians more time to engage in direct patient care. Over the past decade, there have been increasing concerns raised about the extent of physician burnout. Long work hours have been suggested to be a cause of physician burnout but the data suggests that physicians are less satisfied despite working fewer hours than in the past.

What about nurse practitioners and physician assistants?

The large number of retiring physicians coupled with reduced physician work week hours indicates that the overall supply of physician services is declining. Over the past decade, this has been offset by an increase in nurse practitioners. It is far less expensive to train a physician assistant or nurse practitioner (6 years education post-high school) compared to a physician (11 to 16 years education post-high school, depending on specialty). Many services traditionally performed by physicians can be equally or near-equally performed by PAs and NPs. However, specialized medical care and complex procedures still require the additional training and experience of physician specialists and so there is a limit to the degree that PAs and NPs can substitute for physicians.

Are the numbers good or bad?

The reduction in physician work hours is both good and bad. It indicates an improvement in a profession that has historically been seen as arduously time-consuming. But it also implies reduced availability of physician services to the general population. In the future, reduction in administrative and paperwork time coupled with strategic utilization of NPs and PAs could allow physicians to enjoy a good lifestyle while still ensuring that Americans have access to the best possible healthcare.

May 2, 2023

Supplemental oxygen is one of the most commonly prescribed treatments in the hospital. Patients with acute or chronic lung disease depend on supplemental oxygen to stay alive. As a natural and necessary substance, oxygen would seem like a safe thing to prescribe for patients that need it. But sometimes too much of a necessary thing can be harmful. So, when is too much oxygen bad for patients? We can take a lesson from vitamins.

The supplemental vitamin industry is based on the tenet that if our bodies need a small amount of a vitamin to live, then it must follow that by supplementing larger and larger amounts of that vitamin, our bodies will function better and better. Americans love this concept and we spend $30 billion every year on supplements. But frequently, too much of a good thing turns out to be a bad thing. Take Vitamin A – it is necessary for normal health and without vitamin A, we can lose night vision and become immunocompromised. A small amount of vitamin A in our diet keeps our bodies functioning normally. But if a person ingests too much vitamin A from supplements or from a dietary source that is high in it (such as polar bear liver), death can result from vitamin A toxicity. The same is true for water: drinking too little and a person dies of dehydration but drink too much and a person dies of hyponatremia. For decades, we thought oxygen was somehow different and that it was always better to err on the side of prescribing too much oxygen than risk prescribing too little. It turns out that we were wrong.

There are a lot of different oxygen delivery options: home concentrators, portable concentrators, compressed oxygen gas tanks, liquid oxygen, etc. In the hospital, oxygen is usually delivered through a medical gas panel that will have outlets for medical grade oxygen, regular air, and wall suction. The oxygen that comes out of these outlets is generally at a maximum pressure of 55 PSI. Conventional oxygen delivery devices (nasal cannulas and simple face masks) in the hospital usually are capped at a maximum flow rate of 15 liters per minute. Heated high flow nasal cannulas can deliver very high oxygen flow rates of up to 60 liters per minute. Mechanical ventilators can blend pure oxygen with room air to achieve anywhere from 21% to 100% inhaled oxygen concentrations (FiO2). As a result, physicians can adjust the amount of supplemental oxygen that a patient receives from a very small amount to a very large amount.

In the past, when EMS personnel would bring a patient with shortness of breath to the emergency department, they would put a mask on that patient and turn the oxygen up to a 15-liter flow rate to achieve close to 100% inhaled oxygen. In the hospitals, doctors would prescribe a high flow of oxygen and then not decrease that flow rate until a patient’s condition was improving and it was time to wean the oxygen down. In our emergency departments, ICUs, operating rooms, and hospital wards, we would give the patient as much oxygen as it took to raise their blood oxygen saturation to 100% and leave the oxygen at that flow rate for hours or days. But it turns out that we were probably harming patients by doing so.

In the past several years, there have been studies showing that prescribing excessively high oxygen flow rates can worsen patient outcomes in adults with lung disease. A study published in this week’s JAMA extended those findings to children. This study looked at 1,567 children, ages 1 – 4 years old, at 14 hospitals in Australia and New Zealand who had respiratory failure. The children were randomly assigned to receive either standard oxygen therapy or high-flow oxygen therapy. The high-flow group received as high as 40 liters per minute, depending on body weight. The standard-flow group received up to 2 liters per minute that was titrated to keep the blood oxygen saturation above 92%. The children receiving high-flow oxygen had a significantly longer length of hospital stay (1.77 vs. 1.50 days), longer duration of time on oxygen (1.07 vs 0.75 days), and higher rate of admission to the ICU (12.5% vs 6.9%).

Other studies have shown that excessive supplemental oxygen can be harmful in adults. A study of 429 patients given supplemental oxygen after resuscitation for cardiac arrest found that 40.7% of those receiving a conservative oxygen flow rate died but 50% of those receiving a liberal (i.e., higher) oxygen flow rate died. A meta-analysis study of 16,037 critically ill patients treated with conservative vs. liberal oxygen therapy found that those treated with liberal oxygen therapy (higher flow rates) had a higher mortality than those treated with conservative oxygen therapy (lower flow rates). A 2022 study of inpatients receiving supplemental oxygen during COPD exacerbations found that those who had arterial oxygen saturations of 88 – 92% had the lowest mortality rate and those patients given greater amounts of supplemental oxygen to maintain arterial oxygen saturation > 92% had significantly higher mortality rates.

For patients having out-of-hospital cardiac arrest, the conservative oxygen approach may be risky, however, A 2022 study from Australia found that cardiac arrest patients treated by EMS personnel with a conservative oxygen strategy (targeting a blood oxygen saturation of 90 – 94%) had a higher mortality rate than those patients treated with a liberal oxygen strategy (targeting a blood oxygen saturation of 98 – 100%). Because there was a tendency to frequently undershoot the oxygen saturation in conservative oxygen therapy group, it is possible that even brief periods of low oxygen saturations can be harmful in patients immediately after a cardiac arrest which may have resulted in the higher mortality rate.

Adverse effects of too much oxygen

So, if oxygen is so necessary for us, how can it harm us? There are several effects of excessive oxygen that can result in harm:

• Oxygen toxicity. High oxygen concentrations can damage lung cells by oxidant injury resulting from the production of substances such as superoxide anion, hydroxyl radical, and hydrogen peroxide. We sometimes see this in the intensive care unit in respiratory failure patients who require 100% oxygen concentrations for long periods of time who get into a vicious cycle of pneumonia requiring supplemental oxygen and then the high oxygen concentrations cause further lung damage resulting in the patients needing even higher oxygen concentrations in order to get enough oxygen into the blood stream to keep the body’s organs alive. These patients can end up with permanent lung scarring and never get off of the mechanical ventilator. Lung damage depends on how long a person is exposed to high concentrations of oxygen – brief periods are less harmful than breathing very high concentrations of oxygen for longer periods of time. This mainly applies to those patients in the ICU left on 60% – 100% inhaled oxygen for several days.
• Drug-induced lung disease. Many drugs can cause damage to the lung resulting in inflammation and scar in the lungs (interstitial lung disease). Certain drugs are particularly likely to do this when combined with high concentrations of inhaled oxygen. The two biggest offenders are bleomycin (used in cancer chemotherapy) and amiodarone (used in heart rhythm disorders). When a patient is breathing room air, these drugs are usually safe but when breathing high concentrations of oxygen, these drugs can become very toxic. Bleomycin is particularly notable because the risk of pulmonary toxicity from breathing high oxygen concentrations can persist for many years after the patient was given bleomycin.
• Radiation-induced lung disease. Excessively high amounts of radiation to the chest can cause interstitial lung disease but moderate amounts can be used safely to treat lung cancer. However, when a patient is using high oxygen concentrations, those moderate amounts of radiation can cause interstitial lung disease. Thus, like amiodarone and bleomycin, oxygen therapy can increase the risk of radiation-induced pulmonary fibrosis.
• Adsorptive atelectasis. Room air contains about 21% oxygen and 78% nitrogen. Oxygen readily passes from the alveoli of the lungs (air sacks) into the blood stream but nitrogen does not cross as easily. Because of this, nitrogen in the air helps keep the alveoli of the lungs propped open, similar to using PEEP (positive end-expiratory pressure) on a mechanical ventilator, thus preventing atelectasis. When a person breathes very high concentrations of oxygen, the nitrogen in the alveoli gets “washed out” and as a result, the lungs are more prone to developing atelectasis that can in turn worsen oxygenation. Once again, the patient can get into a vicious cycle of worsened blood oxygen levels leading to the doctor increasing the supplemental oxygen concentration leading to worsened atelectasis leading back to worsened blood oxygen levels.
• Carbon dioxide retention. Our breathing rate is determined by the blood oxygen level (PO2) and carbon dioxide level (PCO2). Patients with COPD are sometimes less sensitive to rising carbon dioxide levels and more dependent on the blood oxygen level to determine respiratory rates. In this setting, by giving too much supplemental oxygen, the patient can lose their respiratory drive and begin to hypoventilate, resulting in a high blood carbon dioxide level. Although this effect on PCO2 levels is usually small, it can be important when combined with other things that can suppress the respiratory drive centers, such as sedatives and opioids.
• Longer oxygen weaning times. When a patient with acute respiratory failure in the hospital begins to improve, the doctor will generally order the respiratory therapist to wean the oxygen off (or back down to the patient’s normal home supplemental oxygen flow rate). The respiratory therapist will then reduce the oxygen flow rate by 1-2 liters per minute, wait a couple of hours, recheck the blood oxygen saturation, reduce the flow rate by another 1-2 liters, etc. It takes less time to wean oxygen off if a patient is on less oxygen to begin with than if they are receiving higher flow rates of oxygen. The doctors generally wait until the oxygen is completely off (or down to the normal flow rate the patient uses at home) before the patient is allowed to be discharged. As a result, longer oxygen weaning times can sometimes lengthen the patient’s hospital stay.

So, what is the right amount of supplemental oxygen?

Our bodies require oxygen to live. If we don’t get enough oxygen, our tissues become damaged. On the other hand, too much oxygen is also bad for our bodies. Here is what we can do in our hospitals to draw the right balance between not enough and too much:

1. In emergencies, it is better to give too much than to give too little. In settings such as cardiac arrest, myocardial infarction, and stroke, brief periods of low blood oxygen levels can worsen clinical outcomes. If too much emphasis is placed on using the least amount of supplemental oxygen, there is a greater risk of undershooting the oxygen flow rate and causing low blood oxygen levels for seconds or minutes. Situations where this applies include during transport of cardiac arrest or stroke patients by emergency squads, during CPR for cardiopulmonary arrest, and during the first hours of a myocardial infarction or stroke.
2. Target a blood oxygen saturation of 88 – 92%. After the initial resuscitation, titrate the supplemental oxygen flow rate (or the FiO2 on the mechanical ventilator) to the least amount necessary to keep the oxygen saturation between 88 – 92%. For years, physicians have written orders to “Wean supplemental oxygen to maintain oxygen saturation > 92%”. This is probably too high and our default oxygen weaning orders in our electronic medical records need to be revised.
3. In the ICU, use other measures to improve oxygenation when patients need more than 60% FiO2. This could include increasing the PEEP on the ventilator, using prone ventilation, giving sedatives, controlling fever, or using neuromuscular blockade.
4. Avoid unnecessary use of 100% FiO2 during surgery. In the past, anesthesiologists would sometimes leave patients on 100% FiO2 throughout surgical operations so that there would be no risk of the patient desaturating during surgery. Not only can this increase the risk of post-operative absorptive atelectasis, but it can be particularly harmful if patients have previously received sensitizing drugs or radiation. One of my colleagues was a physician who had been treated with bleomycin years previously for lymphoma. He underwent an elective gall bladder surgery and was left on 100% FiO2 during the procedure. He developed acute lung injury post-operatively and died from bleomycin-induced acute oxygen toxicity.
5. We need pharmacologic 2,3-DPG. 2,3-Disphosphoglycerate (2,3-DPG) is a chemical in red blood cells that helps hemoglobin molecules release oxygen. Physiologically, this can result in a  shift in the oxy-hemoglobin dissociation curve to the right. This means that oxygen can pop off of hemoglobin easier, allowing the red blood cells to release more of their oxygen to tissues that need it. In other words, 2,3-DPG allows the tissues to get by when there is less oxygen in the blood. Normally, red blood cells release about 30% of their oxygen when they pass through tissues; 2,3-DPG allows them to release more than 30%. What is important to our bodies is not the amount of oxygen in the blood but rather the amount of oxygen that our tissues get. 2,3-DPG allows our tissues to function normally when the blood oxygen saturation is lower. If we could increase red blood cell 2,3-DPG levels pharmacologically, we could use lower amounts of supplemental oxygen in patients with acute respiratory failure.
6. Not everyone with a low oxygen saturation needs supplemental oxygen. Insurance will not pay for supplemental oxygen unless a patient has an oxygen saturation of < 89%, either at rest, during exercise, or during sleep. But that does not mean that every patient with brief oxygen saturations < 89% needs supplemental oxygen. The LOTT study showed that COPD patients with oxygen saturations 89 – 92% at rest that desaturated to 80 – 90% with exercise had no benefit from supplemental oxygen, including death rates, hospitalization rates, COPD exacerbations, and quality of life. In my own clinical practice, I have also been hesitant to prescribe oxygen in hyperemic patients who are at fire risk, including those who smoke, have smokers in the home, or use gas cooking stoves. These patients are often more likely to be injured or die from fire caused by supplemental oxygen than they are to die from hypoxemia without supplemental oxygen.

It’s time for auto-titrating oxygen delivery devices

Auto-titrating oxygen devices adjust the oxygen flow rate based on a patient’s oxygen saturation. A 2019 meta-analysis showed that compared to manual oxygen titration, these devices shortened hospital length of stay by shortening oxygen weaning times. A 2020 study of outpatients found that by using auto-titrating oxygen devices, patients had improved 6-minute walk distances and improved dyspnea. It is not surprising that these devices are effective in the hospital – a respiratory therapist cannot be in a patient’s hospital room 24 hours a day to constantly adjust the supplement oxygen flow rate but the auto-titrating oxygen delivery device can. Not only can these devices reduce excessive supplemental oxygen flow rates in hospitalized patients, but they can also reduce excessive oxygen flow rates in outpatients.

The cost to hospitals to buy and implement auto-titrating oxygen delivery devices would be considerable. However, these costs could potentially be off-set by shortened hospital length of stays and by reduction in utilization of respiratory therapists. These devices would also take some of the guesswork out of ordering oxygen flow rates for hospitalized patients resulting in a more standardized and consistent use of supplemental oxygen by hospitalists and intensivists. In addition, there is a potential for a reduction in hospital mortality rates by avoiding excessively high amounts of supplemental oxygen.

There is both an art of medicine and a science of medicine. It is time to harness science to guide the use of supplemental oxygen.

January 26, 2023

“Observation status” was originally conceived of as a way to watch patients presenting to the emergency department for a few hours in order to determine whether or not they needed to be admitted to the hospital. The purpose was to reduce unnecessary inpatient hospitalizations and thus reduce overall healthcare costs. But there has been a creep in the use of observation status as well as the healthcare bureaucracy to administrate it. The result is that observation status has now increased overall healthcare costs. It is time to consider eliminating observation status in the United States.

How did we get here?

Prior to 1960, emergency departments were staffed by residents and general practitioners. The doctor who saw you in the ER was generally the same doctor who took care of you in the hospital. Emergency medicine became a specialty in 1968 with the creation of the American College of Emergency Medicine. The first emergency medicine resident began training in 1970 and the first board examination in emergency medicine was offered in 1980. The result was that the doctor that took care of a patient in the ER was no longer the same doctor who took care of them once they were admitted to the hospital. In 1983, DRGs were first used to determine the amount of money that Medicare would pay hospitals for inpatient admissions for any given diagnosis. After the introduction of DRGs, it became immediately clear that there needed to be some way of determining which patients were sick enough to warrant admission to the hospital from the emergency room, otherwise, the hospitals would be incentivized to admit as many people as possible, even if they were not very sick.

Initially, that determination was left to the emergency room physician. But that ER doctor needed to find an inpatient physician who would agree to admit the patient. During my residency, we had a designated “medical admitting resident” each day who would make the decision about which patients were sick enough to require admission. Some residents got the reputation of being “sieves”, meaning that they would admit everyone that the ER physician called them about whereas other residents got the reputation of being “walls”, meaning that they would block admissions from the ER unless the patients were at death’s door. You always wanted to be on-call at night with a resident who was a “wall” because that meant you would have to do fewer history and physical exams, your inpatient service census would be lower, and you might actually get a few hours of sleep that night.

In order to provide some rules for which patients warranted inpatient admission, Medicare directed that patients who could be sent home within 24 hours should be observed in the ER rather than admitted to the hospital. However, keeping a lot of patients in ER beds for 24 hours was impractical so hospitals started putting those observation patients in regular hospital beds to avoid congestion in the emergency department. The unintended consequence was that this simply led to keeping patients in the hospital for longer than 24 hours, just so they could be classified as inpatients. This was especially a problem with outpatient procedures when many hospitals kept patients overnight for procedural recovery and then billed Medicare for both the outpatient procedure plus an inpatient hospital admission. Medicare countered in 2002 by agreeing to pay hospitals specifically for observation stays in order to provide an alternative to inpatient admission for those patients who were only mildly ill or who needed extra time to recover from an outpatient procedure. Initially, the only diagnoses that could be billed as observation stays were heart failure, chest pain, and asthma. In 2008, Medicare began paying for observation stays for all diagnoses.

In parallel with the development of reimbursement policies for observation stays, Medicare began policing hospital admissions by using the RAC (recovery audit contractors). RAC auditors would review the charts of patients who had been admitted to the hospital and if the auditor determined that there was not sufficient documentation in the chart to justify inpatient admission, Medicare would collect penalties from the hospital for overpayment. By 2014, the RAC program had collected $2.3 billion from hospital overpayments. One of the most common reasons by RAC auditors when denying an inpatient admission was that “…the patient could have safely and effectively been treated as an outpatient.” The auditors were often incentivized to deny admissions since more denials often led to bigger bonuses for the auditors. As a result, the denials were frequently capricious and arbitrary. 25 years ago, a coder for a commercial insurance company confided in me that her supervisors told her to randomly deny every 10th hospital admission because hospitals usually found that it was too expensive to contest denials. Fear of RAC audits resulted in physicians and hospitals increasing the use of observation status in order to avoid the risk of being penalized for an unnecessary inpatient admission.

A second strategy employed by Medicare was to create a list of surgical procedures that were classified as “Medicare Inpatient-Only Procedures”, meaning that those operations required an inpatient admission. Any surgical procedure not on the list was to be classified as an outpatient procedure unless there were extenuating circumstances that uniquely required a patient to be admitted to the hospital. Medicare pays much more if a surgery is performed as an inpatient (Medicare Part A) than if it is performed as an outpatient (Medicare Part B). Over time, the Medicare Inpatient-Only list shrank as more and more surgical procedures were reclassified as being appropriately done as outpatient and not requiring of hospital admission. Thus, knee and hip replacement surgeries were initially considered to require inpatient admission  but are now considered to be outpatient procedures.

Medicare also changed its definition of observation stays to be any condition that requires the patient to be in the hospital for “less than 2 midnights”. Although it could be argued that this gave hospitals longer than 24 hours to treat an observation status patient and send them home, the 2-midnight definition was somewhat arbitrary. For example, a patient presenting to the emergency department at 11:00 PM would spend 25 hours in the hospital before crossing 2 midnights but a patient presenting at 1:00 AM would spend 47 hours in the hospital before meeting the 2-midnight definition. However it is not how many midnights a patient actually spent in the hospital that Medicare auditors used when deciding whether to deny a hospital admission. Instead, it is whether the auditor believed that had the patient been managed appropriately, that patient could have been sent home before 2 midnights have elapsed. For example, if a patient came to the emergency department on Saturday evening with chest pain but the hospital could not do a cardiac stress test until Monday morning (2 midnights later), the auditor would still deny an inpatient admission since if the hospital offered 7-day a week stress tests, they could have sent the patient home on Sunday (after 1 midnight).

The observation vs. inpatient status bureaucracy

In order to avoid losing money from admission denials, hospitals started to go to great lengths to insure that the medical record contained sufficient documentation to justify every hospital admission. This was greatly facilitated by the development of electronic medical records that permitted realtime review of each patient’s hospital stay to ensure that the patient’s chart had appropriate documentation to meet Medicare’s requirements to bill that hospital stay as an inpatient admission. Some of the measures that hospitals now take in order to oversee their hospital admission practices include:

1. Physician training. When a patient is admitted to the hospital from the emergency room, the admitting physician has to enter an order directing that the patient is an “inpatient” or “observation” admission. This means that the physician has to estimate how long the patient will need to be in the hospital at the very beginning of the hospital stay and that estimated length of time dictates whether a patient will be inpatient or observation status. We now train residents in how to correctly estimate length of stay. For example, patients admitted for chest pain, syncope, and dehydration are generally observation status unless there are extenuating circumstances.
2. Nurse admission reviwers. Hospitals will generally hire a group of nurses or other healthcare workers to review every patient’s chart on a daily basis to determine if the medical record documentation justifies inpatient admission. These nurses get special training in the Medicare inpatient admission requirements. If the patient’s chart does not contain the proper documentation, the nurse’s first step is usually to contact the physician since frequently, all that is needed is an extra sentence or two in the medical record describing how sick the patient actually is. If that does not resolve the issue, the next step is to contact a physician admission advisor.
3. Physician admission advisors. Many times, the admitting physician is uncertain whether or not a patient’s illness justifies an inpatient admission order. Or the nurse reviewer’s determination is different from the physician’s admitting order for observation vs. inpatient status. For this reason, hospitals employ physicians whose main job is to arbitrate inpatient and observation orders. Often, this will be a private physician review company where the physician reviewers can access patient charts on a daily basis. Larger hospitals usually do this internally by hiring some of its own emergency medicine physicians or hospitalists to dedicate a certain number of hours per day reviewing admissions.
4. Hospital medical directors. When another layer of physician review of how to classify a patient’s admission is required, it then goes to the hospital medical director. Even for a small hospital, this is usually several charts to review every week. It takes about 15-20 minutes to do one of these reviews and then contact the admitting physicians to try to talk them into changing an admission order from inpatient to observation or vice-versa. Frequently, it requires the medical director to either enter an administrative note in the electronic medical record or to send the hospital utilization review office a letter so that the hospital has a documentation  paper trail in the event of a Medicare denial. More often, the medical director is sent charts for patients who had an inpatient admission order but were discharged before 2 midnights had elapsed. This is a red flag for Medicare auditors. We then have to provide written documentation for why the patient should be billed as an inpatient. Sometimes, this is easy, for example, when a patient dies from their condition in the hospital before 2 midnights elapse. But more frequently, it is because the admitting physician legitimately believed that the patient would need to be in the hospital for at least 2 midnights when that patient first arrived at the hospital. Several years ago, I attended a Medicare seminar and one of the medical administrators from Medicare told us that when this happens, we should use the phrase “The patient had an unexpectedly rapid recovery and was able to be discharged after less than 2 midnights”. Pretty much every time a patient in inpatient status was discharged before 2 midnights, one of our hospital medical directors would review that chart and send the billing office a letter using that phrase.
5. Pre-admission testing consultation. Patients who are planned to undergo a surgical procedure will frequently be sent for pre-operative medical consultation by an internist, family physician, or specially trained advance practice provider. Although designed to identify medical co-morbidities that could increase the risk of complications during surgery, these consultations are increasingly being used to determine whether or not a given patient’s surgery should be classified as an inpatient or an outpatient surgery. And most importantly, these consultations ensure that there is sufficient documentation in the electronic medical record to justify an inpatient procedure. For example, most knee replacement surgeries are now considered outpatient procedures. But if the chart documents that the patient has sleep apnea requiring CPAP, brittle diabetes, COPD requiring supplemental oxygen, and a history of vomiting after anesthesia, then that patient’s knee replacement can be done as an inpatient and the hospital gets paid considerably more. Surgeons are trained to be experts in surgery but are not trained in the nuances of co-morbid medical illnesses that they do not normally manage. Consequently, the surgeon’s outpatient notes often do not contain documentation of the significance of those medical co-morbidities and whether they are severe enough to warrant an inpatient admission for the surgery. That is why the pre-op medical consultation is so highly valued. If the surgeon admits the patient as an inpatient to do the surgery and then discharges that patient before 2 midnights pass, the chart once again gets sent to the medical director so that a letter containing the phrase “The patient had an unexpectedly rapid recovery and was able to be discharged after less than 2 midnights” is sent to the utilization review and billing offices for a documentation paper trail in the event of an admission denial by Medicare.
6. Utilization review staff. Every hospital employs a large number of personnel devoted solely to coding, billing, and utilization review. Before a bill goes out to Medicare or a commercial insurance company, these staff will do a final review to ensure that all of the proper documentation justifying an inpatient admission is present in the chart, including physician admission advisor notes and hospital medical director correspondence.There will also usually be personnel whose only job is to work denials when Medicare or an insurance company denies an inpatient admission. These personnel will then prepare and submit documentation contesting that denial in hopes of overturning the denial and getting paid for the hospital stay.
7. Attorneys and peer reviewers. When Medicare or an insurance company refuses to overturn an admission denial after the billing staff contest the denial, the next step is to turn to the legal system. This usually starts by paying an independent physician reviewer to opine whether the patient’s hospital stay should be classified as inpatient or observation. Next, hospital attorneys get involved by contacting Medicare attorneys about the denial. Sometimes, contested denials require adjudication, requiring more attorney time.
8. Medicare staff. On the other side of the bill, Medicare and commercial insurance companies employ large numbers of staff to review charts to decide whether they think that hospitalizations should be inpatient or outpatient.

So, for any given patient’s hospitalization, there is an army of Medicare nurse reviewers, physician reviewers, utilization review staff, and attorneys that face off against an army of the hospital’s  nurse reviewers, physician reviewers, utilization review staff, and attorneys. In the end, more money is sometimes spent battling an admission denial than is actually paid to the hospital for the admission. Hospitals are willing to occasionally spend the excessive cost to contest a denial since it sends Medicare or the insurance company a signal that the hospital will not go down easily for future admission denials. It is kind of like a basketball coach throwing a tantrum about a penalty call in order to try to dissuade the referee from calling future penalties.

The net result of all of this is that the United States has created an enormous bureaucracy devoted to preventing and contesting hospital admission denials.  So, why don’t hospitals just classify more patients as being in observation status and avoid all of the expense of justifying inpatient status? The reason is money.

The finances of inpatient vs. observation status

The genesis of observation status was to reduce Medicare costs by eliminating unnecessary hospital admissions. For any given diagnosis, hospitals get paid much less if a patient is designated to be in observation status than if that same patient is designated to be in inpatient status. Overall, the reimbursement is about 1/3 less for observation stays. In other words, Medicare can reduce overall Medicare costs by pressuring hospitals to put more patients in observation status. The same holds for commercial insurance companies.

Until several years ago, Medicare also paid doctors less if patients were in observation status. However, it became clear to Medicare that this was incentivizing doctors to preferentially use inpatient status rather than observation status. And since doctors are the ones who write the admission orders, Medicare realized that it could reduce inpatient admissions by removing the physician financial incentive to put patients in inpatient status. Because the physician professional fees are much, much less than the hospital fees, by paying physicians the same whether a patient was in observation or inpatient status, Medicare would pay a little more to the doctors but would pay a whole lot less to the hospitals.

But the biggest savings to Medicare in observation status is that it transfers much of the cost of the hospital stay from Medicare to individual patients. This is because Medicare considers observation stays as outpatient visits. Outpatient services are billed to Medicare Part B but inpatient services are billed to Medicare Part A. This is hugely important to patients because patients have much higher co-pays and deductibles for their Part B charges than their Part A charges.

• Medicare Part A covers inpatient admissions including a semi-private room, nursing care, medications, meals, and tests done during inpatient admissions. Part A also covers skilled nursing facility care, home health care, and hospice care. Medicare Part A is free to Americans over age 65 who have previously worked at least 10 years (or have a spouse who worked 10 years). There are no monthly premiums and no annual deductible. The amount that Medicare covers depends on the length of stay of the hospital admission:
  • $1,600 deductible per admission
  • Days 1-60: Part A covers in full
  • Days 61-90: patients are responsible for $400 per day co-pay, either by co-insurance or out of pocket if the patient lacks co-insurance
  • Days 91-lifetime reserve limit days: patients are responsible for $800 per day co-pay, either by co-insurance or out of pocket if the patient lacks co-insurance
  • After lifetime reserve limit days (total of 60 over the entire lifetime): Part A pays nothing and patients (or their co-insurance) are responsible for the entire costs
• Medicare Part B covers hospital outpatient charges and physician professional charges. Unlike Part A, there is a monthly premium for Part B of $165/month with higher premiums for those with higher incomes. There is an annual deductible amount of $226. Patients also have additional deductibles and co-pays that are either paid by secondary insurance or out of pocket if there is either no secondary insurance or the insurance policy has limited benefits:
  • 20% co-pay for all physician charges
  • 20% co-pay for hospital outpatient charges (hospital room, nursing care, meals)
  • 20% co-pay for x-rays and procedures
  • Part B does not cover medications so the patient (or their Part D insurance) is responsible for medication charges during observation stays

The net result of these differences is that the patient will have greater out of pocket expenses for an observation stay than for an inpatient admission. This is especially true for the 7.5% of Americans over age 65 who are enrolled in Medicare Part A only and do not enroll in Medicare Part B – these patients pay the entire cost of their observation stay out of pocket.

Another financial implication of observation vs. inpatient stays is in skilled nursing facility (SNF) coverage. Medicare Part A pays for 100% of SNF charges for up to 20 days (there is a $200/day co-pay for days 21-100). However, Medicare will only pay for SNF care if a person first has an inpatient hospital stay of at least 3 days. Medicare will not pay for SNF care after an observation stay. If a patient is in observation status (or has an inpatient stay of < 3 days) and gets transferred to a SNF, the patient is responsible for all of the SNF charges.

Why observation status is really, really dumb

CMS absolutely loves observation status. It reduces Medicare costs by paying the hospitals less for any given diagnosis and it also reduces Medicare costs by transferring much of the costs directly to the patient. This allows CMS officials to report to Congress that they are reducing federal spending on healthcare. Congressional representatives can then report to voters that they are reducing government spending. But there is no such thing as free healthcare… the cost of healthcare does not go away, it just gets transferred to the patient. The individual American ends up with more out of pocket costs for co-pays, medication charges, and SNF costs that would have otherwise been covered by Medicare Part A had their hospital stay been inpatient status as opposed to observation status. So, in the long run, the average American does not save any money by being in observation status.

Nationwide, 16% of all hospital stays are observation stays and 84% are inpatient stays. But this percentage can vary widely from hospital to hospital. A tertiary care referral hospital will usually have a lower percentage of observation stays since its patients tend to be sicker with more complex medical problems. On the other hand, a community hospital, especially one that cares for underserved patients, will have a higher percentage of observation stays, typically 25% or more. About the best a hospital can hope for is to break even on observation patients – most hospitals actually lose money on observation stays.

It’s bad enough that observation status does not really save money by transferring the cost of care to the individual patient rather than Medicare. The worst part about observation status is that it actually increases U.S. healthcare expenses. Not only do hospitals have to spend an enormous amount of money justifying inpatient admissions and working inpatient denials, but Medicare spends an enormous amount of money paying staff who police admissions in order to deny inpatient admission charges.

The net result is that observation status represents the epitome of U.S. healthcare dysfunction. It has led to an enormous bureaucracy devoted entirely to deciding whether Medicare or individual patients should pay for hospital care. And that bureaucracy is enormously expensive.

How can we fix this?

Americans pay way more for healthcare than people in any other country. In 2021, the average per capita healthcare cost in the U.S. was $12,914. It will undoubtedly be much higher in 2023. One of the contributors to this is too much of the healthcare costs go into trying to decide whether Medicare or the individual American will be responsible for paying for healthcare. Getting rid of the observation status designation can reduce U.S. per capita healthcare costs. Here is how to do it:

1. Create low-acuity DRGs. For conditions that are currently commonly managed by observation status (chest pain, syncope, dehydration, etc.), CMS can create inpatient DRGs that pay the hospital less, thus simulating the amount that CMS would have paid for an observation stay.
2. Require a modest Part A co-pay for hospitalization. The biggest argument against eliminating observation status is that Medicare costs would go up since co-pay costs currently paid by patients would go back to Medicare. The solution to this would be to require a small co-pay for hospitalization days 1-60. The amount to keep Medicare’s annual budget neutral could be as little as $10 or $20 per day.
3. Outpatient should mean outpatient. How in the world we ever got to the point that we define outpatient care as needing to be in the hospital for more than 2 midnights is baffling. Either a patient needs to be in the hospital or they don’t. I once had a admission denial for a patient in respiratory failure admitted from the ER to our ICU on a mechanical ventilator. The Medicare reviewer said that in his opinion, I should have been able to correct the respiratory failure, extubate the patient, and discharge her before 2 midnights passed. Really?
4. Eliminate the SNF 3-day rule. The whole idea behind the 3-day rule was that Medicare wanted to see if a patient really needed SNF care before it would pay for it. But the unintended consequence is that if one of our patients needs to go to a SNF, we have to figure out a way to admit them to the hospital for at least 3 days first. This means that we have to wait until they fall at home and break their hip or wait until they get septic from an infected decubitus ulcer if they are unable to get out of bed. For patients undergoing surgery, such as a knee replacement, we have to keep them in the hospital for 3 days after their surgery before they can go to a SNF for rehabilitation, even if they live alone and cannot walk after their operation.
5. Eliminate the observation industry. By eliminating observation status, hospitals would no longer have to spend money on nurse admission reviewers, physician admission advisors, and medical directors who laboriously review charts for inpatient justification. Hospitals could reduce their utilization management staff and Medicare could reduce its admission reviewer staff. Hospitals, patients, and Medicare would have less need for attorneys to contest admission denials. Yes, a lot of people would lose their jobs but the overall U.S. healthcare costs would drop.

Elimination of the observation status designation would make everyone happy. Patients would not be surprised by unexpectedly high hospital bills. Doctors would not have to spend time entering unnecessary documentation in their hospital notes to justify why a patient warrants an inpatient admission. Hospitals would not have to pay as much for staff to oversee admission determination. A fundamental concept of industrial engineering is that the more complex a process is, the more energy it takes to keep that process working. Observation status has created a terribly complex process. It is time to simplify the U.S. admission process.

January 22, 2023

Every year, the Centers for Medicare & Medicaid Services (CMS) penalizes hospitals with excessively high readmission rates. The monetary penalty for every U.S. hospital in 2023 was recently released by CMS. In theory, a higher penalty should indicate lower quality and vice-versa. However, the methodology used in calculating the penalties is complex and nuanced with the result that the readmission penalty may not be entirely reflective of a hospital’s overall quality of care.

Background

One of the provisions of the Affordable Care Act of 2010 (aka “Obamacare”) was to direct CMS to penalize hospitals with high rates of patients requiring readmission to a hospital within 30 days of an inpatient hospital stay. The first penalties were assessed in 2013; this is the 11th year of the penalty program. The calculations are only made for a few specific diagnoses and are based on data during 3 previous years. CMS calculates the amount of every hospital’s penalty in October or November each year and then that penalty is applied to the following year’s Medicare payments. The maximum penalty is 3%, meaning that for a hospital receiving the maximum penalty, CMS will reduce the amount that it pays that hospital for all of the Medicare services the hospital charges by 3% the next calendar year. Because many hospitals operate on razor-thin margins, even a relatively small reduction in what it gets paid by Medicare can be financially devastating. This is especially true for hospitals that operate on a July to June fiscal year, such as most academic medical centers, that can find themselves with an reduction in Medicare payments in the middle of the fiscal year. This can make it very difficult for these hospitals to accurately forecast their annual budgets since they do not know what they will get paid from Medicare services during the second half of their fiscal year.

You can look up every U.S. hospital’s readmission penalty for 2023 here. CMS uses a 4-step process to determine the amount of each hospital’s penalty.

The actual equation that CMS uses to calculate each hospital’s penalty is complex with the result that even most hospital administrators do not fully understand it:

Lurking behind the equation are a lot of subtleties that affect how the public should interpret the readmission penalty.

COVID affected the calculations

Normally, CMS looks at historical readmission data from between 2 and 5 years in the past. Thus, normally, CMS would base the 2023 readmission penalty on data from July 2018 to June 2021. This has always been a source of criticism since the penalty is based on what a hospital did 5 years ago rather than what it has done more recently in the past 2 years. Consequently, a hospital with poor readmission performance 5 years ago could have a large readmission penalty even if its readmission performance was stellar during the past 2 years.

The first 6 months of the COVID pandemic was a trying time for most hospitals. Many exceeded their maximum inpatient capacity. To care for inpatients, many had to recruit doctors and nurses who did not normally provide inpatient care. Because of this, CMS excluded all data from January 2020 to July 2020. As a result, rather than being based on 3 years of historical readmission data, this year’s penalty is based on 2.5 year of data.

A second effect of COVID was on the diagnoses used for penalty determination. Normally, CMS looks at readmission rates only for patients with one of six diagnoses: acute myocardial infarction, heart failure, pneumonia, COPD, coronary artery bypass surgery, and hip & knee replacement surgery. COVID disproportionately affected patients coded with pneumonia. As a result, CMS dropped pneumonia as one of the diagnoses used for readmission calculations. Therefore, the penalties were based on 5 diagnoses this year rather than 6.

A third effect of COVID on the readmission penalty calculation was that any patient with COVID as a primary or secondary admission diagnosis was eliminated from the hospital’s readmission calculation. Thus, a patient admitted with an acute myocardial infarction who was found to also have COVID on admission was excluded from the hospital’s data.

A fourth effect of COVID was on comorbidity determination. CMS adjusts every individual patient for that particular patient’s medical co-morbidities. So, for example, a patient with an admission for COPD who requires mechanical ventilation is expected to have a higher readmission rate than a COPD patient who does not require mechanical ventilation. Similarly, a patient undergoing knee replacement who is over age 65 and has diabetes is expected to have a higher readmission rate than a knee replacement patient who is younger than age 65 and not diabetic. This year, CMS added history of COVID within the past year as one of the co-morbidities used in the readmission calculation for all five of the readmission diagnoses. Thus, a patient admitted with COPD who had a COVID infection 8 months previously would be expected to have a higher 30-day readmission rate than a COPD patient who had never had COVID in the past.

Not all hospitals are included

CMS excludes about 40% of U.S. hospitals from the readmission penalty program. These include pediatric hospitals, Veterans Administration hospitals, psychiatric hospitals, rehabilitation hospitals, long-term acute care hospitals, and critical access hospitals. In addition, a hospital must have had more than 25 eligible patients for each of the 5 diagnoses. Thus, a hospital that only performed 24 coronary artery bypass surgeries during the 2.5 year period would not be subject for readmission penalties for CABG surgeries. CMS also excludes all hospitals in Maryland from readmission penalties because of an agreement between CMS and Maryland.

It is impossible for hospitals to monitor their readmission rates

Every autumn, hospitals await the CMS report on their readmission data with no advance knowledge of what the hospital’s readmission rate will be. These are sent to the hospital as a “Hospital Specific Report”. For most other quality metrics, hospitals can continuously monitor their performance internally. For example, any hospital should be able to determine on any given day what their mortality rate, C. difficile incidence, and emergency department wait times are. But readmission rates are unique. Medicare looks at admission to any hospital within 30 days of an inpatient discharge, not just the the hospital that the patient was originally admitted to. The original hospital will know if a patient gets admitted again to that hospital but has no way of knowing if a patient gets admitted to some other hospital. For example, if a patient is discharged from the Ohio State University Medical Center, OSU can track any readmissions to an OSU hospital. However, if that patient gets admitted to a non-OSU hospital in Cincinnati, OSU will not know about it. On the other hand, Medicare gets billed by every hospital that a patient is admitted to so Medicare will know whenever a patient is admitted to any hospital in the United States. This phenomenon has little impact on small, rural or community hospitals since a patient admitted to that hospital will likely return to that same hospital given that it is the only hospital in the region. But for tertiary care or referral hospitals, patients often live hundreds of miles away and readmissions are more likely to occur at their local community hospital rather than at the tertiary care hospital. Thus a tertiary care hospital will have no idea what its readmission rate performance is until CMS sends out the Hospital Specific Reports.

Hospitals normally institute a continuous quality improvement process for quality metrics. This requires real-time monitoring of that quality metric so that the hospital can continuously change its procedures and policies to make their quality outcomes better. This turns out to be difficult for reducing 30-day readmissions because the readmission data that Medicare gets is 2-5 years old. To make an analogy, imagine how difficult it would be for a coach to improve his or her basketball team if the coach did not know the outcome of each game until 5 years after it was played.

All hospitals are not treated the same

One of the main criticisms of the initial formula that CMS used in the first years of the readmission penalty was that hospitals that cared for a large number of poor people were disproportionately penalized compared to hospitals caring for a largely affluent patient population. Poor individuals are less likely to have insurance, less likely to be able to afford medications, less likely to have transportation for doctor office visits, and less likely to have a primary care physician. All of these factors contribute to higher hospital readmission rates but these are factors that are largely not under the hospital’s control. In response to this criticism, several years ago, CMS changed the methodology used in readmission calculation to adjust for the percentage of poor and underserved patients that each hospital cares for. The current methodology uses the percentage of “dual-proportion” patients. This is based on the percentage of Medicare patients that also have full Medicaid benefits. Medicaid is used as a marker for low-income patients. CMS divides U.S. hospitals into one of five quintiles based on the percentage of a hospital’s dual proportion patients. Quintile #1 includes hospitals with fewer than 14% of its Medicare patients having dual coverage with Medicaid. Quintile #5 includes hospitals with more than 31% of its Medicare patients having dual coverage with Medicaid.  The breakdown of hospitals based on their percentage of dual proportion is seen in the graph below:

All hospital stays are not treated the same

Medicare classifies each patient’s hospital stay as either an “inpatient” stay or an “observation” stay. The rules for how to classify any given patient are complicated but in general, a patient who is expected at the time of arrival to the hospital to be in the hospital for less than 2 midnights is considered to be in observation status. Overall in the U.S., about 84% of hospital stays are designated as inpatient and about 16% are designated as observation. The financial difference in the two types of hospital stays is very significant, both for the hospital and for the patient. An observation stay is considered an outpatient visit and is thus subject to Medicare Part B billing rather than Medicare Part A billing. This means that the patient in observation status is responsible for all medication charges and is responsible for a 20% co-pay of the cost of the hospital stay. CMS pays the hospital much less for an observation stay than for a regular inpatient stay. Observation stays are less expensive for Medicare because much of the healthcare costs are passed on to the patient.

Because observation stays are considered outpatient visits from a Medicare perspective, these hospital stays are not included in the hospital readmission calculation. For a readmission to count, both the initial hospital stay must be an inpatient stay and the second hospital stay within 30 days must also be an inpatient stay. This is also the same when CMS calculates a hospital’s mortality rate – only inpatient stays are included and deaths occurring when a patient is in observation status do not count against the hospital’s mortality rate.

Hospitals can “game the system” by putting certain patients in observation status in order to improve their readmission data and their mortality data. For example, take a patient who is admitted for a COPD exacerbation on February 1st. That patient then comes back to the hospital on February 20th after having a cardiopulmonary arrest following a drug overdose. The patient is intubated, receives mechanical ventilation, and placed in the ICU but it is clear that the patient has had severe brain and heart damage and is not expected to live beyond 24 hours. The person overseeing the hospital’s quality data will advise the ICU physician to admit the patient as an observation stay so that the hospital stay does not count as a readmission and so that the death does not count as an inpatient mortality. On the other hand, the person overseeing the hospital’s finances will advise the ICU physician to admit the patient as an inpatient stay so that the hospital gets paid more from Medicare. Whichever of the two hospital administrators is most persuasive (or most vocal) will usually win out. The result is that hospitals that more liberally designate patients as being in observation status can lower their CMS readmission penalty. The goal of Medicare auditors is to pay hospitals as little as possible so they will penalize hospitals who put patients in inpatient status who should really be in observation status. However, those auditors do not care if a hospital puts patients in observation status who should really be in inpatient status since it saves Medicare money.

This can also be an effective strategy for hospitals that have a low volume of patients with one of the five diagnoses used in the readmission calculation. For example, if a hospital has 24 heart failure inpatient admissions over a 3-year period, then putting the next heart failure patient in observation status ensures that the hospital will not have any heart failure readmission penalty since there would still be fewer than 25 heart failure inpatient admissions during that 3-year period. The cost to the hospital is that they might get paid $4,000 less by putting that patient in observation status rather in inpatient status. But by avoiding a 0.2% readmission penalty for all medicare charges for the next year, that hospital might avoid a total $80,000 penalty. That is a $76,000 net return on investment for putting that one patient in observation status rather than inpatient status!

For some hospitals, it is cheaper to pay the penalty

In the United States, the average hospital has 19.8% of revenue from Medicare, 13.1% from Medicaid, and 68.4% from private commercial insurance. The hospital with the highest annual net patient revenue in the U.S. is New York Presbyterian Hospital at $5.7 billion. However, the average U.S. hospital’s total annual patient revenue is much lower at about $200 million. Thus, the average hospital has annual Medicare revenue of about $40 million ($200 million x 19.8%). A maximum Medicare readmission penalty of 3% would therefore be about $1.2 million for that average hospital. Only 17 hospitals were fined the full 3% for 2023 and only 231 hospitals will pay more than 1% penalty. 25% of hospitals will pay no penalty at all. The average hospital penalty is 0.43%.

Given that the average hospital has $40 million in annual Medicare revenue and that the average hospital has a 0.43% Medicare penalty in 2023, that average hospital will have a $172,000 penalty. Implementing a readmission reduction program in a hospital can be very costly. It requires hiring data analysts to monitor readmissions, instituting costly discharge transition clinics, and increasing the percentage of patients in observation status. For many hospitals, the total cost to reduce readmissions sufficiently to avoid a CMS penalty can be considerably more than the expense of the penalty. In general, the larger the hospital and the higher the percentage of Medicare patients in a hospital’s payer mix, the more likely it will make financial sense for a hospital to devote a lot of money into a readmission reduction program. Furthermore, because the readmission penalty is based on the hospital’s performance between 2-5 years previously, it will take 5 years before money spent today on a readmission reduction program will fully affect the annual CMS penalty. And it is likely that CMS will continue to revise the readmission penalty formula so that the formula will look considerably different over the next 5 years.

How can the process be improved?

Medical care in the United States is more expensive than anywhere else in the world and it is essential for our economy that we reign in healthcare costs. Because hospitalizations are expensive, reducing unnecessary hospital admissions is central to controlling those healthcare costs. Readmissions to the hospital within 30 days of hospital discharge are frequently avoidable if processes are in place to ensure that patients get appropriate outpatient care. This includes filling medication prescriptions, keeping office appointments with medical providers after discharge, access to outpatient physical & occupational therapy, etc. Penalizing hospitals for excessive readmissions is one way to reduce costs by incentivizing hospitals to institute processes that reduce hospital readmissions. However, after eleven years of the CMS Hospital Readmissions Reduction Program, it is clear that the program can do better. Some specific improvements include:

1. Provide hospitals with real-time readmission data. This is probably the single most important change that CMS can make and it really should not be terribly difficult. Ideally, hospitals should know what their current readmission rates are every month so that the hospital can employ continuous improvement processes to reduce those readmissions. The current model of basing the penalty on a hospital’s readmission rates from 2-5 years in the past makes improving readmissions very difficult. Ideally, CMS should provide every hospital with its current rolling 3-year average readmission rate and this should be updated monthly.
2. Eliminate observation status hospital stays. Currently, hospitals spend an enormous amount of money to determine whether any given patient should be in observation status or inpatient status. Medicare loves observation status because CMS does not have to pay as much to hospitals for patients in observation status as opposed to inpatient status. Instead, those additional costs are passed on to the individual patient. So, the net overall cost to the country as a whole is the same, regardless of whether a patient is in observation or inpatient status. When the overhead expense of monitoring and policing observation stays is included, the overall cost of having observation status actually increases the country’s overall healthcare costs. When it comes to readmission rates, some hospitals game the system by preferentially putting readmitted patients in observation status instead of inpatient status. It is time to eliminate observation status and simply pay hospitals for patient stays, regardless of whether or not the patient’s hospital stay crosses two midnights.
3. Base the penalty on the overall readmission rate rather readmission rates for only 6 diagnoses. Every hospital is different. Not all hospitals perform coronary artery bypass graft surgery and not all hospitals perform knee & hip replacement surgery. Currently, hospitals focus their readmission rate reduction strategies on just the 6 conditions that CMS penalizes them on. Savvy hospitalists know that they can readmit patients who have had a stroke, diverticulitis, or a drug overdose every week without having to worry about any penalty. By using overall readmission rates (for all diagnoses), the quality process will be simpler for hospitals and will benefit all patients and not just those patients who have one of the 6 conditions that CMS currently uses in readmission penalty determination. However, CMS would need to determine a different method of risk-adjustment for comorbidity since the current method is by using specific comorbidities for each of the 6 eligible diagnoses.
4. Eliminate hospital exceptions. Currently, about 40% of U.S. hospitals are not subject to readmission penalties. This is understandable for pediatric hospitals (few, if any, Medicare patients) and Veterans Administration hospitals (funded by the VA and not by CMS). Psychiatric hospitals are excluded because they do not normally admit patients with the 6 conditions that CMS bases the readmission penalty on. Long-term acute care hospitals, rehabilitation hospitals, and critical access hospitals are also excluded. Ideally, CMS should use data for both Medicare and Medicaid patients since it can track readmissions for both groups. By focusing on total readmission rates rather than the 6 currently used diagnoses, many of the currently exempted hospitals can be included in the readmission reduction program. However, there may need to be different readmission rate benchmarks for psychiatric hospitals, long-term acute care hospitals, etc.

January 4, 2023

A reader recently emailed me to ask: “How do you define a hospitalist FTE?” It turns out that it is a great question with a very nuanced answer. Twenty years ago, an FTE was whatever a physician wanted it to be. Physician earnings were directly tied to physician billing and so a physician would work as much as they wanted in order to generate the income that they wanted. But over the past 2 decades, revenue from physician professional services has not changed appreciably – in 2002, Medicare reimbursed physicians at $36.20 per RVU; in 2022, an RVU was worth $34.61.

To put that in perspective, $1.00 in 2002 is worth $1.66 in 2022 whereas an RVU is now worth $1.59 less than it was in 2002! In order to keep physician incomes constant, hospitals have had to increasingly subsidize physicians. As a result, most physicians are now hospital-employed, rather than independent practitioners. This is especially true for hospitalists who rarely, if ever, are able to support their full salary on billings alone. In the past, the physicians defined what working full-time constitutes but today, it is the hospitals that define what working full-time means for a hospitalist.

Hospitals typically subsidize hospitalist groups based on the number of FTEs (full-time equivalents) that are required in order to cover the hospital’s inpatients. But defining exactly what an FTE is can be complicated and often a source of disagreement between the hospital and the hospitalist group. There are a number of equally valid ways of defining “full-time” and no one definition works best in every hospital. There are several steps to determine the best model in your hospital.

Step 1: Determine the number of patients per hospitalist per day

The number of patients each hospitalist should see per day will vary considerably from hospital to hospital and from nursing unit to nursing unit. There are 19 factors to consider when determining this number as outlined in a previous post:

• • • Case mix index
    • Residents versus no residents
    • Admitting service versus consultative service
    • Presence or absence of advance practice providers
    • ICU versus general ward patients
    • Day shift versus night shift
    • Observation status versus regular inpatient status
    • Ease of documentation
    • Shared electronic medical record with primary care physicians
    • Non-clinical duties
    • Shift duration (hours)
    • Hospitalist experience
    • Patient geographical location within the hospital
    • Average length of stay
    • Inpatient census variability
    • RVU productivity
    • Quality of case management
    • Local hospitalist employment market
    • Patient demographics

There has to be flexibility, however, and rigid adherence to a given number of patients is a recipe for dissatisfaction on both the part of the hospital and the part of the hospitalist. If the hospital inpatient census falls, then the hospital will be unhappy that each on-duty hospitalist is not seeing enough patients. On the other hand, if the inpatient census surges, then the hospitalists will be unhappy since they have to see more patients than they agreed on in their contracts. Many hospitals will have a “risk-call” hospitalist each day who is on standby to come in to work if needed when the inpatient census is higher than normal.

Step 2: Determine how the hospitalists will be scheduled

Early in the hospitalist era, scheduling was simple: a shift was 12 hours long and there were two shifts – a day shift and a night shift. Hospitalist schedules have gotten a lot more complex in recent years as outlined in a previous post. Now, hospitalists often have 8-hour short day shifts and evening swing shifts to cover ER admissions in the early evenings. As a result, scheduling hospitalists has become much more complex. Here are some of the scheduling models:

The 12-hour shift model. This was the original hospitalist scheduling model and typically will have two 12-hour shifts per day, a day shift and a night shift. The day shift is typically 6:00 AM to 6:00 PM or 7:00 AM to 7:00 PM. The night shift starts when the day shift is over. Day shifts and night shifts are treated equally but since night shifts are considered less desirable by most hospitalists, there needs to be a “shift differential” to provide extra payment for covering night shifts. Many hospitals will also provide additional pay for hospitalists who work on holidays. Because most patient care (work rounds, interdisciplinary rounds, daily charting, discharges, family meetings, etc.) occurs during the day shift, hospitals will typically have 1 night shift hospitalist for every 3 -4 day shift hospitalists. High acuity patient care areas, such as the intensive care unit, may require 1 night shift hospitalist for every 1 day shift hospitalist.

The long-shift, short-shift model. In this model, one or more hospitalists works the entire 12-hour day shift but other hospitalists leave earlier in the day, after their work is done. The short-shift hospitalists check out to one of the long-shift hospitalists when leaving. The long-shift hospitalist is then responsible for any admissions that come in later in the day. Some hospitalist groups will have the short-shift hospitalists continue to take phone calls from nurses, the lab, and consultants after they leave the hospital; other hospitalist groups will have the long-shift hospitalists cover calls. Some hospitals will have a specific check-out time for the short-shift hospitalists, for example, 3:00 PM. Other hospitals will have the short-shift hospitalists check out whenever their work is completed, whether that be 1:00 PM or 5:00 PM. An advantage of this model is that it avoids having a lot of hospitalists sitting around doing nothing in the late afternoon, after all of their work is done. In addition, this model is very attractive to hospitalists with children, since they can be home when the kids get out of school.

The swing-shift model. In most hospitals, the peak in admissions from the emergency department occurs between 3:00 PM and midnight. After midnight, the admissions slow down, the inpatients go to sleep, and the hospitalist workload drops. To optimize patient care coverage, some hospitals will create a “swing-shift” to cover the surge in admissions during the evening. Every hospital’s pattern of ER admission is different so swing-shifts could be from 3:00 to midnight, 5:00 to 10:00, etc.

The comprehensive services model. In many hospitals, the hospitalists do more than just serve as the attending physician for inpatients. For example, they may perform medical pre-operative consultation in an outpatient pre-admission testing clinic. They may provide medical consultation for surgical inpatients. They may have a designated “triage attending” to serve as a liaison between the hospitalist services and the emergency department or the outside referring hospitals. Or they may provide on-site supervision of infusion centers. In these situations (except for triage attending), the duration of a shift is determined by whenever the work is done, rather than by a specific time of day or number of hours. In general, these other services require fewer than 12 hours per day. These types of services are often attractive to hospitalists with young children since they are generally able to get home earlier than they would with a traditional 12-hour inpatient hospitalist shift.

Step 3: Determine how a 100% FTE will be defined

Once the hospital has determined how many inpatients a rounding hospitalist should cover and how the hospitalists are to be scheduled, the next step is to determine what will constitute a 100% FTE hospitalist. There are several ways of defining an FTE.

The shifts per month model. This works best when all of the hospitalists work 12-hour shifts. A full-time hospitalist is typically defined as 15 or 16 shifts per month (180 – 192 shifts per year). It generally takes about a half hour to check out at the end of every shift with the result that a 12-hour shift is really a 12.5-hour shift. This works out to about 43 – 46 hours per week on average. Some hospitals will grant additional time off for vacations and CME with the result that full-time may be fewer shifts per year, for example, 170 shifts.

The hours per month model. This model works when there there are different hospitalist shifts of varying durations. In this model, a hospitalist may be scheduled for shifts of a variety of durations up to some pre-agreed upon number of total hours per month. This results in a great deal of scheduling complexity and often requires considerable effort by the scheduler to ensure equity among the hospitalists. Many jobs define an FTE as 40 hours per week, however, most physicians work more than that. Although physician time surveys vary, most find that physicians average closer to 50 hours per week. If we extrapolate from the 15 – 16 twelve hour shifts per month model that results in 43 – 46 hours per week, then this would equate to 2,236 – 2,392 hours per year. Rigid adherence to a specific number of hours per year is difficult. Unlike other hospital employees, hospitalists do not punch in and out on a time clock. There are always some days when a hospitalist needs to stay in the hospital past the end of their shift to finish charting, complete the H&P on a late admission, or provide care for a critically ill patient. In addition, some hospitalists may check-out early to one of their peers once they complete their daily work.

The number of billed wRVUs model. If you look in the annual MGMA physician salary survey, you can find the mean, median, 25th percentile, 75th percentile, and 90th percentile of work RVUs  produced by physicians in every specialty. Using wRVUs as a general guide of FTE productivity can be useful for many specialties but as discussed in a previous post, it is inadvisable to pay individual hospitalists by the wRVU. Nor is it advisable to use wRVU targets to define an FTE. If wRUVs are used to benchmark hospitalist productivity, the RVU targets need to be for the entire hospitalist group and not for individual hospitalists. There is too much variation in RVU production intrinsic to different types of hospitalist shifts – fewer RVUs with night shifts, more with ICU shifts, and none for triage attending shifts. In other words, rather than requiring each of your 10 hospitalists to produce 4,300 wRVUs per year, instead require the entire group of hospitalists to produce 43,000 wRVUs per year.

The traditional workweek model. Most outpatient physicians define full-time as traditional office hours, working Monday through Friday, 8:00 – 5:00. With physician offices often closed on weekends, evenings, and holidays, this works fairly well for outpatient medicine. This model is harder to apply to hospitalists because illnesses requiring inpatient care are just as likely to occur on weekends and holidays as they are on weekdays. Therefore, hospitalists need to cover every day of the year. Nevertheless, some hospitals will have a core group of hospitalists who cover Monday through Friday day shifts. Part-time hospitalists or moonlighters cover weekends. And nights are either covered by home call, by inpatient advance practice providers, or by nocturnists. This model can sometimes work in smaller hospitals that care for lower acuity patients but is impractical in larger hospitals. The weekday hospitalists typically take care of their daily rounds and any admissions. They then leave the hospital in the afternoon, after their work is done. A typical full-time hospitalist in this model might work 46 weeks with 4 weeks of vacation, a week of CME, and a week for holidays. This equates to 230 working days per year.

The academic hospitalist model. In many teaching hospitals, the attending physicians on medical inpatient services are hospitalists who oversee care provided by internal medicine, family medicine, or pediatric residents. In this situation, the residents typically cover a given inpatient service for 4-week blocks. The attending hospitalist typically covers the teaching service daily for 2 weeks, although at some hospitals, the hospitalist covers the service for shorter (1 week) or longer (4 week) blocks. Because residents are in the hospital to perform H&Ps and care for any acute medical problems, the hospitalist can often leave the hospital after rounding with the resident and completing charting. This results in the hospitalist typically being in the hospital for 5 – 8 hours per day. The attending hospitalists generally provide back-up coverage to the residents at night by home call, either individually for their particular inpatient service or on a group rotational night call basis. Unlike the traditional workweek model, the academic hospitalist model generally requires both weekday and weekend coverage in order to ensure continuity of patient care and continuity of resident education. Thus, full time is considered less than 46 weeks and may be anywhere from 6 months (182 days per year) to 8 months (243 days per year) of service time.

Step 4: Determine how a part-time FTE will be defined

Once there is agreement between the hospital and the hospitalists on what will constitute a full-time FTE, it then becomes easier to assign a percent effort to part-time physicians and to determine how those part-time hospitalists will be paid.

For compensation of hospitalists who work less than 100% FTE, the easy answer is to make their base pay the same percent as their FTE. However, that can pose more cost to the employer since there are certain employer-paid expenses that are fixed, regardless of whether a hospitalist is 100% or 70% FTE. For example, the employer’s portion of health insurance premiums and life insurance premiums is the same for part-time employees as it is for full-time employees. Similarly, the employer’s cost of recruitment and credentialing is the same whether the hospitalist is 100% or 70%. In other words, it costs the employer more to have 2 hospitalists who each work 50% FTE than to have 1 hospitalist who works 100% FTE. Most hospitals are willing to cover those higher costs in order to keep high-performing hospitalists who wants to work part-time, particularly if there is a reasonable chance that the hospitalist will eventually return to 100% FTE in the future. For example, an experienced hospitalist who is a parent who wants to cut back to 70% for a few years until his/her child is older.

One size does not fit all

From the above discussion, it is clear that no one single model is best for all hospitals. Each hospital (and each hospitalist group) must examine its own unique inpatient service coverage needs in order to select the definition of “full-time” that fits best. From the hospital standpoint, it is important to be flexible and work with the hospitalists to be sure that they are happy with the model. From the hospitalist standpoint, it is important to ensure that a model that optimizes their work-life balance does not interfere with optimal patient care.

Because hospital censuses ebb and flow from year to year and because new hospitalists are hired from year to year, it is important that every hospital re-examines how full-time is defined periodically to ensure that the agreed upon model best fits the dynamic nature of inpatient medicine.

December 10, 2022

It has now been 3 years since the SARS-CoV-2 virus first infected humans in Wuhong, China in December 2019. By now, most Americans have either had a COVID-19 infection or have been vaccinated against COVID. So, what type of mask should we be wearing in the hospital and in public places?

Our two most powerful weapons against COVID are vaccination and mask-wearing but mandates for both have been been unpopular in the United States. Since the peak of the pandemic, most mask mandates in public areas have been relaxed. However, mask mandates are still in place in many hospitals. Furthermore, as of the writing of this post, there are 25,380 patients with COVID infection currently admitted in U.S. hospitals. Consequently, healthcare workers continue to have regular exposure to infected persons. So, what kind of mask should our healthcare workers wear?

Mask-wearing is not new. Surgeons and other operating room personnel have worn masks during surgeries for more than a century. However, the purpose of wearing masks in the OR is to prevent the surgeon’s respiratory secretions from infecting the patient’s incision and not to prevent the patient’s incision from infecting the surgeon. A simple surgical mask is very effective in catching respiratory droplets if the surgeon sneezes, coughs, or talks during an operation.

Before COVID, there were situations when healthcare workers did wear masks to prevent getting infected by patients, for example, when caring for patients infected with influenza (where transmission is usually by respiratory secretion droplets). A regular surgical mask is effective in preventing infection from respiratory secretion droplets but for smaller particles, an N95 mask is necessary. For 35 years, I wore N95 masks whenever I was caring for patients infected with tuberculosis or when doing bronchoscopy on patients suspected of having TB. It only takes inhaling  one tuberculosis bacteria to become infected with TB. A respiratory secretion droplet is about 5-10 μm in size whereas a tuberculosis bacteria is about 2 μm in size. A regular surgical mask will usually stop the droplets but will not stop a TB bacteria, whereas an N95 mask will.

For an N95 mask to be effective, it must have a tight seal to the face so that air cannot get between the edges of the mask and the skin. Everyone’s face is shaped a little differently and not all brands of N95 masks fit all faces equally well. Healthcare workers are normally required to be “fit-tested” to determine which N95 mask provides an acceptable seal against the face. Passing the fit test meant that all of the inhaled air went through the mask material and did not leak between the mask and the skin. At our hospital, we required healthcare workers who worked with patients with suspected tuberculosis to be fit tested every year. A number of years ago, the Occupational Safety and Health Administration (OSHA) put out a requirement stating “The employer shall not permit respirators with tight-fitting facepieces to be worn by employees who have facial hair that comes between the sealing surface of the facepiece and the face…“. In other words, men with beards were not supposed to use N95 masks.

Like influenza, the COVID virus is usually transmitted by respiratory secretion droplets. So, in theory, a surgical mask should be sufficient. However, there are certain situations when an infected patient’s respiratory secretion droplets can be broken down into smaller particles, in which case, a surgical mask could be insufficient and an N95 mask may be more effective. Such situations include performing endotracheal intubation and bronchoscopy. Having had a beard for 40 years, I had repeatedly passed my annual N95 fit tests despite having a beard but when COVID hit, I shaved it in order to be in compliance.

We now have several studies performed over the past 3 years of the pandemic to give us better guidance of what type of mask is most effective in preventing healthcare workers from becoming infected. Unfortunately, each of these studies have potential limitations and the studies have given mixed results.

A word about KN95 masks. An N95 mask is a NIOSH-approved mask that filters out at least 95% of particles that are 0.3 micons in size. In reality, a well-fitting N95 mask actually filters out 99.8% of particles that are 0.1 micons in size, which is also the size of a COVID virus particle. A KN95 mask is a mask approved by the Chinese government (and not NIOSH). A study of KN95 masks found that 70% of them did not meet the NIOSH standard for effectiveness. For hospital purposes, a KN95 mask can be considered as effective as a surgical mask but not as effective as an N95 mask

One of the earliest studies comparing surgical to N95 masks was published in JAMA in 2009 and randomized 446 nurses caring for patients with acute febrile respiratory illnesses to wear either surgical masks or N95 masks during patient care. An equal percentage of the nurses were diagnosed with influenza during the study and the conclusion was that the type of mask worn did not make a difference. Another study published 10 years later in JAMA in 2019 also found that among 1,993 healthcare workers randomized to wearing surgical or N95 masks, there was no difference in the incidence of influenza or other respiratory infections.

There have been several studies comparing masks to prevent COVID infection. A 2021 study examining the effectiveness of N95 versus surgical masks based on analysis of droplet size characteristics concluded that a surgical mask was theoretically sufficient to prevent COVID infection in low-virus environments but N95 masks would be theoretically more effective in virus-rich environments. A 2021 study comparing the fit of cloth, surgical, KN95, and N95 masks found that N95 masks had better fit factor scores than the other masks and that KN95 masks had similar fit factor scores as surgical masks. The study also concluded that fit is critical to the level of protection offered by masks. A 2021 meta-analysis of 8 studies involving N95 versus surgical masks found that N95 masks were more effective than surgical masks in preventing healthcare workers from becoming infected with a variety of respiratory viruses, including COVID. A 2022 study of 3,259 healthcare workers who were randomly assigned to use filtering facepiece class 2 masks (analogous to N95 masks) versus surgical masks found no overall effect on the incidence of COVID infection; however, those healthcare workers with > 20 contacts with COVID patients did have fewer COVID infections if they were wearing the filtering face piece class 2 masks. A 2022 MMWR report involved a retrospective review of 1,528 COVID-infected persons and 1,511 case controls and found that people who reported wearing N95 masks in public were less likely to become infected than those who wore surgical masks in public places. However, a major limitation of this study is presumably those wearing N95 masks were also more likely to take other precautions to avoid COVID than those who chose to wear a regular surgical mask.

The most recent study was published this week in Annals of Internal Medicine. 1,009 healthcare workers in Canada, Israel, Egypt, and Pakistan were randomized to use N95 or surgical masks during patient care for 10 weeks. Overall, there was no difference in the incidence of COVID infection between the two groups but it is notable that most of the subjects (71%) were in Pakistan or Egypt with only 29% in Canada or Israel. In Pakistan and Egypt, 82% of subjects had previously been infected by COVID before the study whereas in Canada and Israel, only 3% of subjects had a previous COVID infection. There was a trend for N95 masks to be more effective than surgical masks in Canada and Israel; however, because of the low number of subjects in those countries, the results did not reach statistical significance. Because past COVID infection confers some immunity to reinfection, it is possible that the failure of N95 masks to have superior protection over surgical masks was due to the very high percentage of healthcare workers with immunity from previous infection in Pakistan and Egypt.

The bottom line: What mask should you wear?

Medicine is a very dynamic science and recommendations change as new clinical studies are published. As a result, today’s medical dogma is tomorrow’s medical malpractice. Based on the available information a few general recommendations can be made:

1. In hospital areas where the prevalence of COVID infection is low, wearing regular surgical masks is sufficient to prevent healthcare workers from becoming infected. This means parts of the hospital that provide care for non-COVID patients, such as general medical and surgical floors, cafeterias, public areas, and offices.
2. In hospital areas where there is likely to be a high number of viral particles in the air breathed by a healthcare worker, N95 masks are preferable. This would include areas where the viruses are likely to be aerosolized, such as during intubation or bronchoscopy, particularly in rooms with stagnant airflow, such as those lacking sufficient number of air exchanges per hour. In locations where the virus is in respiratory droplets (as opposed to being aerosolized), surgical masks may be sufficient. Because several studies have shown that the viral loads are similar among symptomatic versus asymptomatic patients with COVID infection, the decision of whether to wear an N95 mask should not be based on the severity of the patient’s infection.
3. Wearing surgical masks by healthcare workers is sufficient to prevent workers with asymptomatic COVID infection from infecting patients. This has been a concern for the care of patients at high risk of severe COVID infection, such as those patients who are immunocompromised, obese, diabetic, or elderly. These patient are at particular risk if in contact with a maskless infected healthcare worker. Because daily testing of all healthcare workers caring for these patients is impractical, preventative mask-wearing is prudent.
4. Because viral loads are just as high in asymptomatic patients as in symptomatic patients, hospital visitors should wear surgical masks inside of the hospital as long as the prevalence of COVID infection in the community remains substantial or high. What constitutes “high” is a matter of opinion but the CDC defines “low” as up to 10 cases/100,000 population, “moderate” as between 10-50/100,000, “substantial” as 50-100/100,000 population, and “high” as greater than 100/100,000 population. Currently, the U.S. as a whole has a case rate of 91/100,000 population. However, hospitals are not like the country as a whole – currently 5.6% of all Ohio inpatients have COVID infections for an inpatient rate of 5,600/100,000. Because inpatients with COVID infection are more likely to have family and friends who are also infected (from home and workplace transmission), the probability of encountering a hospital visitor with a COVID infection is a lot higher than encountering a waiter in a restaurant with a COVID infection.
5. Healthcare workers who prefer to wear an N95 mask (but are not required) should be permitted to do so. Because of non-statistically significant trends in some studies suggesting a slight benefit of N95 over surgical masks, those healthcare workers who perceive greater safety with the N95 masks should be allowed to wear them if desired and if supplies permit.
6. Healthcare workers who are required to wear N95 masks should be fit-tested annually. Because ill-fitting N95 masks lose most of their protective benefit, there needs to be assurance that the N95 mask that the worker wears actually does what it has the potential to do. It is reasonable for hospitals to also require fit-testing for those employees who want to wear an N95 mask but are not required to wear an N95 mask, especially if the hospital is paying for the masks.
7. KN95 masks are an acceptable alternative to wearing a surgical mask but should not be used in clinical situations when N95 masks are required. Given the reduced effectiveness of KN95 masks compared to N95 masks, a KN95 mask cannot be viewed as equivalent to an N95 mask. It may, however, be better than a surgical mask.

What about outside of the hospital? At the grocery store or airport, I generally see only 5-10% of people wearing masks. At athletic events, that percentage is even lower. The risk of becoming infected in a public place is dependent on how likely you are to come in contact with someone’s exhaled viruses. That in turn depends on the prevalence of COVID infection in the community, the number of people in an enclosed space, the size of the enclosed space, the duration of time that you are in that space, the amount of singing or shouting in the space, and the ventilation of the space. Based on viral load studies, it does not matter whether an infected person in the space is symptomatic or asymptomatic. In situations when those variables indicate a higher risk of COVID, wearing a surgical mask is prudent. Those people who feel more secure wearing a KN95 mask or an N95 mask should do so.

As for me, N95 masks kept me from getting TB despite caring for patients with tuberculosis for decades. They also kept me from getting COVID infection despite intubating and performing bronchoscopy on COVID patients in our intensive care unit. My fit-tested N95 mask is sort of like a security blanket for me. So, I think I’ll keep wearing it in public areas for now. When the case rate gets below 50/100,000 population, I’ll reconsider. And when the case rate gets below 10/100,000, I’ll feel safe without any kind of mask in public.

December 2, 2022