Estimating the costs of physician turnover in hospital medicine

Abstract

Background and Objective

Costs of physician turnover are lacking for specialties organized around a site of care. We sought to estimate the cost of physician turnover in adult hospital medicine (HM).

Design, Setting, Participants

A retrospective cohort study within a large integrated health system between July 2017 and June 2020. To understand likely variation across the country, we also simulated costs using national wage data and a range of assumptions.

Main Outcome and Measures

Direct costs of turnover borne by our department and institution and indirect costs from reduced hospital billing. In our simulation, we measured costs per hired hospitalist.

Results

Between July 2017 and June 2020, 34 hospitalists left the practice, 97 hospitalists were hired, and a total of 234 hospitalists provided adult care at six hospitals. Direct costs of turnover totaled $6166 per incoming physician. Additional clinical coverage required at times of transition was the largest expense, followed by physician time recruiting and interviewing prospective candidates. The salary difference between outgoing and incoming hospitalists was cost‐saving, while reduced billing would add to indirect costs per hire. In our simulation using national wage data, programs hiring one hospitalist would spend a mean of $56,943 (95% CI: $27,228–$86,659), programs hiring five hospitalists would spend a mean of $33,333 per hospitalist (95% CI: $9375–$57,292), and programs hiring 10 hospitalists would spend a mean of $30,382 per hospitalist (95% CI: $6877–$53,887).

Conclusions

The financial cost of turnover in HM appears to be substantially lower than earlier estimates of the cost of turnover from non‐hospitalist specialties.

INTRODUCTION

Annual physician turnover is higher in hospital medicine (HM) than in many other specialties, with a mean turnover rate of 10.9% in 2019. ¹ Recruiting and hiring physicians to replace a departing physician can carry substantial costs. ² , ³ However, data on the cost of hospitalist turnover are lacking.

Absolute estimates of the costs of physician turnover have ranged from $88,000 to $1,000,000 per physician, and relative estimates suggest that turnover costs are around 2–3 times a physician's salary. ² , ³ , ⁴ , ⁵ , ⁶ However, those estimates predate the field of HM and are largely focused on primary care physicians. Unlike roles that provide longitudinal care for a patient panel, roles organized around a site of care, such as HM or emergency medicine, should not experience a loss of volume due to physician turnover. As a result, the costs of turnover for hospitalists should be substantially lower than for panel‐based physicians. Physicians new to a health system may be less efficient, and time spent on non‐clinical activities can incur costs, but estimates of these costs are lacking. They are also likely to be small compared to lost revenue due to turnover.

Moreover, as practices grow, the costs of turnover may evolve. Some costs (e.g., holding an orientation event or advertising) are fixed regardless of the number of physicians hired, while others (e.g., credentialing effort or travel for recruitment visits) vary with the number of physicians recruited. The costs of hospitalist turnover may therefore change as HM groups grow, and the marginal costs of turnover may be more modest at large employers.

In order to inform recruitment strategies, retention strategies, and interventions meant to reduce turnover, we set out to estimate the costs of hospitalist turnover in a large academic HM group.

METHODS

The Cleveland Clinic Department of HM is among the largest and oldest in the country and provides inpatient care at multiple hospitals, including a large tertiary referral center and five smaller community hospitals in Northeast Ohio. Within that group, we set out to estimate the costs of turnover, including those borne by the department and by the institution. We considered costs from multiple sources, including direct costs related to recruiting, hiring, onboarding/orienting, moonlighting coverage, administrative time for scheduling, and indirect costs from decreased clinical productivity. This study was deemed exempt by the Cleveland Clinic Institutional Review Board.

To estimate the cost of time spent by recruitment personnel, we pro‐rated the salary of our recruitment professional according to the proportion of physicians hired that joined the Department of Hospital Medicine. In doing so, we assumed that the proportion of time spent recruiting physicians who were not hired did not differ between HM and other specialties.

Our department also pays two professionals specifically for clinical hospitalist recruitment. We pro‐rated the salary of the recruitment director over this period by the 0.2 FTEs of effort dedicated to recruiting. We considered the full salary of the recruitment coordinator a relevant expense.

To capture the departmental expense of holding recruitment meetings through the year, we assumed a cost of $125 per hour (the moonlighting reimbursement during this time period) for each of seven physicians involved for each meeting held. We used the average salary of the administrative personnel who attended to tally the cost of administrative time related to recruitment meetings.

Each physician candidate was interviewed by other hospitalists during recruitment. We tallied the amounts paid to each staff member for time spent interviewing prospective hires.

We obtained the estimated cost of credentialing each new physician from our central physician staffing office, including the cost of visas or other legal services.

Actual relocation costs, which are based on submitted receipts, were unavailable. Instead, we assumed that each hospitalist eligible for relocation expenses (based on distance from the candidate's previous home to our hospital and initial title when hired) incurred the maximum allowable relocation expense.

As part of onboarding, each incoming physician attends institutional orientation, training in the use of our electronic health record (EHR), and a departmental orientation. Our central staffing office provided the average per capita cost of institutional orientation and EHR training. For departmental orientation, we summed the cost of physician leadership time required to hold those events and divided the total by the number of hired physicians.

To estimate the change in staffing costs related to turnover, we considered salary differences and additional costs of clinical coverage during transition periods. To estimate salary differences between incoming and outgoing hospitalists, we obtained the salaries of each physician leaving and assumed that, had the departing physician continued, he/she would have received a 2.5% annual salary increase typical of our institution over the time period studied. We subtracted the actual salaries of incoming physicians from the expected salaries of outgoing physicians to estimate the difference in salary. Existing hospitalists are tasked with orienting new hires to clinical services; if those shifts are beyond salaried obligations, they are paid as moonlighting. We tallied all moonlighting expenses paid to existing physicians to orient new hires. We did not consider the cost savings that could have resulted from clinical coverage at moonlighting rates rather than at a pro‐rated mean hospitalist salary for an equivalent number of shifts, nor did we consider differences in benefits, which are paid to all salaried hospitalists regardless of moonlighting shifts.

Our department predominantly recruits for clinical positions. Some new hospitalists start with small time allocations for resident education, but clinician‐investigators are recruited through a different center.

We theorized that newly hired physicians might be less productive, as measured by relative value units (RVUs), than established hospitalists. To estimate changes in clinical productivity with experience, we obtained physician‐level daily billing data for January 2018 through June 2020. We then used piecewise regression to identify an optimal cutpoint (in days after the physician was hired) at which RVUs billed per shift plateaued. Next, we used a multilevel (clustered by physician), generalized linear model, with a log link function to estimate the daily RVUs generated per clinical day as a function of day since the physician's start date. We controlled for the day/night breakdown of each physician's workload, because physicians working night shifts may have different billing opportunities from those working days, and the proportion of days and nights may differ with time since appointment. Lastly, we used a multilevel linear regression with an indicator variable for whether the shift was before or after the RVU plateau to quantify the mean expected difference between a shift in the initial onboarding period and a shift after that initial period. Based on the predicted daily RVUs, we calculated the revenue expected from those shifts, assuming a 7 on/7 off schedule and the 2020 Medicare conversion factor of $36.09 per RVU. ⁷ Analyses were performed in Stata (version 16).

We normalized all costs for an individual hospitalist by summing costs at the department level and dividing by the number of physicians hired over this period. To illustrate how turnover costs might differ at other centers, we also simulated turnover costs under a range of assumptions based on national wage data.

RESULTS

Between July 2017 and June 2020, our practice interviewed 194 hospitalists and hired 97. We had 34 hospitalists leave, and a total of 234 hospitalists who provided adult care at 6 hospitals in Northeast Ohio. Direct costs of turnover totaled $6166 per incoming physician (Table 1). Moonlighting expenses paid to existing hospitalists to orient newly hired physicians was the largest expense, while the salary difference between outgoing and incoming hospitalists was cost‐saving. Physician time spent recruiting and interviewing prospective candidates was the next‐largest category of expense. Newly‐hired hospitalists had a mean of 1.7 years of post‐residency experience, compared to 2.9 years for departing hospitalists.

Table 1.

Direct costs of turnover per hired hospitalist

Cost category	Subcategory	Mean per hired hospitalist
Recruiting	Salary of recruitment professionals	$254
	Advertisements/job postings	$17
	Department recruitment director salary	$1377
	Department recruitment coordinator salary	$1171
	Recruitment meetings, including physician‐leadership time	$947
	Physician time interviewing prospective hospitalists	$711
Hiring	Credentialing	$100
	Visas and legal fees	$1242
	Relocation	$1822
Staffing costs	Salary difference between outgoing and incoming hospitalists	–$5561
	Moonlighting coverage paid for clinical orientation of new hires	$3731
Onboarding/orientation	Institutional orientation	$44
	EMR training	$14
	Department orientation	$122
Scheduling time	Additional administrative FTEs to incorporate new physicians into clinical schedules and workflow	$174

Our billing dataset included 33,698 total physician‐days across 117 physicians between January 2018 and June 2020. Figure 1 shows the mean RVUs billed on each day, as a function of days since each physician's start date.

Figure 1.

Mean relative value units billed per day in the first six months of physician employment.

In our piecewise regression, RVUs per shift increased until a cutpoint of approximately 25 days. Billed RVUs in the first 25 days of employment are shown in Figure 1. A hospitalist would be expected to bill a mean of 10.3 RVUs per clinical day (95% CI: 9.7–10.9) in the first 25 days of employment, after which they would be expected to bill a mean of 13.0 RVUs per clinical day (95% CI: 12.6–13.5). Over the first 25 days of employment, a hospitalist would be expected to bill 33.2 fewer RVUs (95% CI: 27.7–38.6) than a hospitalist after that time period. Using 2020 Medicare reimbursement, this amounts to a reimbursement difference of $1197 (95% CI: $1001–1393). Figure 2 shows expected RVUs per clinical day in the first 25 days of employment.

Figure 2.

Expected relative value units per shift in the first 25 days of hospitalist practice.

In our simulation over a range of plausible inputs, costs at our institution were within the 95% confidence interval for the number of physicians hired. Institutions hiring fewer hospitalists are likely to incur higher costs of turnover: under our assumptions, programs hiring one hospitalist would spend a mean of $56,943 (95% CI: $27,228–$86,659), programs hiring five hospitalists would spend a mean of $33,333 per hospitalist (95% CI: $9375 ‐ $57,292), programs hiring 10 hospitalists would spend a mean of $30,382 per hospitalist (95% CI: $6877–$53,887), and programs hiring 20 hospitalists would spend a mean of $28,906 per hospitalist (95% CI: $5601–$52,211). Complete results of our simulation are included in the Supporting Information.

DISCUSSION

Physician turnover is widely reported to be expensive, largely based on studies of specialties organized around caring for a panel of patients. ² , ³ , ⁵ , ⁸ In this observational analysis from a hospitalist practice at a large health system, we estimated the costs of turnover in adult HM to be much smaller than earlier estimates from non‐hospitalist specialties.

Multiple reasons underlie the differences between our measurement and earlier, higher estimates. First, lower recruitment costs are an expected result of hospital expansion and growth. Large health systems have infrastructure and personnel dedicated to recruitment, hiring and onboarding, and the marginal cost of one additional physician given those structures is small. Merging hospitals have long cited consolidation of such support infrastructure as their rationale. ⁹

Second, earlier studies focused on practices organized around patient panels, rather than around a site of care like HM or emergency medicine. To provide clinical services and generate RVUs, a primary care physician must have a panel of regular patients. Similarly, office‐based specialists rely on referral networks and long‐standing relationships with primary care physicians for their clinical productivity. Disruption of a panel or referral network can take time to rebuild, and patients who leave a practice during such a disruption may never return. In contrast, hospitalist billing relies solely on the inpatient census. Patients present to the hospital because they are severely ill, not to see a specific hospitalist; there is no reason to think that replacing one hospitalist with another would substantially impact the inpatient census. Reduced billing resulting from hospitalist turnover, therefore, relates only to the period of decreased productivity associated with onboarding and increasing familiarity with hospital billing procedures, rather than a period of rebuilding a patient panel or referral network. Similarly, because almost every patient admitted to a hospitalist is a new patient, turnover creates no loss in efficiency due to hospitalists having to get to know their patients, as would occur in primary care. Because reduced billing and loss of future revenue were among the largest costs in earlier analyses of turnover, this likely accounts for the biggest difference between our results and earlier estimates. ² , ¹⁰

Third, technology has likely reduced recruitment costs, just as it has reduced costs in many sectors. One way to interpret the failure of a market (labor or otherwise) to perfectly match supply with demand is to invoke “friction” costs—the inefficiencies and obstacles that make transitions more difficult or complicated, and thereby make every purchase or hiring process less efficient. ¹¹ Our practice spent less on recruitment than physician practices in earlier studies, which may reflect the same technological processes—online advertising, virtual interviews, and the like—that have reduced such friction costs throughout the economy.

These changes mirror broader shifts in the economy in the past two decades. In the same way that Uber and Lyft aggregate rider demand for transportation while commoditizing drivers, ¹² health systems with a direct patient relationship can aggregate inpatient demand and commoditize clinical services. Although physicians are highly‐paid employees rather than independent contractors, and physician guilds limit the supply of available employees, from the perspective of an employer's balance sheet, any available employee who can accomplish the work at hand is equivalent. To a ride‐sharing company, any driver who can get a passenger to the destination while leaving the passenger satisfied is essentially equivalent; to a hospital, any physician who can provide high‐quality care and quickly discharge a satisfied patient after an acute illness is essentially equivalent. In fact, moonlighting used to overcome staffing shortfalls during times of physician turnover can be thought of as supply‐demand matching—a licensed, credentialed workforce able to flexibly fill a need for clinical care, much as surge pricing matches driver supply with rider demand. Moreover, the same tools that enable virtual communities across the globe allow recruitment across the country. Although medicine is limited by state licensure and not a purely commoditized profession, the internet has clearly increased recruiters' ability to reach qualified candidates. Finally, the US economy has become more concentrated in the past two decades; the same rationales of reducing redundant support operations have been cited in support of mergers and acquisitions within healthcare as in other industries. ¹³

Each of these changes brings advantages. Compared with physicians who provide longitudinal care for a patient panel and who must consider the personal costs of leaving longstanding doctor‐patient relationships, hospitalist jobs have lower barriers to entry and exit. Hospitalists may therefore be more readily able to leave one institution for a more desirable position elsewhere, matching supply with demand or, as in the moonlighting costs observed here, to meet temporary needs for additional clinical care. As a result, hospitalist salaries may be more sensitive to market forces than those of office‐based physicians. Lower barriers to entry or exit may incentivize health systems to create more desirable positions—a desirable workplace represents a distinct advantage in a field with such low barriers.

These changes carry downsides as well. First, commoditizing physicians or other professionals obscures the different ways in which we ought to value those professionals and their work; not all goods should be treated as commodities. ¹⁴ Collegiality and community are precious, even if consequentialist frameworks fail to value them. ¹⁴ Second, although mergers and acquisitions can reduce redundant support operations, hospital consolidation is associated with reduced subsequent wage growth among nurses and pharmacists; the same labor market power could be used to reduce wage growth for other professions with healthcare‐specific expertise, including physicians. ¹⁵ Third, high turnover may incur second‐order costs, such as longer hours, low morale, burnout, loss of mentorship, or others, any of which could further contribute to turnover and/or reduce opportunities for hospitalists' personal growth. Such costs and risks are difficult to quantify but may have important adverse consequences on care, perhaps mediated by burnout. If taken to an extreme, such an approach could lead to catastrophic failure, where a health system is incapable of providing inpatient care due to staffing shortages and is forced to contract with an outside vendor for hospitalist coverage.

This analysis has important limitations. First, RVUs are a measure of hospital income, but likely do not reflect overall clinical contributions. RVUs may not capture skilled diagnostic reasoning or work to build a consensus treatment plan among multiple involved subspecialties. Patients can benefit from many physician services, and not all are reflected in the RVU scale. Second, we have estimated the monetary shortfall between a hospitalist in the initial period of employment and a more experienced hospitalist using Medicare reimbursement, but payment per RVU varies among private insurers, between private insurers and Medicare, and across state Medicaid plans. ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ Our analysis does not incorporate differences in hospital revenue from payer mix; however, given the small loss in RVUs related to turnover, this is unlikely to account for a substantial difference. Third, these data derive from one large health system, and may not generalize to other large systems in different geographies or to smaller health systems. In our simulation (included in the Supporting Information), costs of turnover in HM are unlikely to approach previous estimates from panel‐based specialties; only when hospitals with high recruitment costs hire one hospitalist would the cost of turnover in HM approach previous estimates of the cost of turnover, and even then only the lowest of previous estimates from panel‐based specialties. ² , ³ Still, we lack confirmatory data from other centers. Ideally, future studies from other institutions would illuminate how the costs of turnover vary with faculty size, geography, and other parameters. Until then, our simulation suggests that the low costs of turnover at our institution are in part from distributing fixed recruitment costs over large numbers of hired physicians, and in part from lower costs at our institution compared to those expected across the country. Fourth, we have not captured intangible costs or estimated the costs of rare‐but‐catastrophic events that could result from intangible harms to medical practices, such as practice collapse due to en masse turnover. The probability and consequences of such events are difficult to calculate; our data cannot inform the risk of such events. ²⁰ , ²¹ In spite of these limitations, we believe these data more accurately represent the costs of hospitalist turnover than earlier studies. Importantly, they help to explain why turnover in HM remains high and why HM programs often hire physicians, such as those applying for subspecialty fellowships, who are very likely to leave within a few years.

If the costs of turnover estimated here are replicated elsewhere, we would encourage reflection among specialties organized around a site of care regarding their mission and purpose. As with other administrative expenses, the costs of turnover do not improve patient health outcomes, and reducing them is to the good. ²² , ²³ But the lack of frictional costs could also allow exploitative departments to cycle through physicians, knowing that replacing a burned‐out physician is not prohibitively expensive on a balance sheet. Our profession must take care not to apply the logic of markets where other values should prevail. ¹⁴ Technology is morally neutral, and the monetary costs of recruitment and turnover are but one consideration when building a workforce.

In summary, the direct and indirect costs of turnover in HM appear to be substantially lower than previous estimates from outpatient practices. These differences may stem from organization around a site of care rather than a patient panel, from different hiring practices at integrated healthcare systems, and from the beneficial effects of technology in aiding recruitment. Building physician groups that are both sustainable and sustaining will require careful ongoing focus and deliberation on the goals of our profession, not simply the logic of markets.

CONFLICT OF INTEREST

The authors declare no conflict of interest.

Supporting information

Supplementary information.

Pappas MA, Stoller JK, Shaker V, Houser J, Misra‐Hebert AD, Rothberg MB. Estimating the costs of physician turnover in hospital medicine. J Hosp Med. 2022;17:803‐808. 10.1002/jhm.12942