Estimating the Attributable Cost of Physician Burnout in the United States

Occupational burnout is a syndrome characterized by 3 key dimensions: emotional exhaustion, feelings of cynicism and detachment from work, and a sense of low personal accomplishment (1, 2). The prevalence of burnout among physicians is high relative to the general working population: In a 2014 study, approximately 54% of physicians reported at least 1 symptom of burnout, almost twice the rate of the general U.S. working population (3, 4). Recent studies have begun to provide a more complete picture of the challenges physician burnout presents to the nation's health care delivery system. Systematic reviews have documented associations between physician burnout and negative clinical outcomes as well as unfavorable productivity-related outcomes (5, 6). For example, studies have found that burned-out physicians have higher rates of self-reported medical errors (79) and their patients have poorer clinical outcomes (10, 11). Physicians with burnout are more likely to report an intention to reduce their work hours or to leave medical practice altogether (1214). They also have higher absenteeism rates (13). Recent research has uncovered the organizational roots of burnout (15, 16), and health care executives have begun to recognize the urgency of this problem. A group of 10 CEOs of leading U.S. health care organizations unanimously concluded that physician burnout is a pressing issue of national importance (17) and called on other leaders to commit to addressing it. Despite the recent public interest in this subject and literature suggesting that burnout has the potential to be a major problem, only a few studies (18, 19) have attempted to quantify its economic magnitude in the form of easily understandable metrics. As a result, policymakers cannot holistically assess the extent of the burnout problem and develop appropriate policy responses, nor are leaders of health care organizations equipped to make informed decisions when determining whether to invest scarce resources into programs to mitigate burnout. In this study, we undertook a cost-consequence analysis to investigate the economic burden associated with physician burnout. We used cost as a metric because it is easily understandable by policymakers and organizational leaders and is typically an important data point they can use to make informed decisions, develop organizational strategy, and effect change. We followed a standard approach used by cost-effectiveness studies (20, 21): We constructed a mathematical model linking measureable inputs to the output of interest, estimated values for the input parameters from several data sources, and ran them through the model to estimate the value of the output. This study's contributions are 2-fold. First, we introduced a model to estimate the cost associated with burnout in a given population of physicians. Second, we used the model to estimate the annual burnout-attributable costs for the United States as well as for a hypothetical 1000-physician organization whose distribution of age and specialty segments matched the national averages. We used published data sources to estimate the model's input parameters, which reflect our best attempt to synthesize the findings of recent research on the effect and prevalence of physician burnout. Nevertheless, ideal data were not always available, and some parameters had to be extrapolated. Methods Estimation Approach Our cost-consequence analysis (Figure 1) simulated a hypothetical population of U.S. physicians stratified into 6 segments comprising 2 age groups (<55 years and 55 years) and 3 specialty groups (primary care physicians, surgical specialties, and other specialties). The definitions of the segments were chosen to be consistent with previous studies and available data (18, 22); segment sizes were set to match the distribution of U.S. physicians from the 2013 American Medical Association Physician Masterfile (22). In this study, we focused on 2 costly organizational outcomes: turnover and reduction in clinical hours. These were chosen over other productivity metrics because they directly affect the net supply of clinical capacity, which in turn is an important consideration for strategic planning at both a national level (from a health policy perspective) and the level of individual organizations (from a managerial perspective). The model's primary output was the cost attributable to burnout, that is, the difference in costs for these outcomes as observed and the corresponding costs if physicians were not burned out. Further details of the model and input estimation methodology are reported in Supplement 1. Supplement 1. Supplemental Appendix Figure 1. Cost-consequence model used to estimate the cost attributable to physician burnout. Input Parameters and Data Sources All costs described in the present study were already inflation adjusted to 2015 dollars by using the medical care component of the Consumer Price Index (23). Although data sources for the inputs were not found through a formal systematic search, we generally prioritized studies that were recent, were published in peer-reviewed journals, directly measured the parameters, and contained segment-specific estimates. Table 1 summarizes the values of selected model parameters. Table 1. Summary of Input Parameters and Estimated 95% CIs Burnout Prevalence Burnout prevalence was estimated from a 2014 national survey of 6880 physicians that assessed level of burnout and short-term career plans. Details of this survey were reported previously (4, 24). Single-specialty studies (2527) and other less rigorous studies (28) have identified similar prevalence estimates. Odds Ratios and Outcome Prevalence The outcome prevalence and odds ratios for intended reduction in professional effort were estimated from the 2014 survey (4, 24). For the outcome of physician turnover, annual turnover statistics were estimated from a 2013 survey conducted by Cejka Search and the American Medical Group Association (29). To the best of our knowledge, only 2 studies have been published that investigated the association between burnout and actual physician turnover: 1 from the Cleveland Clinic (30), and the other from Stanford University (31). Our base analysis used fixed-effects meta-analytic inverse-variance weighting of the odds ratios from these 2 investigations. The estimated I 2 statistic was 44.6%, suggesting moderate heterogeneity between the studies. We note, however, that this statistic is known to be insensitive when only 2 analyses are combined. Conversion Parameters For the outcome of reduced clinical hours, we had to estimate conditional probability parameters that mapped from intended to actual reduction in clinical hours. We extrapolated these parameters by assuming that these conditional probabilities were the same as those mapping from intended to actual physician turnover, estimated them from the Stanford study (31), and adjusted them to 1-year probabilities. Cost Parameters We accounted for 2 cost components associated with physician turnover. The first component was the cost associated with physician replacement; the second was the lost income from unfilled physician positions. Physician replacement cost was broken down further into 3 subcomponents, which were estimated separately: search costs, hiring costs, and physician startup costs (termed friction costs by economists). These were estimated from, respectively, a 2015 report from the Association of Staff Recruiters (ASPR) (32), a 2016 report by a search firm (33), and a 2004 study of physician turnover in a U.S. academic medical center (34). Lost income from unfilled positions was included only in the organizational-level analysis. We excluded this component from the national-level analysis because at that level the lost income from physicians leaving 1 organization is gained by the new organization they join, unless the physician leaves medical practice permanently. In the latter scenario, this estimate would be conservative. We estimated this component as the difference between physicians' collections and compensation by using industry benchmarking data collected by the Medical Group Management Association (35), adjusting this difference for the average duration of vacancy obtained from the ASPR report (32). To estimate the cost of physicians reducing their clinical hours, we adjusted the net cost of turnover by a fraction that represented the average percentage difference in weekly work hours between burned-out and nonburned-out physicians. This fraction was estimated by analyzing primary data from the 2014 physician survey (24). Sensitivity Analysis We conducted 3 groups of sensitivity analyses: a rerun of the model using alternate modeling assumptions, univariate sensitivity analyses, and multivariate probabilistic sensitivity analyses. In the first group, we focused on assessing the effect of using alternative data sources for our estimates for some model parameters and varying implicit model assumptions. In particular, for the odds ratio of burnout and actual turnover, we assessed the effect of using only results from either the Cleveland Clinic or the Stanford study. The second and third groups of analyses aimed to assess the model's robustness to perturbations in the inputs within their ranges of uncertainty (reported in Table 1). In the latter group, we used 100000 random draws with standard distributional assumptions (Table 1). Additional details and results are reported in Supplement 1. Results Using the base-case model, we estimated that approximately $4.6 billion a year related to physician turnover and reduced productivity is attributable to physician burnout in the United States. As Figure 2 shows, estimated turnover costs were generally higher than costs of reduced productivity across all segments. Burnout-attributable costs tended to be greater in the younger segment of physicians (those aged <55 years). Figure 2. Estimated annual cost (in 2015