Hospital-physician integration and clinical volume in traditional Medicare

1 INTRODUCTION

Over the last decade, American hospitals have acquired physician practices in large numbers and have become major employers of physicians.^1-4 As of 2018, hospitals owned 31.2% of physician practices, up from 13.6% in 2012.² The shift toward greater integration between hospitals and physicians has raised concerns among policy makers, including the Medicare Payment Advisory Commission and the Federal Trade Commission.^{5, 6} These concerns have been validated by empirical work that has associated hospital-physician integration with increases in prices and spending.^7-14

Yet hospital-physician integration, of which hospital employment of physicians is an important form,¹⁵ may cause ripple effects that go beyond prices and spending. The literature has just begun to reckon with an array of additional potential consequences, including consequences for physicians and care teams. For example, Whaley and colleagues demonstrated that even though hospital-employed physicians generated sizable revenue gains for their hospital employers, physicians did not receive higher compensation.¹⁶ Barnes and colleagues found that integration affected clinical labor within practices; notably, they found that hospital-integrated practices were more apt to employ nurse practitioners and physician assistants.¹⁷

The above findings are suggestive that integration could change physicians' clinical workloads, yet no empirical work has examined it directly. One important and underappreciated potential benefit of integration is reduced administrative hassle for physicians, which could translate to seeing more patients. Given anecdotes from physicians that private practice involves excessive time on logistics (e.g., billing insurers) and not enough time treating patients, integration could outsource all the business functions, paperwork, and regulatory compliance to the hospitals' centralized system. This could result in a beneficial re-allocation of physicians' time toward delivering care. Such a development would be good for the hospital system (more billable services), good for physicians (less administrative headache), and good for patients (increased access and more available appointments). These benefits may be most valuable in primary care: primary care physicians (PCPs) are especially important to hospitals since they play a key role in hospital strategies for clinical excellence and business performance, including accountable care organizations and value-based purchasing; indeed, hospitals have directly employed many PCPs in recent years.¹⁸ Moreover, if integration frees up time to see more patients, this would represent an improvement in primary care access and a substantial unappreciated benefit of hospital-physician integration.

To fill these gaps, we asked: what is the effect of hospital-physician integration on physician clinical volume? We studied this question in the context of traditional Medicare, the largest health care insurer in the United States. We hypothesized that integration would enable PCPs to increase their volume of clinical activities. We created a unique dataset by linking multiple physician characteristics files to 7 years of traditional Medicare claims data over a period that exhibited significant integration activity (2010–2016). The resulting sample included over 70,000 U.S. PCPs. Using a difference-in-differences (event study) framework, we measured changes in volume before and after integration among physicians who became hospital-integrated during the study window and compared them to physicians who remained non-integrated. Our results speak to the wide range of effects that hospital-physician integration exerts on the delivery system, with implications for the 38 million patients enrolled in—and the 99% of physicians who participate in—the traditional Medicare program.^{19, 20}

2 METHODS

2.3 Statistical analysis

Following the precedent of other strong study designs in this literature, we evaluated the effect of integration using an event study with integration as the event.^28-30 We centered each physician's integration start dates around t = 0, where t < 0 corresponded to the pre-integration period. Physicians integrated as early as 2011 and as late as 2016. Others did not integrate at all; these comprised our control group.

In our main specification, we estimated an event study model in which we interacted treatment status with each relative time period to create a model fully saturated with treatment leads and lags, omitting the last pre-period.³¹ This specification avoided the concerns raised by Goodman-Bacon and others about two-way fixed effects models with varied treatment start times³²; our approach explicitly modeled variation in treatment effects over time to avoid the pitfalls of those models. We controlled for the calendar year of the observation as well as the competitiveness of the hospital market and the physician market. To provide a summary of our results, we estimated the overall treatment effect by averaging over all lagged effects using a linear combination, that is, an observation-weighted average of the treatment effects.^{31, 33} We used physician fixed effects to address unobserved time-invariant physician characteristics that may have been correlated with integration and clinical volume.³⁴ In all specifications, we estimated linear models with standard errors clustered by physicians. See Technical Appendix T1 for further detail.

We estimated a standard event study model, plotting the regression coefficients and 95% confidence intervals (CIs) of the treatment effect leads and lags before and after treatment in Figure 1.³¹ These approaches enabled us to verify that the pre-period trends were parallel.

3 RESULTS

3.2 Heterogeneous effects

We found modest treatment effect heterogeneity by time period and by practice size. The first full year of integration exhibited the largest declines, with subsequent years showing less extreme effects (Figure 1). This pattern was true for the outcomes of claims volume (e.g., year 1: −295, compared to year 6: −241), work RVUs, and patients treated (Appendix 5). This could have occurred if integrating physicians took time to adjust to a new practice environment, ramping up their billing over time. However, even 5 and 6 years after integration, their volume remained below that of comparison physicians. For professional revenue, the gap between comparison physicians and integrating physicians in the post-period was approximately constant over time.

We tested for heterogeneity by practice size (results for work RVUs shown in Figure 2; see Appendix 6 for other dependent variables) by running our model on subsamples. We defined small practice physicians as those in the lowest quartile of the distribution (1 or 2 physicians). We defined large practice physicians as those in the highest quartile (32 or more physicians). Work RVUs fell more among small-practice physicians. Claims and professional revenue fell by about the same amount for small and large practice physicians (Appendix 6). For most outcomes, the percentage decrease among large practices was larger than that of small practices since pre-period levels were lower among large practices (Table 2). We also estimated a model in which we interacted the treatment term with an indicator for small practices. The coefficient on the interaction term was negative and significant at p < 0.01 for the dependent variables of claims, revenue, and work RVUs (while not significant for patients).

3.3 Facility fees

Vertical integration was associated with reduced professional revenue in our main analysis. This was partly a mechanical byproduct of the Medicare payment system: many physicians, once integrated, would not bill the same practice expense components for reimbursement that they did when independent. However, the hospital became eligible to receive facility fees for their integrated physicians' labor; we calculated the value of these fees.³⁵ We used the Medicare OPPS facility fee payment rates for each service code. We applied these payment rates to every service code used by each physician in each year in our sample. We summed the facility fee reimbursement value for all services that a physician performed within a hospital outpatient place of service code. For example, the facility fee reimbursement for a visit with a patient was $92.53 in 2014. If a physician performed 100 visits from a hospital outpatient department, their value for this dependent variable in 2014 would have been $9253. This allowed us to construct a comprehensive measure of the annual facility fee value associated with each physician's labor for our full study period. In Figure 3, we show the results of our main specification using this dependent variable.

Integration increased the value of facility fees by $28,556 per physician per year (95% CI: 26,909 to 30,203). When excluding the transitional year (t = 0), the increase was closer to $34,000. In the main analysis, the loss of professional revenue came to about $29,000 per physician per year. The results from the present analysis imply that facility fees almost entirely offset the loss in professional revenue despite major reductions in per-physician billing.

4 DISCUSSION

In this study, hospital-physician vertical integration reduced PCPs' clinical activity among traditional Medicare patients. Integrated physicians saw fewer patients and billed fewer claims. They also generated less professional revenue. We further calculated that these losses in professional revenue were almost completely offset by the extra revenue from Medicare's outpatient facility fees. Together, these findings imply that integration changes physicians' clinical work patterns in ways that could make it more difficult for patients enrolled in traditional Medicare to access primary care services, though it does not appear to result in losses in traditional Medicare revenue for hospital systems.

Previous work has demonstrated that integration affects spending, prices, and the cost of treatment.^{14, 28, 34} We add to the literature by showing that integration also affects PCPs' clinical volume. Contrary to our expectations—that is, that integration would free up physicians to deliver more clinical care—our work showed reduced volume among Medicare patients (we explore several interpretations of this finding below). Our results also add insight into how hospitals are recouping their investments in physician practice acquisitions. Specifically, despite anecdotes that employment contracts in this decade's wave of hospital-physician integration are more closely tied to physician productivity than contracts from the integration boom of the 1990s and early 2000s, hospitals still do not appear to be getting their physicians to do more clinical services, at least not as measured by greater billing in traditional Medicare. Prior work has shown that per-patient Medicare spending increases after physicians integrate.^{11, 13, 34} Squaring that finding with our results—that per-physician billable activity decreases—implies that patients of integrated physicians are getting referred with greater frequency to additional clinicians, to higher-cost sites of service, or to additional services like lab or diagnostic tests. Such an explanation fits with recent work which demonstrated that Medicare spending on labs and imaging (and even inappropriate imaging) increased after PCPs integrated.^{36, 37}

Our results are consistent with several interpretations, each of which has different implications both for physician workload and access for Medicare patients. First, reduced Medicare fee-for-service billing activity following integration could be explained by the desire for part-time work. Larger systems might offer employees more flexibility than small practices. Though we did not have data on hours worked, we did have data on the total number of days on which physicians billed visits under the Medicare program.³⁸ Before integration, physicians billed on about 139 days; this fell by about 13 days, or 9%, after integration (Technical Appendix). This is not conclusive of, although is consistent with, part-time clinical work, which is fairly commonplace among physicians; during the study period, an estimated 1 in 5 physicians nationally worked part-time.³⁹ This pattern is also consistent with physicians seeking hospital employment for changes in professional roles, notably teaching or research, which would reduce billed services. To unpack this possibility analytically, we compared those joining teaching hospitals to those joining non-teaching hospitals. The results between them were not significantly different, implying that teaching responsibilities were not a primary driver of reduced billing (Technical Appendix). Some physicians might also seek non-academic professional roles, such as administrative responsibilities, specialized clinical roles involving less traditional billing, or leadership and service roles on hospital boards and committees. We could not test this here, but research with interviews or surveys could improve our collective understanding. If physicians enjoy these changes in professional roles and the possibility of part-time work, integration could improve physicians' overall work satisfaction. At the same time, reallocating physician time away from clinical roles reduces the supply of physician care available to patients.

A second interpretation is that integration might have changed nothing except the type of Medicare insurance in which the patients were enrolled. If so, integration affected neither physician workload nor patient access. Our study window corresponded to a time of increased Medicare Advantage enrollment, and integrated health systems might have been more inclined to move into the MA market than independent practices. Some health systems (although, in 2016, only 74 nationwide) have even developed their own MA plans.⁴⁰ If physicians' patients were being moved into the hospitals' MA plans, it could explain some of the reduced billing in traditional Medicare. We observed a positive but minimal association between integration and MA enrollment (a 1 percentage point increase in county-wide physicians integrated was associated with a 0.007 percentage point increase in MA penetration). Given this result, we doubt these results would be fully explained by patients switching into MA, but this relationship should be examined in future work.

A third interpretation is that our results reflected changes in the payer mix that disadvantaged Medicare patients. The implications would be that physician workload was reallocated but unchanged in total; access among Medicare patients deteriorated; and access improved among, for example, commercial patients. Once integrated, hospitals may prefer that their physicians treat lucrative commercial patients instead of Medicare patients. Given that commercial reimbursement is about twice that of Medicare for inpatient services and about triple that of Medicare for outpatient services, this interpretation has ample economic rationale.⁴¹ Hospital-integrated physicians might also treat more Medicaid⁴² or uninsured patients; these subjects have been little examined in the literature. Our exploratory analysis of this issue, though far from definitive, did not indicate significantly larger declines in Medicare volume among physicians practicing in areas with more commercial patients compared to those with fewer. Given that this test may not be particularly sensitive, we recommend further research.

A fourth interpretation is that integration reduces physician labor in aggregate. Physicians might continue to treat the same payer mix and engage in similar non-clinical activities but work fewer hours and treat fewer patients across the board. There is economic rationale for this also: hospital-integrated physicians likely face smaller incentives to deliver high volumes of care and see many patients. Prior examples of integration brought hospitals major financial losses in part for this reason.⁴³ Physicians employed by hospitals generally receive a greater share of their compensation via salary compared to physicians in other practice settings.⁴⁴ Reduced incentives on the margin could lead, as found here, to declines in clinical volume when hospitals employ physicians. This may improve physician work satisfaction while reducing the aggregate supply of physician labor. We point out, though, that if clinical workload declines, it is not a guarantee of better work-life balance, even though lifestyle benefits are sometimes discussed as a potential reason for hospital employment.^{15, 16, 45} Several lines of evidence cast doubt on the work-life balance explanation: recent research showed that, among full-time employees, physicians working for hospital-owned practices worked about as many hours as those working for physician-owned practices.¹⁶ Those hours could, in principle, be more enjoyable, but other lines of evidence suggest that there are bureaucratic costs associated with integration⁴⁶; that hospital priorities do not always align with work attributes that physicians find satisfying⁴³; and that zero-burnout practices tended to be clinician-owned rather than hospital-owned.⁴⁷ Interestingly, we found that on a per-claim basis, integrated physicians billed about 13% fewer HCPCS codes, but that RVUs per claim increased by about 10%. This implies that once integrated, physicians or their organizational support staff shift away from lower-RVU services and toward higher-RVU services.

Most of the above interpretations imply reductions, one way or another, in the supply of physician clinical services available to Medicare patients, and the evidence presented here is consistent with reduced supply. Our physician-level analysis was not designed to test patient-level access associated with integration; it is possible that the reduction in integrated physicians' Medicare clinical volume was fully offset by non-integrated physicians or other clinicians absorbing the difference. However, prior work found initial evidence that the integration of a patient's PCP was associated with a per-patient decline in office visits.³⁰ Since hospital-physician integration is rapidly becoming the dominant organizational form in health care delivery, we think it is important to further explore whether this increasingly consolidated landscape exerts contractionary effects on the supply of primary care.

We note several limitations. Our findings apply only to Medicare fee-for-service patients. Results from other payers (e.g., Medicare Advantage, commercial insurers, or Medicaid) may differ; this is an area we recommend for future research. However, one advantage of our data source is that it covers a wide and important swathe of patients, with results pertinent to 38 million patients. Second, we have focused on PCPs. While this setting is clinically important and enabled us to achieve a large sample size, effects may differ among medical specialists or surgeons. Third, if physicians left the workforce, they also left our data: to the extent that integration may have caused physicians to retire or stop treating patients altogether, our estimates would only understate the total reduction in clinical volume that we found here. Some physicians may be predisposed to certain kinds of work settings that would reduce billing productivity. To account for this, our models used physician fixed effects, but time-varying changes in such predispositions are potential unmeasured confounders. If such confounders persisted, our results may not generalize to all unintegrated physicians. Our sample appeared to be representative of the practicing U.S. PCP population (Appendix 2B), although given the data source, we likely exclude physicians who overwhelmingly treat commercial patients and physicians with less stable attachment to the workforce.

Our study points to several promising areas for research. Most obviously, our study could be replicated with other insurers (e.g., Medicare Advantage, Medicaid, or commercial insurance); while outside our scope and the capacity of our data, this would be a welcome complement. Further, studies of physician time use over clinical and administrative tasks could provide useful evidence on the possibilities articulated above, and whether physician practice style, such as longer but fewer visits, is at play.³⁸ Last, whether hospital integration saves physicians from burnout, or, by contrast, increases unsatisfying attributes of the work environment, deserves further exploration.

5 CONCLUSION

Hospital-physician integration affects not only prices and spending but also the composition of and/or the quantity of physician clinical workload. As systems become more integrated, researchers should continue to examine the contexts in which integration may improve physician job satisfaction and increase patient access.

CONFLICT OF INTEREST STATEMENT

As Associate Editor for Urology, Dr. Brent K. Hollenback receives support from Elsevier. Drs. Brady Post, Edward C. Norton, and Andrew M. Ryan have no conflicts to declare.