Hospital ownership and admission rates from the emergency department, evidence from Florida

Abstract

Objective

In light of Department of Justice investigations of for‐profit chains for over‐admitting patients, we sought to evaluate whether for‐profit hospitals are more likely to admit patients from the emergency department.

Data Sources

We used statewide visit‐level inpatient and emergency department records from Florida's Agency for Healthcare Administration for 2007–2019.

Study Design

We calculated differences in admission rates between for‐profit and other hospitals, adjusting for patient and hospital characteristics. We also estimated instrumental variables models using differential distance to a for‐profit hospital as an instrument.

Data Collection/Extraction Methods

Our main analysis focuses on patients ages 65 and older treated in hospitals that primarily serve adults.

Principal Findings

Adjusted admission rates among patients ages 65 and older were 7.1 percentage points (95% CI: 5.1–9.1) higher at for‐profit hospitals in 2019 (or 18.8% of the sample mean of 37.8%). Differences in admission rates have remained constant since 2009.

Conclusion

Our results are consistent with allegations that for‐profit hospitals maintain lower admission thresholds to increase occupancy levels.

What is known on this topic

• False Claims Act lawsuits have alleged that for‐profit hospitals over‐admit patients from the emergency department.

• Estimates of differences in admission rates between for‐profit hospitals and non‐profit/public hospitals from prior studies are inconsistent, but the studies were not designed to address this issue.

What this study adds

• We find that admission rates are almost 20% higher at for‐profit hospitals in Florida.

• Differences in admission rates between for‐profit hospitals and non‐profit/public hospitals have been steady over time.

1. INTRODUCTION

Patients make almost 150 million visits to the emergency room annually (AHRQ 2021). At each visit, physicians must decide whether to admit the patient to the hospital or send her home. The decision has major consequences for health and spending. Hospitals receive higher payments from insurers—$3000 to $7000 per encounter ¹ , ² —for patients who are admitted as inpatients versus treated in the emergency department or observation status. Given the extra reimbursement—and potential for higher profits—hospital managers may be tempted to pressure physicians to admit patients. According to the Department of Justice, some have. For‐profit hospital chains Community Health Systems (CHS), Prime Healthcare Foundation, and Health Management Associates (HMA) collectively paid $419 million to the government to resolve allegations that they billed Medicare for unnecessary admissions. These cases were resolved between 2014 and 2018 and addressed claims filed as late as 2013. Recently, allegations have surfaced that HCA Healthcare over‐admits patients. A False Claims Act case is pending, and U.S. Rep. Bill Pascrell has requested that the government investigate HCA's admissions practices. ³ , ⁴

The False Claims Act is a Civil War‐era law that allows individuals to sue companies that are defrauding the federal government. ⁵ Whistleblowers in successful suits receive between 15% and 30% of the resulting penalties or settlements, which can amount to millions of dollars. The Department of Justice has the option of joining, or “intervening”, in False Claims Act suits, and the Department takes the lead in most successful cases.

According to whistleblowers' filings, Department of Justice press releases, and media reports, ⁶ managers at for‐profit hospitals used a number of tools to encourage physicians to admit patients. They either discouraged the use of observation status or encouraged physicians to admit patients from observation status. ³ , ⁷ Administrators set up systems and defaults to favor admission. For example, some HCA administrators provided physicians with pre‐filled admission forms. ³ HMA hospitals used a proprietary emergency department software package that required physicians to make additional clicks if they wanted to discharge a patient the system had flagged as a candidate for admission. Management developed admission rate targets for physicians and sent feedback reports assessing physicians' performance relative to targets and peers. Physicians whose rates were below target were pressured to increase admissions and, if they did not improve, fired. ³ , ⁷ The admission decision is difficult to standardize or evaluate on a case‐by‐case basis. Patients present with a wide range of conditions, and non‐medical factors, such as patients' housing situations, can influence the admission decision. ⁸ However, case volumes are large, and patients are more or less randomly assigned to physicians in the emergency department. These features allow administrators to use yardstick measures to flag and pressure physicians at the lower end of the admission rate distribution.

In this paper, we estimate whether admission rates from the emergency department are higher at for‐profit hospitals. We focus on admission, rather than admission or use of observation status. While admission and observation status are medical substitutes in some cases, the distinction has important implications for Medicare payment. Observation care is billed as an outpatient service, generally at much lower rates than inpatient care. False Claims Act lawsuits against for‐profit hospitals focused on excess admissions.

In most hospitals, emergency medicine physicians and hospitalists (physicians who specialize in providing inpatient care) jointly decide whether to admit patients from the emergency room. The decision is complex, and physicians have a wide degree of latitude to admit patients, keep them in observation status, or send them home. Patients present with a wide range of conditions, and physicians must also consider non‐medical factors, such as patients' living situation and mental status (Baker et al. 2021). The profitability of admitting patients from the emergency department depends on the payment differential between inpatient and outpatient care, the direct cost of care in each setting, the “shadow cost” of admission, and the impact of admission on penalties under the Hospital Readmissions Reduction Program and other federal programs. Insurers pay more for inpatient care. When hospitals are near capacity, managers must consider how the admission of one patient could displace the admission of another, potentially more profitable patient (the “shadow cost”). Hospitals with higher admission rates are liable for greater penalties under Medicare's Hospital Readmissions Reduction Program (McWilliams et al. 2019), which began in 2012. In our case, the relevant question is not whether inpatient care or observation status is more profitable on average. Rather, the question is whether admission is more profitable for the marginal patient (i.e., the patient potentially affected by differences in hospitals' admission thresholds). Since the marginal patient is probably healthier than the average admitted patient, hospitals earn higher profits from the admitted marginal patient.

Although hospital chains' settlements with the Department of Justice may seem like prima facie evidence that they over‐admit patients, or at least did so in the past, healthcare providers face tremendous pressure to settle False Claims Act lawsuits even if facing erroneous or exaggerated allegations. Litigation is costly, and places providers at risk for paying treble damages plus up to $25,000 in penalties per false claim. False Claims Act cases have historically relied more on review of patient records and management practices rather than rigorous data analysis. It is also possible that for‐profit hospitals over‐admitted patients in the past (almost all cases addressing unnecessary admissions have involved for‐profit hospitals), but, having observed the success of the Department of Justice in extracting large settlements from CHS, Prime, and HMA, have discontinued this behavior. The goal of our study is to determine whether patients treated in the emergency departments of for‐profit hospitals are more likely to be admitted from the emergency department compared to patients treated at other types of hospitals.

Our work is related to the literature on differences between for‐profit and non‐profit healthcare providers. For‐profit and non‐profit hospitals have different governance and capital structures, pursue different objectives, and differ in their ability to fully appropriate profits. These differences may affect how patients are treated. An implicit promise not to exploit their informational advantage relative to patients is, in the view of Arrow ⁹ and others, the raison d'être for the predominance of non‐profits in the hospital industry. In our case, non‐profit hospitals may decline to admit patients even when doing so would be profitable. A commitment to providing high‐quality care may attract physicians and patients, either directly or via referral, to the hospital. Managers in for‐profit hospitals may also be more likely to exert control over physicians' decisions because, unlike in non‐profit hospitals, there is a residual claimant for firm profits who is motivated to intervene. ¹⁰ , ¹¹ Prior studies have found that patients treated in for‐profit hospitals receive costlier care and are more likely to receive expensive (and presumably profitable) new treatments (Greenwood et al. 2017; Horwitz 2005; Morris et al. 2017; Sen et al. 2014; Sloan et al. 2001; 2003). However, spending for heart attack patients is lower in areas with more for‐profit hospitals (Kessler and McClellan (2002). Non‐profit and for‐profit hospitals respond similarly to changes in payment levels (Duggan 2002; Cook and Averett 2007), and, adjusting for hospital location and patient sorting, they are similar in terms of their provision of charity care and unprofitable services (Capps et al. 2020).

2. METHODS

2.1. Data and sample selection

We have data on the universe of emergency department and hospital admissions from the state of Florida from the first quarter of 2007 to 2019. These data are compiled and made available to researchers by Florida's Agency for Health Care Administration. ¹² The data include the patient characteristics typically found in billing data. Florida is home to a large number of for‐profit hospitals, and some of the allegations regarding excess admissions have focused on hospitals in the state, making Florida a good setting for assessing the relationship between ownership and admission rates.

Our main analysis uses data from 2019. The data include patients' age, sex, zip code, diagnoses, insurance type, and procedures. They report whether the patient sought care on a weekend but do not report dates of service, other than a quarter of the year, or physician identifiers that allow us to measure physician‐level admission rates. We linked these data to the RAND Hospital Data (which is compiled from hospitals' Medicare cost reports), ¹³ the Area Resource File, a list of urgent care centers in Florida, and latitude and longitude coordinates for patient zip code centroids.

We included patients who sought care in the emergency departments of short‐term general acute care hospitals in Florida and listed a Florida address as their residence. We excluded patients for whom age was missing, patients with home zip codes outside of Florida, patients treated in hospitals with fewer than 1000 emergency department encounters per year, patients treated in specialty hospitals (for example, psychiatric hospitals), and patients whose zip codes could not be merged to latitude or longitude coordinates or emergency department visit rates from the Area Resource File (see Table S1 for details).

2.2. Outcome variable and hospital ownership

Our outcome variable was an indicator of whether the patient was admitted as an inpatient. We identified inpatient records where the patient was admitted from the emergency department using the Condition Code variable. In our baseline analysis, we did not consider patients as “admitted” if they were treated in observation status. However, we also report differences by hospital type in the proportion of patients who were either admitted or treated in observation status, as indicated by having positive observation charges, in a sensitivity analysis. We categorized hospitals' ownership using the RAND Hospital Data and our own review of hospitals' websites. In our main analysis, we compared all for‐profit hospitals to non‐profit and other hospitals. (Non‐profit and public hospitals have similar admission rates).

2.3. Analysis

We estimated the impact of ownership type on admission rates using linear probability models. Our main model takes the following form:

$$\mathrm{Pr}\left({\mathit{\text{ADMIT}}}_{\mathit{ih}}=1\right)=\alpha +\theta {\mathit{FP}}_h+X_i\beta +D_i\pi +W_h\delta +{\epsilon }_{\mathit{ih},}$$ (1)

where i indexes emergency department visits and h indexes hospitals. The unit of observation is the emergency department visit (the data do not allow us to identify patients with multiple visits). We clustered standard errors at the hospital level.

The main parameter of interest is θ, the coefficient on an indicator for whether the hospital is for‐profit. There are three groups of covariates: patient and visit characteristics, X _i , patient diagnoses, D_i, and hospital characteristics other than ownership, W_h. We also estimated versions that exclude covariates for patient and hospital characteristics to show how the addition of covariates changes the estimated effect of receiving treatment at a for‐profit hospital.

Patient and encounter‐level characteristics include age, sex, race, enrollment in Medicare Advantage (or, for patients ages 30–64, payer type), an indicator for whether the visit occurred on a weekday, travel distance between patients' zip code centroids and hospitals (a proxy for severity), and the number of urgent care centers in a five‐mile radius based on patients' home zip code (a measure of the availability of substitutes for emergency department care). Covariates for patients' diagnosis are indicators for patients' presenting or admitting diagnosis, that is, the reason they sought care. We did not use patients' principal or secondary diagnoses as controls because admitted patients probably undergo a more intensive work‐up. Hospitals and physicians will detect and record more and different diagnoses for patients who are admitted. Hospitals may use different standards for coding patients' diagnosis, and diagnosis is a poor indicator of emergency department patient acuity. ¹⁴ We categorized patients' presenting diagnosis using the Agency for Healthcare Research and Quality's multi‐level Clinical Classification Software diagnoses. ¹⁵ We grouped diagnoses categories that account for smaller shares of visits into an “other” group.

Hospital‐level variables include patient volume (the average number of emergency department visits per year), indicators for whether the hospital offers percutaneous coronary intervention and coronary artery bypass graft surgery (proxies for hospitals' ability to treat seriously ill patients), and the number of emergency department visits per person in hospitals' county (a proxy for area‐level differences in the propensity to use emergency department care).

There are large differences in admission rates across age groups, and so we stratified the analysis by age group. We restrict additional analyses, described below, to patients ages 65 and older. They are of special interest given the implications of differences in admission rates for the Medicare program and False Claims Act enforcement. We excluded patients ages 29 and younger. Admission rates are low (~5%) in this age group.

To illustrate heterogeneity in hospital admission patterns, we also estimated a version that excludes any covariate for ownership type and calculated the difference between observed and predicted admission rates for each hospital.

We also conducted an analysis of trends using data from 2007 to 2019. We estimated a least squares model to assess trends in the difference in admission rates between for‐profit and other hospitals. The model includes year indicators and year indicators interacted with ownership (for‐profit versus other). Covariates (for example, age) are held constant across years. We did not include controls for patient diagnosis, because we observed that the proportion of patients admitted with non‐specific diagnoses (for example, abdominal pain) at HMA/CHS hospitals increased dramatically around 2012, suggesting that these hospitals changed their coding practices.

2.4. Sensitivity analysis

Estimates of differences in admission rates between for‐profit and other types of hospitals may be biased by unobserved differences in patient or market characteristics. For example, if for‐profit hospitals are located in more affluent areas, estimates could be biased upwards (if, for example, patients in these areas are more likely to use alternatives to emergency department care) or downwards (if, for example, patients receiving care in non‐profit and public hospitals are more likely to be admitted due the presence of social risk factors). We estimated several alternative models to assess the influence of omitted variables on our estimates. First, we estimated an instrumental variables model. The instrument is patients' differential distance to a for‐profit hospital (following McClellan et al. ¹⁶ Konetzka et al. ¹⁷ and many other papers on hospital outcomes). The differential distance is the distance between a patient's closest for‐profit hospital and the closest other (i.e., non‐profit, public) hospital. The instrument is valid under the assumption that unobserved patient characteristics are unrelated to where patients live in relation to for‐profit and other hospitals. We limited the sample to patients with a differential distance in the interval −5 miles to +5 miles, so that the sample is restricted to patients for whom small shifts in residential location, say from a differential distance of −3 miles to +1 mile, may have a big impact on hospital choice. We estimated the model using two‐state least squares. We clustered standard errors at the zip code level (differential distance varies by zip code). Second, we estimated a model restricted to patients receiving care at for‐profit hospitals that are within five miles of a non‐profit hospital (and vice versa). Estimates from this model should be less prone to bias due to differences in market characteristics between for‐profit and non‐profit hospitals.

We estimated a model where we excluded patients treated in the emergency department who expired, left against medical advice, or were transferred to another short‐term hospital. We also estimated a model that includes a separate indicator for public hospitals to confirm that admission rates at non‐profit and public hospitals are similar. We used linear probability models in our baseline specifications to speed the analyses of our large dataset and facilitate the instrumental variables analysis. However, we re‐estimated the baseline model using probit regression to confirm that estimates are not sensitive to the linear specification. We estimated a model where we classified patient diagnosis based on the primary diagnosis rather than the presenting or admitting diagnosis. As previously mentioned, we estimated a model where the outcome was admission or placement in observation care. We estimated models separately for each single‐level Clinical Classification Software category where the total number of emergency department visits in 2019 was 10,000 or greater.

3. RESULTS

Table 1 shows differences in covariates for visits at for‐profit and non‐profit hospitals. Most differences are small‐to‐modest in magnitude. One exception is the proportion of encounters where the presenting or admitting diagnosis falls into the “other” category. The difference could reflect true differences in patients' conditions or, alternatively, differences in coding practices. Given that patients in this sample are ages 65 or older, most are enrolled in either traditional Medicare or Medicare Advantage. Table S2 presents a detailed description of payer mix by hospital ownership for all age groups. Differences in payer mix between for‐profit hospitals and other types of hospitals are small.

TABLE 1.

Comparison of the characteristics of patients treated at the emergency departments of for‐profit hospitals vs. other types of hospitals (including admitted patients) in the state of Florida in 2019, patients ages ≥65.

	Treated at
	For‐profit	Other a	Difference
For‐profit	100.0	0.0	100.0
Admitted	40.9	36.1	4.9
Patient/encounter characteristics
Age	0.4	0.4	0.0
Male (%)	68.4	68.6	−0.2
White (%)	17.2	15.8	1.4
Hispanic (%)	42.6	40.4	2.1
Medicare advantage (%)	72.7	73.1	−0.4
Weekday (%)	875.3	980.1	−104.7
Travel distance in miles	0.4	0.4	0.1
Urgent care centers within 5 miles	5.2	5.1	0.2
Presenting diagnosis (%)
Nervous system	7.5	6.5	0.9
Cardiac	14.9	14.9	−0.1
Lung	12.9	11.9	1.1
Digestive	5.3	5.8	−0.5
Genitourinary	5.1	5.1	−0.1
Skin	2.5	2.4	0.2
Musculoskeletal	13.7	12.0	1.8
Fractures	10.9	10.5	0.4
Symptoms	16.8	14.6	2.2
Other	10.4	16.2	−5.8
Hospital/county characteristics
ED volume (10,000 s)	134.7	167.3	−32.6
Offers PCI (%)	84.6	81.9	2.7
Offers CABG (%)	62.8	52.1	10.6
County‐level ED visit rate	0.12	0.13	0.0
N	840,335	1,415,597

Note: The sample includes patients ages ≥65 who visited the emergency department in 2019.

Abbreviations: CABG, coronary artery bypass graft surgery; ED, emergency department; PCI, percutaneous coronary intervention.

^a The "other" category includes non‐profit, public, and teaching hospitals.

Figure 1 presents coefficient estimates of admission‐rate differences between for‐profit and other hospitals (full coefficient estimates are presented in Table S3). Among patients ages 65 and older, the admission rate at for‐profit hospitals is 5.9 percentage points (15.7% of the average admission rate of 37.9%) higher than at other hospitals (model 1). Predicted admission probabilities from the model fell into the range of −14.8% to 85.1%, but 99% were in the range of 2.2%–67.4%.

FIGURE 1.

Adjusted difference in admission rates between for‐profit versus non‐profit/public hospitals among patients treated in the emergency departments of Florida hospitals in 2019. LPM, linear probability model; IV, instrumental variables model estimated via two‐stage least squares. Error bars represent 95% confidence intervals. The numbers above the error bars are the point estimate (the percentage point difference in admission rates) and, in parentheses, the percent difference expressed as a percent of the overall sample mean. Models control for patient and hospital characteristics. The size of the circles indicates emergency department visit volume: ○ <10,000, ○ 10,000–15,000, ○ ≥10,000.

The estimate from the instrumental variables model is 6.2 percentage points (model 2). We performed instrumental variables balance tests: comparisons of the covariates between patients with a differential distance of less than 0 (closest hospital is a for‐profit) versus greater than 0 (see Table S4). Most differences in patient/encounter characteristics and diagnoses, while statistically significant, are not economically significant. There are some differences in hospital characteristics, but these are to be expected given that differential distance is defined with respect to hospital ownership. The first‐stage F‐statistic was 56 (see Table S5 for first‐stage coefficient estimates).

Estimates for patients ages 50–64 and 30–49 are smaller in magnitude but similar in percentage terms. However, the 95% confidence intervals of estimates from the instrumental variables models include zero.

The coefficient from a model that limited the sample to patients ages 65 and older treated at for‐profit hospitals that were within 5 miles of a non‐profit was 4.5 percentage points (p = 0.02; see Table S6). Excluding non‐admitted patients who left against medical advice, expired, or were transferred to another short‐term hospital (about 3% of the sample of patients ages 65 and older) did not affect the coeffect on profit status (5.9 percentage points (p < 0.001; see Table S7). The coefficients from a model that included a covariate for public hospitals were 6.3 percentage points (p < 0.001; see Table S8) for for‐profit hospitals and 1.0 percentage points (p = 0.53) for public hospitals. The marginal effect of for‐profit ownership on admission rates from a probit model was 5.9 percentage points (p < 0.001; see Table S9). The marginal effect from a model where patient diagnosis reflects patients' primary diagnosis was 5.0 percentage points (p < 0.001; see Table S10). We find that patients treated in for‐profit hospitals are 3.6 percentage points more likely to either be admitted or treated in observation status (p = 0.01; see Table S11). We identified 45 Clinical Classification Software condition categories where there were 10,000 or more admissions in 2019. Adjusted differences between admission rates for for‐profit and other types of hospitals were positive for 40 conditions and significant at the 5% level for 27.

Figure 2 displays hospital‐level differences between observed and predicted admission rates. Differences are Winsorized at ±20 for purposes of display. We display results separately for three large, for‐profit hospital systems. Thirty‐one of System A's 44 hospitals (70.5%) had admission rates above expected levels, and all of System B's 10 hospitals had higher‐than‐expected admission rates. Seventeen of System A's 20 hospitals (85.0%) had higher‐than‐expected rates. Among non‐profits, 21 of 76 hospitals (27.6%) had higher‐than‐expected admission rates.

FIGURE 2.

Hospital‐level differences in observed minus predicted admission rates for patients treated in Florida hospitals in 2019. FP, for‐profit. Each circle represents a hospital. For each hospital, we calculated the difference between observed and expected admission rates. We calculated prediction admission rates using a linear probability model that controls for patient and hospital characteristics. Differences were Winsorized at −20/+20 percentage points.

Figure 3 presents coefficients on the interactions of year and ownership, which can be interpreted as percentage point differences in admission rates between for‐profit and other hospitals. The difference in admission rates increased from 2.6 percentage points in 2007 to 5.1 percentage points in 2009 and 5.6 percentage points in 2012. A Wald test rejects the hypothesis that the coefficients are equal (p < 0.01). The differences are stable thereafter.

FIGURE 3.

Adjusted difference in admission rates between for‐profit and other hospitals in Florida by year, patients ages ≥65. The figure plots coefficients from a regression where an indicator for ownership (for‐profit vs. other) is fully interacted with year indicators. The model does not include a constant. The outcome is an indicator equal to one if the emergency department patient was admitted.

4. DISCUSSION

We find that patients are much more likely to be admitted as inpatients from the emergency department if they receive care at a for‐profit hospital. Differences in admission rates are large; our preferred specification indicates that admission rates are 7 percentage points (or 19%) higher at for‐profit hospitals for patients ages 65 and older. Given that 70% of all inpatient admissions originate in the emergency department, even modest changes in admission rates can have large effects on hospital use and spending. Estimates of differences in admission rates between for‐profit hospitals and other hospitals are consistent across a wide range of specifications and samples. Estimates differ by a large margin across conditions. It is unclear if these reflect true differences in admission thresholds by condition or differences in coding practices.

A handful of prior studies have examined hospital‐level variation in admission rates. Estimates of the impact of treatment at the emergency department of a for‐profit hospital on admission rates are ~0, ¹⁸ 1.8 percentage points, ¹⁹ and 6.5% (about 2.5 percentage points) for Medicare beneficiaries. ²⁰ Pracht et al. ²¹ found that admission rates were slightly higher in for‐profit hospitals among patients with mild‐to‐moderate trauma. These studies were not designed to assess the impact of hospital ownership. One counted patients in observation status as “admitted”, ¹³ and several included controls for comorbidities or patients' discharge diagnosis, ¹¹ , ¹³ which may bias estimates because admitted patients probably undergo a more intensive work‐up. The papers all use data from before or around the time when HMA and CHS reached settlements with the Department of Justice, and so it is unclear if the results are generalizable to more recent time periods. One exception is a recent report from the Service Employees and Industrial Union. ²² Using Medicare claims, the report estimates that admission rates at HCA hospitals were 5 to 10 percentage points higher than expected rates based on patient demographics, principal diagnosis, and rural/urban location. Our results are consistent with the service union's study.

We find that the difference in admission rates between for‐profit and other hospitals increased around 2008. The timing of the increase is consistent with the Department of Justice finding that the scheme to increase admissions at HMA hospitals began “in or around” September 2008. ²³ HMA and CHS became aware they were being investigated in 2010. Prior studies of False Claims Act investigations have found strong deterrence effects, ²⁴ , ²⁵ , ²⁶ —both in targeted (i.e., “specific deterrence”) and non‐targeted providers (i.e., “generalized deterrence”)—but we find that admission rate differences did not decline in the wake of these investigations and resulting multimillion‐dollar settlements. The profits hospitals earn by admitting more patients may have outweighed the future expected costs of penalties. Admission rates were declining from 2010 onwards in both for‐profit and other hospitals. It may be difficult to detect deterrence effects against a backdrop of secular changes in practice patterns.

Non‐random sorting of patients between hospitals may bias comparisons of admission rates. About 30% of elderly patients arrive at the emergency department in an ambulance. ²⁷ (Our data do not record arrival mode.) Emergency Medical Service protocols instruct ambulance drivers to take patients to the nearest “appropriate” hospital, which in most cases would be the closest hospital. ²⁸ Ambulances will bypass the closest hospital in the event that patients are diagnosed with conditions (for example, stroke) that benefit from treatment at specialized units.

Biases in comparisons of admission rates may also arise from local differences in the use of substitutes for emergency department care, such as primary care, urgent care, and self‐care. These affect the pool of patients who use emergency care. Urgent care centers in particular are a close substitute. Some studies find that they siphon‐off low acuity patients. ²⁹ , ³⁰ , ³¹ , ³² Wang et al. ²⁰ find that the degree of substitution is small: there is one less low acuity emergency department visit for every 37 urgent care center visits. Xu and Ho ³³ find that the entry of freestanding emergency departments does not affect the number of visits to hospital‐based departments.

We find that estimates of the difference in admission rates between for‐profit and other hospitals are similar between models that do and do not adjust for covariates. Also, estimates are similar between regression models based on cross‐sectional and instrumental variables designs. Together, these findings increase confidence that admission rate differences are not biased by unobserved variables. However, our regression models only adjusted for broad groups of conditions.

Our analysis is limited to hospitals in Florida. However, we have no reason to believe that the multi‐state corporations that operate hospitals in Florida maintain different admission thresholds by state. The Service Employees and Industrial Union analysis, ¹⁵ which found similar results, used national Medicare data. False Claims Act allegations have included hospitals in Georgia, California, and other states. ³ , ⁷

We cannot determine empirically whether the excess admissions at for‐profit hospitals were completely unnecessary, low value, or cost‐effective. But if the excess admissions were cost‐effective, that begs the question of why non‐profit hospitals, facing the same payment rates and payer mix as for‐profit hospitals, would under‐admit patients while for‐profit hospitals would not. The most logical explanation for admission rate differences is that some admissions at for‐profit hospitals are medically unnecessary or low value.

The proportion of patients who present to the emergency department and are admitted to observation status has increased over time. ³⁴ Hospitals' use and capacities of their observation units vary widely. ³⁵ , ³⁶ The key issue in False Claims Act investigations is whether patients are admitted. Medicare pays more for admitted patients even if they would have received exactly the same care if they spent their entire stay in observation status. Whether admitted patients passed through observation status or were admitted directly is of secondary importance to determining whether for‐profit hospitals over admit patients. However, we find that patients treated in for‐profit hospitals are more likely to either be admitted or be placed in observation status. Hospitals may use observation status as a feeder for inpatient admissions. Note that Medicare billing contractors may retroactively switch a claim from inpatient to observation status if they determine that a hospital improperly classified an encounter as an inpatient stay. We cannot identify these changes in the data.

Health services researchers should be mindful of hospital‐level variation in admission rates when conducting studies using emergency department‐only or inpatient‐only data. Inpatients in hospitals with lower thresholds for admission will have a lower average acuity than inpatients in hospitals with a higher bar for admission. Differences in admission thresholds may also impart bias to publicly‐reported quality measures, such as 30‐day mortality rates for patients hospitalized for pneumonia. We expect that hospitals with higher admission rates will have lower mortality rates.

FUNDING INFORMATION

No funding to report.

Supporting information

Data S1. Supporting Information.

Howard DH, David G. Hospital ownership and admission rates from the emergency department, evidence from Florida. Health Serv Res. 2024;59(2):e14254. doi: 10.1111/1475-6773.14254