An updated model of rural hospital financial distress

Abstract

Purpose

To create a model that predicts future financial distress among rural hospitals.

Methods

The sample included 14,116 yearly observations of 2311 rural hospitals recorded between 2013 and 2019. We randomly separated all sampled hospitals into a training set and test set at the start of our analysis. We used hospital financial performance, government reimbursement, organizational traits, and market characteristics to predict a given hospital's risk of experiencing one of three financial distress outcomes—negative cash flow margin, negative equity, or closure.

Findings

The model's area under the receiver operating characteristic curve (AUC) equaled 0.87 within the test set, indicating good predictive ability. We classified 30.55% of the observations in our sample as lowest risk of experiencing financial distress over the next 2 years. In comparison, we classified 32.52% of observations as mid‐lowest risk of distress, 26.40% of observations as mid‐highest risk, and 10.52% of observations as highest risk. Among test set observations classified as lowest‐risk, 5.78% experienced negative cash flow margin within 2 years, 1.50% experienced negative equity within 2 years, and zero observations experienced closure within 2 years. Within the highest‐risk group, 61.57% of observations experienced negative cash flow margin, 43.02% experienced negative equity, and 3.33% experienced closure.

Conclusions

Given the ongoing challenges and consequences of rural hospital unprofitability, there is a clear need for accurate assessments of financial distress risk. The financial distress model can be used by researchers, policymakers, and rural health advocates as a screening tool to identify at‐risk rural hospitals for closer monitoring.

AN UPDATED MODEL OF RURAL HOSPITAL FINANCIAL DISTRESS

Background and new contribution

From January 2005 to May 2024, 219 rural hospitals closed or converted to a facility without inpatient services. ¹ Hospital financial distress is often cited as a main contributor to closure ² and thus is a key target for policies aimed to protect rural health care access. In 2017, Holmes, Kaufman, and Pink published research on the “Financial Distress Index (FDI),” a novel tool designed to describe rural hospital financial distress risk. ³ The model for the FDI was developed and validated using data for 2466 rural hospitals from 2000 to 2013 and could be used to predict the probability of rural hospital financial distress within 2 years. Since the model was originally developed, there have been many changes in the operating environment that may have affected the risk of financial distress among rural hospitals, thus prompting the need for model revision.

Recent changes affecting rural hospitals include the following: (1) More patients are insured by Medicare Advantage plans. Medicare Advantage plans are projected to enroll a majority of nonmetropolitan beneficiaries in 2 years, and the rate of growth continues to be greater in nonmetropolitan counties (10.5% growth per year) than in metropolitan counties (7.2% growth per year) ⁴ ; (2) More rural hospitals are being acquired by systems. A recent study found that, between 2010 and 2018, 17% of rural hospitals merged, most commonly with organizations from outside their local geographic market ⁵ ; (3) A greater percentage of patient revenue comes from outpatient sources. For the average rural hospital, the percent of revenue coming from outpatient services increased from 66.5% in 2011 to 74.2% in 2019⁶; (4) Reported uncompensated care is decreasing. Previous research has found that Medicaid expansion has resulted in decreased uncompensated care among rural hospitals. ⁷

The aim of this study was to create an FDI model that recognizes recent changes in the operating environment and uses financial performance, government reimbursement, organizational traits, and market characteristics to predict the risk of rural hospital financial distress 2 years hence. Rural hospital profitability continues to be a challenge, and having a model that can accurately predict financial distress and potential closure helps programs and policymakers better target resources to prevent closure or convert to other models of care.

METHODS

Framework of rural hospital financial distress

The following goals were established for the research methods: (1) the model should be based in financial theory and previous empirical findings on hospital financial distress; (2) a quantitative approach should be used to assess model calibration and external validity; (3) the model should produce stable results and refrain from “overfitting” the training data; (4) the model should be able to individually identify the risk of financial distress for each rural hospital; (5) the model should have high face validity; (6) the model should be parsimonious; (7) the model should use publicly available data and easily accommodate annual updates to these data.

The conceptual framework for the study extends the original model of rural hospital financial distress. ³ Net income is defined as total revenue minus total expenses over a specific period of time, and equity is defined as total assets minus total liabilities. Net income and equity are related to each other at time period t through the following equation, assuming no dividends are paid out:

$$\mathrm{Equit}{\mathrm{y}}_{t+1}=\mathrm{Equit}{\mathrm{y}}_t+\mathrm{Net}\mathrm{Incom}{\mathrm{e}}_{t+1}.$$ (1)

Equation (1) indicates that profitability (i.e., net income > zero) leads to a growth in equity and that unprofitability (i.e., net income < zero) leads to a decline in equity. Unprofitability and negative equity can lead to additional, stronger signals of financial distress such as hospital closure.

In our framework, we assumed each hospital has a probability of financial distress that is determined by key hospital‐level predictor variables and the prevalence of a given financial distress outcome. We started by considering the overall prevalence of a given financial distress outcome (i.e., unprofitability is more likely to occur than negative equity, negative equity is more likely to occur than closure). From there, we modified the probability of financial distress based on key hospital‐level predictor variables. Under this approach, we characterized the risk of financial distress via a latent, continuous index. Given a value of the index, the probability of each financial distress outcome is independent and differs only due to a constant term determined by the overall prevalence of the outcome. This set‐up is described in Equation (2) below.

$$\mathrm{Pr}\left(y_{h,k,t+2}=1\right)=f\left(X_{h,t}\beta +{\varphi }_k\right).$$ (2)

In Equation (2), y_{k,h,t+ 2} is an indicator variable that equals one if hospital h experiences financial distress outcome k in year t + 2 and equals zero otherwise. The probability of y_{k,h,t +2} equaling one is given as a function of the characteristics of hospital h in year t and the overall prevalence of financial distress outcome k.

Design and sample

We used a longitudinal study design to examine rural hospital financial distress. Our sample included observations of general care, nonfederal rural hospitals from 2013 to 2019 recorded in the Healthcare Cost Reporting Information System (HCRIS). ⁸ HCRIS includes annual cost reports from Medicare‐certified hospitals. Cost reports are generally based on fiscal rather than calendar years, and thus we standardized the reports to calendar years by assigning the calendar year representing the most days from the cost reporting period.

Each cost report includes data on financial performance, utilization, and facility characteristics. We also identified the rurality of each hospital using criteria outlined by the Federal Office of Rural Health Policy in the Health Resources & Services Administration starting in fiscal year 2022. ⁹ Namely, we considered a hospital to be rural if the hospital was located in a nonmetro county, OR a metro Census tract with a Rural‐Urban Commuting Area (RUCA) code between 4 and 10, OR a large area metro Census tract of at least 400 square miles in area with a population density of 35 or less per square mile and a RUCA code between 2 and 3, OR an outlying metro county without an urbanized area.

Measures

We considered three types of financial distress outcomes—negative cash flow margin, negative equity, and closure. We gathered data on the first two outcomes from HCRIS. Negative cash flow margin (see Table A1 for full definition) is a stronger signal of financial distress than other profitability ratios because a negative value indicates use of depreciation to cover operating expenses. Sustained unprofitability can lead to negative equity (when total liabilities > total assets), otherwise known as accounting insolvency. A hospital with negative equity is at greater risk of closure. We identified hospital closures using data from the North Carolina Rural Health Research Program. ¹

To predict rural hospital financial distress, we considered variables from four domains—financial performance, government reimbursement, organizational traits, and market characteristics (Table A1). Variables related to financial performance included hospital profitability, uncompensated care, outpatient revenue, and performance on benchmarks established by the Critical Access Hospital Measurement and Performance Assessment System (CAHMPAS; see Table A2). ¹⁰ Variables related to government reimbursement included Critical Access Hospital (CAH) status, Medicare outpatient payer mix expressed as a percentage of all outpatient charges, ratio of Medicare Advantage and cost plan days to traditional Medicare acute care days, Medicaid‐to‐Medicare fee index, ¹¹ , ¹² , ¹³ , ¹⁴ and Medicaid payer mix expressed as a percentage of all patient charges. Organizational traits of interest included ownership, size (as measured by net patient revenue), and system affiliation. Lastly, market characteristics of interest included competition, market size, economic condition, and Medicare Advantage penetration rate; we describe the methodology for defining hospital markets in the Appendix. Several of the above predictors were present in the 2017 version of the FDI model ³ ; others were added as new variables to increase the model's robustness, given the aforementioned changes in the operating environment.

Based on findings from the 2017 FDI model, we hypothesized that increased hospital profitability, increased CAHMPAS benchmark performance, CAH status, increased Medicaid‐to‐Medicare fee index, increased hospital size, and increased market size would be associated with lower probability of financial distress. In contrast, we hypothesized that hospital for‐profit ownership status, increased market competition, and increased market poverty would be associated with greater probability of financial distress. ³ As for the newly introduced model predictors, we hypothesized that increased uncompensated care would be associated with greater probability of financial distress and that increased outpatient revenue as a percentage of total revenue would be associated with lower probability of financial distress. This latter hypothesis was driven by ongoing shifts away from rural‐based inpatient care ¹⁵ and the financial challenges of maintaining inpatient services within small rural hospitals. ¹⁶ Furthermore, we hypothesized that increased Medicare Advantage and cost plan days (or increased Medicare Advantage market penetration), and increased Medicaid payer mix would be associated with greater probability of financial distress. Recent reports have raised concerns about denied or delayed payments from Medicare Advantage Organizations to hospitals, ¹⁷ , ¹⁸ and Medicaid reimbursement rates have consistently been lower than rates provided by Medicare or commercial insurers. ¹⁴ , ¹⁹ Although increased Medicaid payer mix could signal decreased uncompensated care, our hypothesis on Medicaid payer mix and financial distress assumes that uncompensated care is held constant.

Directions of association for the remaining predictors—Medicare outpatient payer mix and system affiliation—were not obvious ex ante. Increased hospital Medicare outpatient payer mix could be associated with lower probability of financial distress among low‐volume CAHs that are reimbursed by Medicare at 101% of allowable cost for services. ²⁰ However, among non‐CAHs, Medicare reimbursement rates could be substantially less than those provided by commercial insurers. ¹⁹ Furthermore, joining a health system has been linked to improved rural hospital operating margin in previous research, ²¹ but rural hospitals struggling with financial outcomes could be more likely to seek system affiliation to remain viable. This latter occurrence could obscure the relationship between system affiliation and financial distress among our sample.

Statistical analysis

We randomly separated all sampled rural hospitals into a training (i.e., development) set and test (i.e., validation) set at the start of our analysis. Next, we ran a probit regression model of financial distress with hospital‐clustered standard errors using the training set, outcomes, and predictors described above. Model predictors in year t or earlier were used to predict financial distress outcomes 2 years hence (i.e., t + 2). We used a probit rather than logit model for two reasons. First, under a probit framework, the linear combination of predictors has an intuitive interpretation as a Z‐score for a standard normal distribution (in contrast, the linear combination must be interpreted as the log‐odds of financial distress under a logit framework). Second, specification testing ²² suggested that a probit model was a better fit for the data compared to a logit model.

We handled missing data through complete case analysis (see Table A3 for further details), and we checked for multicollinearity by calculating the variance inflation factor of each predictor. ²³ In addition, to account for multiple financial distress outcomes of interest, we used a “stacked” dataset that included one observation for each combination of distress outcome and hospital‐year. For example, a hospital measured in the year 2019 would have three observations in the stacked dataset, one observation that used an indicator for negative cash flow margin in the year 2021 as the binary outcome, one observation that used an indicator for negative equity in the year 2021 as the binary outcome, and one observation that used an indicator for closure by year 2021 as the binary outcome. The model accounted for the stacked nature of the dataset by including a term for “financial distress outcome type” (${\varphi }_k$ in Equation 2). Although the stacked dataset requires a given predictor variable to have the same functional relationship with each distress outcome, we reasoned that this constraint was less important than the resulting gains in model simplicity and parsimony. A parsimonious model of general financial distress will facilitate sorting of highest‐ versus lowest‐risk hospitals and increase model ease‐of‐use by policymakers and rural interest groups.

After running the probit regression model, we calculated the latent FDI and predicted probability of financial distress for each hospital‐year observation in both the training and test sets. We then used the predicted probabilities to check the calibration of our model. First, we calculated the area under the receiver operating characteristic curve (AUC) ²⁴ for our test set. The AUC represents the probability that a randomly chosen hospital observation with financial distress would have a greater predicted probability of distress than a randomly chosen hospital observation without distress. Although context‐dependent, previous research has often considered AUC values between 0.6 and 0.7 to represent poor predictive ability, values between 0.7 and 0.8 to represent fair predictive ability, values between 0.8 and 0.9 to represent good predictive ability, and values above 0.9 to represent excellent predictive ability. ²⁵ , ²⁶ , ²⁷ Next, we sorted the test set observations from lowest to highest predicted probability of distress and grouped them into one of 20 equal‐width probability groups, (0, 5%], (5, 10%], …, (95, 100%]. We then plotted the observed rate of financial distress within each group against the group's average predicted probability. A well‐calibrated model should show that the observed rate correlates strongly with the predicted probability. ²⁸

Lastly, to assist with implementation of our model we classified each hospital into one of four risk categories (“lowest,” “mid‐lowest,” “mid‐highest,” or “highest”) based on their relative risk of experiencing financial distress within 2 years. Lowest‐risk hospitals were defined as rural hospitals with a lower than 10% predicted probability of unprofitability. Among nonlowest‐risk hospitals, hospitals were classified as mid‐lowest risk, mid‐highest‐risk or highest risk based on their probability of experiencing negative equity. Hospitals with mid‐lowest risk have a lower than 10% probability of negative equity, hospitals with mid‐highest risk have at least a 10% probability of negative equity (but lower than 25% probability), and hospitals with highest risk of financial distress have at least a 25% probability of negative equity. We based these risk thresholds on criteria established by the 2017 FDI model, with some added modifications to further prioritize model sensitivity.

RESULTS

Descriptive statistics

Our sample initially included 15,402 yearly observations of general care, nonfederal rural hospitals with a cost report with a reporting period between 360 and 366 days. Of these observations, 91.7% had complete data on all predictor variables and at least one financial distress outcome of interest. Thus, our final sample included 14,116 yearly observations of 2311 unique rural hospitals from 49 states (our sample did not include any rural hospitals from Rhode Island; see Table A4 for the distribution of hospitals by state). Among the sample, 22.3% of observations exhibited negative cash flow margin, 10.5% exhibited negative equity, and 0.7% exhibited closure within 2 years. Regarding other measures of financial performance, the mean value for total margin among sampled hospital observations was 2.3% (standard deviation [SD] = 11.4), the mean value for outpatient revenue to total revenue was 72.4% (SD = 12.7), the mean value for uncompensated care as a percentage of operating expenses was 5.5% (SD = 4.8), and the mean percent of CAHMPAS benchmarks met was 50.9% (SD = 25.0). In terms of government reimbursement, 56.8% of observations were from CAHs, and the mean (SD) values for Medicare outpatient payer mix, the ratio of Medicare Advantage and cost plan days to traditional Medicare days, Medicaid‐to‐Medicare fee index, and Medicaid payer mix were 30.3% (9.6), 20.9% (42.4), 75.4% (13.8), and 15.2% (9.1), respectively. As to organizational traits, 11.3% of observations were from for‐profit hospitals, 39.7% of observations were from system‐affiliated hospitals, and the mean (SD) net patient revenue was $49,277,012 ($58,357,953). Furthermore, our examination of market characteristics showed that the mean (SD) distance to the nearest hospital with more than 100 beds was 42.9 (39.8) miles, the mean (SD) market share was 29.1% (21.1), the mean (SD) market population was 36,897 (29,783), the mean (SD) percent of market population in poverty was 16.2% (5.5), and the mean (SD) market Medicare Advantage penetration rate was 21.0% (13.6) (Table 1).

TABLE 1.

Prevalence and distributions of rural hospital financial distress indicators and predictors for hospital‐year observations (2013–2019).

Variable (variable type)	General formula	Mean (standard deviation) or % with outcome
Distress indicators	Year t + 2	n = 14,116 (unless otherwise noted)
Negative cash flow margin (binary)	((net income—contributions, investments, and appropriations + depreciation + interest expense)/(net patient revenue + other income—contributions, investments, and appropriations)) × 100%	22.2% (n = 13,515)
Negative equity (binary)	equity = total assets—total liabilities	10.5% (n = 13,532)
Closure (binary)	Calculate number of days between NCRHRP's listed closure date (either complete or converted closure) and midpoint of year t hospital cost report.	0.7%

Distress predictors	Year t a
Financial performance
Total margin (continuous)	(net income/total revenue) × 100%	2.3% (11.4)
Total margin, year t – 1 (continuous)	See above entry for “Total margin”	2.3% (11.1)
Total margin, year t – 2 (continuous)	See above entry for “Total margin”	2.5% (10.9)
Outpatient revenue to total revenue (continuous)	(total outpatient revenue/total patient revenue) × 100%	72.4% (12.7)
Uncompensated care as a percentage of operating expenses (continuous)	((charity care + bad debt)/operating expense) × 100%	5.5% (4.8)
Percent CAHMPAS benchmarks met (continuous; see Table A2 for list of benchmarks)	(number of nonmissing b CAHMPAS benchmarks met/total number of nonmissing b CAHMPAS benchmarks) × 100%	50.9% (25.0)
Government reimbursement
CAH status (binary)	Not applicable	56.8%
Hospital Medicare outpatient payer mix (continuous)	(hospital Medicare outpatient charges/hospital total outpatient charges) × 100%	30.3% (9.6)
Ratio of Medicare Advantage and cost plan days to traditional Medicare acute care days (continuous)	((Medicare Advantage days + Medicare cost plan days)/ (Traditional Medicare acute care days)) × 100%	20.9% (42.4)
Medicaid‐to‐Medicare fee index for “All Services” (continuous)	See Refs. (11), (12), and (14)	75.4% (13.8)
Medicaid payer mix (continuous)	(Medicaid charges/total patient charges) × 100%	15.2% (9.1)
Organizational traits
For‐profit status (binary)	Not applicable	11.3%
Net patient revenue c (continuous)	((net patient revenue)/10,000,000) × (365/days in period)	4.9 (5.8)
System affiliation (binary)	Not applicable	39.7%
Market d characteristics
Miles to nearest hospital with > 100 beds c , e (continuous)	Calculate geodesic distance from selected hospital to nearest hospital with > 100 beds.	42.9 (39.8)
Market share (continuous)	First, identify all ZIP codes in the selected hospital's market. Next, calculate the percent of visits and discharges from these ZIP codes that attended the selected hospital.	29.1% (21.1)
Market population c (continuous)	First, estimate the population of each ZIP code in the market using Census population data and HUD USPS County‐to‐ZIP crosswalk files. Next, sum the population estimates for all market ZIP codes.	36,897 (29,783)
Percent market population in poverty (continuous)	First, estimate the proportion of population in poverty for each ZIP code using Census poverty data and HUD USPS ZIP‐to‐County crosswalk files. Next, calculate the average poverty rate among all ZIP codes in the market, weighting by ZIP code population size.	16.2% (5.5)
Medicare Advantage penetration rate (continuous)	First, estimate the Medicare Advantage penetration rate for each ZIP code using CMS Medicare Advantage penetration data and HUD USPS ZIP‐to‐County crosswalk files. Next, calculate the average penetration rate among all ZIP codes in the market, weighting by ZIP code population size.	21.0% (13.6)

^a Unless otherwise stated.

^b The median number of missing CAHMPAS benchmarks was 0 (out of 11 total benchmarks). The average observation had 0.7 missing benchmarks.

^c This variable is logged when used in our probit regression model.

^d See Appendix for further details on market definition.

^e We selected miles to nearest large hospital as a measure of market competition rather than a more general measure such as the Herfindahl‐Hirschman Index (HHI). Calculation of the HHI requires identification of all competing firms in a market, which is difficult in rural settings. Rural hospitals must compete with other nearby rural hospitals as well as larger urban hospitals potentially located multiple hours away. Thus, adapting the HHI to rural settings is nontrivial.

Abbreviations: CAH, Critical Access Hospital; CAHMPAS, Critical Access Hospital Measurement and Performance Assessment System; CMS, Centers for Medicare & Medicaid Services; HUD, United States Department of Housing and Urban Development;NCRHRP, North Carolina Rural Health Research Program; USPS, United States Postal Service.

Probit regression

The probit regression model yielded multiple statistically significant predictors of rural hospital financial distress (Table 2). Since the coefficients of the probit are not directly interpretable, and since we have a variety of covariates, we present the average marginal effects. For continuous covariates, we present the average marginal effect resulting from a one standard deviation change in the covariate. A standard‐deviation increase in percent total margin was associated with a 2.01 percentage point decrease in the probability of financial distress (Average Marginal Effect [AME] = −2.01, p < 0.001), and a standard‐deviation increase in percent CAHMPAS benchmarks met was associated with a 3.62 percentage point decrease in the probability of financial distress (AME = −3.62, p < 0.001). Furthermore, CAH status was associated with a 3.16 percentage point decrease in the probability of financial distress (AME = −3.16, p < 0.01), a standard‐deviation increase in Medicare outpatient payer mix was associated with a 0.89 percentage point decrease in the probability of financial distress (AME = −0.89, p < 0.05), system affiliation was associated with a 2.06 percentage point increase in the probability of financial distress (AME = 2.06, p < 0.01), and increased logged net patient revenue was associated with decreased probability of financial distress (AME = −3.50, p < 0.001). In terms of market characteristics, a standard‐deviation increase in the percent market population in poverty was associated with a 1.09 percentage point increase in the probability of financial distress (AME = 1.09, p < 0.01). The remaining predictors in our model were not statistically significant (p ≥ 0.05). Calculation of the variance inflation factor for each predictor suggested no issues with multicollinearity.

TABLE 2.

Probit regression model of rural hospital financial distress.

Description	Variable	Average marginal effect	Standard error
Financial performance
Profitability	Percent total margin, year t	−2.01 ***	0.29
	Percent total margin, year t – 1	−0.67 *	0.28
	Percent total margin, year t – 2	−1.64 ***	0.24
Outpatient revenue to total revenue	Percent outpatient to total revenue	−0.49	0.38
Uncompensated care	Uncompensated care as a percentage of operating expenses	0.33	0.28
Benchmark performance	Percent CAHMPAS benchmarks met (2 years)	−3.62 ***	0.40
Government reimbursement
CAH status	CAH indicator variable	−3.16 **	1.08
Hospital Medicare outpatient payer mix	Percent Medicare outpatient payer mix	−0.89 *	0.38
Medicare Advantage plus cost plan days to traditional Medicare days	Percent Medicare Advantage and cost plan days	−0.06	0.56
Medicaid‐to‐Medicare fee index for “All Services”	Fee index expressed as a percent	−0.62	0.35
Medicaid payer mix	Percent Medicaid payer mix	0.02	0.37
Organizational traits
Ownership	For‐profit indicator variable	2.09	1.20
Hospital size	Net patient revenue (log)	−3.50 ***	0.57
System affiliation	System affiliation indicator variable	2.06 **	0.74
Market characteristics
Competition	Miles to nearest hospital with more than 100 beds (log)	−0.45	0.59
	Market share	−1.59	0.84
Market size	Total population (log)	−0.41	0.72
Economic condition	Percent market population in poverty	1.09 **	0.35
Medicare advantage penetration rate	Percent Medicare Advantage penetration	−0.66	0.46
Additional model terms
ϕ NegativeEquity		−11.73 ***	0.93
ϕ Closure		−21.06 ***	0.85

Note. For nonlogged continuous variables, the Average Marginal Effect reflects the change in predicted probability of financial distress associated with a one‐standard‐deviation increase in the predictor variable.

^* p < 0.05.

^** p < 0.01.

^*** p < 0.001.

Abbreviations: CAH, Critical Access Hospital; CAHMPAS, Critical Access Hospital Measurement and Performance Assessment System.

Testing model calibration

The AUC equaled 0.87 within the test set, indicating good predictive ability. ²⁷ In addition, observed rates of financial distress within the test set were similar to those predicted by our model (Figure 1). For instance, among 10,697 hospital‐year observations with a predicted financial distress probability between 0% and 5%, approximately 1.3% of observations actually experienced financial distress after 2 years. Thus, for this set of observations, the model was well calibrated and accurately identified these hospitals as having a low risk of financial distress. Although the correlation between observed rate of financial distress and predicted probability weakens for observations with the highest predicted risk of distress (possibly due to limited sample size within this range), there is strong congruence for most hospital‐year observations.

FIGURE 1.

Observed vs. predicted percentage of rural hospitals in financial distress, test subsample.

Note. The sample size next to each data point reflects the respective number of observations in the test subsample of the “stacked” dataset; larger sample sizes correspond to larger data points in the above graph. The stacked dataset included one observation for each combination of distress outcome and hospital‐year.

We classified 30.55% of the 14,116 hospital‐year observations in our sample as lowest risk of experiencing financial distress over the next 2 years. In comparison, we classified 32.52% of observations as mid‐lowest risk of distress, 26.40% of observations as mid‐highest risk, and 10.52% of observations as highest risk. Consistent with previous calibration tests, observations in groups with greater predicted risk also had greater rates of observed financial distress outcomes (Figure 2). Among test set observations classified as lowest‐risk, 5.78% experienced negative cash flow margin within 2 years, 1.50% experienced negative equity within 2 years, and zero observations experienced closure within 2 years. Within the mid‐lowest‐risk group, the percent of observations that experienced negative cash flow margin increased to 15.78%, the percent of observations that experienced negative equity increased to 5.63%, and the percent of observations that experienced closure increased to 0.27%. The rates of financial distress were even greater among observations in the mid‐highest‐risk group, as 37.94% experienced negative cash flow margin, 15.66% experienced negative equity, and 0.85% experienced closure. Finally, observations in the highest‐risk group experienced the most financial distress outcomes; 61.57% experienced negative cash flow margin, 43.02% experienced negative equity, and 3.33% experienced closure.

FIGURE 2.

Observed rates of financial distress outcomes by FDI risk category, test subsample.

Abbreviation: FDI, Financial Distress Index.

DISCUSSION

Revisiting research goals

At the outset of the research, we established several goals for our model of rural hospital financial distress. The first goal was to base the model in financial theory and previous empirical findings on financial distress. We accomplished this goal by using an accounting perspective to define financial distress in our conceptual framework and by using empirical findings on financial distress from earlier research ³ to identify key predictor variables. The second goal was to use a quantitative approach to assess model calibration and external validity. Figures 1 and 2 demonstrate our fulfillment of this objective with quantitative analysis and an external test sample. The third goal stated the model should produce stable results and refrain from “overfitting” the training data. Our calculation of the variance inflation factor for each predictor variable suggested no stability issues due to multicollinearity. Furthermore, our model demonstrated good predictive performance in the test subsample, suggesting we did not overfit the model using the training data. The fourth goal stated the model should be able to individually identify the risk of financial distress for each rural hospital; Figures 1 and 2 demonstrate our fulfillment of this objective. The fifth goal stated the model should have high face validity. Although completion of this goal is subjective, we believe that model results shown in Table 2 and Figures 1 and 2 are generally consistent with anecdotal reports from rural hospital executives and thus demonstrate sufficient face validity. Hospitals with worse financial metrics had a greater estimated risk of future financial distress (Table 2), and hospitals with a greater estimated risk of future financial distress generally had greater rates of observed financial distress (Figures 1 and 2). The sixth and seventh goals stated the model should be parsimonious and use publicly available data, respectively. Model parsimony is also dependent on subjective interpretation, but we consider the model's set of variables to be manageable, particularly considering their public availability. The one model component that is not publicly available is our hospital market definition (see Appendix), which had to be constructed using Centers for Medicare & Medicaid Services claims data ²⁹ ; however, subsequent updates can take the market as fixed in the short run and perform the full update periodically.

Model predictors

Our research focused on the total predictive ability of our full model rather than statistical inference on individual predictor variables. However, under the caveat that our model is not designed to determine causality, considering the findings for each predictor in the context of our hypotheses could be helpful for future studies of rural hospital financial distress. We hypothesized that increased hospital profitability, increased outpatient to total revenue, increased CAHMPAS benchmark performance, CAH status, increased Medicaid‐to‐Medicare fee index, increased hospital size, and increased market size would be associated with lower probability of financial distress. Each of these predictors was indeed associated with lower probability of financial distress, consistent with our hypotheses; the associations for hospital profitability, CAHMPAS performance, CAH status, and hospital size were also statistically significant (p < 0.05; Table 2). We further hypothesized that increased uncompensated care, increased Medicaid payer mix, hospital for‐profit ownership status, increased market competition, and increased market poverty would be associated with greater probability of financial distress. Although market poverty was the only statistically significant predictor from this group, all predictors from the group exhibited associations with greater probability of financial distress, consistent with our hypotheses.

In addition, we hypothesized that hospitals serving a greater proportion of Medicare Advantage patients (relative to traditional Medicare patients) would be associated with greater probability of financial distress. Counter to our hypothesis, we found that both Medicare Advantage measures had an inverse but nonsignificant association with the probability of financial distress in our training sample. Although previous reports have raised concerns about denied or delayed payments from Medicare Advantage Organizations to hospitals, ¹⁷ , ¹⁸ empirical research published by Henke and colleagues since the start of our study also found that Medicare Advantage penetration was associated with increased financial stability and lower probability of closure among rural hospitals. ³⁰ Given the mixed findings of recent literature on this topic and the continued growth in Medicare Advantage enrollment, ³¹ future studies should consider exploring the relationship between Medicare Advantage penetration and hospital financial distress in more depth. Recent growth in Medicare Advantage enrollment could be particularly important if there is a threshold at which greater proportions of Medicare Advantage patients increase the probability of financial distress.

Lastly, we noted that directions of association for Medicare outpatient payer mix and system affiliation were not obvious ex ante. We found that increased hospital Medicare outpatient payer mix was significantly associated with lower probability of financial distress. Given that over half of the rural hospitals in our sample were CAHs, this finding could be driven by favorable cost‐based reimbursement provided to CAHs by Medicare for facility services. ²⁰ Moreover, we found that system affiliation was significantly associated with increased probability of financial distress. This result appears to conflict with previous research findings that linked health system affiliation to improved rural hospital operating margin, ²¹ but the discrepancy could be explained due to selection bias. Rural hospitals in financial distress could be more likely to seek system affiliation to remain viable, and this “selection” into a health system could obscure the relationship between system affiliation and financial distress among our sample.

Comparison to the 2017 FDI model

The FDI model from this research includes multiple new predictors not available in the original model from 2017. ³ However, several predictors are shared between the early and new versions of the model. Both models found that increased total margin, increased CAHMPAS benchmarks met, hospital CAH status, increased net patient revenue, and decreased market poverty were significantly associated with lower probability of rural hospital financial distress. Additional predictors shared between both models, including Medicaid‐to‐Medicare fee index, hospital for‐profit status, distance to nearest 100‐bed hospital, market share, and market population, demonstrated consistent directions of association with financial distress across both models but differed in terms of statistical significance (i.e., precision of the effect estimates). These differences could be due to inclusion of additional predictors in the new model that exhibit some degree of correlation with the original predictors.

Limitations

The FDI model has several limitations worth noting. First, our sample included general care, nonfederal rural hospitals; model findings may not generalize to federal Indian Health Service hospitals located in rural areas. In addition, the model is not able to capture all relevant financial distress outcomes (e.g., bankruptcy) or predictors, including detailed information on a hospital's number and type of medical staff, changes in local and state‐specific polices that could affect hospital earnings, community efforts to assist financially struggling hospitals, and unexpected changes in hospital management and personnel. Data on many of these variables are not widely available; if such variables were available, our classification of a hospital's financial distress risk could change. Furthermore, the model is trained on data from 2013 to 2019. Given the broad impact of the 2020 COVID‐19 pandemic and related public health emergency funding on rural hospitals, ³² , ³³ the model may not be as accurate in predicting financial distress when using data from the early 2020s. However, given that the emergency funds were temporary and rural hospital profitability may be returning to prepandemic levels, ³⁴ we reason that the FDI model is still broadly applicable to current and future rural hospital data. Lastly, the model is not intended or sufficient to determine causality. Although the model is useful as a predictive tool, additional research is needed to test whether identified predictors have causal effects on hospital financial distress. Omitted variables could confound associations between the model's predictors and outcomes of financial distress.

CONCLUSIONS

Hospital financial distress and closures have been long‐standing concerns within the field of rural health. Given the ongoing challenges and consequences of rural hospital unprofitability, there is a clear need for accurate assessments of financial distress and closure risk. The FDI model can be used by researchers, policymakers, and rural health advocates as a screening tool to identify at‐risk rural hospitals for closer monitoring.

CONFLICT OF INTEREST STATEMENT

This research was supported by the Federal Office of Rural Health Policy (FORHP), Health Resources and Services Administration (HRSA), U.S. Department of Health and Human Services (HHS) [cooperative agreement # U1CRH03714‐10‐00].The information, conclusions, and opinions expressed in this manuscript are those of the authors and no endorsement by FORHP, HRSA, HHS, or The University of North Carolina is intended or should be inferred.The study sponsors had no role in the design or completion of this research project or the decision to submit the research for publication.

Supporting information

Supporting Information

Malone TL, Pink GH, Holmes GM. An updated model of rural hospital financial distress. J Rural Health. 2025;41:e12882. 10.1111/jrh.12882