Cancer Surgery Outcomes after Implementation of a Global Budget Model in Rural Hospitals

Introduction

Rural hospitals face persistent and growing financial pressures that threaten their ability to sustain essential services. Since 2010, over 150 rural hospitals have closed in the United States.¹ Compared with urban facilities, rural hospitals operate with smaller patient volumes, more unfavorable payer mixes, and lower operating margins.^2-5 These structural pressures heighten rural hospitals’ vulnerability to revenue fluctuations and are associated with reduced care capacity and lower quality of care.^3,6,7

These vulnerabilities have important implications for rural residents with cancer. Surgery is part of recommended treatment for many of the most prevalent localized solid tumor cancers.⁸ Surgical oncology requires timely diagnosis, preoperative evaluation, and coordination across multiple specialties. Resources necessary to support access to high-quality surgical care may be disrupted when local hospitals face financial instability or closure.^9-11 While complex cancer surgeries are often centralized at high-volume facilities, many surgeries still take place in community settings, and the stability of rural hospitals matters for all patients given their role in facilitating referrals and participating in management of perioperative care.^11,12 Rural hospital financial health remains closely tied to access to timely, high-quality surgical care and related outcomes.^14-16

To address concerns about the financial stability of rural hospitals, Pennsylvania implemented the Pennsylvania Rural Health Model (PARHM) from 2019 to 2024, a global budget initiative administered through the Center for Medicare & Medicaid Innovation (CMMI).¹⁷ Under PARHM, hospitals entered global budget contracts with individual payers that guaranteed each hospital a fixed revenue stream that was not dependent on volume. PARHM represented one of the first U.S. efforts to apply a global budget model specifically to rural hospitals.

Global budget models are designed to stabilize hospitals and promote financially sustainable care delivery, but evidence regarding their impact on quality of complex surgical care is limited.¹⁸ Surgical services are among the largest contributors to hospital revenue under fee-for-service payment, creating incentives to retain volume locally. By decoupling revenue from volume, global budgets may facilitate referral to higher-volume centers while preserving local access to cancer care through more financially stable rural hospitals, supporting timely diagnosis, care coordination, and postoperative follow-up. These potential mechanisms are particularly consequential for cancer surgery, where outcomes are associated with surgical volume and referral to specialized centers, as well as timeliness and quality of perioperative care. Whether global budgets improve or worsen quality of care remains unclear.¹⁶

To date, no studies have directly examined the impact of global budgets on surgical oncology outcomes in rural contexts. This study addresses that knowledge gap by evaluating how implementation of PARHM influences the quality of cancer surgery for patients residing in rural HSAs. Leveraging the staggered implementation of rural hospitals into this alternative payment model and linked cancer registry and inpatient discharge data, we assess whether global budgets were associated with changes in 30-day and 90-day mortality and readmissions among surgical oncology patients.

Methods

Policy context and study design

The Pennsylvania Rural Health Model (PARHM) was a voluntary global budget payment program developed by the Commonwealth of Pennsylvania and the Center for Medicare & Medicaid Innovation (CMMI).¹⁷ Participating hospitals received a fixed annual revenue from each participating payer, paid in regular installments, regardless of patient volume. This budget was calculated using a hospital’s historical net patient revenue, with adjustments for inflation, demographic changes, service line shifts, and patient transfers. Medicare and Medicaid, including managed care plans for Medicare or Medicaid beneficiaries, were mandatory participants, while commercial payer participation was voluntary. Participating hospitals were also required to submit and implement multi-year “Hospital Transformation Plans” (HTPs) that address community health needs, including care coordination, enhance preventive care, and address social determinants of health.¹⁷

Pennsylvania identified 66 critical access and acute care hospitals as eligible for PARHM participation, of which 62 remained open and PARHM-eligible through the study period. Eligibility was based on hospital location in counties classified as rural under the Center for Rural Pennsylvania’s definition. These hospitals were located in 60 Hospital Service Areas (HSAs), which are geographic areas defined by the Dartmouth Atlas of Health Care as regions where most residents receive hospital care locally. The first cohort of 5 hospitals entered PARHM in January 2019, followed by additional cohorts in 2020 (8 hospitals) and 2021 (5 hospitals). Each participating hospital was the sole hospital in their HSA. In addition, 44 hospitals that were eligible to join PARHM, which served 42 distinct HSAs, did not participate in PARHM.

We leverage these features of PARHM’s implementation in a stacked difference-in-differences (DiD) design. We took a population perspective in examining whether PARHM was associated with changes in outcomes among cancer patients undergoing surgery for rural patients living in areas served by PARHM-participating hospitals, for whom global budgets were most likely to affect provision of surgical cancer care. Specifically, we compared changes in outcomes among patients in HSAs with PARHM-participating hospitals (before and after program implementation), to changes during the same period among residents of HSAs whose local hospitals were eligible for but did not join PARHM.

Data

This study used linked administrative and registry data from the Pennsylvania Health Care Cost Containment Council (PHC4) Inpatient Discharge Database and the Pennsylvania Cancer Registry (PCR), respectively, covering 2016 through 2023.^19,20 PHC4 data provide visit-level inpatient discharge records from all licensed Pennsylvania hospitals. The PCR contains detailed information on cancer diagnoses, tumor characteristics, diagnosis dates, and surgical treatment dates derived from medical records.

We identified patients newly diagnosed with cancer who underwent inpatient surgical treatment in Pennsylvania hospitals. Our focus is on those who received cancer-directed surgery for one of 11 common cancer types: brain, breast, bladder, colon, esophageal, liver, lung, pancreas, prostate, rectal, and stomach. These cases are identified based on tumor site recorded in the PCR. The linked PHC4/PCR data include patient-level covariates; age, sex, race, ethnicity, cancer type, comorbidities (measured by the Elixhauser Index), and primary insurance type.

We supplemented these data with additional area-level socioeconomic measures, including the Area Deprivation Index (ADI) and variables from the American Community Survey (ACS), as well as county-level physician supply and hospital bed measures from the Area Health Resources File (see Supplemental Digital Content (SDC) Table S1).^21-23 National ADI rankings were assigned at the census tract level and categorized into Pennsylvania-specific quartiles to capture relative socioeconomic disadvantage within the state.

Analytic Sample

Our analytic sample consisted of patients diagnosed with one of the 11 specified cancers who underwent inpatient surgical treatment in a Pennsylvania hospital between 2016 and 2023. We limited the sample to patients who, at the time of their surgery, resided in a Hospital Service Area (HSA) containing at least one PARHM-eligible hospital; of these HSAs, 18 had hospitals that participated in PARHM. We retained only each patient’s first lifetime cancer diagnosis and excluded individuals with a second cancer diagnosis within a year, missing diagnosis or surgery dates, those diagnosed at death, and patients with overlapping inpatient stays.

Outcomes

We examined the following outcomes: 30-day mortality, 90-day mortality, 30-day readmission, 90-day readmission, and two composite indicators that capture the occurrence of either mortality or readmission within 30 or 90 days of the surgical date.

Treatment and comparison groups

We used an intention-to-treat (ITT) approach that compared patients residing in HSAs with at least one hospital that joined PARHM during the study period (treatment group) versus HSAs with PARHM-eligible but non-participating hospitals (comparison group). PARHM was designed to influence healthcare delivery within rural service areas and may affect referral patterns, care coordination, and pre- and post-operative care received locally. Thus, treatment assignment was based on patient residence rather than the hospital where surgery occurred. Attributing patients to hospitals based on HSA rather than location of surgical care is important because PARHM could have influenced where patients received surgical care, which is one of the pathways through which PARHM could have affected outcomes. Accordingly, our ITT approach estimates the effect of PARHM implementation on rural residents undergoing cancer surgery who reside in PARHM-participating HSAs, rather than the effect of receiving surgical care at a PARHM-participating hospital.

Weighting

Participation in PARHM was voluntary, and recent research highlights that participation decisions were shaped by hospitals’ financial and strategic considerations.^24,25 To account for differences between participating and eligible but non-participating hospitals and their local markets, we constructed propensity score (PS) weights. Weights were constructed using a penalized logistic regression model that included the characteristics of hospitals and local communities (see SDC Appendix 1). Final weights were calculated at the HSA level and applied to all regression analyses.

Statistical analysis

We used a stacked DiD design with the surgical case as the unit of analysis. Patients were attributed to HSAs based on where they lived at the time of their surgery for the ITT analysis. Each HSA was classified as treated in the year that a participating hospital within the HSA began PARHM participation, defined as event year 0. We aligned HSAs by event time and constructed stacked datasets for cohort-specific comparisons; each stack contained patients in the cohort of treated HSAs and patients in all comparison HSAs. Treated hospitals entered PARHM in one of three cohorts: 2019, 2020, or 2021. Each cohort of adopters was analyzed as a separate sub-experiment (“stack”), and the cohort of comparison HSAs could contribute to each of the three stacks. These stacks were pooled to estimate the average treatment effect on the treated (ATT) across all treated cohorts. This approach mitigated bias arising from differential timing and non-parallel trends across cohorts. Pooled difference-in-differences estimates characterize differential changes in outcomes from the pre- to post-intervention periods between treatment and comparison HSAs.

We estimated separate models for each outcome. Each model adjusted for the following patient-level covariates: age, sex, race, ethnicity, cancer type, cancer stage, and comorbidity burden (Elixhauser index) and area-level covariates: median household income, educational attainment, and ADI ranking. Race was conceptualized as a social rather than biological construct and included as a covariate to account for structural and healthcare access factors that may influence cancer treatment and outcomes.^26-28 Models included stack-by-year fixed effects to account for time trends specific to each stack and stack-by-HSA fixed effects to account for repeated use of comparison HSAs across stacks.²⁹ All estimates were weighted using the participation propensity weights; SDC Table S2 & Figure S1 show that weighting substantially reduced standardized mean differences across most covariates. Standard errors were clustered at the HSA-level.³⁰

The validity of the DiD design relies on the parallel trends assumption. To assess this assumption, we estimated event study models that include a series of event-time dummies for the years before and after PARHM implementation. This enabled assessment of pre-treatment trends and visualization of the dynamic effects of the intervention over time. To strengthen the robustness of our findings, we applied the Honest DiD estimator developed by Rambachan and Roth (2023),³¹ which provides confidence bounds for DiD estimates that account for different possible violations of the parallel trends assumption. SDC Appendix 2 provides additional methodological detail.

Results

Study Population

After exclusions, the final analytic sample included 17,048 patients. Figure 1 displays the inclusion/exclusion process. Patients attributed to PARHM and non-PARHM HSAs had similar baseline characteristics (Table 1). Baseline outcome measures were nearly identical across groups. Age, race, and insurance distributions were also comparable, although patients residing in an HSA with a PARHM participating hospital were slightly more likely to be female (42.2% vs 39.0%) and Medicaid-enrolled (10.3% vs 9.1%). At the area level, treatment HSAs had somewhat higher social disadvantage (mean ADI percentile 74.0 vs 69.9) and lower physician supply (156 vs 204 physicians per 100,000 residents).

Figure 1: CONSORT Diagram Sample Inclusions and Exclusions.

Flow of patients from the Pennsylvania Health Care Cost Containment Council (PHC4) inpatient discharge database linked to the Pennsylvania Cancer Registry (PCR), 2016–2023. The diagram shows application of eligibility criteria, exclusions for missing or inconsistent data, and final analytic sample of 17,048 cancer surgery admissions among residents of hospital service areas with a Pennsylvania Rural Health Model (PARHM)–eligible hospital.

Table 1. Baseline Characteristics of Patients Undergoing Cancer Surgery in Hospital Service Areas With and Without PARHM Participation.

Variable	Category	Non-PARHM	PARHM (Pooled)	Cohort 1 (2019)	Cohort 2 (2020)	Cohort 3 (2021)
Number of Surgical Stays, N		13,511	3,537	421	1,951	1,165
Age	Under 55, N (%)	2,482 (18.37)	671 (18.97)	82 (19.48)	366 (18.56)	223 (19.14)
	55–64, N (%)	3,834 (28.38)	1,010 (28.56)	124 (29.45)	571 (29.27)	315 (27.04)
	65–74, N (%)	4,347 (32.17)	1,117 (31.58)	137 (32.54)	599 (30.70)	381 (32.70)
	75–84, N (%)	1,899 (14.06)	492 (13.91)	55 (13.06)	269 (13.79)	168 (14.42)
	85+, N (%)	655 (4.85)	160 (4.52)	11 (2.16)	97 (4.97)	52 (4.46)
Race/Ethnicity	White, N (%)	13,037 (96.49)	3,428 (96.92)	414 (98.34)	1,883 (96.51)	1,131 (97.08)
	Black, N (%)	325 (2.41)	72 (2.04)	4 (0.95)	50 (2.56)	16 (1.37)
	Other Race, N (%)	149 (1.10)	37 (1.05)	3 (0.71)	18 (0.92)	18 (1.55)
	Hispanic, N (%)	226 (1.67)	133 (3.76)	3 (0.71)	107 (5.43)	23 (1.97)
Female, N (%)		5,270 (39.01)	1,492 (42.18)	150 (35.63)	822 (42.13)	520 (44.64)
Insurance Type	Medicare, N (%)	7,081 (52.41)	1,845 (52.16)	205 (48.69)	1,019 (52.23)	621 (53.30)
	Medicaid, N (%)	1,234 (9.13)	366 (10.35)	48 (11.40)	198 (10.15)	120 (10.30)
	Private Insurance, N (%)	4846 (35.87)	1,261 (35.65)	152 (36.10)	705 (36.14)	404 (34.68)
	Uninsured, N (%)	99 (0.73)	22 (0.62)	3 (0.71)	12 (0.62)	7 (0.60)
	Other Insurance, N (%)	251 (1.86)	43 (1.22)	13 (3.09)	17 (0.87)	13 (1.12)
Unweighted Comorbidity Index, Mean (SD)b		2.855 (1.846)	2.748 (1.797)	2.703 (1.74)	2.671 (1.806)	2.894 (1.792)
Hospital Beds per County, Mean (SD)		307.8 (251.9)	308.5 (469.5)	128.2 (139.6)	411.6 (505.6)	201.1 (70.4)
Physician Supply, Mean (SD)c		204.3 (357.4)	155.8 (72.0)	131.3 (60.8)	175.9 (84.9)	131.0 (28.7)
ADI Percentile, Mean (SD)d		69.9 (16.9)	74.0 (13.6)	66.4 (16.6)	73.4 (14.3)	77.7 (9.0)
Income per Capita, Mean (SD)		28,615 (7,169)	28,595 (6,390)	28,012 (6,631)	29,431 (6,772)	27,405 (5,351)
Share Below Poverty, Mean (SD)		0.123 (0.079)	0.122 (0.070)	0.109 (0.049)	0.119 (0.069)	0.133 (0.078)

^aComposite outcomes include death or readmission within the specified time frame.

^bComorbidity index refers to the Elixhauser comorbidity score.

^cPhysician supply is expressed as physicians per 100 000 county residents.

^dADI indicates Area Deprivation Inde

Overall

Table 2 reports difference-in-differences estimates for changes in the proportion of admissions followed by 30-day mortality or readmission. Estimates are on the 0-1 scale. For example, in treatment vs. comparison HSAs, the differential change in the proportion of admissions followed by 30-day mortality was 0.010, or 1 percentage point, which was not statistically significant (95% CI: −0.014 to 0.034, P=0.391). Hospital adoption of PARHM also was not associated with statistically significant changes in readmission or composite outcomes.

Table 2. Difference-in-Differences Estimates of the Association Between PARHM Implementation and Postoperative Outcomes.

Outcomec	Estimated effect of PARHMb (95% CI)	Non-PARHM Pre-Treatment Mean (SD)	p-value
30-Day Mortality	0.010 (−0.014 to 0.034)	0.041 (0.198)	0.391
30-Day Readmission	−0.007 (−0.045 to 0.030)	0.159 (0.366)	0.695
30-Day Composite a	0.003 (−0.038 to 0.044)	0.188 (0.390)	0.875
90-Day Mortality	0.011 (−0.018 to 0.039)	0.073 (0.260)	0.460
90-Day Readmission	−0.016 (−0.058 to 0.027)	0.239 (0.427)	0.465
90-Day Composite a	−0.002 (−0.044 to 0.039)	0.276 (0.447)	0.906

^aComposite outcomes include death or readmission.

^bModels adjusted for patient- and area-level covariates and included hospital service area and year fixed effects. Propensity score weights were applied, and standard errors were clustered at the hospital service area level.

^cOutcomes are expressed as the proportion (0–1) of surgical admissions experiencing each event. Estimated effects represent the differential change in this proportion associated with PARHM implementation.

By Cohort

Results were consistent across the different adoption cohorts, with no significant changes observed in 30- or 90-day mortality, readmission, or composite outcomes (Figure 2). Point estimates for the 2019 cohort were larger, particularly for the 30-day composite outcome, but confidence intervals were wide and included zero. Estimates for the 2020 and 2021 cohorts were smaller in magnitude and frequently negative, though also imprecise.

Figure 2. Results of Difference-in-Differences Models by PARHM Cohort.

Panels A–F show adjusted difference-in-differences estimates of the association between Pennsylvania Rural Health Model (PARHM) implementation and 30-day and 90-day mortality, readmission, and composite outcomes, stratified by hospital entry cohort (2019, 2020, 2021). Estimates are adjusted for patient- and area-level covariates and weighted using propensity scores. Error bars indicate 95% CIs. No statistically significant differences were observed across cohorts for any outcome.

By Cancer Type

Across cancer types, PARHM participation was also generally not associated with statistically significant changes in 30- or 90-day mortality, readmission, or composite outcomes (Figure 3). Among patients undergoing bladder cancer surgery, PARHM was associated with a differential reduction in the proportion of admissions followed by 30-day mortality (β = −0.037; 95% CI, −0.064 to −0.010). Among patients undergoing stomach cancer surgery, PARHM was associated with a reduction in the proportion with 30-day readmissions (β = −0.313; 95% CI, −0.622 to −0.004), and among patients receiving liver cancer surgeries, there was a reduction in the proportion with 90-day readmissions (β = −0.213; 95% CI, −0.425 to −0.001). In contrast, for patients undergoing colon cancer surgery, PARHM was associated with an increase in the 90-day composite outcome (β = 0.039; 95% CI, 0.003 to 0.074). However, these isolated findings did not form a consistent pattern across the eleven cancer types.

Figure 3. Results of Difference-in-Differences Models by Cancer Type.

Panels A–F show adjusted difference-in-differences estimates of the association between Pennsylvania Rural Health Model (PARHM) implementation and 30-day and 90-day mortality, readmission, and composite outcomes, stratified by cancer site. Models were adjusted for patient- and area-level covariates and weighted using propensity scores. Error bars indicate 95% CIs. No consistent patterns were observed across cancer types.

Event-Study and Sensitivity Analyses

Event study models showed no evidence of differential pre-trends, supporting the parallel trends assumption (SDC Figure S2). Post-implementation estimates were imprecise, and confidence intervals overlapped the null across all time periods. Event time periods are not weighted by the number of surgeries, meaning later post-treatment periods (e.g., the fourth period after implementation) reflect data only from the earliest and smallest (2019) adoption cohort, leading to wider confidence intervals. Sensitivity analyses using HonestDiD indicated that results were robust to potential deviations from the parallel trends assumption (SDC Figure S3).

Discussion

This difference-in-differences study of patients undergoing cancer surgery found that implementation of a PARHM, a global budget model for rural hospitals, was not associated with measurable changes in surgical oncology outcomes. Across 30-day and 90-day mortality, readmission, and composite measures, effect estimates were statistically indistinguishable from zero, and consistently null across hospital cohorts and most cancer types. Although a few cancer type associations reached statistical significance, these findings did not follow a consistent pattern and should be interpreted cautiously given smaller subgroup sample sizes (SDC Table S3) and the rarity of some outcomes.

These results add to an accumulating body of research on PARHM, which found that the program was associated with limited changes in hospital utilization. For example, Bourne et al. (2024) reported no reductions in potentially avoidable hospital utilization following adoption, and Bourne et al. (2025) found no differential change in the likelihood of patients bypassing nearby rural hospital for elective surgeries.^32,33 In complementary work, Sabik et al. (2025) found that PARHM was not associated with changes in the location of surgical oncology care in Commission on Cancer accredited, National Cancer Institute designated, or high-volume hospitals for the 11 cancers included in this study.³⁴ This suggests that greater referral to cancer-focused facilities, a potential mechanism for improved surgical outcomes, was not consistently realized under PARHM. Altogether, this evidence suggests that PARHM may not have substantially altered the referral pathways through which global budgets were hypothesized to influence cancer surgery outcomes.

These results may reflect important characteristics of the scope and implementation of PARHM. PARHM was primarily designed to preserve rural access to care and hospital financial viability rather than directly target surgical oncology care. Few, if any, PARHM-participating hospitals developed transformation plans specifically focused on surgical care.³⁵ Additionally, PARHM was designed as a time-limited intervention: at the program’s inception, the initial demonstration period was planned for 5 years. The program’s time-limited nature might have dampened hospitals’ incentives to pursue long-term structural changes aligned with global budget incentives. Finally, early implementation periods coincided with the COVID-19 pandemic, which disrupted elective surgical volumes, strained staffing, and altered referral patterns.³⁶ These factors may have constrained the model’s ability to influence surgical outcomes within the implementation period or the study window.

In this context, the maintenance of surgical oncology outcomes may be reassuring, suggesting no measurable deterioration in short-term surgical outcomes during a period of financial and operational stress. However, the mixed evidence of financial stabilization reported by Chatterjee et al (2025) underscores that participating hospitals’ financial stability remained uncertain.³⁷ PARHM concluded at the end of 2024, shortly before substantial Medicaid changes that will shape the future of rural hospital financing. Research on Maryland's hospital global budget program – the other notable effort to adopt hospital global budgets – found some changes in surgical care centralization and outcomes, but mixed evidence for cancer surgeries.^38,39 Furthermore, Maryland's model was not focused on rural hospitals.^40,41 Over the longer term, sustained investment in rural hospital infrastructure, care coordination, and quality improvement will be necessary to translate financial reforms into measurable gains in clinical outcomes.

Limitations

This study has several important limitations. First, participation in PARHM was voluntary, raising the possibility of selection bias. Although we applied propensity score weights to balance observable differences between treatment and comparison areas, unmeasured factors may still differ systematically between PARHM and non-PARHM hospital service areas, potentially biasing our estimates. Second, our analysis focuses on short-term mortality and readmission. These indicators of surgical quality may not capture other important dimensions of care quality. In addition, our statistical power to detect modest changes in rare outcomes, such as post-operative mortality, may be limited, particularly in cancer-specific subgroup analyses and for less common cancer type. Additionally, many patients received definitive surgical care outside PARHM-participating hospitals, which may have reduced direct exposure to the intervention and attenuated measurable effects. Fourth, the study period spans the COVID-19 pandemic, which disrupted care delivery in multiple ways. Although our models include year fixed effects to adjust for secular trends, residual confounders from pandemic-related shocks may still affect our estimates. Finally, the generalizability of our findings is constrained by the unique design and implementation of PARHM within a single state. Results may not translate to other rural contexts or to global budget models that differ in design, scope, or enforcement mechanisms.

Conclusion

This study found that adoption of a global budget program by rural Pennsylvania hospitals was not associated with changes in surgical oncology outcomes for rural patients. The absence of measurable changes should be taken in context of this global budget program’s primary objective of stabilizing rural hospitals, rather than targeting surgical care improvements. As policymakers consider the future of global budgeting in rural hospitals, incentives for long-term investments in surgical quality improvement may be necessary to convert financial stability into better patient outcomes.