Optimal Location for Centralization of Hospitals Performing Pancreas Resection in California

Abstract

This study investigates the optimal locations for centralization of hospitals performing pancreaticoduodenectomy in California.

There is an extensive body of research in complex cancer surgery that asserts a volume-outcome association in which patients are safer and have improved survival when their surgery is performed at hospitals and by surgeons with higher volume experience.^1,2 This volume-outcome relationship has been especially true after resection for pancreatic cancer.³ As a consequence, organizations have advocated for minimum-volume standards and, as such, a centralization of complex cancer surgery operations. However, concerns about decreasing access to complex cancer surgery have been raised.⁴ Given this concern, we modeled what an optimal location-allocation market for hospitals performing pancreatic resections would look like, with the goal of maximizing market share while minimizing cannibalization and eliminating low-volume centers.

Methods

The California Office of Statewide Health Planning database, an all-payer claims database with complete capture of all hospital stays, was used to identify patients who underwent pancreaticoduodenectomy (PD) in 2016. California is one of the largest states with a mix of both urban and rural communities, making it an acceptable model for analyzing health care markets and patient travel patterns. Hospitals were identified as high volume (≥20 PD) and low volume (<20 PD). Patients were geocoded at the zip code level, and hospitals were geocoded using their addresses. The Network Analyst Location-Allocation function in ArcGIS Desktop, version 10.6 (ESRI), was used to generate the geospatial model. The model was optimized to maximize attendance with minimum number of facilities, while not exceeding facility capacity, assuming only hospitals that performed PD in 2016 exist. Hospital capacity was set to 200 so as to not exceed realistic capacity at any hospital. This study was approved by the Ohio State University institutional review board and the California Committee for the Protection of Human Subjects, which waived informed patient consent for deidentified data.

Results

One thousand fifty-six patients from California who underwent PD in 2016 in 1 of 104 hospitals were identified; 15 centers (14.4%) were identified as high volume (≥20 PD). In a real-life existing scenario, 912 patients (86.4%) traveled 90 minutes or less to their destination hospital. In the optimal location-allocation model (OM), 972 patients (92%) would need to travel 90 minutes or less to the closest hospital capable of performing PD. Median travel time in the existing scenario was 22.7 minutes (interquartile range [IQR], 11.1-43.9). In the OM, median travel time for 100% coverage was 22.9 minutes (IQR, 13.2-44.3). Importantly, in the existing scenario, 607 patients (57.3%) received care at a high-volume hospital vs 100% in the OM. Median hospital PD in the existing scenario was 26 (IQR, 9-53) compared with 97.5 (IQR, 73-117) in the OM (Table). Mortality was higher at low-volume hospitals (4.0%; 95% CI, 2.2-5.8) compared with high-volume hospitals (1.2%; 95% CI, 0.3-2.0). Maps were produced to provide a visual representation of the PD hospitals selected along with the patients covered with a 90-minute travel time around each hospital (Figure).

Table. Comparison of Patients and Hospitals in California vs Optimal Location-Allocation Model^a.

Patient and Hospital Characteristics	No. (%)
	2016	Optimal Location-Allocation
Patients (n = 1056)
Travel 90 min or less	912 (86.4)	972 (92.0)
PD at HVH	605 (57.3)	1056 (100)
Time to destination, min, median (IQR)	22.7 (11.1-43.9)	22.9 (13.2-44.9)
Hospitals (n = 104)
HVH	22 (21.1)	10 (100)
PD, median (IQR)	26 (9-53)	97.5 (73-117)

Abbreviations: HVH, high-volume hospital; IQR, interquartile range; PD, pancreaticoduodenectomy.

^aLocation-allocation model is set to optimize locations for hospitals performing PD while maximizing attendance, assuming only hospitals that performed PD in 2016 exist (eg, no new hospitals). Each hospital’s capacity was set to 200 so as to not exceed capacity at any hospital.

Figure. Map of Identified Optimal Location Hospitals Performing Pancreaticoduodenectomy (PD) and Patients Who Underwent PD in 2016.

Location-allocation model is set to optimize locations for hospitals performing PD while maximizing attendance, assuming only hospitals that performed PD in 2016 exist (eg, no new hospitals). Each hospital’s capacity was set to 200 so as to not exceed capacity at any hospital. The H symbol represents identified optimal hospital locations, gray diamonds represent all patients who underwent PD in 2016, and gray lines from diamond to H symbol indicate patients who would need to travel 90 minutes or less to nearest identified optimal hospital.

Discussion

Although opponents of centralization for complex surgical procedures raise concerns of limiting access to care, data derived from simulation models of optimal location-allocation demonstrated that patients in California would have no increased travel burden to undergoing PD at a high-volume center. Furthermore, as a consequence of centralizing care at optimal locations, every patient in California would be assured that they undergo PD at a high-volume center. Additionally, because the location of centers performing PD would be optimized so as not to cannibalize from each other, the median number PD would increase from 26 to nearly 100. This increase in PD performed at higher-volume centers could result in improved outcomes while reducing complications and mortality.^1,2,3 Additionally, spillover effects may be achieved through standardization and limiting variation, leading to reduced spending and increased value. A mechanism for achieving a scenario similar to the one modeled in this study would be through a certificate of need.⁵ Alternatively, as a condition of reimbursement, payors may require a minimum annual hospital volume, similarly to previous Centers for Medicare & Medicaid Services models.⁶ As hospital consolidation continues to increase, a third approach could have systems take advantage of their growing networks by shifting operations from low-volume hospitals to nearby high-volume hospitals in an effort to consolidate operations improve outcomes and value. Ultimately, the best approach to assuring the highest quality of care for patients undergoing PD may be a multifaceted one, with special consideration for unintended consequences, such as creating barriers based on payment type, protecting against increased cost to payors and patients, and creating mechanisms to assure access with no added burden for the rare patient who may live an extraordinary distance from a high-volume hospital.

Several limitations should be considered when interpreting the results. While the California database allowed for 100% capture, with complete evaluation of travel for all patients receiving surgery at California-licensed facilities, the data were limited to 1 state and therefore may not be generalizable to other geographically distinct states. While possible reasons for traveling could include personal preferences, level of education, financial constraints for both medical and nonmedical expenses, or referral practices of diagnosing clinicians, we were not able to define specific reasons why patients chose the hospital they did for their operation. Finally, for patient privacy, patient’s location was geocoded to the zip code; therefore, there may be small variations on the true travel times.