Frequencies and costs for 15 090 patient encounters were classified according to this cost-effectiveness model. Hospitals were first stratified based on safety-net burden group (see the Safety-Net Burden subsection of the Methods section). Patients were then categorized based on their severity of illness, mortality, and complications or comorbidities according to their pertinent diagnosis related group (DRG) code as well as any readmission.
Bars represent the reduction in costs at high-burden hospitals (HBHs) as the rates of complications or comorbidities (diagnosis related group codes 405 and 406) (A) and mortality (B) are reduced to half their current rates (dotted dark gray line). Costs after performance improvements are shown below the bars. Costs for low-burden hospitals (LBHs) and medium-burden hospitals (MBHs) are shown by dotted blue lines.
Dotted lines indicate the starting costs at hospital groups; bars, the change in costs with redistribution. A, A simulated scenario is shown in which patients from high-burden hospitals (HBHs) are shifted to low-burden hospitals (LBHs). B, A simulated scenario is shown in which patients from HBHs are shifted equally to LBHs and medium-burden hospitals (MBHs). Redistribution was done according to severity of illness rates in the HBH group. Costs before and after redistribution are shown.
Shown are high-burden hospital (HBH) savings (A) and overall cost savings (B) when hospital performance metrics are reduced compared with patient redistribution. LBHs indicates low-burden hospitals; MBHs, medium-burden hospitals.
eTable. Costs of Patient Characteristics and Outcomes per Safety-Net Hospital Group
Customize your JAMA Network experience by selecting one or more topics from the list below.
Go DE, Abbott DE, Wima K, et al. Addressing the High Costs of Pancreaticoduodenectomy at Safety-Net Hospitals. JAMA Surg. 2016;151(10):908–914. doi:10.1001/jamasurg.2016.1776
What is the best way to reduce costs of complex surgery at safety-net hospitals?
Using a decision analytic model, this study found that reducing complications would have a negligible effect but that redistributing patients away from safety-net hospitals for complex surgery may have the greatest effect on cost reduction.
Future research and policy should focus on improving processes of care at safety-net hospitals and potentially reallocating complex patients to more cost-effective hospitals.
Safety-net hospitals care for vulnerable patients, providing complex surgery at increased costs. These hospitals are at risk due to changing health care reimbursement policies and demand for better value in surgical care.
To model different techniques for reducing the cost of complex surgery performed at safety-net hospitals.
Design, Setting, and Participants
Hospitals performing pancreaticoduodenectomy (PD) were queried from the University HealthSystem Consortium database (January 1, 2009, to December 31, 2013) and grouped according to safety-net burden. A decision analytic model was constructed and populated with clinical and cost data. Sensitivity analyses were then conducted to determine how changes in the management or redistribution of patients between hospital groups affected cost.
Main Outcomes and Measures
Overall cost per patient after PD.
During the 5 years of the study, 15 090 patients underwent PD. Among safety-net hospitals, low-burden hospitals (LBHs), medium-burden hospitals (MBHs), and high-burden hospitals (HBHs) treated 4220 (28.0%), 9505 (63.0%), and 1365 (9.0%) patients, respectively. High-burden hospitals had higher rates of complications or comorbidities and more patients with increased severity of illness. Perioperative mortality was twice as high at HBHs (3.7%) than at LBHs (1.6%) and MBHs (1.7%) (P < .001). In the base case, when all clinical and cost data were considered, PD at HBHs cost $35 303 per patient, 30.1% and 36.2% higher than at MBHs ($27 130) and LBHs ($25 916), respectively. Reducing perioperative complications or comorbidities by 50% resulted in a cost reduction of up to $4607 for HBH patients, while reducing mortality rates had a negligible effect. However, redistribution of HBH patients to LBHs and MBHs resulted in significantly more cost savings of $9155 per HBH patient, or $699 per patient overall.
Conclusions and Relevance
Safety-net hospitals performing PD have inferior outcomes and higher costs, and improving perioperative outcomes may have a nominal effect on reducing these costs. Redirecting patients away from safety-net hospitals for complex surgery may represent the best option for reducing costs, but the implementation of such a policy will undoubtedly meet significant challenges.
Our group has recently demonstrated that safety-net hospitals perform complex surgery that has not only inferior outcomes but also significantly higher costs than hospitals with a lower safety-net burden.1 Notably, this cost difference was not affected by adjusting for patient factors, which implies that there may be intrinsic characteristics of safety-net hospitals that contribute to increased surgical costs. Safety-net hospitals are defined either as having a stated mission to maintain an “open door” to all patients, regardless of the ability to pay, or having a substantial share of uninsured, Medicaid, and otherwise vulnerable patients.2 These hospitals always have been crucial to their communities and thus subsidized by the government for providing high levels of underreimbursed or nonreimbursed care.3
However, safety-net hospitals are disproportionately threatened by federal cuts to these subsidies3,4 as well as cost-reduction strategies being initiated by payers.5,6 It has previously been shown that hospital payer mix7 and financial health8-11 both affect patient outcomes, which poses a real threat to the sustainability of these institutions. If hospital reimbursement is increasingly linked to value-based purchasing and safety-net hospitals are at an inherent disadvantage due to payer mix and hospital resources, how can these hospitals survive in an increasingly margin-focused health care industry?
In this study, we aimed to understand how the variation in surgical outcomes after pancreaticoduodenectomy (PD) affects costs based on hospital safety-net burden. Pancreaticoduodenectomy was chosen as a model of complex, high-risk surgery with ample opportunity for value improvement. We hypothesized that improvements in hospital performance would reduce costs. However, due to the significant disparity in outcomes at safety-net hospitals,1 we anticipated that centralizing care will likely have the largest effect on reducing overall cost.
The University HealthSystem Consortium (UHC) is an alliance of 95% of the nation’s major not-for-profit academic medical centers, including 118 centers and 298 of their affiliated hospitals. The Clinical Database/Resource Manager is an administrative data set that contains patient demographic; financial; International Classification of Diseases (Ninth Revision); procedure; and short-term outcome data provided by member medical centers.1,12
We created a data set of patients undergoing PD between January 1, 2009, and December 31, 2013, based on the International Classification of Diseases, Ninth Revision code 52.7 (n = 15 802). The following patient encounter information was collected from the UHC: insurance type (private, government, or other), in-hospital mortality, total direct cost, discharge disposition (home, rehabilitation, or other), and 30-day readmission. Patient severity of illness (SOI) in the UHC (moderate [SOI 2], major [SOI 3], and extreme [SOI 4]) is derived from a proprietary formula, which is based on all payer–refined diagnosis related groups (DRGs), and has been validated in a nationwide data set that includes 8.5 million discharges from more than 1000 hospitals.13 Medicare severity DRGs (MS-DRGs) were then used to stratify based on patient risk, with MS-DRG codes 405, 406, and 407 corresponding to major, minor, or no complications or comorbidities. The University of Cincinnati’s Institutional Review Board approved this study and did not require informed consent because the data were blinded across multiple institutions.
For this study, primary outcomes of interest were postoperative mortality, 30-day readmissions, and cost of surgical care. Total direct costs are calculated from hospital charges using institution-specific Medicare cost-to-charge ratios and then adjusted with federally reported area wage indexes to normalize regional variation in labor cost.14,15 Therefore, for the purposes of these analyses, “cost” is reflective of financial consequences from a societal or payer perspective.
To study the effect of safety-net burden on surgical outcomes, hospitals were stratified as previously described.16,17 All patient encounters (n = 16 121 161) from 2009 to 2013 were included in the analysis. Hospitals (n = 262) were assigned a safety-net burden, defined as the proportion of all patient charges during the study period that were Medicaid covered or uninsured. Hospitals were then stratified into the following 3 groups of safety-net burden16: low-burden hospitals (LBHs) were in the lowest quartile (0.0%-14.3%), medium-burden hospitals (MBHs) were in the middle 2 quartiles (14.4%-36.8%), and high-burden hospitals (HBHs) were in the highest quartile (36.9%-100.0%).
Based on clinical outcomes and cost data from the UHC, a cost-effectiveness model (TreeAge Pro 2015, release 1; TreeAge Software, Inc) was constructed (Figure 1). First, hospitals were categorized according to safety-net burden. For each hospital group (LBHs, MBHs, and HBHs), patients were categorized according to their SOI (moderate, major, or extreme), perioperative mortality, complications or comorbidities based on MS-DRG codes, and readmission. Costs from the index hospitalization and readmission hospitalization (when present) were included in the model.
Sensitivity analyses were performed to simulate scenarios of improved hospital performance based on the rate of complications and comorbidities (MS-DRG codes 405 and 406) and mortality. These rates were decreased for each SOI group to 50% below the base case. These values were chosen to represent extremes of performance improvement or patient comorbidity reduction. Further sensitivity analyses simulated redistribution of care by shifting patients between safety-net hospital groups. High-burden hospital patients were centralized to LBHs in one scenario and equally to LBHs and MBHs in another. Severity of illness was preserved during redistribution such that patient risk would not be artificially altered by the severity distribution at the new hospital group.
Statistical analysis was performed using software programs (SAS, version 9.4; SAS Institute Inc and SigmaPlot; Systat). An α level of .05 was used for all significance tests.
Overall, HBHs had the lowest patient volume (9.0% at HBHs vs 28.0% at LBHs vs 63.0% at MBHs), but they had the greatest proportion of patients with extreme SOI (18.8% at HBHs vs 11.4% at LBHs vs 12.8% at MBHs). These hospitals also had the highest rates of perioperative mortality (3.7% at HBHs vs 1.6% at LBHs vs 1.7% at MBHs), any complication or comorbidity (85.6% at HBHs vs 84.0% at LBHs vs 84.2% at MBHs), major complication or comorbidity (42.0% at HBHs vs 33.6% at LBHs vs 34.2% at MBHs), and readmission (23.5% at HBHs vs 17.8% at LBHs vs 18.8% at MBHs). The total counts and frequencies are listed in greater detail in the Table.
As demonstrated previously, surgical costs also increase with hospital safety-net burden. Pancreaticoduodenectomies cost $35 303 per patient at HBHs compared with $25 916 at LBHs and $27 130 at MBHs (detailed breakdown costs are listed in the eTable in the Supplement). For every patient group except mortality in extreme SOI (SOI 4) cases, HBHs had the highest costs. However, mortality was a rare event, and its associated cost was similar across hospital groups ($62 994 at HBHs vs $64 732 at LBHs vs $63 553 at MBHs).
One-way sensitivity analyses were performed to simulate a scenario in which HBHs improve their performance in response to financial penalties. First, rates of complications or comorbidities (MS-DRG codes 405 and 406) were reduced to half their current values and evaluated in terms of the effect on cost (Figure 2A). This situation could represent a hypothetical scenario in which hospitals improve preoperative care and reduce comorbidities or improve postoperative care and reduce complications. The greatest reduction in cost resulted from a 50% reduction in complications or comorbidities in extreme severity cases. Costs of HBHs dropped from $35 303 to $30 696 per patient, a reduction of $4607.
A second scenario of improving hospital clinical outcomes was modeled by decreasing the mortality rate at HBHs (Figure 2B). In this situation, there was a negligible effect on cost, where the greatest cost savings (resulting from extreme severity cases) were $63 per patient. In both of these cases, the greatest cost reduction at HBHs ($30 696) still remains much greater than costs at LBHs and MBHs ($25 916 and $27 130, respectively).
Given the cost disparity between safety-net hospital groups, the effect of centralizing care on cost was evaluated. A one-way sensitivity analysis was performed by redistributing all patients from HBHs to LBHs (Figure 3A). This redistribution was done according to their SOI at HBHs so that patient risk would not be artificially reduced. In this scenario, former HBH patient costs were reduced from $35 303 per patient to a postredistribution LBH cost of $26 228 per patient, a reduction of $9075. The postredistribution LBH cost increased due to the greater burden of extreme severity cases from redistributed HBH patients. Figure 3B shows the same analysis but for a scenario in which all HBH patients were shifted to LBHs and MBHs equally. This situation resulted in an even greater cost reduction from $35 303 per patient to $26 148, or $9155 in cost savings. Moreover, this analysis represented a cumulative cost savings during the 5-year study period of $12 496 575 among this population of academic medical centers.
From this study, we found that HBHs have inferior outcomes and higher costs associated with PD compared with lower-burden hospitals. Using a cost-effectiveness model, we simulated scenarios in which HBHs were able to reduce their rates of complications, comorbidities, or mortality. While these findings had a modest effect on the cost of PD, we found the greatest cost reduction by centralizing care to hospitals with a lower safety-net burden. Redistributing HBH patients to LBHs and MBHs reduced the costs of these patients by more than $9000, resulting in a global cost savings of up to $699 per patient (Figure 4).
Given our findings, it appears that there are 3 potential options for reducing the increased costs of complex surgery at safety-net hospitals. First, we could accept the cost of caring for the most vulnerable patients of our society. Is the maintenance of safety-net hospitals in our country a form of social service? Given the socioeconomic challenges facing these patients, is it reasonable to ask the hospitals to adequately implement cancer screening, prehabilitation, enhanced recovery pathways, telehealth, and other innovative programs to improve efficiency and reduce costs? Perhaps we should accept that maintaining an open door to all patients, regardless of medical or financial condition, is simply an expensive enterprise. However, this option would require additional resources to sustain the financial burden of this mission.
Another solution is to invest more resources in understanding and improving the qualities of safety-net hospitals that are associated with increased costs for complex surgical care. It is known that these hospitals perform worse on timeliness, patient-centeredness, and equity of treatment,18 which may correlate with differences in human resources. Identifying areas for improved efficiency may certainly have a positive effect on cost.19 Perhaps with more financial resources, safety-net hospitals could invest in the same care improvement initiatives as more profitable hospitals.20,21 Also, the vulnerable patients at safety-net hospitals may have unique needs that require different approaches than are used for commercially insured patients,22,23 and further research could highlight these opportunities. Indeed, these approaches may be the best options for improving value at safety-net hospitals because payment penalties will likely exacerbate the disparities in care that currently exist.24-26 However, our findings indicate that these efforts must have a drastic effect on reducing costs to bridge the cost inequities that exist between HBHs and lower-burden safety-net hospitals. In a sensitivity analysis of our model, we found that reducing complications or comorbidities at HBHs to 12.5% of their current rate would equal the cost savings of redistributing these patients. If a 50% reduction in complications or comorbidities is unlikely, then a reduction of almost 90% is simply unrealistic.
Finally, we could redistribute patients requiring complex surgery to higher-performing centers that likely have a lower safety-net burden. From the hospital perspective, this redistribution may be unwelcome given that the LBHs and MBHs in our model would see an increase in charity and undercompensated care3 as well as more complicated patients, which may increase their surgical costs. There are also ethical and logistical considerations. Low-income patients are disproportionately affected by the travel and missed work associated with medical care,27-29 and centralizing care will likely make these patients travel even farther, limiting the number of patients who will do so.30,31 These limitations have forced some to question the usefulness of centralizing care.32-34 Despite these concerns, our findings show that the cost disparity between safety-net burden groups is most reduced by the redistribution of patients to lower-burden hospitals. The scenarios in our analysis represent the upper limits of redistribution, and future work should focus on discovering a practical patient volume that would maximize cost savings and reduce unintended consequences.
Our group previously found that HBHs perform complex surgery that is more expensive independent of patient characteristics.1 However, that analysis was only able to adjust for patient age, sex, socioeconomic status (quintiles based on zip code–level US census data), and SOI (score range, 1-4). In the present study, we stratified patients by SOI score and DRG code (major, minor, or no complication or comorbidity). There are a myriad of patient-specific factors that cannot be appreciated with research based on national administrative databases. However, these factors are the types of data that currently inform federal policy and health care reimbursement. The Comprehensive Care for Joint Replacement model, which takes effect this year, adjusts bundled payments by only considering DRG codes and the presence or absence of a hip fracture.35 Without adequate patient-level risk adjustment, it will be impossible to fully appreciate the complexity of the patient populations at safety-net hospitals. Therefore, reimbursement penalties will continue to represent a disadvantage to these centers.
There are limitations to this study. First, as previously mentioned, we cannot adequately adjust for the variation in patient complexity at these centers. The use of DRGs in our model, even when combined with a validated SOI illness score, simply does not capture the granular aspects of each patient’s medical and socioeconomic characteristics. We also cannot assume that readmission rates at these hospital groups would remain the same after the implementation of a policy that forced centralization of care. It is likely that patients who are forced to travel farther for care and who already have limited resources will end up being readmitted to a hospital closer to their home and not to the operative hospital. Also, this data set is limited to academic hospitals and their affiliates. However, a large proportion of safety-net hospitals are urban academic centers; therefore, our conclusions can be reasonably extrapolated to safety-net hospitals nationwide.1,36 Last, our model simulated changes in surgical care by adjusting the rates of specific DRG codes, whereas actual quality improvement initiatives would be multifactorial and likely have a different effect. However, we simulated a reduction in all complications or comorbidities by 50% at all safety-net hospitals, which is likely a greater improvement than is realistically achievable.
We have shown that improving clinical outcomes after complex surgery at safety-net hospitals can help mitigate their increased costs but have demonstrated that redistributing patients to less expensive hospitals may have a much larger effect on cost. This redistribution of patients would undoubtedly meet many challenges, including resource constraints of the population served at safety-net hospitals, the acceptance of high-risk patients at hospitals with lower safety-net burden, and the ethical and logistical challenges of such policies. More work is needed to find an optimal extent of redistributing care in conjunction with quality improvement measures to help sustain the institution of safety-net hospitals in our country. Otherwise, these hospitals and the vulnerable population they serve will face increasing challenges and disparities in health care in the coming years.
Accepted for Publication: April 18, 2016.
Corresponding Author: Richard S. Hoehn, MD, Cincinnati Research in Outcomes and Safety in Surgery (CROSS), Department of Surgery, University of Cincinnati School of Medicine, 231 Albert Sabin Way, Mail Location 0558, Cincinnati, OH 45267-0558 (firstname.lastname@example.org).
Published Online: July 27, 2016. doi:10.1001/jamasurg.2016.1776.
Author Contributions: Dr Hoehn had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Abbott, Hoehn.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Go, Hoehn.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Go, Abbott, Wima.
Obtained funding: Abbott, Shah.
Study supervision: Abbott, Hoehn.
Conflict of Interest Disclosures: None reported.
Funding/Support: This study received funding from the Department of Surgery, University of Cincinnati School of Medicine.
Role of the Funder/Sponsor: The funding source had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.