Association of Insurance Type With Inpatient Surgical 30-Day Readmissions, Emergency Department Visits/Observation Stays, and Costs

Objective: To assess the association of Private, Medicare (MC), and Medicaid/Uninsured (MU) insurance type with 30-day emergency department visits/observation stays (EDOS), readmissions, and costs in a safety-net hospital (SNH) serving diverse socioeconomic status patients. Background: MC’s hospital readmission reduction program (HRRP) disproportionately penalizes SNHs. Methods: This retrospective cohort study used inpatient National Surgical Quality Improvement Program (2013–2019) data merged with cost data. Frailty, expanded operative stress score, case status, and insurance type were used to predict odds of EDOS and readmissions, as well as index hospitalization costs. Results: The cohort had 1477 Private; 1164 MC; and 3488 MU cases with a patient mean age 52.1 years [SD = 14.7] and 46.8% of the cases were performed on male patients. MU [adjusted odds ratio (aOR) = 2.69, 95% confidence interval (CI) = 2.38–3.05, P < 0.001] and MC (aOR = 1.32, 95% CI = 1.11–1.56, P = 0.001) had increased odds of urgent/emergent surgeries and complications versus Private patients. Despite having similar frailty distributions, MU compared to Private patients had higher odds of EDOS (aOR = 1.71, 95% CI = 1.39–2.11, P < 0.001), and readmissions (aOR = 1.35, 95% CI = 1.11–1.65, P = 0.004), after adjusting for frailty, OSS, and case status, whereas MC patients had similar odds of EDOS and readmissions versus Private. Hospitalization variable cost %change was increased for MC (12.5%) and MU (5.9%), but MU was similar to Private after adjusting for urgent/emergent cases. Conclusions: Increased rates and odds of urgent/emergent cases in MU patients drive increased odds of complications and index hospitalization costs versus Private. SNHs care for higher cost populations while receiving lower reimbursements and are further penalized by the unintended consequences of HRRP. Increasing access to care, especially for MU patients, could reduce urgent/emergent surgeries resulting in fewer complications, EDOS/readmissions, and costs.


INTRODUCTION
Medicare (MC) introduced the Hospital Readmission Reduction Program (HRRP) as a quality of care metric [1][2][3] and to reduce healthcare expenditures. 4 HRRP reduces MC payments for hospitals with higher than predicted 30-day readmissions by risk-adjusted models for select medical and surgical admissions. 4 Reduced payments are applied to all fee-for-service MC diagnosis-related groups for the fiscal year. In Fiscal Year 2019, 75% of all hospitals were penalized. 3 Although penalties are capped at up to 3%, 4 this represents a significant burden on healthcare systems, especially for safety-net hospitals (SNHs) serving low-socioeconomic status (SES) populations.
HRRP succeeded in decreasing hospital readmissions. 5 However, HRRP has fostered several unintended consequences with multiple publications questioning whether readmissions are an appropriate quality metric. [6][7][8] First, pressure to reduce 30-day readmissions may incentivize hospitals to increase observation stays or discharges from emergency department (ED) visits, rather than having readmissions. 1,2 These practices may have increased mortality for heart failure and pneumonia. 9 Second, 30-day readmission risk can be attributed to patient factors 10 outside of a provider's control, such as frailty 11 and social risk factors 12 beyond dual MC-Medicaid eligibility. 3 These factors are not included in HRRP risk adjustment models, [13][14][15][16] penalizing hospitals serving more frail and/or socially disadvantaged patients. Frail and low-SES patients have higher rates of urgent or emergency surgeries, 17 which are associated with worse outcomes. 18,19 SNHs serve vulnerable populations 17 and may provide the same quality of care, 20 but are the most penalized by HRRP. 2,12,21 Stratifying hospitals by dual MC-Medicaid eligibility reduces SNH penalties. 3 However, 91.6% of low-SES/high-burden hospitals experienced penalties and the typical SNH still receives a 0.28% reduction in MC payments. 3 Multiple studies classifying hospitals as having a high or low safety-net burden found that high-burden hospitals have higher readmission rates and higher costs, 17,22 likely as a result of the population they serve.
However, the National Academy of Medicine states these factors need to be studied within healthcare systems serving a wide range of SES patients, rather than across hospitals with vastly different patient populations, to understand whether low-SES patients have worse outcomes due to lower quality of care or factors beyond provider control. 23,24 Insurance status/type is a patient-level social risk factor, commonly used as a proxy measure of patient SES to predict surgical outcomes, with widely varying results. 25,26 Variables such as insurance type, 25,26 case status (elective, urgent, or emergent surgery), 18 frailty, 19,27 and increased surgical-induced physiologic stress 27 have been associated with adverse outcomes across surgical specialties. Despite the developing literature on these topics, many studies [28][29][30] are limited by their use of administrative claims data, which lack detailed data on patient risk factors and outcomes. [31][32][33] Variations in assigning ICD-9/10 codes across institutions further adds to inaccuracies. 31 Others use charge data 34 or costto-charge ratios, 35 not the actual, variable costs associated with patient healthcare use. Finally, prior studies focus on either readmissions 29,34 or ED visits 30,35 but do not include both factors.
We designed this study using high-quality, nurse-abstracted data for preoperative risk factors and complications from the National Surgical Quality Improvement Program (NSQIP) 31,32 enriched with electronic health record (EHR)/billing data using actual variable costs to examine these associations including ED visits/observation stays (EDOS) and readmissions using the HRRP definition of 30 days from the date of discharge. We assessed the association of insurance type for inpatient surgical procedures with (1) actual variable costs of the index hospitalization, (2) 30-day EDOS, (3) 30-day readmissions, and (4) variable costs of EDOS and readmissions. Our cohort comes from one healthcare system serving patients over a wide range of SES, as recommended by the National Academy of Medicine to study outcomes within a healthcare system to understand the impact of social risk factors. 23,24 We hypothesize that after risk adjustment for clinical and surgical factors, patients with Medicaid/ Uninsured (UN) compared to Private insurance will be associated with increased 30-day EDOS, 30-day readmissions, and index hospitalization costs.

Study Population and Data
This retrospective cohort study used data on all patients undergoing inpatient procedures present in the 2013-2019 American College of Surgeons NSQIP at an academic medical center and SNH following STROBE Reporting Guidelines. NSQIP registry provides standardized definitions of preoperative risk factors and complications 31 and was used for cohort identification. The Institutional Review Board (IRB) of the University of Texas Health San Antonio approved this study and consent was waived.

Expanded Operative Stress Score) Assignment, Case Status, and 30-Day Complications
The operative stress score (OSS) estimates surgical-induced physiologic stress of procedures across surgical specialties based on CPT codes by assigning a score ranging from 1 to 5 with 1 and 5 representing very low and very high physiological stress, respectively. We used the expanded OSS 19 with 2343 CPT codes providing improved case coverage for nonmajority male populations compared to the original OSS. 27 After excluding cases without an expanded OSS assigned to the principal CPT code, OSS was assigned using the highest score for all available procedures within each case. 19 Case status was determined from NSQIP variables with urgent cases being defined as "no" responses to elective and emergency variables. 18 Any complication was defined using the 20 NSQIP variables defining postoperative complications that occurred 30 days after the date of the index surgery (Supplemental Table 1, http:// links.lww.com/AOSO/A201).

30-Day EDOS and Readmissions
NSQIP only tracks patients for 30 days after surgery and contains 30-day readmission variables from the date of surgery. We merged NSQIP with EHR data to determine readmissions and EDOS within 30 days of discharge from the index procedure's hospitalization, to be consistent with the HRRP definition of 30-day readmissions.

Insurance Type and Cost Data
The identified, local NSQIP data were merged with EHR and managerial accounting data to determine insurance type and cost of the index hospitalization, readmissions, and EDOS. Insurance type was categorized based upon billing data for the encounter supplemented by EHR data and defined as (1) Private insurance including Tricare and Workers Compensation; (2) MC; and (3) Vulnerable including dual enrollment in MC/Medicaid, Charity Care, self-pay, or county indigent care programs (Supplemental Table 2, http://links.lww.com/AOSO/A201). "Other" included encounters billed to the Veterans Administration, Department of Corrections, or self-pay with >1% of charges collected and were excluded (n = 87).
We defined variable costs as costs related directly to patient care occurring during the encounter, such as supplies and salaries, and included direct variable costs that vary directly with the quantity of resources provided for patient care. Variable costs were derived using direct measurements from a bottom-up approach, rather than calculated estimates derived from charges. Hospital fixed costs, outpatient, and professional fees were not included. We used variable costs, as fixed costs are not directly related to patient care and vary between hospitals. 39 The natural logarithm of variable costs was used, as previously described, 40 after adjusting costs to 2019 dollars using the Personal Health Care Index. 41 Exclusions Cases were excluded due to (1) missing expanded OSS coverage of principal CPT code, (2) missing variables used to calculate the RAI other than cognitive decline, (3) other insurance status, and (4) missing or inaccurate cost variables (including readmissions external to the index hospital).
Patient mortality resulting in no or reduced chances of subsequent EDOS and readmissions were excluded, as previously described. 42 Cases were excluded due to (1) death during the index hospitalization, (2) discharge to another acute care hospital, (3) discharge against medical advice, (4) death within 30 days of discharge when discharged to Hospice or Home on Hospice, and (5) death within 30 days of discharge without a 30-day EDOS or readmission. Two sensitivity analyses were performed adding exclusion groups 4 and 5 and 1 to 5 to the analysis to determine whether the association of insurance type was robust to cohort selection.

Study Outcomes
Our clinical outcomes were the association of insurance type with (1) any complication, (2) EDOS, (3) readmissions, and (4) index hospitalization length of stay (LOS) on EDOS and readmission adjusted for frailty, OSS, and case status. Secondary analyses assessed the association of urgent/emergent cases with insurance type adjusted for frailty. Finally, the probabilities of EDOS and readmissions were calculated for lowest-risk versus highest-risk scenarios.
Our cost outcomes compared variable costs for the index surgery hospitalization using 3 discrete groups: (1) no 30-day EDOS/readmission, (2) 30-day readmission, and (3) 30-day EDOS without a 30-day readmission, patients with both a readmission and EDOS were assigned to the readmission group. We also assessed the variable costs of the first readmission and EDOS for each index case.

Statistical Analysis
Categorical data was summarized using count and percentage, with continuous data using mean and standard deviation (SD). Chi-square tests and F tests were used to test for difference between groups for categorical and continuous variables. Kruskal-Wallis tests were used for the skewed LOS and variable costs for 1) index hospitalization, 2) EDOS and 3) readmissions. Logistic regression analyses were performed for case status and complications adjusting for a combination of RAI, OSS, case status, and insurance type. We calculated probabilities for lowest-risk versus highest-risk scenarios stratified by RAI/ frailty from these final models. Natural logarithms were used to normalize the skewed LOS and variable costs, which reduces the impact of extreme values, as previously described. 40,43 Percent change/relative difference was calculated using the exponential function; %change = (e Estimate coefficients − 1) × 100. Analyses were performed using R version 4.1.0 (2021-05-18).

Population Demographics
Our cohort consisted of 6129 cases of inpatient procedures at a major urban SNH (Supplemental Fig. 1 (Table 3).

Lowest-Risk and Highest-Risk Probabilities for EDOS and Readmissions
The probabilities of EDOS and readmissions were estimated for the lowest-risk and highest-risk groups stratified by frailty ( Table 4). As frailty increased, the difference between lowest and highest risk decreased for EDOS. For instance, robust patients at lowest-risk (Private, elective surgery, no complications) versus highest-risk (MU, urgent/emergent surgery, any complication) had an EDOS probability of 7.4% and 20.5%, respectively, a difference of 13.1%. In contrast, as frailty increased, the difference between lowest-and highest-risk increased for readmissions. Very frail patients at lowest risk versus highest risk had a readmission probability of 7.8% and 44.8%, respectively, a difference of 37.0%.

Increased Index Hospitalization Variable Costs among Patients with a 30-Day EDOS or Readmission
The %change of variable costs was decreased for robust compared to normal patients (Table 5; M1-M3). The %change was lower for OSS1-2 and higher for OSS4 and OSS5 compared to OSS3 cases. MC (14.8%) and MU patients (7.8%) had higher %changes in variable costs compared to Private (Table 5; M1), but MU patients had similar %change to Private after adjusting for urgent/emergent cases (Table 5; M3). Patients with a readmissions or EDOS had a 37.2% and 12.8% higher %change, respectively, than those without either an EDOS or readmission as the reference group (Table 5; M1). However, after adjusting for complications, EDOS and readmission groups had similar costs to the reference group (Table 5; M2).
Two sensitivity analyses increased the cohort to 6170 and 6324 cases showed similar results for insurance type for urgent/ emergent case status, any complication, EDOS, readmissions (Fig. 1), and index hospitalization variable costs (Supplemental Fig. 2, http://links.lww.com/AOSO/A201) compared to our study cohort of 6129 cases.

Similar EDOS and Readmission Costs Between Private, MC, and MU patients
The %change of the first EDOS and first readmission variable costs was similar for MC and MU compared to Private (Supplemental Table 4, http://links.lww.com/AOSO/A201). The %change of the first EDOS or readmission variable costs increased for cases experiencing any complication by 33.5% and 50.3%, respectively, and was decreased for robust compared to normal patients.

Mean and Median Index Hospitalization Variable Costs by Insurance Type and OSS
Descriptive statistics for mean and median dollars for the index hospitalization were reported due to the right skewed variable costs. Variable costs increased with increasing OSS and were significantly different between insurance types overall and for all OSS groups except OSS5 (      The use of HRRP as a metric for low-quality care 1-3 has been heavily criticized. [6][7][8] Readmissions are largely driven by patient factors, 10 urgent/emergency surgeries, 17 and complications arising after discharge. 44 SNHs have higher readmission rates 3 and have also been shown to provide similar quality of care. 20 Lower readmission rates may not indicate high-quality care as decreased readmission rates can be accompanied by increased EDOS. 28,29 Failure to appropriately readmit patients can increase mortality, 1,2 especially for patients originally admitted for heart failure and pneumonia. 9 Readmissions in the current study were consistent with increased index hospitalization costs and LOS for patients undergoing coronary artery bypass graft surgery. 45 Our study demonstrated the clinical significance of insurance type and case status. Robust patients at lowest-risk (Private, elective surgery, no complications) versus highest-risk (MU, urgent/emergent surgery, any complication) had an EDOS probability of 7.4% and 20.5%, respectively, a difference of 13.1%. In contrast, readmission probabilities were highest in very frail patients with a difference of 37.0% for lowest-versus highest-risk scenarios. We speculate that robust, low-SES patients are more likely to use the ED for routine/primary care, 30 whereas older and frail patients are more likely to be readmitted secondary to their multiple medical comorbidities. SNH care for low-SES populations with Medicaid or no insurance. Our study shows the increased probabilities and increased costs of EDOS and readmissions borne by SNH.
Using readmissions as a care quality metric has serious consequences. Multiple studies demonstrate that SNH and academic medical centers receive disproportionate HRRP penalties 13,14,21 ; further widening disparities between SNH, caring for vulnerable populations, and hospitals serving affluent communities. 21,46 Poor communities exist under the burdens of inequalities in insurance coverage, education, and poverty while being treated at hospitals with less funding/resources, 14,21 limiting their access to care. Our study shows that patients having a readmission or EDOS have higher index hospitalization variable costs which become similar after adjusting for complications. Additionally, increased rates and odds of urgent/emergent procedures in MU patients also drive increased complications, costs, and EDOS/ readmissions. Thus, SNH operate under a triple burden; (1) higher index hospitalization costs, (2) increased readmissions and EDOS encounters leading to more costs, and (3) penalties for worse outcomes. Like SNH, academic medical centers also serve as healthcare providers for low-SES patients, and this trend continues to grow, 47 making the healthcare system under study a particularly strong example of SES's effect on surgical outcomes.
EDOS have similar associations as readmissions, with MU patients having increased index hospitalization costs and LOS. Unlike readmissions, care provided in many ED visits could be accomplished in lower-cost settings, suggesting that many ED visits are preventable. 48 However, providers often advise their patients to go to the ED for surgical FIGURE 1. Cohort and sensitivity analyses of adjusted odds ratios for MC and MU insurance groups compared to Private for case status, any complication, 30-day EDOS and 30-day readmissions. aOR for MC and MU Insurance groups (reference Private group) for the study cohort and two sensitivity analyses. Cases were excluded from the cohort due to (1) death during the index hospitalization, (2) being discharged to another acute care hospital, (3) discharge against medical advice, (4) death within 30 days of discharge when discharged to Hospice or Home on Hospice, and (5) death within 30 days of discharge without a 30-day EDOS or readmission. Two sensitivity analyses were performed adding exclusion groups 4 and 5 and 1-5 to the analysis to determine whether the association of insurance type was robust to cohort selection. Readm indicates readmissions complications. 49 Postoperative ED visits occurred in 17.3% in MC patients within 30 days after hospital discharge and 4.4% of patients had multiple ED visits. 50 Low-SES patients often use the ED for routine care, 30 creating yet another cost burden for SNH.
Consistent with prior studies, we observed variation in index hospitalization costs by insurance type 40 with complications (94% change) having the highest associated costs. 22 In this study, MU insurance patients had increased index hospitalization costs that were similar to Private after adjusting for urgent/ emergent cases, whereas MC patients had 11.2% change even after adjusting for case status. Mean variable costs for urgent/ emergent cases were substantially higher for all OSS groups versus elective cases. Insurance plans pay at different rates, and uninsured patients provide minimal, if any, revenue. MC and MU groups had increased urgent/emergent cases, complications, and costs, yet provide lower reimbursement than Private insurance.

Limitations
This study is a retrospective review and did not establish causal relationships. NSQIP provides a random sample of a broad range of surgeries but does not include all procedures for a healthcare system. Encounters occurring at outside hospitals may be missing from our data set leading to misclassification of cases without an EDOS or readmission. Only the first readmission and EDOS costs were examined; MU patients might have increased readmission and EDOS encounters and costs if all 30-day encounters were included. The data are derived from one healthcare system which may restrict generalizability. However, as this study has no equivalent published in the literature, this study can affect the national debate on presentation acuity and access to care in public policy and risk adjustment.

CONCLUSIONS
Increased rates and odds of urgent/emergent cases in MU (aOR = 2.69) and MC (aOR = 1.32) populations drive increased odds of complications, index hospitalization costs, and LOS versus Private insurance patients. However, despite similarly increased odds of complications in MC (aOR = 1.22) and MU (aOR = 1.24) patients, 30-day EDOS, and readmissions odds were only increased in MU compared to Private patients. SNH care for higher cost populations while receiving lower reimbursements and are further penalized by the unintended consequences of value-based programs such as HRRP. Increasing access to care, especially for MU patients, could reduce urgent/emergent surgeries resulting in fewer complications, EDOS, readmissions and costs.