Medicaid insurance status predicts postoperative mortality after total knee arthroplasty in state inpatient databases

The SID is a database compiles inpatient information from nonfederal, as well as nonpsychiatric hospitals. Each SID observation represents a unique hospital admission as our unit of analysis. Hospitalization and discharge information was extracted for the SID on patients ≥18 years of age, using 2007–2014 data from New York, Florida, Maryland and Kentucky. Additionally, 2007–2011 data for California was extracted from the SID, Healthcare Cost and Utilization Project (HCUP) database. HCUP has established measures used to ensure database quality [41].

We identified patients in SID from New York, Florida, Maryland and Kentucky who underwent primary TKA from January 2007 through December 2014 using the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) code 81.54. Patients were divided into two groups based on whether they had a unilateral-TKA (UTKA) or bilateral-TKA (BTKA). Records with UTKA contained ICD-9-CM code 81.54 once, whereas BTKA had this code present twice [42]. We excluded patients who underwent concomitant revision of knee replacement (ICD-9-CM 81.55) or who had more than two entries for procedure code 81.54. Additionally, records were excluded if age, gender, postoperative disposition or insurance status were absent.

The following patient variables were collected: Hospital length of stay (LOS), ICD-9-CM diagnoses and procedures, insurance type or expected payer, dates of admission and discharge, and patient disposition. Additionally, the following surgical and hospital variables were collected: date and location (state) hospital where the procedure was performed, the hospital Core Based Statistical Area and hospital surgical volume. The SID marks diagnoses as present-on-admission (POA) in order to indicate co-morbidities present prior to surgery from perioperative complications.

We stratified patients into five groups based on payer type (or in some cases, expected payer): Medicare (fee-for-service as well as Managed Care patients), Medicaid (also fee-for-service as well as Managed Care patients), Uninsured (including self-pay and if no-charge was reported), Other (including government programs, Title V, CHAMPUS, CHAMPVA and Worker's Compensation) and private insurance (private HMOs and PPOs, commercial carriers, as well as Blue Cross).

The Elixhauser co-morbidity index was used to measure patient co-morbidities [43], only selecting those diagnoses marked as POA. We relied on the Van Walraven's method for calculating a numeric score based on this index, which we then converted into a categorical variable [44].

The total number of observations obtained for the years 2007–2014 for New York, Maryland, Florida and Kentucky, and 2007–2011 for California, was 922,819. Within these states, 199,141 TKAs were performed in 2007 with a peak of 136,884 performed in 2010. The number of TKAs then decreased, with 95,737 performed in 2014. In Tables 2 and 3, respectively, we contrast patient characteristics and hospital characteristics by insurance type using bivariate analysis.

Our primary outcome measure was in-hospital mortality, measured by the unadjusted rate and adjusted odds ratio (OR).

Secondary outcomes included hospital LOS and postoperative complications, measured as incidence rate ratios and ORs, respectively. The following postoperative complication groups were of interest: intraoperative/procedure related, infectious, wound, pulmonary, urinary, gastrointestinal, cardiovascular and systemic.

Patients were first stratified in cohorts by insurance type, to contrast demographic and POA co-morbidities across strata. The unadjusted rates of LOS, postoperative complications, in-hospital mortality and total charges were calculated and compared by insurance status. Analysis of variance was used to compare continuous variables, and the Pearsons Chi-squared test or Fisher's exact test were used to compare categorical variables. Variables that violated assumptions of normality were converted to nonparametric equivalents.

We fit marginal logistics regression models using GEEs with exchangeable correlation in order to measure the influence of insurance type on our primary postoperative outcome in-hospital mortality while controlling for demographic characteristics, co-morbidities and potential confounders. Clustering occurs when individual hospitals repeatedly contribute observations, and GEE models take this into account. We reported ORs with their 95% CI. Separate models were developed to analyze our primary outcome of interest (in-hospital mortality) and secondary outcomes (overall and by category post-TKA complication rates). Potential confounders were accounted for in our models, by including demographic characteristics and co-morbidities (significance indicated by p < 0.05) and patient characteristics such as race, age, gender and insurance type. Additionally, the models included the van Walraven modified Elixhauser co-morbidity index (divided into three categories: less than zero, zero and more than zero), median household income based on patient zip-code, procedure year and procedure state.

The adjusted effect of insurance type on hospital LOS was examined by fitting a generalized linear mixed model to log-transformed LOS (LOS has a non-normal distribution). We reported regression coefficients with a 95% CI, denoting significance levels using the asterisk system noted above. As with the GEEs, generalized linear mixed models take clustering into account in cases where individual hospitals repeatedly contribute observations.

In order to account for potentially unmeasured confounders, the regression models were subjected to several sensitivity analyses. We used the Wald statistic to identify the most significant covariates, and our models were re-estimated after removal of the covariate, so long as there was not a significant attenuation of the effect for Medicaid insurance (<10% attenuation of estimated odds for each outcome) and maintained statistical significance after re-estimation. In each model, the most significant covariate was age (in years).

Statistical significance was set at an alpha of <0.05, and all p-values are two-sided. Statistical tests and analysis were conducted using SAS version 9.4 (SAS Institute, NC, USA).

Our primary outcome measure was unadjusted and adjusted in-hospital mortality by insurance type. During the time period studied, total in-hospital mortality was less than 655. By payer group, in-hospital mortality totaled 528 for Medicare patients, 17 for Medicaid and 98 for privately insured patients. The unadjusted in-hospital mortality was statistically significantly lower in privately insured patients compared with the Medicare and Medicaid population (p <0.0001). We tabulated the unadjusted differences in in-hospital mortality and readmission rates between primary insurance types in Table 4.

The effect of primary payer status on in-hospital mortality was estimated using a logistic regression model, with results shown in Table 5. After controlling for patient characteristics, surgical factors, state and hospital type, Medicaid patients had a 73% higher odds of in-hospital mortality compared with privately insured patients (OR: 1.73; 95% CI: 1.01–2.95). Self-pay or no charge patients had higher odds of 166% (OR: 2.66; 95% CI: 1.03–6.83). Hence, we refute the null hypothesis that there is no difference in in-hospital mortality between payer groups following TKA.

In-hospital mortality is a rare event. However, our findings were corroborated (and likely mediated by) our secondary outcomes, further exploring the effect of primary payer status on patient outcomes after TKA. Compared with privately insured patients, individuals insured by Medicaid were more likely to experience any postoperative complications (OR: 1.25; 95% CI: 1.18–1.33), pulmonary complications (OR: 1.26; 95% CI: 1.16–1.36), infection-related complications (OR: 1.51; 95% CI: 1.35–1.68), intraoperative complications (OR: 1.25; 95% CI: 1.03–1.52), longer LOS (OR: 1.09; 95% CI: 1.08–1.10) and greater total charges (OR: 1.03; 95% CI: 1.02–1.04).

The hospitals in our dataset were divided into quartiles based on patient volume, with the first quartile representing the lowest volume. Of all Medicaid patients in this study, 44.3% were treated at hospitals in the first quartile. In comparison, 22.3% of privately insured patients were treated at such hospitals. In-hospital mortality was greatest for hospitals in the first quartile. Using hospitals in the first quartile as the reference, the OR for in-hospital mortality for the second quartile was (OR: 0.64; 95% CI: 0.52–0.79), (OR: 0.68; 95% CI: 0.54–0.86) for the third quartile and (OR: 0.42; 95% CI: 0.31–0.57) for the fourth quartile.

Stratifying the data by state in our sensitivity analyses led to similar inferences (data not shown). There was no significant difference in the adjusted OR for in-hospital mortality between each state and Florida, except in the case of California, which had a significantly lower odds of in-hospital mortality (OR: 0.56; 95% CI: 0.45–0.70).

We performed additional sensitivity analyses: the risk-adjusted OR for Medicaid and LOS after removing age (OR: 1.08; 95% CI: 1.07–1.10) and total charges in 2016 after removing age (OR: 1.03; 95% CI: 1.02–1.04) were still significantly greater for Medicaid compared with private insurance. Risk-adjusted ORs were again calculated after removing California, and the incidence rate ratio for Medicaid patients was still significantly greater for LOS (OR: 1.07; 95% CI: 1.06–1.09) and for total charges (OR: 1.02; 95% CI: 1.01–1.03) (Table 6).

This is the first direct comparison between Medicaid and private insurance type, demonstrating increased mortality after TKA in underinsured patients while controlling for patient characteristics, co-morbidity, risk factor and age. This is deeply concerning, given that the data are from several state databases, including several diverse large states (California, Florida, Maryland, Kentucky and New York), drawing from a significant portion of the US populations and hence likely representing a general trend across the country.

Beyond describing and lamenting disparate outcomes driven by social determinants of health, we need to explore potential mechanisms with a view to design and implement effective countermeasures [49]. We suspected that hospital volume may be an important mediator, which could explain the observed disparities between Medicaid and private insurance patients. Racial/ethnic minorities and Medicaid patients are less likely to be treated at high-volume hospitals when compared with White, privately insured patients [31,50,51]. Patients treated at low-volume hospitals have an increased risk for readmission following TKA [40], and other studies show that high-volume surgeons and institutions have lower rates of complications and mortality following orthopedic surgery [52–54]. Our study corroborates this exploratory hypothesis, in that LOS, in-hospital mortality and the odds of any complication were greater in the hospitals in the first quartile for hospital volume when compared with greater volume hospitals. The differences in mortality and complication rates between payer groups demonstrated in our study may in part be due to the fact that Medicaid patients are largely treated at low-volume hospitals. This generated a new exploratory hypothesis that hospital volume and quality interact with high risk and lower SES to compound morbidity and mortality after major joint replacement, a hypothesis that should be investigated and confirmed in subsequent studies. Additional factors that may contribute to poorer outcomes for Medicaid patients include proximity of patient home to hospitals, longer waiting period for surgical appointments and lower baseline range of motion and function, and such factors should be investigated as well.

Mediated for example through (implicit) bias, the racial makeup of Medicaid patients may play a role in the disparities presented in this study. Overall health insurance coverage in 2016 for non-Hispanic Whites was 93.7%, while for Blacks coverage was 89.5%, Asians 92.4% and Hispanics 84.0% [55]. In terms of Medicaid coverage, Blacks make up 34.1% of the insured population, Hispanics make up 31.1%, and Whites makes up 16.9% as of 2015 [56]. A systematic review showed that racial and ethnic minority groups are at a greater risk for postoperative complications and mortality compared with White patients for spinal and joint replacement procedures [57,58].

Of note, the Hospital Readmissions Reduction Program, under the ACA, has since 2012 reduced payments for Inpatient Prospective Payment System hospitals with excess 30-day readmissions [59]. Such penalization may disproportionally affect hospitals that treat racial/ethnic minority groups and patients of lower SES, since the Hospital Readmissions Reduction Program model does not differentiate hospitals based on the SES of the patient population [60]. Instead, the Center for Medicare and Medicaid Services policies could promote the preferential admission of the highest risk patients in the high-quality and high-volume institutions.

We believe this is the most up-to-date, largest recent study examining the affect of insurance type on patient mortality following TKA. Previous studies on TKA focused on disparities in Medicare patients only, were outdated, did not group by insurance type, or were mainly concerned with racial/ethnic disparities (Table 1). Our sample size is large and was extracted from several diverse states, thus allowing us to control for many potential confounding variables.

Our hypothesis, formulated a priori, that Medicaid insurance status is associated with a higher rate of postoperative mortality following TKA when compared with privately insured patients, builds on previous studies investigating disparities between Medicaid patients and privately insured patients undergoing other orthopedic interventions [2,11,12].

Nevertheless, the present study corroborates previous research examining disparities in surgical outcomes depending on insurance status [61], including in orthopedic operations [10–12,23]. LaPar et al. conducted a large retrospective analysis of 893,658 patients from the Nationwide Inpatient Sample database, examining the role of primary payer status on outcomes in several surgical procedures, including 230,000 THAs, and found that those insured by Medicaid had higher odds of unadjusted mortality, higher costs and greater LOS compared with private insurance patients [62]. Our study of 922,819 cases similarly found that in-hospital mortality, LOS and total cost were higher for Medicaid patients.

One strength of our study comes from its generalizability. The HCUP dataset is a large, diverse sample that we extracted from several of the most populated states, allowing us to draw useful conclusions that can be generalized to the rest of the country. Since our study takes a large number of patient and hospital factors into account, the results can be applied broadly. Conversely, the results should not be construed to implicate bias by any individual or discrimination in any institution. Instead, further investigation at the provider and institutional level is warranted in order to expose the mechanisms driving patient disparities based on insurance type in our healthcare delivery system [49].

The results maintained statistical significance when controlling for state, hospital characteristics and patient demographics in our sensitivity analyses. The results also maintained significance independent of inclusion and exclusion of potential confounders, modeling choices and assumptions. Stratification by state still demonstrated higher mortality in Medicaid populations. Our results are compelling and consistent, considering that Medicaid insurance status was associated also with higher complication rates and remained so across several statistical models and sensitivity analyses. We report many p-values that are highly statistically significant (<0.005), but we acknowledge that statistical significance is less meaningful in such a large sample, since significance is quickly reached in studies with such a large number of observations [63,64].

We acknowledge several limitations of our study. Although we attempted to control for confounding variables, we cannot exclude that confounders were unaccounted for in our analyses; these confounders could be contributing to poorer postoperative outcomes for Medicaid patients. Patient risk factors that confer poorer outcomes may be proportionally higher in the Medicaid population, such as obesity, lower level of education, lower income level and decreased English proficiency and may contribute to a higher risk of adverse outcomes in Medicaid patients [65,66]. Studies examining social determinants of health in patients undergoing orthopedic surgeries are limited, but those that have been conducted found that Medicaid patients undergoing orthopedic procedures are more likely to smoke tobacco and have a higher rate of obesity [67,68]. Our results may underestimate the rate of adverse events in patients undergoing TKA since patients are only accounted for if they present with complications to the hospital where the procedure was performed. Intraoperative information is not contained in the database, preventing us from studying whether poorer outcomes for Medicaid patients are associated with particular intraoperative events. Proper event coding in the HCUP database is dependent upon providers using particular medical terminology in free-text case notes [69], and failure to do so leads to missing and misclassified data.

Perioperative and intraoperative factors may also play a role in the disparities observed in the present study. In anesthesia, social determinants of health appear to play a role in the quality of care provided [49]. The use of regional anesthesia in primary joint arthroplasty has been shown to lead to more favorable patient outcomes [70–77,78]. No large-scale studies have been performed investigating the disparities in the use of regional techniques between racial and ethnic groups, but a study by Memtsoudis et al. using a database of 382,236 THA and TKA patients found that neuraxial anesthesia was used in 17% of Black patients and 25% of White patients [70]. Similarly, a study by Patorno et al. using the same database found that in cases of hip fractures, regional anesthesia was used more often in White patients (16%) compared with Black patients (13%) [79]. Postoperative care and pain management also contain significant disparities between racial and ethnic groups [1,65,80–82], with minorities suffering from higher pain scores [83], longer wait times to receive pain treatment [84] and obtaining lower amounts of pain medication compared with White patients [85].