Article Text
Abstract
Background and objectives Inpatient shoulder arthroplasty is widely performed around the USA at an increasing rate. Medicaid insurance has been identified as a risk factor for inferior surgical outcomes. We sought to identify the impact of being Medicaid-insured on in-hospital mortality, readmission, complications, and length of stay (LOS) in patients who underwent inpatient shoulder arthroplasty.
Methods We analyzed 89 460 patient discharge records for inpatient total, partial, and reverse shoulder arthroplasties using data from the Healthcare Cost and Utilization Project’s State Inpatient Databases for California, Florida, New York, Maryland, and Kentucky from 2007 through 2014. We compared patient demographics, present-on-admission comorbidities, and hospital characteristics by insurance payer. We estimated multilevel mixed-effect multivariate logistic regression models and generalized linear models to assess insurance’s effect on in-hospital mortality, readmission, infectious complications, cardiac complications, and LOS; models controlled for patient and hospital characteristics.
Results Medicaid-insured patients had greater odds than patients with private insurance, other insurance, and Medicare of inpatient mortality (OR: 4.61, 95% CI 2.18 to 9.73, p<0.001) and 30-day and 90-day readmissions (OR: 1.94, 95% CI 1.57 to 2.38, p<0.001; OR: 1.65, 95% CI 1.42 to 2.38, p<0.001, respectively). Compared with private insurance, other insurance, and Medicare patients, Medicaid patients had increased likelihood of developing infectious complications and were expected to have longer LOS.
Conclusions Our study supports our hypothesis that among inpatient shoulder arthroplasty patients, those with Medicaid insurance have worse outcomes than patients with private insurance, other insurance, and Medicare. These results are relatively consistent with previous findings in the literature.
Statistics from Altmetric.com
Introduction
Total shoulder arthroplasty (TSA) is widely performed across the USA for treatment of osteoarthritis, rheumatoid arthritis, and rotator cuff tear arthropathy1; approximately 50 000 TSAs are performed in the USA annually, with annual growth rates of 7%–13%.2 From 1998 to 2008, there was a 2.5-fold increase in the number of inpatient shoulder arthroplasty procedures performed in the USA.1 Additionally, reverse shoulder arthroplasty (RSA) and partial shoulder arthroplasty (PSA), performed for similar indications as TSA, also have increasing utilization, with estimates of more than 20 000 RSAs and 15 000 PSAs performed annually in the USA.3
Previous literature in orthopedics has identified disparities in utilization and outcomes of joint arthroplasty by primary insurance payer status, race, and socioeconomic status. However, limitations in the TSA literature exist with respect to having a large sample size, recent data, and thorough examination of racial and socioeconomic factors (table 1).1 2 4–11 Primary payer status, specifically having Medicaid insurance, is increasingly being viewed as a perioperative risk factor for adverse outcomes after orthopedic surgeries, as it is linked with increased inpatient mortality, complications, increased length of stay (LOS), and readmissions.5 12 13 Given the recent expansions of Medicaid in many states following the implementation of the Affordable Care Act14 15 (resulting in 74.2 million enrolled through October 2017),16 surgical outcomes disparities based on insurance provider have serious implications for the health of US citizens.12 17 While disparity in outcome is a multifactorial problem, a patient’s insurance is one of the modifiable risk factors.
The goal of our retrospective cohort study was to further investigate disparities in outcomes of inpatient shoulder arthroplasty across different regions of the USA through analysis of the State Inpatient Databases (SID), Healthcare Cost and Utilization Project (HCUP), Agency for Healthcare Research and Quality (AHRQ) data from five geographically diverse states—California, Florida, Kentucky, Maryland, and New York.18 Our study sought to examine the association of primary insurance payer, particularly Medicaid insurance, on outcomes of TSA/RSA/PSA. We separately examined the effects of insurance payer on in-hospital mortality, 30-day/90-day readmission, complications, and hospital LOS. We hypothesized that patients insured by Medicaid would have increased adjusted odds of perioperative adverse events (including mortality) and longer expected LOS compared with patients with private insurance, other insurance, and Medicare in a population undergoing a shoulder replacement.
Methods
Study database and population
We analyzed discharge records of patients aged ≥18 undergoing inpatient TSA, PSA, and RSA using data from the HCUP SID (AHRQ) for five states from a period spanning 8 years (2007–2014).18 Data from 2007 to 2014 were available for Florida, Kentucky, Maryland, and New York; data from California were available from 2007 to 2011. The SID is an administrative database inpatient discharges in a given state; collectively, the SID contains approximately 97% of all inpatient discharges from community hospitals in the USA.18 The database contains information on procedures performed; diagnoses; hospital characteristics including procedure volume; patient demographics including admission and discharge date; insurance payer; a categorization of the median income of the patient’s residential Zip code (our measure of socioeconomic status); comorbidities identified by the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes; present-on-admission (POA) indicators for each diagnosis; complications; hospital LOS; total charges; and discharge disposition. The database does not contain information on cause of death. Records are unique by hospital admission and can be linked to subsequent readmission records.12
Shoulder arthroplasties were categorized into procedure types: TSA (ICD-9-CM code 81.80), PSA (81.81), and RSA (81.88). Revision shoulder arthroplasties were not included in our analysis; we only included records with mutually exclusive codes for TSA, PSA, or RSA.
Primary payer was classified into five categories: Medicare (including fee-for-service and managed care patients), Medicaid (including fee-for-service and managed care patients), uninsured (including self-paying patients or with no charge reported), other payer (such as worker’s compensation, military, other government insurers), and private insurance (including Blue Cross Blue Shield, commercial carriers, private Health Maintenance Organizations (HMOs), and private Preferred Provider Organizations (PPOs)).12 In our multivariate models, we collapsed the primary payer categories of private insurance, other insurance, and Medicare into a single category (collectively referred to as “non-Medicaid”), in order to compare patients with those insurance types with populations insured by Medicaid and the uninsured, individually. The collapsed category served as the reference in multivariate models. While we were particularly concerned with outcomes between Medicaid and the pooled group that comprised private insurance, other insurance, and Medicare, we isolated uninsured patients to avoid conflating their outcomes with insured patients.
Smoking status was classified into three categories: current smoker (30.51), former smoker (V15.82), or never smoked (neither current nor former smoker). A modification of the Elixhauser comorbidities was used as a measure of POA comorbidities.19 This continuous measure was recoded into a three-category variable; cut-off points were determined by visual inspection of a histogram and a lowess-smoothed curve, with inpatient mortality as the outcome.
Outcomes
The primary outcomes of our study were inpatient mortality, 30-day readmission, and 90-day readmission. Our secondary outcomes were binary indicators for each of the following: cardiac complication (including supraventricular arrhythmia, myocardial infarction, postoperative stroke, deep venous thrombosis, pulmonary embolism); infectious complication (including sepsis, urinary tract infection, postoperative wound infection); and intraoperative complication (including puncture/laceration, bleeding complication). The ICD-9-CM codes for the aforementioned complications are listed in table 2. We also analyzed a count outcome, the hospital LOS.
Statistical analyses
We excluded any patient with a missing value on age (or age under 18), or with missing values on gender, discharge disposition, or insurance payer. Bivariate analyses were performed and with patients categorized by primary payer. Categorical variables were compared between groups with a Pearson’s χ2 test or Fisher’s exact test, as appropriate. Continuous or ordinal variables were compared between groups using a one-way analysis of variance, Mann-Whitney U test, or Kruskal-Wallis test, as appropriate. Predictors were selected for inclusion in multivariate models if they were significantly different between payer groups at an alpha level of <0.25 and/or hypothesized a priori to be predictors of the outcome (eg, age, sex, race/ethnicity, median household income quartile of the patient’s Zip code, hereafter referred to as “median income”). Multilevel mixed-effects multivariate logistic regression models were used to predict binary outcomes of inpatient mortality; 30-day and 90-day readmission; and cardiac, infectious, and intraoperative complications.20 Multilevel mixed-effects generalized linear models with negative binomial distributions were used to predict the count outcome of hospital LOS.21 In all models, records were grouped by hospital, which served as the level 2 variable. For binary outcomes, model discrimination was assessed with the area under the receiver operating characteristic (ROC) curve.
In order to better qualify the impact of insurance status as compared with other social determinants of health (race/ethnicity, median income) and patient POA comorbidities on our findings, we reran our multivariate models in a series of stratified models by race/ethnicity, median income quartile, and our three-category recode of the van Walraven score to account for these differences. Due to difficulties in multilevel model convergence, stratified models were not clustered on hospital.
Readmission data were available only for certain states and years: California 2007–2011, Florida and New York 2007–2014, and Maryland 2012–2014. Two sensitivity analyses were conducted to account for the lack of data from California (missing years 2012–2014). The first sensitivity analysis was conducted with data only from 2007 to 2011, and the second with data from all years omitting California. We conducted an additional sensitivity analysis that included only elective procedures.
Data are expressed as percentage, OR (95% CI) for logistic regression models, or incidence rate ratio (IRR, 95% CI) for generalized linear models with negative binomial distributions. Use of the term “expected” in the reporting of IRRs should be interpreted as solely a statistical descriptor. It does not pertain to expected LOS as defined by Diagnosis Related Groups. Cells in bivariate tables with <11 observations are masked per HCUP requirements. In spite of this masking requirement, statistical tests were able to be conducted. All p values are two-sided, and statistical significance is evaluated at the 0.05 alpha level. Data analysis was conducted in SAS V.9.3 and in Stata SE V.15.
Results
Patient and hospital characteristics
Table 3 shows patient demographics, POA comorbidities, and hospital characteristics by payer status. After applying our exclusion criteria (which resulted in 948 records being dropped, or 1.04% of the total sample), our sample included 44 941 patients who underwent inpatient TSA, 24 432 who underwent inpatient PSA, and 20 087 who underwent inpatient RSA between 2007 and 2014 in California, Florida, Kentucky, Maryland, and New York. These three procedures resulted in a total sample size of 89 460. Nearly three-quarters of the sample (69.9%) was insured by Medicare, followed by private insurance (22.4%), Medicaid (2.2%), and other insurance (4.8%). Less than 1% of the sample was uninsured. A majority of the sample was white (84.9%), and 68.0% were undergoing elective shoulder arthroplasty. A larger number of patients with private insurance were treated at hospitals in the highest quartile of procedure volume than Medicaid patients (28.8% and 16.6%, respectively).
Figure 1 shows the interconnectedness between primary insurance payer status, race/ethnicity, and median income by presenting a breakdown of payer status by race/ethnicity and median income quartile. More black patients were insured by Medicaid (9.2%) than white patients (1.5%), Hispanic patients (6.2%), and patients of other races/ethnicities (5.2%). Of the patients, 25.5% of Other races/ethnicities, 22.5% of white, 21.7% of black, and 16.6% of Hispanic patients were insured by private insurance. Of black patients, 44.4% lived in the lowest quartile of median income, followed by Hispanic patients (28.2%), other races/ethnicities (18.8%), and whites (17.7%). Of patients of other races/ethnicities, 33.6% lived in the top median income quartile, and more white patients (27.8%) than black (11.5%) and Hispanic (18.2%) lived in the top quartile of the median income.
Outcomes
Outcomes by primary payer group are found in table 4. Fewer than 157 patients (less than 0.2% of the sample) died during their hospital stay, including 120 Medicare patients, 15 private insurance patients, and <11 Medicaid and <11 other insurance patients (note: the total number is masked due to certain payer groups fewer than 11 outcomes having n<11). Overall 30-day readmission was 4.8%, with a higher percentage of Medicaid (7.7%), uninsured (6.5%), and Medicare (5.3%) patients readmitted than those with private insurance (2.9%). Overall 90-day readmission was 9.6%, with more Medicaid (14.9%), Medicare (10.4%), and uninsured (9.6%) patients readmitted compared with those with private insurance (7.0%). The median LOS across all groups was 2 days (IQR: 2–3).
Multivariate results
Primary outcomes
In adjusted models in which we controlled for patient characteristics (insurance payer, race/ethnicity, gender, age, POA comorbidities, median household income of the patient’s residential location (in quartiles by state), procedure type, and hospital factors (state, year, hospital procedure volume of our included surgeries)), Medicaid patients had four times greater odds than those with private insurance, other insurance, and Medicare of inpatient mortality (OR: 4.61, 95% CI 2.18 to 9.73, p<0.001; table 5). Medicaid patients also were significantly more likely than patients with non-Medicaid insurance to be readmitted up to 30 days (OR: 1.94, 95% CI 1.57 to 2.38, p<0.001) and 90 days (OR: 1.65, 95% CI 1.42 to 1.93, p<0.001) postoperatively. Uninsured patients were also more likely than those with non-Medicaid insurance to be readmitted up to 30 days postoperatively (OR: 1.59, 95% CI 1.09 to 2.31, p<0.05), although they did not have significantly different odds from those with non-Medicaid insurance in terms of 90-day readmission. Model discrimination for the outcome of in-hospital mortality was high (area under the ROC curve: 0.86).
In stratified analyses by individual categories of race/ethnicity, median income quartile, and a three-category recode of the van Walraven comorbidity score, not all groups showed significantly different odds of an outcome measure for Medicaid patients compared with non-Medicaid patients, suggesting effect measure modification (online supplementary tables 1–3 / SDC 1, 2, 3). However, the incongruity between select results of main and stratified analyses likely reflects a lack of statistical power due to select subgroups with small numbers. Small subgroups included black patients (n=3641), Hispanic patients (n=4993), and other races/ethnicities (n=2830 patients); models suffered further sample attrition in the case of a lack of variation in the outcome.
Patients insured by Medicaid had significantly greater adjusted odds of inpatient mortality than those with non-Medicaid insurance in samples of white patients (OR: 4.17, 95% CI 1.49 to 12.39, p<0.05), Hispanic patients (OR: 13.76, 95% CI 2.23 to 84.99, p<0.01), in the lowest quartile of the median income (OR: 3.76, 95% CI 1.15 to 12.30, p<0.05), the top quartile of the median income (OR: 12.17, 95% CI 2.28 to 64.84, p<0.01), and in a population of patients of high scores on a categorical recode of the van Walraven score (OR: 4.41, 95% CI 1.93 to 10.10, p<0.001; online supplementary table 1 / SDC 1) only.
Supplemental material
Medicaid patients had greater adjusted odds of readmission in a population of white patients only, and patients on Medicaid in the lowest quartile of the median income were not significantly more or less likely than those with non-Medicaid insurance to be readmitted up to 30 days postoperatively (online supplementary table 2 / SDC 2). The same results applied to Medicaid patients readmitted up to 90 days postoperatively (online supplementary table 3 / SDC 3). In all median income quartiles except for the lowest, the results of Medicaid patients were consonant with the findings present in the main models: patients on Medicaid had higher adjusted odds of being readmitted up to 30 and 90 days than patients with non-Medicaid insurance; the same was observed for samples in each tertile of the van Walraven score.
Supplemental material
Supplemental material
In a sensitivity analysis of elective procedures only, Medicaid patients were not significantly more likely than patients with non-Medicaid insurance to die in-hospital (OR: 2.11, 95% CI: 0.26-17.07), but they were more likely to be readmitted up to 30 and 90 days postoperatively (30-day OR: 1.81, 95% CI: 1.32-2.48, p < 0.001; 90-day OR: 1.43, 95% CI: 1.14-1.80, p < 0.01) (Unpublished data).
Secondary outcomes
Medicaid-insured patients had twice the odds of developing an infectious complication than those with non-Medicaid insurance (OR: 2.33, 95% CI 1.65 to 3.28, p<0.001; table 5). LOS was expected to be higher among Medicaid patients (IRR: 1.33, 95% CI 1.29 to 1.37, p<0.001) and uninsured patients (IRR: 1.30, 95% CI 1.23 to 1.36, p<0.001) than non-Medicaid-insured patients.
Results from stratified models indicated that the odds of having an infectious complication were significantly greater for Medicaid patients compared with non-Medicaid patients in a population of white patients (OR: 2.65, 95% CI 1.70 to 4.14, p<0.001; online supplementary table 4 / SDC 4), in the second and third quartiles of the median income (OR: 2.66, 95% CI 1.33 to 5.30, p<0.01; OR: 4.36, 95% CI 2.35 to 8.12, p<0.001, respectively), and in the lowest and highest tertiles of the van Walraven score (OR: 2.80, 95% CI 1.51 to 5.20, p<0.01; OR: 2.11, 95% CI 1.32 to 3.38, p<0.01, respectively). Stratified models corroborated the findings of the main models for the outcomes of LOS, intraoperative complication (online supplementary table 6 / SDC 6), and cardiac complication (online supplementary 7 / SDC 7), with the exception that in a population of black patients those with Medicaid were not expected to have longer LOS than those with non-Medicaid insurance (IRR: 1.05, 95% CI 0.96 to 1.14; online supplementary table 5 / SDC 5).
Supplemental material
Supplemental material
Supplemental material
Supplemental material
In a sample of elective procedures only, Medicaid patients were not more likely than patients with non-Medicaid insurance to experience an infectious complication during their hospital stay (OR: 1.83, 95% CI: 0.94-3.56) (unpublished data).
Discussion
Our study’s results support our hypothesis that among patients who undergo inpatient TSA, RSA, and PSA, those with Medicaid insurance have poorer outcomes compared with the population of patients with private insurance, other forms of insurance, and Medicare. Results from multivariate models showed that those with Medicaid had a four fold increase in the odds of inpatient mortality, as well as increased odds for 30-day and 90-day readmissions, infectious complications, and longer LOS compared with those patients with types of insurance other than Medicaid. Given the demographic diversity of selected states studied, the overall number of patients in our sample, and the ability to control for patient variables (including insurance payer, race/ethnicity, gender, age, comorbidities, median income of the patient’s residential location, procedure type, and hospital factors), our results are robust.
Our results are consistent with previously described findings within the field of orthopedics, which show that insurance payer status impacts an array of perioperative outcomes; differences in complication rates, LOS, readmission, and mortality have all been associated with insurance payer status.5 12 13 Browne et al 22 examined patients undergoing hip or knee arthroplasty from 2002 until 2011 using the HCUP Nationwide Inpatient Sample (NIS) database and found that patients insured by Medicaid, compared with non-Medicaid patients, had increased odds of in-hospital infection (OR: 1.7, 95% CI 1.3 to 2.1) and increased mean LOS (3.7 days vs 3.3 days, p<0.01). Notably, there was no difference found in mortality between the two groups.22 Xu et al retrospectively examined 295 572 patients from 2007 through 2011 in the HCUP SID who underwent total hip replacement and found that patients insured by Medicaid had an increased risk of mortality (OR: 2.25, 99% CI 1.01 to 5.01) as well as increased rate of complications, increased 30-day and 90 day readmissions, and longer LOS as compared with private insurance.12
However, there are limited data available on the effects of insurance provider on outcomes after shoulder arthroplasty. Li et al,5 in a retrospective study of 103 290 patients who underwent shoulder arthroplasty from 2004 to 2011, found that patients who had Medicare, Medicaid, or no insurance had significantly higher rates of medical and surgical complications than matched controls with private insurance.5 Privately insured patients had an overall complication rate of 10.5%, which was significantly lower than complication rates of 16.9% and 20.3% for Medicaid/uninsured patients and Medicare patients, respectively. Overall mortality was not found to be associated with insurance status. While the overall complication rates in our study were lower for privately insured patients as well as Medicaid patients (4.5% and 6.7%, respectively), Li et al 5 included any adverse surgical event as a complication, which may explain their overall higher complication rate.5 Additionally, Li et al 5 conflated Medicaid enrollees with being uninsured; we differentiated between these two groups.
While our main findings show a clear effect of insurance payer as an independent predictor of our outcomes, we acknowledge that the relationship between insurance payer and outcomes is potentially mediated by many preoperative comorbidities and demographic characteristics (as highlighted by our stratified analysis). Research has shown that Medicaid patients have worse perioperative outcomes after shoulder arthroplasty due to a multitude of interrelated reasons: racial minorities are often discriminated against, which limits their access to expedient and comprehensive care23; intraoperatively, racial minorities are at an increased risk of adverse events related to patient safety, which speaks to subtle institutional biases that may be ingrained into healthcare processes.24 In addition, patients insured by Medicaid oftentimes have more preoperative comorbidities than patients insured by other groups. A recent analysis comparing Medicaid patients with privately insured patients found that when controlling for age, Medicaid patients were significantly more likely to have diabetes, chronic obstructive pulmonary disease, peripheral vascular disease, and other diseases.25
Prior studies have also demonstrated less resource utilization, decreased preoperative screening, and delayed diagnosis in populations of Medicaid patients primarily due to cost.26 As such, it is plausible that less access to preventative and preoperative care places the burden on the hospital admission, where resource utilization is often greater and more costly to manage higher rates of perioperative morbidity.22
Inferior surgical outcomes for those insured by Medicaid are not isolated to the field of orthopedics, as multiple previous studies have demonstrated increased mortality and morbidity in other surgical specialties based on payer status. LaPar et al 27 in 2010 identified 893 658 patients from 2003 through 2007 using the NIS database who underwent a variety of major inpatient surgeries (including total hip replacement) across multiple surgical subspecialties. They found that those insured by Medicaid had increased risk of mortality compared with privately insured patients (OR: 1.97, 95% CI 1.84 to 2.10), as well as longer LOS, increased complications, and increased total costs.27
Our study expands on previous studies due to the inclusion of nearly 90 000 more recent shoulder arthroplasty cases derived from a diverse study population present in the states analyzed. Our analysis encompasses records from disparate regions of the country (California, Florida, Kentucky, Maryland, and New York) whose populations collectively encompass more than a quarter of the entire population of the USA.28 Moreover, our study accounts for confounding introduced by preoperative comorbidities and demographic factors. Results from stratified analyses revealed effect modification, with significant differences in outcomes between Medicaid and private insurance groups in select demographic groups only.
Nonetheless, there are several important limitations to our study. First, our data were obtained from a large administrative database using diagnostic and procedure codes. Our data are limited by the accuracy and detail of coded patient hospital stays and by data not captured by available measures (eg, access to providers, timing from diagnosis to treatment, and medications, among others).29 Second, as with any retrospective study, our data are subject to misclassification. A third limitation is that our data set only captured inpatient claims, which excludes shoulder arthroplasties performed in the ambulatory setting. Although inpatient TSA represents a majority of TSAs performed in the USA (97% inpatient from 2005 to 20124), the incidence of outpatient TSA is on the rise and represents a potential for significant cost savings.2 Future studies should examine healthcare disparities in outcomes and utilization of outpatient versus inpatient shoulder arthroplasty.
Outcomes disparities in shoulder arthroplasty based on payer in the USA have significant implications for the overall health and economic welfare. With annual healthcare spending reaching approximately $3.2 trillion in 2015 (17.8% of gross domestic product), closing the gap in healthcare disparities represents a potential for cost savings.30
Conclusions
In conclusion, this study provides further evidence of disparities in perioperative outcomes after shoulder arthroplasty based on patients’ primary payer status. Specifically, Medicaid patients have an increased likelihood of inpatient mortality, readmission, infectious complications, and longer LOS compared with privately insured patients. Disparities in outcomes and access to healthcare are pervasive throughout all fields of medicine; identification of and discussion about these results are the first steps in correcting these outcomes and improving healthcare within the USA.
References
Footnotes
Presented at This work was presented in part at the American Society of Regional Anesthesia and Pain Medicine World Congress, April 2018, in New York City.
Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.
Competing interests None declared.
Patient consent Not required
Ethics approval The study was approved by the Institutional Review Board at Weill Cornell Medical College.
Provenance and peer review Not commissioned; externally peer reviewed