Schizophrenia and potentially preventable hospitalizations in the United States: a retrospective cross-sectional study

Hospital discharge data were obtained from the Nationwide Inpatient Sample (NIS), the largest all-payer inpatient care database in the US, comprised of about 1,000 hospitals across 35 states, with stratified sampling designed to represent all US census regions. The NIS is stratified by geographic region (Northeast, Midwest, West, or South), location (urban or rural), teaching status (teaching or non-teaching), ownership (public, private not-for-profit, or private investor-owned), and bed size (small, medium, or large). It contains data on age, sex, race, length of stay, and ICD-9-CM codes for up to 15 diagnoses and 15 procedures. Discharges are linked to information on hospital characteristics from the American Hospital Association’s Annual Survey Database.

The outcomes of interest were ACS hospitalizations. Based upon work by Billings [30], the Agency for Healthcare Quality and Research (AHRQ) developed a set of indicators called the Prevention Quality Indicators (PQIs) to identify ACS hospitalizations. We used AHRQ’s PQI software [31] to create indicators of four acute and nine chronic ACS hospitalizations. Acute ACS hospitalizations included those with primary diagnoses of bacterial pneumonia, urinary tract infection, dehydration and perforated appendix [31]. Chronic ACS hospitalizations included hospitalizations with primary diagnoses of congestive heart failure (CHF), chronic obstructive pulmonary disease (COPD), asthma, diabetes mellitus long and short term complications, hypertension, angina without procedure, diabetes related lower-extremity amputation, and uncontrolled diabetes [31]. Composite overall, acute, and chronic ACS hospitalizations were also measured.

The primary exposure of interest was secondary diagnosis of schizophrenia (ICD-9-CM code 295). Potential confounders were included in the model based on previous research in this field, differences found in Table 1, and so those that remained significant after adjusting for other covariates for the vast majority of PQI outcomes. Measured covariates included age, sex, race, insurance type (uninsured hospitalizations are included in the ‘other’ category), median household income quartile for patient zip-code, hospital region, hospital teaching status, urban/rural designation, hospital bed size, hospital volume, admission source, hospitalization characteristics (elective vs. non-elective, medical versus surgical, against medical advice, and death in hospital) and 12 medical co-morbidities (Table 1). As race data was missing for nine states, missing race was included as a unique category in the race variable. Measures of medical comorbidity were generated using AHRQ’s Clinical Classifications Software [39].

The study sample included hospital discharges with a primary medical or surgical diagnosis for years 2003–2008 among adults aged 18–64 years. As universal Medicare coverage for persons aged 65 years and older may affect the likelihood of ACS hospitalization among older adults, this age group was excluded from analysis. Hospitalizations included in the sample reflect the risk set for a particular ACS hospitalization, as defined by AHRQ. For example, the risk set for perforated appendix includes hospitalizations with a primary diagnosis of appendicitis [40]. To protect patient confidentiality, the NIS does not have identifiers to link hospitalizations to unique individuals. The Johns Hopkins Institutional Review Board deemed that this study met criteria for exemption.

Differences in nationally weighted proportions of patient, hospitalization, and hospital characteristics among hospitalizations with and without secondary schizophrenia diagnosis were tested using Wilcoxon rank sum tests for continuous variables and chi-square tests for categorical variables (Table 1). We also determined the nationally weighted proportions of ACS hospitalizations stratified by age group and secondary diagnosis of schizophrenia (Table 2).

To test the association between schizophrenia and ACS hospitalizations, we developed logistic regression models to obtain odds ratios (ORs) comparing the likelihood of hospitalization for an ACS condition in admissions with versus without secondary diagnosis of schizophrenia. Odds ratios were calculated for the four acute and nine chronic ACS hospitalizations identified by AHRQ, as well as for composite overall, acute, and chronic ACS hospitalizations. Models adjusted for the covariates listed above (Figure 1). Analyses accounted for the NIS sampling scheme by nesting within stratum and hospital and incorporated discharge weights. While analyzing subpopulations, we included all individuals outside the subpopulation to create the correct standard errors. Since the level of observation was hospitalization and multiple hospitalizations per patient were likely, survey design based variance estimators were used to account for within-respondent correlation. Since a year interaction term did not affect results, years 2003 to 2008 were combined. All tests were two-sided and p-values were considered significant at the 0.05 level. Analyses were conducted with SAS 9.2 and SAS-Callable SUDAAN 10 software (Research Triangle Institute).

The study sample included 182,423 and 15,092,914 medical and surgical hospitalizations for adults with and without schizophrenia, respectively (Table 1). Hospitalizations with a secondary diagnosis of schizophrenia were more likely to be non-white, have a low-income zip code of residence, be discharged from a teaching or low-volume hospital, and be emergency, medical (versus surgical), or non-elective admissions. Over 82% of admissions had either Medicaid or Medicare (among adults aged 18–64 years, Medicare coverage indicates receipt of Social Security Disability Insurance). Hospitalizations with schizophrenia had higher proportions of several co-morbid medical conditions, including diabetes mellitus and substance abuse.

Table 2 shows the proportions of medical and surgical hospitalizations accounted for by ACS hospitalizations. Unadjusted ORs of overall, acute, and chronic ACS hospitalizations were higher for admissions with secondary diagnosis of schizophrenia compared to those without. Unadjusted ORs of all four individual acute and six of the nine chronic ACS conditions measured, including CHF, COPD, asthma, diabetes mellitus long and short-term complications, and uncontrolled diabetes were higher for hospitalizations with versus without secondary diagnosis of schizophrenia. The proportion of ACS hospitalizations among medical and surgical hospitalizations was highest for perforated appendix, reflecting this indicator’s small risk set that includes only admissions with appendicitis.

In a nationwide US sample, schizophrenia was associated with 17% increased odds of any ACS hospitalization (Figure 1). Schizophrenia was associated with increased odds of acute ACS hospitalization (OR 1.34, 95% CI 1.31-1.38). Schizophrenia was also associated with increased odds of chronic ACS hospitalization (OR 1.04, 95% CI 1.01-1.06), however this result appears to be driven by the heightened odds of hospitalization for four chronic ACS conditions: COPD (OR 1.52, 95% CI 1.46-1.59), asthma (OR 1.14, 95% CI 1.08-1.20), diabetes mellitus short-term complications (OR 1.42, 95% CI 1.33-1.50), and uncontrolled diabetes (OR 2.34, 95% CI 2.20-2.50). In contrast, schizophrenia was associated with decreased odds of hospitalization for congestive heart failure, diabetes mellitus long-term complications, angina without procedure, and diabetes related lower-extremity amputation. Schizophrenia was not associated with hospitalization for hypertension.

Our study has several limitations. The NIS data does not link hospitalizations to individuals, so there may have been more than one hospitalization per person. While ACS hospitalizations have been shown to be associated with poor access to high quality primary care at the population level, [32] this is only one possible cause of ACS hospitalizations. The ACS hospitalization indicators were designed to signal potential areas of poor access to high quality care that need further investigation; while we have suggested possible explanations for the patterns observed, the indicators are not designed to identify underlying causes and our national data may mask differences by region or health system. Our results therefore suggest overall patterns in the association between schizophrenia and potentially preventable hospitalizations in the US. Measurement error, resulting from unidentified or unreported diagnoses of schizophrenia, PQI indicators, or medical comorbidities may bias results. Residual confounding from medical illness not reported during hospitalization for patients with schizophrenia could help explain the association between schizophrenia and ACS hospitalization, and our administrative database was unable to capture measures of disease severity or lifestyle factors, such as smoking, that may increase susceptibility to ACS hospitalizations. Nonetheless, our study was the first to report an association between schizophrenia and ACS hospitalizations in a nationally representative US sample, as well as the first to describe differences in this association across various acute and chronic ACS conditions.