Diabetes Prevalence Among Racial-Ethnic Minority Group Members With Severe Mental Illness Taking Antipsychotics: Double Jeopardy?

This retrospective cohort study examined administrative, pharmacy, and billing data from California Medicaid (Medi-Cal) and the Client and Service Information (CSI) system from January 1 to December 31, 2009. The CSI system utilizes encounter-based data to track state- and county-funded outpatient mental health service use in California. After obtaining approvals as previously described (9), the California Department of Health Care Services combined these databases and provided a unique deidentified data set to investigators.

The following criteria characterized the cohort (N=19,364): age ≥18 and <68, utilized community mental health services, prescribed an antipsychotic medication at least once during the study period, enrolled in Medi-Cal, not dually eligible for Medicare, and either screened for diabetes (glucose-specific fasting serum test [CPT codes 82947, 82948, 82950, or 82951] or hemoglobin A1c (CPT code 83036]) or had preexisting diabetes during the study period. The dually eligible population was excluded because Medicare laboratory billing data were unavailable. Of note, diabetes screening guidelines for anyone taking second-generation antipsychotic medications were distributed to health plans before the study period (6). However, there was variable enforcement of the guidelines and no formal policies on diabetes screening of antipsychotic-treated adults.

The primary outcome measure was diabetes, defined by a diabetes diagnosis (ICD-9 codes 249.xx, 250.xx, 253.xx, and 648.xx) or an antidiabetic prescription claim during the study period. [A table in an online supplement to this report lists the antipsychotic classes and specific medications.] The database included additional variables, such as age, gender, race-ethnicity, county type, psychiatric diagnosis, and substance use disorder diagnoses. Individuals with multiple psychiatric comorbidities were classified hierarchically, and counties were dichotomized into rural or urban, as previously described (9).

Descriptive statistics were used to show the distribution of demographic and clinical characteristics by diabetes status. To facilitate comparisons with the general population, we estimated adjusted diabetes prevalence by using direct standardization to the distribution of the adult U.S. population by age, sex, and race-ethnicity on the basis of U.S. Census Bureau data for 2009. In this procedure, overall diabetes prevalence was estimated by a weighted average of sample diabetes prevalence within each age, sex, and racial-ethnic stratum, with weights given by the proportions of the overall U.S. population accounted for by each subgroup.

Adjusted Poisson regression models were used to estimate the overall effects on prevalence of diabetes of gender, race-ethnicity, age, county type, psychiatric diagnosis, substance use diagnosis, and health care utilization. We used a directed acyclic graph to summarize hypothesized causal relationships between variables in order to distinguish the confounders and mediators of each predictor. Models specific to each predictor of interest were used to estimate overall effects, by adjusting for variables that influence the predictor variable and diabetes (confounders) but excluding factors on the causal pathway between them (mediators). Models excluding confounders would provide biased estimates, whereas models including mediators would at best estimate direct effects via other pathways. Poisson regression was used as the best approximation of prevalence ratio. Because California delegates delivery of mental health services to its counties, we used robust standard errors to account for clustering of outcomes by county and to accommodate use of a Poisson model for a binary outcome. As previously described, seven rural counties were grouped, and one county was excluded (9).

Given the low diabetes screening rates we have previously documented (9), estimating diabetes prevalence in the screened sample could have overestimated prevalence, because patients with diabetes risk factors, some unmeasured, were almost surely overrepresented the screened sample. To address this, we used inverse weighting based on available covariates to remove some but probably not all of the selection bias. In addition, to provide a conservative lower bound, we estimated prevalence by using the entire sample, which we recognize is biased low by omission of cases among the unscreened [see online supplement].

The prevalence of diabetes in this sample was 32.0% (N=6,195 of 19,364). Table 1 shows adjusted differences in prevalence on the basis of patient characteristics. Compared with the rate among white patients, diabetes prevalence was higher among black patients (adjusted prevalence ratio [APR]=1.24) and Hispanic patients (APR=1.28) (p<.001). In addition, compared with the rate among patients with schizophrenia, diabetes prevalence was significantly lower among patients with bipolar disorder (APR=.88) and major depressive disorder (APR=.83) (p<.001).

After direct standardization to the demographic distribution of the U.S. population, the adjusted prevalence was 27.3%. Differences in prevalence were found by race-ethnicity, with whites having the lowest prevalence (26.5%) and other minority groups having higher rates (blacks, 29.7%; Hispanics, 29.7%; Asians, 27.2%; and other, 28.4%).

With inverse probability weighting to reduce selection bias captured by measured factors (9), the estimated overall prevalence of diabetes was 27.8% [see online supplement]. The standardized lower-bound estimate of diabetes, which was based on the larger parent study that included close to 70% who were unscreened for diabetes, was 9.6% (9).

Our study filled a gap in the literature by documenting a significantly higher risk of diabetes among racial-ethnic minority group members with severe mental illness, compared with their white counterparts, even after the rate was standardized to the U.S. population and the analysis controlled for confounding variables. Our results confirm findings from smaller studies that certain racial-ethnic minority groups with severe mental illness have higher prevalence of diabetes than whites (7). This high diabetes prevalence is concerning because health care disparities are well documented for both racial-ethnic minority populations (10) and populations with severe mental illness (11), making this subpopulation at especially high risk for disparities in care (7,8).

Our study confirms prior findings of high prevalence of diabetes among people with severe mental illness (3). The standardized and adjusted prevalence of diabetes among people with severe mental illness taking antipsychotic medications in California was estimated to be 27.3%, or about 2.5 times that of the general U.S. population (10.2%), according to NHANES data during that time period (2009) (4).

The main limitation was that we could assess prevalence only among those who were tested for diabetes and who had ICD-9 diabetes diagnoses or claims for diabetes medications; thus our findings may overestimate true prevalence, primarily because patients at high suspicion of diabetes are almost surely preferentially screened. In contrast, lack of screening almost certainly leads to missed cases of diabetes, because 25%−50% of all patients with type 2 diabetes in the United States are undiagnosed (12). Because dually eligible (Medicaid and Medicare) patients were excluded, diabetes prevalence may have been underestimated. In addition, we did not have data on antipsychotic medication exposure or the timing of exposure relative to diabetes diagnosis.

Future studies should examine both diabetes incidence and prevalence longitudinally among diverse populations with severe mental illness taking antipsychotic medications. Examining potential racial-ethnic differences more thoroughly may help stratify risk levels within certain high-risk subpopulations. Although accurate understanding of risk is clearly important, recent findings indicate that use of the U.S. Preventive Services Task Force screening guidelines identifies diabetes in only about half of those served in community health centers (13). A large study should be conducted to systematically screen all people with severe mental illness served in community mental health settings to obtain an accurate assessment of missed diagnoses.

Culturally and linguistically appropriate integration models are needed to address the health care needs of minority populations with severe mental illness. As a first step, we recommend systemwide efforts to improve diabetes screening and care for all people with severe mental illness, because such efforts can be effective in reducing disparities. Unfortunately, systemwide efforts are complicated, because the public mental health and general medical care systems often operate in separate digital, financial, operational, and structural silos. Many people with severe mental illness are publicly insured and receive care in community mental health clinics (14). Therefore, we believe that integrating primary care into these public mental health settings may improve health screening in this population. Regrettably, evidence-based models to integrate primary care into behavioral health care have not yet been developed (15).

The prevalence of diabetes among people with severe mental illness taking antipsychotic medications was significantly higher than in the general population. Racial-ethnic minority group members with severe mental illness taking antipsychotic medications had higher prevalence of diabetes compared with whites. Because racial-ethnic minority populations with severe mental illness are extremely vulnerable to health care disparities, they deserve particular public health and clinical attention.

This study was initiated during a state quality improvement project to integrate primary care and mental health care, called the California Mental Health Care Management Program (CalMEND). The project was a collaboration between the California Department of Mental Health and the Pharmacy Benefits Division of the California Department of Health Care Services. The authors acknowledge John Igwe and other CalMEND staff for their assistance in combining administrative databases without compensation and Martha Shumway, Ph.D., for initial coding of the data. They also thank Penelope Knapp, M.D., for assistance in understanding how the California Department of Mental Health’s policies and procedures during the study period may have influenced findings. The authors also thank Nicholas Riano, B.A., for assistance in manuscript preparation.