A Bayesian Approach to Modeling Risk of Hospital Admissions Associated With Schizophrenia Accounting for Underdiagnosis of the Disorder
underdiagnosis, Bayesian analysis
Background/Aims: Schizophrenia is a debilitating serious mental illness (SMI) characterized by a complex array of symptoms. Patients tend to seek treatment only intermittently, contributing to difficulty in diagnosing the disorder. A misdiagnosis may potentially bias and reduce the validity of a study based on recorded diagnoses. It also may impact patient outcomes by delaying receipt of appropriate care. Thus, we present a statistical model to compare the odds of 1-year hospitalization among patients with schizophrenia versus patients with or without other SMIs when schizophrenia is underreported in administrative databases.
Methods: A retrospective study design examined patients seeking care during 2010 in one of 19 care-and-coverage health systems across the United States comprising the Health Maintenance Organization Research Network (HMORN). Bayesian analysis was applied to address the problem of underdiagnosed schizophrenia using a statistical measurement error model that explicitly evaluated the impact of varying assumptions about the extent of underreporting. Results were then compared to a classical multivariable logistic regression model.
Results: Among 87,806 patients, 7.3% had a SMI, including 114 (1.3 per 1,000) diagnosed with schizophrenia. Admission was greatest among patients with schizophrenia (14%), followed by other SMIs (12–13%) and non-SMI patients (8%, P<0.01). Assuming no underreporting, there was an 87% greater relative odds of admission associated with schizophrenia (odds ratio: 1.87; 95% confidence interval: 1.08–3.23). In the Bayesian approach, assuming varying sensitivities, effect sizes were 2–3% lower with credible interval smaller (reduced by 2–4%) than that observed with the classical approach.
Discussion: A delayed diagnosis of schizophrenia can lead to inappropriate treatment and symptom exacerbation, increasing the risk of hospitalization. In the Bayesian approach, reduced association between hospitalization and schizophrenia, as well as other SMIs, was uniformly observed across varying rates of underdiagnosing, meaning prompt diagnosis should lead to lower rates of hospitalization. Although effect sizes may vary across health care systems, the analytical approach has useful applications in other contexts where the identification of patients with a given condition may be subject to underreporting in administrative records.
Stock EM, Stamey JD, Zeber JE, Thompson AW, Copeland LA. A Bayesian Approach to Modeling Risk of Hospital Admissions Associated With Schizophrenia Accounting for Underdiagnosis of the Disorder. J Patient Cent Res Rev 2015;2:139. http://dx.doi.org/10.17294/2330-0698.1191