Can valid cases of schizophrenia be identified in administrative claims data?

Document Type

Conference Proceeding

Publication Date


Publication Title

Pharmacoepidemiology and Drug Safety


Background: Large data sources, such as administrative claims, can be used to better understand the natural history, treatment and outcomes of schizophrenia provided that valid cases can be identified. International Classification of Diseases (ICD) codes or a combination of ICD codes and prescription claims have been used to identify schizophrenia patients, but validation studies of these methods for schizophrenia are limited.

Objectives: To determine if valid cases of patients with schizophrenia can be identified using administrative claims data.

Methods: Claims data from the Henry Ford Health System, an integrated healthcare system serving metropolitan Detroit, Michigan, were used to identify patients aged 18-64 years with schizophrenia from 01/ 01/2009 to 06/30/2014. Potential cases had ≥2 ICD-9 codes (295.x) for schizophrenia disorder in any position, ≥2 claims for an antipsychotic medication, ≥12 months of continuous enrollment pre-index, and ≥6 months of continuous enrollment post-index. Index date was defined as the first 295.x ICD-9 code. Patients with organic cognitive decline or schizoaffective disorder independent of schizophrenia were excluded. Trained medical records abstractors performed a structured review of all relevant fields including inpatient and outpatient records of the electronic medical record (EMR) (e.g. diagnosis fields; free text) to verify the schizophrenia diagnosis ±12 months from the index date.

Results: Of the 145 patients who met inclusion/exclusion criteria, EMR review was completed on a random sample of 111 patients. Of these, 65 had an EMR-confirmed diagnosis of schizophrenia for a positive predictive value (PPV) of 59% (95% confidence interval: 52-64%). Unconfirmed patients had diagnoses of bipolar disorder (N = 25; 54%), major depressive disorder (N = 28; 61%), and/or schizoaffective disorder (N = 3; 7%). These diagnoses may be comorbid with a schizophrenia diagnosis, but no schizophrenia diagnosis was recorded.

Conclusions: Identifying valid cases of schizophrenia in administrative claims data is challenging. There are few published studies of validated claims-based algorithms that identify cases of treated schizophrenia. This study, requiring ≥2 ICD-9 codes and ≥2 prescription claims, did not yield a high PPV for schizophrenia. Reasons may include diagnostic challenges in differentiating psychiatric conditions or comorbid diagnoses where only 1 diagnosis is recorded. Future studies of validated algorithms to identify schizophrenia patients are warranted.




Supplement 2

First Page