(151) Validation of Medical Code Lists for Defining Mental Health Conditions with Primary Care Data: A Mixed Qualitative and Quantitative Study

Publication Number: 482

Background: Defining mental health (MH) conditions with data collected from primary care is challenging. Researchers have addressed this by using validated algorithms and code lists. Algorithms detect and integrate different data types (e.g., diagnosis codes and drug treatment records) to define if the patient has an MH condition. A list of medical codes (e.g., diagnoses and symptoms) usually supplies these algorithms to improve their accuracy. The code lists validation is frequently a qualitative process (e.g., clinical review) and can demand additional external information, increasing costs, logistics and data management.

Objectives: To validate code lists for defining common MH conditions with primary care data, using a mixed qualitative and quantitative approach and without recurring to external data.

Methods: Using a mixed approach, we validated a Read codes list created to define common MH conditions in patients registered in the UK primary care system (The Health Improvement Network (THIN) data). The goal was to select valid Read codes to retain in the list and classify these codes in clusters defining MH conditions. The qualitative step included an in-depth revision of the codes by a panel of six doctors (3 general practitioners and 3 psychiatrists). Simultaneously, the quantitative step included different statistical analyses on THIN data (i.e., exploratory factor analysis (EFA), cluster analysis) for reaching the same goal. The statistical results informed the preliminary qualitative conclusions (e.g., correspondence analysis), generating a final selection and classification. Finally, we contrasted the classification against appropriate drug treatment to verify consistency.

Results: From a preselected list of 2,007 Read codes, a total of 1,638 were selected by all doctors. Later, they agreed on classifying these codes into 12 categories of common MH disorders. From the same preselected list, a total of 1,364 were quantitatively selected. Using data from 497,649 persons who used these Read codes at least once, we performed the EFA, retaining five factors (5 categories) from the unrotated factorial solution. Both classifications showed good correspondence, informing a final classification strongly correlated with drug treatment, for example, for severe mental illnesses and cognitive decline.

Conclusions: We produced a comprehensive set of medical codes lists for 12 common MH conditions validated by a combination of clinical consensus panel and quantitative cluster analysis with cross-validation. The mixed qualitative and quantitative approach maximises benefits from internal data, avoiding extra costs and logistics from external data collection/recovery in validating MH code lists.