Development and validation of a claims-based prediction model for COPD severity.

Background: Administrative claims are an important data source for COPD research but lack a validated measure of patient COPD severity, which is an important determinant of treatment and outcomes.

Methods: Patients with ≥1 diagnosis of COPD and spirometry results from 01/2004-05/2011 were identified from an electronic health records database linked to healthcare claims. Patients were classified into 3 COPD severity groups based on spirometry and Global Initiative for Chronic Obstructive Lung Disease (GOLD) guidelines: GOLD-Unclassified, Mild/Moderate, and Severe/Very Severe. A multinomial logistic regression model was constructed using claims data from 3 months before and after (observation period) the most recent spirometry (index date) to categorize patient COPD severity. A random selection of 90% of patients in each severity level was selected to build the model, and the remaining 10% were used as a validation sample. Model predictions were evaluated for sensitivity, specificity, accuracy, and concordance.

Results: Among 2028 COPD patients who met sample selection criteria, 886, 683, and 459 patients were in the GOLD-Unclassified, Mild/Moderate, and Severe/Very Severe categories, respectively. The final model included age, sex, comorbidities (such as pulmonary fibrosis and diabetes), COPD-related resource utilization (such as oxygen use), and all-cause healthcare utilization. In the validation sample, the model correctly predicted COPD severity for 62.7% of all patients (accuracy for predicting GOLD-Unclassified: 73.5%; Mild/Moderate: 70.6%; Severe/Very Severe: 81.4%) with kappa = 0.41.

Conclusions: The prediction model was developed using clinically measured COPD severity to provide researchers an approach to classify patients using claims data when clinical measures are not available.