Abstract
Background: Large administrative databases are increasingly used to identify patients with chronic conditions, however the best methodology for Chronic Obstructive Pulmonary Disease (COPD) is still debated.
Objective: To develop and validate a predictive model to identify patients with COPD in Lazio region (2,625,102 residents over 45) linking clinical and administrative data.
Methods: From regional hospitalizations and drug prescriptions, through record linkage, we identified patterns of specific drug use (minimum 2 prescription during 12 months) and COPD hospitalizations during a 9-year period in 428 patients with COPD, who attended an outpatient clinic in 2006, and in 2140 people without COPD (selection from outpatients' specialized health care registry). Through a Bootstrap-Stepwise procedure we analyzed COPD associated factors. We validated the algorithm through internal (cross-validation-bootstrap, jack-knife) and external validation (comparison with COPD patients with confirmed diagnosis).
Results: Prevalence of COPD was 7.8%. Factors associated with COPD were prescription of beta2-agonists, anticholinergics, corticosteroids, oxygen, previous hospitalization for COPD and respiratory failure. For each patient we estimated an expected probability to suffer from COPD. Depending on the cut-point of expected probability, sensibility ranged from 74.5 to 99.6%, specificity from 37.8 to 86.2%. We defined a cut-point of 0.30 to identify COPD patients. Applying our algorithm on external COPD patients we succeeded to identify 70%.
Conclusion: The predictive model showed good performance to identify COPD patients confirming the strength of administrative data for monitoring chronic diseases.
- © 2011 ERS