Using clinical databases to evaluate healthcare interventions

Objectives: The aim of this study was to test the feasibility of conducting rigorous, nonrandomized studies (NRSs) of healthcare interventions using existing clinical databases in terms of the following: recruiting a large representative sample of hospitals, identifying eligible cases, matching cases to controls to achieve similar baseline characteristics, making meaningful comparisons of outcomes, and carrying out subgroup analyses. Methods: Data were extracted from the Intensive Care National Audit & Research Centre's Case Mix Programme Database to investigate the impact of management with a pulmonary artery catheter (PAC) in intensive care unit (ICU) patients. Participating ICUs were invited to collect additional data for the analysis. Patients managed with a PAC were matched to control patients on their propensity score. Hospital mortality was then compared between the two groups. Results: Of 117 eligible ICUs, 68 (58 percent) agreed to participate, of which 57 (84 percent) collected additional data. Although a slightly higher proportion of larger ICUs in university hospitals participated, the patient case-mix was similar to that in nonparticipating ICUs. Almost all patients managed with a PAC (98 percent) were successfully matched to patients managed without a PAC. The two groups were similar for baseline characteristics. However, hospital mortality was worse for PAC patients than for non-PAC patients (odds ratio, 1.28; 95 percent confidence interval, 1.06-1.55). Subgroup analysis suggested that the impact of management with a PAC was modified by severity of illness. Conclusions: Rigorous NRSs are feasible if they are based on data from high-quality clinical databases. However, the reliability of estimated treatment effects from such studies requires further investigation.