External adjustment for unmeasured confounders improved drug-outcome association estimates based on health care utilization data

Objectives: Health care utilization (HCU) databases are widespread sources of data for pharmacoepidemiologic investigations. Possible confounders are typically not measured in such databases. We show how to assess the impact of confounders in a study aimed at comparing cardiovascular (CV) risk according to drug regimen prescribed at starting antihypertensive therapy, nominally one agent (monotherapy) or a combination of agents in a unique tablet (fixed-dose combination) or in at least two distinct tablets (extemporaneous combination). Study Design and Settings: A nested case-control study was carried out by including the 209,650 patients from Lombardy (Italy) newly treated between 2000 and 2001. Cases were the 10,688 patients who were hospitalized for CV disease until 2007. Three controls were selected for each case. Logistic regression was used to model the CV risk associated with initial therapeutic regimen. A Monte Carlo sensitivity analysis was performed for accounting unmeasured confounders (hypertension severity and chronic disease score) by means of external adjustment with medical record (MR) data. Results: Compared with patients on fixed-dose combination, those on extemporaneous combination or monotherapy, respectively, had CV risk increased to 15% (95% confidence interval [CI]: 3%, 29%) or 17% (95% CI: 8%, 26%). External adjustment did not modify the risk associated with monotherapy. In contrast, the excess of risk associated with extemporaneous combination was annulled when external adjustment was applied. Conclusion: MR data can be used to assess confounding bias unmeasured from HCU database. Starting antihypertensive therapy with a combination of agents probably reduces the CV risk with respect to monotherapy, even in the setting of primary prevention.