Observational studies using clinical databases, such as the United Kingdom's General Practice Research Database (GPRD), may provide an alternative to clinical trial data for detecting longitudinal changes in blood pressure due to drug exposures that vary over time. Blood pressure data which are measured at variable intervals and are often missing present a particular methodological challenge to the analysis of such studies.Methods
To assess effects on blood pressure, we extracted from the GPRD several cohorts of new drug users of warfarin (n=21 532), ibuprofen (n=92 037), proton pump inhibitors (n=153 695), statins (n=118 704), rofecoxib (n=6399), and celecoxib (n=6217) from 2001 to 2003. Several blood pressure readings were missing either before or after initiating therapy. We compared the results of analyses using a linear mixed model with a pre-post quasi-experimental design, using the multiple imputation approach to account for missing data.Results
There was evidence that the missing blood pressure data were not missing completely at random as subjects with more blood pressure readings tended to have higher recorded values. For statins, the mixed model estimated a change in systolic blood pressure of −3.80mmHg (99% confidence interval (CI): from −3.97 to −3.63), similar to the quasi-experimental model and to the −4.00mmHg estimated from clinical trials. Sensitivity analyses indicate that these estimates are robust. For rofecoxib, the change in systolic blood pressure were 2.20mmHg (99%CI: 1.09–3.32) and 1.21mmHg (99%CI: 0.21–2.22) for the two methods, respectively, again confirming the findings of randomized trials.Conclusion
With appropriate statistical techniques, GPRD blood pressure data can be used to estimate blood pressure changes secondary to drug therapy. Copyright © 2008 John Wiley & Sons, Ltd.