Background: Valid comparative effectiveness research (CER) will require large, electronic health record (EHR) databases and analytics that overcome unmeasured confounding. Unmeasured confounding applies to undetected factors that influence the results of nonrandomized studies in a fashion that leads to incorrect conclusions. Through our prior research, we developed the Prior Event Rate Ratio (PERR) analysis, which overcomes unmeasured confounding in nonrandomized studies by comparing treated with untreated patients before treatment begins. It has been validated by subsequent empirical studies, as well as theoretical statistics. However, an important limitation of PERR is its inability to analyze mortality (since prior events cannot occur).
- Assess an extension of PERR that overcomes unmeasured confounding for the mortality rate.
- Apply this strategy to cardiovascular outcomes in order to enhance PERR’s validity.
- Perform PERR studies to test kidney failure progression.
- A new mortality rate analysis will use the EHR database studies employed in the development of the PERR method. We will expand on our earlier research with PERR to examine what transpires after a study concludes, when both the treated and untreated research participants are no longer receiving medication. We have named this new strategy the Post Study Event Rate Ratio (PSERR). Our preliminary studies suggest the PSERR strategy will work, but rigorous evaluation is required.
- The PSERR approach will be applied to cardiovascular outcomes analyzed previously with PERR. The PSERR results will be compared to the PERR results and also used to adjust study outcomes. This will further establish PERR method validity.
- Empirical examination of progression of renal disease, a continuous outcome, will use the same strategy as our earlier studies by comparing The Health Improvement Network database study outcomes to previously performed randomized controlled trials.
Patient Outcomes: Randomized controlled trials are the gold standard for valid outcomes research but apply only to the specific population studied (i.e., results may not be generalizable). If valid CER using large EHR databases is feasible, the impact will be considerable. Efforts to advance the validity of large EHR database outcomes studies by overcoming unmeasured confounding can have a transformative effect on future health care.