Background: The decision to adopt a treatment is complex, and the doctor, patient, and parents/guardians interact and consider various factors. Some of these factors may predict better or worse patient outcomes regardless of whether the treatment is effective. This problem is known as confounding. Confounding, when present, may complicate explaining the relationship between the outcome and the treatment, as other factors may be responsible for the relationship. Confounding is often present when treatments are selected rather than randomly assigned. Thus confounding, unless properly addressed, may render the results of a study invalid or irrelevant. The proper treatment of confounding gets more complex for studies where patients are clustered, for example, by geographical area of residence, healthcare provider (hospital/clinic or physician), or health plan. The complex situation with confounding occurs commonly in studies using registries, network databases, or electronic health record databases.
Objectives: This study will focus on how to properly address confounding in this complex situation. The proper treatment consists of two parts:
- Accounting for confounding’s effect in relating the observed outcome to the treatment by using available data
- Examining how easily or dramatically the outcome relationship with the treatment changes if confounding is not sufficiently well addressed with the available data
We will develop novel approaches for both parts.
Methods: To facilitate the development of the methodologies, we will use both computer-simulated data and real data examples. The two complement each other for comprehensively evaluating the novel approaches developed in this study.
Anticipated Impact: This study will contribute significantly to PCORnet, the National Patient-Centered Clinical Research Network.
^Mi-Ok Kim, PhD, was affiliated with Cincinnati Children's Hospital Medical Center when the project was initially awarded.