Helping patients make patient-centered treatment decisions requires treatment effect evidence aligned to the circumstances of individual patients. This evidence must include relationships between treatment and the array of possible outcomes that can result. Outcomes can be beneficial to the patient (eg, increased survival or cure rates) or detrimental (eg, increased adverse event risk, pain associated with actual treatment, direct treatment costs, time required for treatment). Randomized controlled trials are usually insufficient to supply this evidence as they are generally only powered to estimate the treatment effect on a single outcome and may not reflect the circumstances of many patients in the real world. Analysis of observational data is an alternative to find treatment effect estimates across patient circumstances and across outcomes. Observed treatments found in these databases are not the result of randomization, but rather of choice. Real-world treatment choices often involve complex assessments of the expected effects of treatment across many outcomes. Failure to account for this complexity when interpreting treatment effect estimates using observational data could lead to clinical and policy mistakes. This study assesses the proper interpretation of treatment effect estimates with observational data when treatments have both beneficial and detrimental effects on outcomes and these effects are heterogeneous across patients.
