Objective: The effect of surgeons' and hospitals' operative volumes on post-operative patient outcomes has been studied for decades and holds important policy implications. However, in many analyses of a general ``volume-outcome relationship,'' no specific intervention is clearly defined. Without a well-defined comparison, effect estimates may not be meaningful. Misinterpretations of such biased results may lead to unintended consequences for resulting policy interventions or patient recommendations. Our objective was to describe an approach to evaluate the effect on post-operative mortality of interventions on patients that specify a range of different surgeon and hospital volumes.
Data source: Fee-for-service Medicare claims (100%), 2011-2016.
Methods: For Medicare patients undergoing pancreatectomy, we specify four different target trials (i.e., hypothetical randomized trials) to estimate the effect on 90-day mortality of assigning patients to surgeons and hospitals with different operative volumes. Previous observational analyses can be viewed as an attempt to emulate our first two target trials.
Principal findings: The emulation of target trials with unrealistic, ill-defined interventions (e.g., those that ignore travel time) shows an apparent decrease in mortality with increasing operative volume. However, once the target trials are specified using interventions that could be possibly implemented in the real world, the effect estimate for operative volume becomes closer to null. We clarify that standard observational analyses are equivalent to the emulation of unrealistic target trials whose results are difficult to interpret and prone to bias.
Conclusions: When studying health systems interventions with observational data, the target trial framework may be of value to health services researchers to articulate causal questions that correspond to well-defined interventions for the real world.
- Manuscript
- R code