Payments to Doctors: Causal Inferences Require a Closer Look

An Analysis Group team responded to a ProPublica report’s provocative suggestion that doctors prescribe more drugs when paid by manufacturers (e.g., with speaking/consulting fees, business travel, meals, royalties, or gifts).

Noting the age-old wisdom that correlation does not establish causation, our team articulated four overarching points of caution.

1. A strong correlation between speaking payments and prescribing may be explained by reasons other than those implied by ProPublica. For example, doctors who have the greatest familiarity with a manufacturer’s product are likely the best spokespeople to impart firsthand experiences with prescribing the drug.
    
2. The ProPublica investigation focused on the relationship between overall payments and overall prescribing. It did not consider disaggregated data or any counterexamples. For instance, increased prescribing by some doctors may have started prior to any speaking engagements. Other doctors might have received high speaking fees but had low levels of prescribing. Still others might have continued to prescribe even after payments stopped. Aggregate analyses might miss insights from granular counterexamples such as these.
   
   
3. Payments to physicians from competing firms dilute the potential impact of those from any one manufacturer. Data from ProPublica as well as the Open Payments federal program provide information on payments to specific doctors from particular drug companies. These rich data sources reveal, for example, that many doctors have received payments from multiple manufacturers in the same therapeutic space. 
   
   
4. The ProPublica study failed to control for the many factors that affect prescribing decisions. These include: drug attributes and publicly available information concerning efficacy, safety, and side effects; disease practice guidelines and compendial listings; reimbursement coverage; physician-specific characteristics (e.g., age, specialty, region, or past experience with the drug); and patient-specific medical circumstances. Without controlling for these and other potential influences on prescribing choices, it is not possible to draw much insight from aggregate correlations.