Origin of Physician Practice Variation: Role of Training and Clinical Trials

Presenter: Bingxiao Wu, Rutgers University
Discussant: Ian McCarthy, Emory University and NBER

ASHEcon, June 12, 2023


Geographic Variation in Health Care Spending

Variation is not just in spending across geographic areas…also exists:

  • quality of care
  • within geographic markets

Within-market Variation in Health Care Spending and Quality

Research Question

Does a physician’s training explain some amount of the observed variation in care?

What does this paper do?

Empirical strategy

  • Transcatheter Aortic Valve Replacement (TAVR)
  • Exploit launch of PARTNER TAVR trials in 22 hospitals in 2007
  • “trained” physicians compared to “matched” physicians with same fellowship location but just before PARTNER trial

Empirical strategy

  • Compare physicians with and without PARTNER training
  • Compare physicians exposed to a trained physician versus not (spillovers)


  1. Trained physicians much more likely to use TAVR
    • 150% more likely to use TAVR when part of PARTNER trial
  1. Untrained physicians “around” trained physicians also more likely to use TAVR
    • 12% more TAVR when practicing in the same zip code as a trained physician

Some thoughts

Minor interpretation question

  • Is “1.5 times more likely” the right conclusion?
  • Or should it be 1.5 times faster?
  • Estimates on volume are much smaller (between 20% and 35%)

Metrics point

  • Some growing skepticism with \(ln(y + \text{c})\) (Mullahy and Norton, 2023)
  • Consider OLS on untransformed outcome, two-part model, or Poisson

TAVR adoption is pretty complicated

  • TAVR first approved only for high-risk patients ineligible for SAVR
  • Later approved for SAVR-eligible high-risk and eventually intermediate risk
  • Hospital infrastructure requirements
  • Hospital and physician volume requirements

TAVR adoption is pretty complicated

  • Requires hospital commitment and likely direct investment
  • Availability depends on physician’s prior experience
  • Are results a reflection of physician skill/preference or…
    • hospital willingness/ability to invest in TAVR program
    • physician switching costs due to differential volume requirements with prior TAVR experience

Suggestions moving forward

  • Incorporate volume and infrastructure requirements when constructing matched pairs
  • Want to make sure hospitals are on equal footing


  1. How do we explain observed variation? Great question with clear policy implications
  2. Good identification strategy to test role of clinical training
  3. Complicated setting but seems doable given data

Thank You!

Ian McCarthy, Emory University & NBER