Homework 2 — Physician Agency and Payment Changes

Instructions

This homework is about how physicians respond to payment incentives, the material from the second module of the course. It is empirical only — the theory is assessed in class through the quizzes and the midterm. Here you work with real data and interpret it as a health economist.

You are expected to use an AI coding assistant. Use GitHub Copilot (free with your Emory GitHub account) or the tool of your choice, and let it write the R or Python. Producing the code is not the skill being graded. Directing the tool, checking that the output makes sense, and interpreting the result in the language of this course is. Read what the assistant gives you, make sure it actually answers the question, and explain what it means.

Submit a rendered notebook (Quarto, R Markdown, or Jupyter) containing your code, its output, and your written answers. Each part is graded on this scale, scaled to the points shown for that part:

  • Full credit: correct, with sound economic reasoning.
  • Partial credit: close, with a minor error or thin reasoning.
  • Little credit: attempted, but the economic reasoning is missing or wrong.
  • No credit: no real work, or output pasted in with no interpretation.

Homework 2 is due by midnight on Friday, October 16.

Q4 — Epidural steroid injection payment cut (15 pts)

In January 2014, CMS reduced its payments to physicians for epidural steroidal injections (CPT codes 62310, 62311, 62318, 62319) as part of adjusting “potentially misvalued services,” then raised them again in 2015. Use medicare-physician-epidural.csv (on the shared OneDrive, linked from Canvas), which has one row per physician and year for 2013 through 2015, with the physician’s specialty, state, the number of injections, and the average Medicare payment.

  1. Compute the average number of injections per physician in each year, 2013–2015, and report the trend.

  2. Does that yearly average tell a clean story about how physicians responded to the 2014 cut? What population is being averaged over in each year, and does it change? Identify at least one reason the year-over-year average could move for a reason that has nothing to do with physician agency.

  3. Find the physicians with the largest drop in injections from 2013 to 2014. A raw ranking of the largest drops can mislead, because some of the biggest reductions come from physicians who barely performed the procedure to begin with, or who leave the data, rather than from a response to the payment cut. Restrict the comparison so it isolates the behavioral response, and explain what you changed and why.

  4. Connect to the model. Payment fell in 2014 and rose in 2015. State what physician-agency theory predicts for utilization in each year, and whether your corrected data are consistent. Where they are not, give the most plausible economic reason.

  5. Plot the corrected set of largest-responders over time and say, in one sentence, whether the 2015 rebound shows up and what that implies about physician responsiveness to payment.

Q5 — Physician coding and the 2021 office-visit change (15 pts)

Physician agency shows up not only in how much care is delivered but in how that care is coded and billed. In January 2021, Medicare raised the payments for office visits (evaluation and management, or E&M, codes) and changed the rules for choosing a visit’s level, letting physicians bill based on time or medical decision-making instead of the older, more demanding documentation requirements. That made the higher-level visits easier to bill. You will look for physicians’ coding response in Georgia’s Medicare data, using medicare-physician-ga-office-visits.csv (on the shared OneDrive, linked from Canvas), which has one row per physician, HCPCS code, and year, with the code’s description, the number of services, and the average Medicare payment. The file holds the office-visit codes along with a few other services, so part of the work is picking out the office visits. For the policy background, see the CMS fact sheet on office/outpatient E&M visits.

  1. (3 points) Find the office-visit codes and their levels. Identify the office-visit E&M codes in the data and how they run from lower to higher level (for example, established-patient visits 99211 through 99215). Confirm your set against the code descriptions in the data, and say briefly how the levels differ.

  2. (3 points) Measure coding intensity. For each year from 2017 to 2023, compute a measure of how high physicians are coding office visits, such as the share billed at the two highest levels (99214 and 99215 for established visits) or the average visit level. Plot it over time.

  3. (3 points) Find the break. Does your measure jump around 2021, when the payment and documentation changes took effect? Use the earlier years as a baseline, and describe the timing and size of any change.

  4. (3 points) Connect to agency. Explain how this pattern reflects physician agency operating on the coding margin. What incentive did the 2021 change create, and how does agency predict physicians respond?

  5. (3 points) Be careful about the claim. Give one other explanation, besides physicians coding more aggressively, that could produce the same upward pattern. What would that mean for how confidently you can attribute the change to the policy?