Rank hip-replacement surgeons on cost and readmissions, adjusted for patient mix

CareScope is a health-analytics firm that helps health systems understand the cost and quality of the care their physicians deliver. A client wants to know which of its surgeons deserve credit for producing high-quality care at a lower cost for hip-replacement patients.

Cost here is the total episode cost: the all-inclusive cost of care over the 90 days that start with the surgery. Quality is defined simply as whether the patient was readmitted to the hospital. There is no fixed definition of “best” and “worst”, except that the best physicians produce high-quality care (few readmissions) at the lowest overall episode cost. Be careful about simple comparisons: physicians are unlikely to react well to a ranking that ignores the characteristics of their patients, and you will need to present the findings back to them in a constructive way.

The data

• episodes — one row per hip-replacement patient (fake patients and doctors).
  • patient_id: unique patient id.
  • operating_physician_id, operating_physician_name: the surgeon. There are 17 physicians; the names are placeholders (borrowed from sports stars), not real doctors.
  • episode_total_cost: the 90-day all-inclusive cost of the episode, in dollars. It is stored as text with thousands separators (e.g. 37,144.02).
  • has_readmission: whether the patient was readmitted to the hospital during the episode.
  • surgical_date: date of surgery.
  • congestive_heart_failure, hypertension_uncomplicated, valvular_disease, obesity, chronic_pulmonary_disease, diabetes_uncomplicated, hypothyroidism, depression: comorbidity flags.
  • gender, age_at_admission: patient demographics.

Your task

1. How different are the physicians’ patient populations, and how do patient characteristics relate to cost and to readmission? Say which differences between physicians you think are the patients and which might be the physician.
2. Build a measure of each physician’s performance on cost that accounts for patient mix, and bring in quality. Explain how you combine the two (for example separate adjusted measures, a composite, or cost among patients without a readmission), keeping in mind that a readmission’s cost is already part of the episode cost. Defend your choices and say what each alternative would reward or penalize.
3. Identify the two best and the two worst performing physicians, with a detailed rationale. How certain are you? Account for the number of patients each physician has, show how the ranking changes under reasonable alternative methods, and say which conclusions you would stand behind and which you would not.
4. What are the limitations of this analysis — patient risk the data can’t see, how patients are assigned to physicians, changes over time, the size of the data — and how would you improve it with more time or more data?
5. How would you present these results to the physicians themselves so they are seen as fair and useful? Anticipate the objections they are most likely to raise and say how your analysis answers them.

What to deliver

A one-page executive summary for CareScope’s clinical lead, who will take the results back to the physicians, followed by an appendix of any length with the methods and the full code needed to reproduce every number.

All company names in this question are fictional.