Physician Response to Prices of Other Physicians: Evidence from a Field Experiment

The experimental treatment for this study was a report listing two numbers for each of six busy ophthalmology practices. The numbers were risk-adjusted, 180-day cost averages for newly referred patients to ophthalmology for both HMO and Medicare Advantage patients. Together with the cost report, the treatment group PCPs also received a cover letter from the CEO of the IPA, briefly explaining the reason for receiving the report and a description of how the costs were calculated.

The cost report was comprised of two parts: the cost table and a set of footnotes with additional information. On the left side of the table were the names of six ophthalmology practices. Below each practice name were the names of the IPA network ophthalmologists associated with each of those practices in smaller, italicized print. On the right were two costs, one for each type of patient, for each practice.

2014-05-05

2014-11-15

My primary outcome is the share of a PCP practice's ophthalmology referrals that an ophthalmology practice receives, measured on a bimonthly basis (two-month periods).

The subjects of the experiment were all PCPs associated with the IPA who practiced in either the family practice or internal medicine specialties in an outpatient (ambulatory) setting. Subjects assigned to the treatment group all received an informational treatment: a letter containing historical average costs of six ophthalmology practices affiliated with the IPA, the ``treatment,'' ``prices,'' or ``cost report.'' Control group subjects did not receive anything. In order to maximize the experiment sample size, all of the IPA's PCPs who were active at the time of the treatment distribution were included if they satisfied minimal criteria: they must have had at least ten claims during each calendar month from August 2013 through January 2014, and made at least one patient referral to ophthalmology during that period. In the end, a total of 93 PCPs -- 35 internists and 58 family practitioners -- were included in the experiment. These physicians were typically organized into group practices. In total, there were 55 included practices, with 24 of them being internal medicine, 30 family practice, and one mixed specialty group.

To account for the organization of PCPs into practices, assignment into treatment and control groups took place at the practice level, so that either all PCPs in a practice were assigned to the treatment group, or none were.

During the experiment, the IPA collected data on PCP referrals as part of its normal operations. This data was generated by the PCPs' activities of seeing and treating patients as part of their usual medical practices in their regular offices. The IPA regularly collects all of this data, and all of the physicians are aware of this data collection. Moreover, none of the physicians were made aware that the distribution of the price information was related to an experiment, thereby minimizing any possible influence of the ``Hawthorne effect.''

Randomization by computer.

Medical primary care practice. While there are multiple individuals in each practice, since I measure the outcome at the practice level the practice is considered the formal subject. So the study is not cluster randomized in that sense.

No

55 PCP practices

55 PCP practices times 6 ophthalmology practices by 6 periods = 1980

28 treatment, 27 control