Data

We conducted a retrospective study of surgical outcomes using a large private insurance claims database, specifically the Truven Health MarketScan Commercial Claims and Encounters Database, for Florida over the period 2008 through 2012, which was the most recent 5-year period available when this research was conducted. Florida was chosen because of its large population overall, its relatively large share of residents aged 65 years and older (19.4% compared with 14.9% for the United States in 2015), and its relatively high proportion of expenditures for physicians and other professional services (30.9% compared with the US average of 27.4% in 2009).²⁶ The MarketScan database contains individual-level, de-identified, health care claims information from employers, health plans, hospitals, Medicare, and Medicaid programs. These databases reflect the real world of treatment patterns and costs by tracking millions of patients as they travel through the health care system offering detailed information about all aspects of care. Data from individual patients are integrated from all providers of care, maintaining all health care utilization connections at the patient level.

There were thousands of different surgical procedures performed at outpatient facilities during the time period, so one of the first tasks was to reduce the number of index surgical procedures for analysis and reporting purposes. We selected outpatient procedures using the following criteria: (1) we only selected procedures that had at least 100 office-based procedure claims over 2008-2012 (ie, at least an average of 20 procedures per year) and (2) a “work relative value unit” (work RVU) of at least 4.0 (ie, relatively complex procedures).²⁷ The volume threshold was selected to eliminate procedures with an insufficient sample of office-based cases for analysis. The focus on relatively complex procedures was applied because, as noted in the “Background” section, theory predicts that differences in outcomes across types of surgery settings are least likely to be observed for low-complexity procedures.²⁸

Procedure complexity was measured using the work RVU, which is a component of the Medicare (Part B) physician payment system based on the Current Procedure Terminology (CPT) code for the procedure. The payment formula contains 3 RVU components: physician work, practice expense, and malpractice expense. A code with a higher work RVU generally takes more time, more intensity, or some combination of the 2. Thus, our underlying assumption in using the work RVU cut-off is that, in general, work RVU approximates procedure complexity. A total of 88 surgical procedures met these criteria (see Table A1 for descriptions). This represents a wider range of procedures compared with the 15 procedures included in the Fleischer et al²⁵ study.

The location of services (physician office, ASC, or HOPD) was determined by the place of service code in the claims. In addition, patients with claims for study procedures were required to have insurance coverage during the month of and the month after the procedure.

The initial study sample consisted of 4 080 800 claims with CPT codes for one of the 88 study procedures in Florida for 2008-2012. From this initial sample, 139 243 claims were excluded because the surgical facility type was not one of the 3 study facility types. After combining claims with the same patient number, CPT code, data of service, and location of service, the sample included 2 807 857 unique surgical procedures. From this total, 148 531 procedures were excluded because the patients did not have insurance coverage in both the month before and the month after surgery, leaving a final study sample of 2 757 016 procedures (see Figure 1).

Sample flowchart. CPT indicates Current Procedure Terminology.