Medical Claims Data
To measure price and quality for outpatient surgical services, this study used 2009–2013 claims data from the HCCI. HCCI, along with the companies providing data to it—Aetna, Humana, and UnitedHealthcare—collects medical and pharmaceutical claims data for approximately 50 million commercially insured individuals. The HCCI data includes observations from every state and each US metropolitan region.
The study sample was restricted to patients aged 18–64 who were enrolled in the same insurance plan for the entire year. For each patient, health-care utilization (the annual number of inpatient hospital, outpatient hospital, emergency department visits, physician office visits, and prescription drug fills) was also measured. To further control for patient risk characteristics, the Charlson Comorbidity Index and indicators for each of the 17 chronic conditions underlying the Charlson risk score were also identified for each patient and year.13 Each procedure was linked to a specific provider using that provider’s National Provider Identifier (NPI).
Identification of Surgical Procedures
From the HCCI database, all claims for three surgical services performed in outpatient settings and related complications to these services were identified according to Current Procedure Terminology (CPT) and International Classification of Disease (ICD-9) codes. In the HCCI data, these three surgical procedures accounted for 2.5% of total medical spending, 21.7% of spending on outpatient surgical services, and 11.8% of the volume of outpatient surgical services. Cataract replacement surgeries were identified by CPT codes 66982, 66983, and 66984. Colonoscopies were identified using CPT codes 44388–44394 and 45378–45385 and ICD-9 codes 45.22, 45.23, 45.25, 45.41, and 45.42. Joint arthroscopy surgeries were identified by CPT codes 29830–29838 (elbow arthroscopy), 29862 (hip arthroscopy), 29870–29887 (knee arthroscopy), 29805–29826 (shoulder arthroscopy), and 29846 (wrist arthroscopy).
Identification of Procedure Price
Procedure prices were identified by each procedure’s bundled allowed amount for both facility and professional fees. This amount represents the negotiated price between the provider and the insurer and captures actual payments by the patient, employer, and insurer. It does not represent the “billed charge” amount, which is often not representative of the actual transacted price but has previously often been used to study provider prices.
Identification of Procedural Complications
For each procedure, complications related to that procedure were also identified and were linked to the index surgical procedure but were required to be separate in time from the primary cataract surgery. Complications related to joint arthroscopy were analyzed at both the 30- and 90-day periods following the index arthroscopy procedure.14 Complications for cataract surgery were identified by secondary surgeries as surrogate markers performed within 90 days of the primary cataract surgery.15 Colonoscopy complications were measured at 30 days and were classified as any procedural complication for one of three categories: cardiovascular, serious gastrointestinal, or other gastrointestinal complications.16 A list of the CPT and ICD-9 codes used to identify each complication is included in Online Appendix Table 1.
Additional Control Variables
From the HCCI data, provider volume for each procedure was calculated and used to control for volume-outcome relationships.17,18 To control for market-level characteristics, additional county-level data from the Area Health Resources File (number of physicians and nurses per capita, share of physicians that are specialists vs. primary care providers, median household income, share of households below poverty line, mean rent, percent of individuals with a college degree, median rent, and share of population enrolled in both Medicare and Medicaid), the American Hospital Association’s Annual Survey (hospital market concentration), the InterStudy survey of insurers (insurer market concentration), and the SK&A survey of physician practices (physician market concentration) were included. A full list of these additional control variables is included in Online Appendix Table 1.
Multivariate regressions were used to measure the risk-adjusted association between procedure price and the probability of procedural complications. To estimate risks of procedural complications, a logistic regression with a dichotomous outcome variable indicating if the procedure was linked to a procedural complication was estimated. To address skewness, prices in the logistic regression were standardized to the mean service-level price in each geographic market. Alternative models estimated prices in $1000 units (Online Appendix Table 3). Separate multivariable regressions were also used to risk-adjust procedure prices. To risk-adjust prices, a generalized linear model with a log link and gamma distribution was used to measure the predictors of each procedure’s price.19,20
Each regression controlled for patient characteristics and demographics, including patient age, gender, industry and insurance plan type, and medical utilization (number of inpatient, outpatient, office-based, and emergency department visits and number of prescription drug fills). Charlson scores and fixed effects for chronic conditions further controlled for patient risk characteristics. Fixed effects for year and month, CPT codes, and the 306 hospital referral regions in the USA (HRR)21 were used to control for temporal time differences (e.g., inflation), specific procedure differences, and time-invariant market-level characteristics. The abovementioned controls for provider volume, provider supply, demographics, and market power were also included. All regressions were estimated using heteroskedasticity-robust standard errors.
Construction of Provider Risk-Adjusted Price and Complication Rates
Each provider’s risk-adjusted price and complication rates were calculated using two steps and separately for each of the three services. First, the patient-level risk-adjusted price and complication rates were calculated by computing the fitted values from the respective multivariate regressions. The price of each procedure was omitted from the logistic regression used to risk-adjust complication rates. Second, each provider’s risk-adjusted price and complication rate were estimated by calculating the mean patient-level risk-adjusted price and complication rate from all patients treated by each provider with at least 50 observations.
Next, the correlation between risk-adjusted provider prices and complication rates was measured. For each procedure, standardized risk-adjusted price and quality scores that show the standard deviation difference between each provider compared to the rest of the providers in each HRR were calculated for each provider with at least 50 patient observations. The standardized scores, often referred to as z-statistics, were computed by dividing the difference between the provider’s mean value of each outcome measure (e.g., risk-adjusted price and complication rate) and the mean risk-adjusted price and complication rate for all providers in the HRR by the respective standard deviations from each HRR.22 To assess the association between risk-adjusted provider price and complication rates, the correlation between these two measures was calculated using a linear regression. Observations were weighted by provider volume.