2.1 Study Population

From a 20% national sample of Medicare claims, we identified men with newly diagnosed prostate cancer between 2016 and 2019, using a validated algorithm with 99.8% specificity and 88.7% positive predictive value [20]. In assessing spillover effects of MA penetration, our analysis included only men covered by Traditional Medicare, and outcomes of those enrolled in MA plans were not assessed. Men included in our analysis had a minimum of 12 months of follow-up and data available through December 31, 2020. We included only those with continuous enrollment in Medicare Parts A and B in the year before and after their prostate cancer diagnosis.

2.2 Medicare Advantage Practice Penetration

An empirical measure of MA penetration (i.e., proportion of Medicare beneficiaries covered by MA) was derived at the urology practice-level and served as the study's primary exposure. First, as described in previous work, we defined urology practice markets (i.e., the geographical boundaries from which a practice draws its patients), using the variable market approach [21-24]. Briefly, we established each practice's market as the minimum radius between zip code centroids of the practice and the patient that captured at least 75% of all Medicare payments to the practice each year. Notably, the practice market was defined by all patients seen in a given year, independent of diagnosis and gender. For practices with markets spanning multiple zip codes, we weighted each zip code based on its contribution of payments to a practice's market.

Using the Medicare Beneficiary Summary File, we then estimated the proportion of patients with MA [MA beneficiaries / (MA + Traditional Medicare beneficiaries)] at the zip code level. We then applied this proportion to each urology practice market, based on the weighted contribution of zip codes previously described. For example, in a practice market spanning three zip codes, Zip Codes A (MA penetration = 20%), B (MA penetration = 50%), and C (MA penetration = 40%), we would calculate their contribution to the practice—60%, 30%, 10%, respectively. Taking the weighted average of its zip codes [$$$ Zipcode\ A\ \left(0.20\times 0.60\right)+ Zipcode\ B\ \left(0.50\times 0.30\right)+ Zipcode\ C\ \left(0.40\times 0.10\right)=0.31 $$$], allowed us to derive the urology practice-level MA penetration.

After determining urology practice-level MA penetration, Traditional Medicare beneficiaries were assigned to practices as per previous methods [25, 26]. Briefly, all urologists providing care to a patient within the first year of a prostate cancer diagnosis were identified using their National Provider Identifier number. If a patient saw more than one urologist in the first year after their prostate cancer diagnosis, the patient was assigned to the urologist having 75% or more of the associated visits. Then urologists were assigned to practices in each year using tax-identification numbers. With links between patients and urologists as well as between urologists and practices, we were able to assign patients to urology practices.

2.3 Primary Outcomes—Quality

Our study assessed two primary outcomes in prostate cancer quality, both measured at the patient level. First, we measured potential overtreatment. Given prostate cancer's protracted natural history, clinical guidelines recommend against treatment in older, sicker patients with a high risk of noncancer mortality [5, 6]. Using a previously validated model with high discrimination (C statistic = 0.82), we estimated noncancer mortality risk and identified a subgroup of men unlikely to benefit from treatment (i.e., noncancer mortality risk > 75% within 10 years of diagnosis) [27].

The second quality measure was the use of confirmatory testing in men initiating active surveillance. We identified men undergoing active surveillance using a previously validated algorithm with 97% specificity and 92% negative predictive value [28]. The men identified in the algorithm are a healthier subset of patients (< 75 years old with Charlson comorbidity index < 3). We considered men to have undergone confirmatory testing if they received at least one magnetic resonance imaging study, prostate biopsy, or genomic test within 12 months of diagnosis. Receiving a confirmatory test within 12 months of diagnosis is recommended by clinical guidelines, as up to 40% of men on active surveillance harbor more aggressive or extensive disease not found on their initial diagnostic biopsy [6, 10]. Confirmatory testing is particularly important in this healthier subset of men who have a greater lifetime likelihood of cancer progression and may eventually require treatment. Incentives afforded by increasing MA penetration to reduce spending may lead to underuse of confirmatory testing in men initiating surveillance.

2.4 Secondary Outcomes—Utilization

As a secondary outcome, we assessed price standardized spending, a global measure of healthcare utilization [22]. We summed price standardized spending, inflation adjusted to 2019 dollars, for Medicare claims associated with a prostate cancer diagnosis code during the 12-month period after diagnosis. We hypothesized that higher practice-level MA penetration would be associated with lower price standardized spending. Finally, we measured the odds of prostate cancer treatment, defined as either radiation therapy or surgery (i.e., radical prostatectomy) in all men regardless of noncancer mortality risk. Incentives to constrain spending in a practice market highly penetrated by MA may reduce treatment and encourage the adoption of more conservative management strategies. Given the potential for these constraining spillover effects, we hypothesized that Traditional Medicare beneficiaries managed by practices with higher MA penetration would have lower odds of undergoing treatment.

Though not statistically significant, we found that practices with higher MA penetration typically had lower price-standardized spending. However, MA penetration was associated with increased odds of treatment (radiation therapy or surgery). Given that the association with MA penetration pointed in opposite directions for our two measures of utilization, we performed a post hoc analysis among men receiving treatment for their prostate cancer. As prior work demonstrates higher Medicare spending among patients treated with radiation therapy [29], we hypothesized that increased MA penetration would be associated with lower use of radiation therapy among patients undergoing treatment—and thereby higher use of surgery.

2.5 Statistical Analysis

We examined clinical characteristics of men by the MA penetration of their managing practice. To assess the relationship between practice-level MA penetration and our four outcomes, we fit a multilevel model for each outcome. The multilevel model allowed us to account for the nested nature of our data (i.e., patients within practices). For our dichotomous outcomes—treatment, potential overtreatment, and confirmatory testing—we used logistic regression models. We fit a negative binomial model for the price standardized spending outcome. In all cases, we modeled MA penetration as a continuous variable (i.e., per 10% increase in MA penetration in a urology practice market). From the adjusted models, adjusted probabilities of dichotomous outcomes and spending were estimated using postestimation commands (i.e., the margins command in STATA). Regression models were adjusted for age, ethnicity, socioeconomic status, Charlson Comorbidity Index, rurality, year of diagnosis, United States Region, and practice type. We characterized practice type as small urology practice (1–2 urologists), medium urology practice (3–9 urologists), large urology practice (10 or more urologists), subspecialty group (only specialty physicians, not limited to urologists), multispecialty group (physicians from multiple specialties, including primary care), and hospital integrated.

Analyses were conducted using STATA version 18 software (College Station, TX). All statistical tests used p < 0.05 to define statistical significance. Our institutional review board deemed this study exempt from regulation.

3.1 Cohort Characteristic and Practice-Level MA Penetration

We identified 41,092 patients meeting inclusion criteria. Median practice-level MA penetration across 6784 practice-years was 33% (IQR 23%–43%) (Figure S1). Table 1 illustrates cohort characteristics, grouping men by the MA penetration quartile of their managing practice. Age and Charlson Comorbidity Index were well balanced between the quartiles. However, patients managed by practices highly penetrated by MA were more likely to live in cities (54% in Quartile 4 vs. 39% in Quartile 1) and the West region of the United States (31% in Quartile 4 vs. 11% in Quartile 1). They were also more commonly diagnosed with prostate cancer in later calendar years (diagnosed in 2019: 35% in Quartile 4 vs. 19% in Quartile 1).

3.2 Primary Outcomes—Quality

After adjusting for patient characteristics, MA penetration was not associated with potential overtreatment (adjusted OR 1.04 (95% CI 0.98, 1.10), p = 0.22, per 10% increase in MA penetration) in those with competing health risks (> 75% noncancer mortality within 10 years of diagnosis). This represents an increase of 0.8% (95% CI −0.4%, 2.0%) in the adjusted probability of potential overtreatment per 10% increase in MA penetration. Among men initiating active surveillance, the adjusted probability of receiving a confirmatory test increased by 1.2% (95% CI 0.0%, 2.3%) for each 10% increase in MA penetration but did not reach statistical significance (adjusted OR 1.04 (95% CI 0.99, 1.10) p = 0.11; Figure 1, Table S1).

3.3 Secondary Outcomes—Utilization

After adjusting for patient characteristics, practice-level MA penetration was not associated with global utilization, as measured by price standardized spending (Incident Rate Ratio 1.00 95% CI (0.99, 1.01), p = 0.49; Table S1). This represents a $54.68 decrease in adjusted price standardized spending per 10% increase in MA penetration (Figure 2), though not statistically significant (95% CI: −$208.82, $99.47). MA penetration was associated with an increase in treatment among all patients (adjusted OR 1.04 (95% CI 1.01, 1.06), p = < 0.001, per 10% increase in MA penetration; Table S1). For each 10% increase in MA penetration, 0.7% (95% CI 0.3%, 1.2%) more patients in a practice would undergo treatment (Figure 1).

Lastly, we modeled the odds of treatment modality, radiation therapy or surgery, among men receiving prostate cancer treatment. Increasing practice-level MA penetration was associated with a decrease in radiation therapy (adjusted OR 0.93 (95% CI 0.89, 0.97), p < 0.001; Table S1). Among men receiving treatment, for every 10% increase in MA penetration, the adjusted probability of undergoing radiation therapy decreased by 1.1% (95% CI −1.8%, −0.4%). Reciprocally, for every 10% increase in MA penetration, the adjusted probability of undergoing surgery increased by 1.1% (95% CI 0.4%, 1.8%; Figure 3).