br and detection and precipitous increases in
and detection, and precipitous increases in the use of MRI based on the proliferation of literature that appeared in recent years. For example, recent estimates from the United Kingdom indicate that prostate MRI was obtained in 51% of men newly diagnosed with PCa in England from April 2015 through March 2016, with a majority performed prior to diagnosis.27
We found considerable sociodemographic variation in the receipt of prostate MRI including racial and
Table 3. Factors associated with observation (versus definitive therapy) among 8144 low-risk prostate cancer patients, 2010-2013
Observation Definitive Therapy Odds Ratio* 95% Confidence Interval
Age at Diagnosis
Year of Diagnosis
Flu Shot Before Diagnosis
Zip Code Median Household Income
Hospital Referral Region Urologist Density (per 100,000)
* All variables were mutually adjusted in the model.
socioeconomic disparities. Although non-white men were less likely than white men to receive prostate MRI, they were also less likely to receive definitive treatment. This finding is consistent with other reports indicating that non-white patients are less likely to be treated for PCa,28 and highlights previously unreported racial disparities in the use of advanced imaging for PCa. As racial disparities
in gene Pimozide testing for cancer have been observed,29 additional study is needed to determine whether unequal use of MRI reflects differences in access or acceptance of its use.
There are several limitations of this work that require discussion. Our study included patients diagnosed from 2010-2013, and practice patterns have continued to
evolve since that time. Although the data used represents the most currently available, anticipated increases in the use of MRI and initial observation for men with low-risk PCa might modify the association of imaging and observa-tional management. Patients in the SEER-Medicare data-base are also older, on average, than men diagnosed in the US with PCa, which may lead to a greater inclination to select observation as management. Further, the observed association between MRI and observation does not imply causality. Therefore, it is possible that MRI did not influ-ence the decision for treatment or observation, but was undertaken as a component of AS. Because we ascer-tained prostate MRI status based on administrative claims, we are unable to account for the interpretation of the imaging studies themselves. As a result, we do not know whether MRI provided clinically accurate disease predic-tions, or how clinicians or patients incorporated prostate MRI data into management decisions.
Although, we attempted to control for confounding associated with the decision to perform prostate MRI using propensity score matching, there might still have been residual confounding due to unmeasured factors. However, we suspect that among otherwise low-risk patients, MRI would be more likely to be obtained in equivocal cases favoring a higher-risk profiles, where such misclassification would mask the association of MRI and observation in patients at average or low risk. Addition-ally, Site-specifix recombination is possible that a small number of MRIs were per-formed for the purpose of treatment planning instead of staging. We tried to reduce misclassification by restricting MRIs to be conducted at least 28 days before treatment and by performing sensitivity analyses with different exclusion criteria, which have not significantly changed our results. Even if such a misclassification exists, we expect that this would also bias in favor of underestimat-ing the association of MRI and observation. As a final consideration, we employed a definition of observation as no definitive treatment within 12 months but lack detail regarding the management received, including AS, watchful waiting, or deferred therapy.
A major strength of our study is the large, population-based cohort of patients with low-risk PCa who received clinical care in the real-world setting. The nationwide Medicare claims data covered a wide spectrum of health services, regardless of where the patients sought their care, therefore providing comprehensive information on the care received by patients. Furthermore, the linked SEER-Medicare database also enabled us to control for many other factors that may influence prostate MRI and treat-ment decisions, such as sociodemographic factors, comor-bidity, and urologist density. In addition to adjusting for these factors in a multivariable logistic regression model, we also conducted propensity score matching followed by conditional logistic regression. The similarity between the findings derived from these 2 different analytical strategies was reassuring.
Efforts to facilitate observational approaches for low-risk PCa are highly valuable to improving the quality of