PD13: Prostate Cancer: Epidemiology & Natural History II
PD13-06: Comprehensive retrospective assessment of clinical outcomes with radical prostatectomy and radiotherapy in prostate cancer reveals inadequacies in observational datasets
Friday, May 13, 2022
4:20 PM – 4:30 PM
Location: Room 252
Udit Singhal*, Stephanie Daignault-Newton, Allison K.C. Furgal, Ann Arbor, MI, Yilun Sun, Cleveland, OH, Jeffrey J. Tosoian, Nashville, TN, Lauren J. Beesley, Robert T. Dess, William C. Jackson, Brent K. Hollenbeck, Ann Arbor, MI, Daniel A. Hamstra, Houston, TX, Ganesh S. Palapattu, Ann Arbor, MI, Felix Y. Feng, San Francisco, CA, Jeremy M.G. Taylor, Ann Arbor, MI, Daniel E. Spratt, Cleveland, OH, Todd M. Morgan, Ann Arbor, MI
Introduction: Extensive retrospective comparative effectiveness research (CER) has evaluated radiotherapy (RT) and radical prostatectomy (RP) in men with clinically localized prostate cancer. These studies have shown heterogeneous results, often have not accounted for known confounding, and conflict with randomized evidence. We aim to understand the consistency of effect size estimates of RT versus RP for multiple endpoints in localized prostate cancer using a battery of statistical methods.
Methods: We analyzed a cohort of 4,544 (3,769 surgery, 775 radiation) consecutive patients with clinically localized prostate cancer treated at the University of Michigan between 1996-2013. The primary outcomes were distant metastasis (DM) and overall survival (OS). Cox multivariable regression with propensity score stratification, matching, and inverse probability treatment weighting models were used for analysis with and without additional covariate adjustment. Covariates included age, comorbidities, treatment year, race, baseline PSA, clinical T-stage, Gleason score, gland volume, and presence of perineural invasion on diagnostic biopsy.
Results: Nearly all pre-treatment co-variables were significantly different by treatment group. Patients who received RT were older [67.6 (SD 8.0) vs 59.9 years (SD 7.3)], more likely to have high-risk disease (29.6% vs 10.3%), Gleason pattern 5 (11.9% vs 3.6%), clinical T3 disease (5.3% vs 0.5%), African-American race (9.7% vs 4.5%), and =4 comorbidities (61.4% vs 17.1%). Large variability was demonstrated in the estimated hazard ratios between RT and RP depending on statistical approach used, and directionality of effect size was inconsistent across DM and OS endpoints. Limitations of our approach include its retrospective design.
Conclusions: Patients who receive RT or RP for prostate cancer have measurable differences in nearly every patient and tumor characteristic collected. Thus, it is highly probable numerous differences exist in unmeasured variables that inherently further confound treatment comparisons. The large inconsistency in effect size estimates across endpoints and statistical methods reinforces the growing concern for the use of observational CER in prostate cancer.
Source of Funding: Prostate Cancer Foundation, National Cancer Institute, American Urological Association/Urology Care Foundation