Background: Recommendations for staging newly diagnosed prostate cancer patients vary between guidelines and literature.
Methods: Our objective was to validate and compare prediction models selecting newly diagnosed prostate cancer patients for bone scan staging. To achieve this, we validated eleven models in a population-based cohort of 10,721 patients diagnosed with prostate cancer between 2005 and 2019. The primary outcome was net-benefit. This was assessed at different balances of conservatism and tolerance, represented by preference ratio and number-willing-to-test (NWT). Secondary outcomes included calibration slope, calibration-in-the-large (intercept), and discrimination measured by Area-under-the-receiver-operator-characteristics curve (AUC).
Results: For preference ratios less than 1:39 (NWT greater than 40), scanning everyone provided greater net-benefit than selective staging. For preference ratios 1:39 to 3:97 (NWT 33-40), the European Association of Urology (EAU) 2020 guideline recommendation was the best approach. For preference ratios 3:97-7:93 (NWT 14-33), scanning EAU high-risk patients only was preferable. For preference ratios 7:93-1:9 (NWT 10-13), scanning only Gnanapragasam Group 5 patients was best. All models had similar fair discrimination (AUCs 0.68-0.80), but most had poor calibration.
Conclusions: We identified three selective staging strategies that outperformed all other approaches but did so over different ranges of conservatism and tolerance. Scanning only EAU high-risk patients provided the greatest net-benefit over the greatest range of preference ratios and scenarios, but other options may be preferable depending upon the local healthcare system's degree of conservatism and tolerance.
© 2022. Crown.