Radiology

PI-RADS v2 score 5 strongly associates with worse prostate cancer outcomes

Across three retrospective cohorts, PI-RADS v2 score 5 associated with markedly higher prostate cancer–specific mortality and worse survival endpoints. In the largest cohort PI-RADS 5 had HR 18.4 for prostate cancer–specific mortality after multivariable adjustment. PI-RADS 5 also associated with metastasis-free survival and biochemical recurrence in surgical and diagnostic cohorts. Main limitation is retrospective design across centers, so use PI-RADS 5 as a high-risk signal but corroborate with clinical and pathology data. 1

Epistemonikos Database of Trials offers high recall and efficient screening for RCTs

In a case study of pharmacological migraine RCTs, single-database recall ranged 98.3–99.4% and combinations reached 99.4–100%. Epistemonikos (EDT) yielded lower number-needed-to-read and similar network-meta-analysis estimates compared with traditional databases. EDT's search and documentation tools were less sophisticated but sufficient for reproducible searches. Implication: include specialized RCT databases like EDT or CENTRAL to reduce screening burden without sacrificing completeness. 2

SMART semantic-CT model differentiates new primary lung cancer from metastasis with AUC 0.81

In 649 pre-treatment CTs after prior malignancy, the SMART model achieved AUC 0.81 for SPLC versus metastasis classification. Emphysema, irregular contour, and spiculation associated with SPLC; peripheral distribution favored metastasis. SMART outperformed radiologists for metastasis classification overall and matched experts in a subset. This is retrospective multicentre data; consider local validation before replacing expert read or altering management. 3