The abstract is read at the publisher; the summary is JClub's.
Rule-derived labels for Escherichia coli urine cultures did not consistently improve cefepime/cefoxitin/ceftazidime probability prediction versus using all reported labels, with equal-weight augmentation worsening Brier scores.
- Why it matters: When reported labels are scarce, weak rule labels may help, but their benefit depends on label quality and whether comparisons properly account for the reported-label budget and calibration.
- What they did: In two EHR datasets (Stanford ARMD and UT Southwestern ARMD-UTSW), logistic regression and histogram gradient boosting were trained on reported labels only or with added rule-derived labels using equal weights, tuning-selected weights, class-proportion adjustment, or prior-probability correction across 10%, 25%, 50%, and 100% development budgets.
- The result: All rule-derived labels were susceptible; equal-weight augmentation produced lower predicted risks than observed prevalence (e.g., Stanford cefepime at 10%: 0.35% vs 4.85%) and higher Brier scores than the all-budget reported-only baseline in every drug/model/budget summary.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.