What precision and recall capture
Recall, the same quantity as sensitivity, is the fraction of true cases that the model flags. Precision, equivalent to positive predictive value, is the fraction of flagged cases that are truly positive. Pushing a model to catch more cases usually drags precision down, because the extra catches include more false alarms, so the two trade against each other as the decision threshold moves.
A precision-recall curve plots this tradeoff across all thresholds, and the area under it summarizes performance in a single number.
Why it beats ROC on rare conditions
When positives are rare, an ROC curve can look reassuringly good because specificity stays high simply by correctly labeling the abundant negatives. Precision is far more sensitive to false positives in that setting, so a precision-recall view exposes weak diagnostic performance that ROC can hide.
This is general educational information, not medical advice. A model metric describes average behavior on a dataset and does not by itself establish that a tool is safe or appropriate for an individual; clinical validation and professional oversight are required.