Table 1.

Example performance metrics.

Variable Type	Example ML-Extracted Variable	Example Performance Metric
Categorical	Diagnosis (yes/no)	Sensitivity Positive predictive value (PPV or precision) Specificity Negative predictive value (NPV) Accuracy Predicted prevalence vs. abstracted prevalence Calibration plots (if applicable)
Date	Diagnosis date	Sensitivity with a ± n-day window PPV with a ± n-day window 1 Distribution of date errors
Continuous	Lab value	Sensitivity, PPV, and accuracy for classifying the result as within vs. outside the normal range Sensitivity, PPV, and accuracy for classifying the result within ±X of the true value Mean absolute error (MAE)

¹: The proportion of patients’ human-abstracted as having the diagnosis that is also correctly identified as having the diagnosis by the model and where the ML-extracted diagnosis date is within ±n days of the abstracted diagnosis date or both abstracted and ML-extracted dates are unknown.