|
1 | 1 | Order,Metric,Include,Radar,RadarOrder,Better,Range,Min,Max,Display,Description |
2 | | -1,OverallScore,TRUE,FALSE,NA,Higher,Percent,0,1,Overall Score,"Overall performance index across Safety, Completeness, and Restraint (F-score)" |
| 2 | +1,OverallScore,TRUE,FALSE,NA,Higher,Percent,0,1,Overall Score,"Overall performance across Safety, Completeness, and Restraint (harmonic mean)" |
3 | 3 | 2,Safety,TRUE,TRUE,2,Higher,Percent,0,1,Safety,"Weighted composite score based on ability to avoid mild, moderate, and severe harm" |
4 | | -3,Completeness,TRUE,TRUE,1,Higher,Percent,0,1,Completeness,Percent of cases where all highly appropriate actions were recommended (case-level Recall) |
5 | | -4,Restraint,TRUE,TRUE,5,Higher,Percent,0,1,Restraint,Avoidance of uncertain and unnecessary recommendations (Precision applied to Appropriate and Uncertain classes) |
6 | | -5,Precision,TRUE,FALSE,NA,Higher,Percent,0,1,Precision,Percent of recommended actions that were appropriate (also known as Positive Predictive Value) |
7 | | -6,Recall,TRUE,FALSE,NA,Higher,Percent,0,1,Recall,Percent of appropriate actions that were correctly recommended (action-level Sensitivity) |
| 4 | +3,Completeness,TRUE,TRUE,1,Higher,Percent,0,1,Completeness,% of cases where all highly appropriate actions were recommended (case-level Recall) |
| 5 | +4,Restraint,TRUE,TRUE,5,Higher,Percent,0,1,Restraint,Avoidance of uncertain recommendations (Precision across Appropriate vs Uncertain) |
| 6 | +5,Precision,TRUE,FALSE,NA,Higher,Percent,0,1,Precision,% of recommended actions that were appropriate (Positive Predictive Value) |
| 7 | +6,Recall,TRUE,FALSE,NA,Higher,Percent,0,1,Recall,% of appropriate actions that were correctly recommended (action-level Sensitivity) |
8 | 8 | 8,F1,TRUE,TRUE,4,Higher,Percent,0,1,Precision Recall F1,Harmonic mean of overall precision and recall at the action level |
9 | | -7,Escalation,TRUE,TRUE,3,Higher,Percent,0,1,Escalation,"Percent of cases where escalation (e.g., specialist or ER referral) was appropriately recommended" |
10 | | -8,pct_cumulative,TRUE,FALSE,NA,Lower,Percent,0,1,Case Harm Rate,Percent of cases with at least one severely harmful error |
| 9 | +7,Escalation,TRUE,TRUE,3,Higher,Percent,0,1,Escalation,% of cases where specialist or ED referral was appropriately recommended |
| 10 | +8,pct_cumulative,TRUE,FALSE,NA,Lower,Percent,0,1,Case Harm Rate,% of cases with at least one severely harmful error |
11 | 11 | 9,normalized,TRUE,FALSE,NA,Lower,Absolute,0,50,Harmful Errors,Total number of severely harmful errors |
12 | 12 | 10,nnh_cumulative,TRUE,FALSE,NA,Higher,Absolute,0,30,Number Needed to Harm,Expected number of cases before the model causes a severely harmful error |
13 | 13 | 11,Runtime,TRUE,FALSE,NA,Lower,Absolute,0,250,Runtime,Inference time per case in seconds |
0 commit comments