Explainations

On the main page, after clicking on tool´s name, further information is displayed. This information on each tool contains, among other items, a “convergent validity” section. Here we would like to provide some explanations on how to interpret data presented within the “convergent validity” section.

In convergent validity we presented some measures such as discrimination ability, calibration, and a so-called golden standard comparison.
Discrimination refers to how well an assessment tool can separate one group from another (for example, avoidable vs. unavoidable care transitions). Discrimination is commonly measured using ROC curves, the latter is a plot of Sensitivity vs. (1-Specificity) over the range of possible cutpoints. If the model has perfect discrimination, the ROC curve should hit the upper left corner of the plot (100% sensitivity and 100% specificity). Discrimination ability can be summarised by area under ROC curve (equivalent to c-statistic), sensitivity/specificity, positive predictive value/negative predictive value, Brier score (combine calibration and discrimination) [1].
Brier scores range from 0 (best performance) to 0.25 (no informative) [2].
C-statistic value below 0.5 indicates poor discriminatory ability. Fair discrimination when ranging from 0.5 to 0.7. A good discrimination when the c-statistic ranges from 0.7 to 1 [3].

Calibration refers to a measure of how well the predicted probabilities agree with the observed probabilities (is a property related to goodness of fit of a model). It can be measured for example by Hosmer-Lemeshow goodness of fit test, Brier score (combine calibration and discrimination) [1]. For Hosmer-Lemeshow goodness of fit test, P ≥ 0.05 implies a good fit, with higher numbers signifying better fit [2].
Golden Standard. We refer to this assessment type as a comparison of an index assessment tool to a “golden standard” (i.e. for example another assessment tool or expert consensus).

Filter options on this website contain “conditions”, “patient types”, “care transition”, “settings”. These filter options are not binding, but are rather provided as a recommendation.