Table 1 Model reporting criteria that prediction studies should report to enable external validation
From: External validation of existing dementia prediction models on observational health data
Category | Reporting criteria | Description |
|---|---|---|
Population settings | Target population definition | Definition or description of the population for which predictions are made. |
Index date | Date at which a patient qualifies for inclusion in the target population. | |
Time-at-risk | Time window in which a model’s predictions are valid relative to the index date. | |
Outcome definition | Definition or description of the outcome to be predicted during the time-at-risk. | |
Statistical analysis settings | Prediction method | Prediction methods in this study are limited to logistic regression and Cox proportional hazard for predicting a binary outcome. |
Predictor definitions | Predictor descriptions or definitions in terms of data source codes. | |
Predictor time window | Time window in which the predictor is assessed. In a special case, a predictor can be assessed in a time window all time prior to index, often reported as “a history of …” using all available prior data of a person. | |
Model specifications | The prediction model, e.g., parameters to construct the model given a prediction method. Alternatively, a risk calculator or nomogram could be reported. We also distinguish here between fully and partially specified models. For example, if no intercept is reported in the case of a logistic regression model, we are still able to construct a simple risk stratification model using only coefficient values. However, this method does not consider the baseline risk of the original model and (re-)calibration will not be assessed. |