Reporting with TRIPOD+AI
Where clinpred’s output answers the methods and results items of the TRIPOD+AI checklist for prediction model development and validation.
TRIPOD+AI (Collins et al. 2024) is the reporting guideline for studies that develop or validate a clinical prediction model, whether by regression or machine learning. The table maps the methods and results topics of its checklist that depend on the analysis to the clinpred output that documents them. Items on the study design, data sources, ethics, and open science are outside any software.
| TRIPOD+AI topic | What to report | Where in clinpred |
|---|---|---|
| Sample size | how the sample size was decided; whether it was adequate | Sample size adequacy table (criteria of Riley et al.) |
| Missing data | how missing data were handled | Missing values: exploration and patterns tables; the method note under Data |
| Data preparation | coding of predictors, transformations | effects table (reference levels, scaling); Model equation (spline knots, coding) |
| Type of model, building | model type, predictor selection, interactions | selection and interaction tables, with their notes (method, criterion, baseline covariates) |
| Internal validation | method, and that it includes all modelling steps | performance note (bootstrap or cross-validation; selection repeated in each fit) |
| Performance measures | discrimination, calibration, clinical utility | Performance tables; calibration plot; decision curve |
| Updating | methods to update the model in validation data | Model updating table and note |
| Heterogeneity / fairness | performance across subgroups or settings | performance per validation set; case-mix table |
| Participants | numbers, outcome events | Data table (per data set, with excluded rows) |
| Characteristics | predictors by outcome; development vs validation | Descriptives; case-mix table |
| Full model | all coefficients, intercept, baseline risk | Model coefficients; Model equation; Export model (JSON) |
| Performance | with confidence intervals; optimism corrected | Performance in the Development Data (apparent, CI, corrected); Performance in the Validation Data |
| Calibration | calibration plot and measures | calibration plot; calibration-in-the-large, slope, O/E, ICI |
| Updating results | the updated model and its performance | Model updating table |
Two habits help a report:
- Report the corrected numbers, not the apparent ones, for the development data, and say how many bootstrap samples (or folds and repeats) were used — the note under the performance table has it.
- Give the model: the equation with the spline formula and the baseline survival at the horizon, or the exported JSON as a supplement. A model that cannot be computed by a reader cannot be validated by anyone else.
References
Collins, Gary S, Karel G M Moons, Paula Dhiman, et al. 2024. “TRIPOD+AI Statement: Updated Guidance for Reporting Clinical Prediction Models That Use Regression or Machine Learning Methods.” BMJ 385 (April): e078378. https://doi.org/10.1136/bmj-2023-078378.