Prediction-model reports
A prediction-model report gives the development of a multivariable prediction model, its validation, or both, in enough detail for a reader to calculate a prediction for one patient and to judge how well the model performs.
The TRIPOD statement
TRIPOD (Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis) recommends what a prediction-model report contains, and marks each item as applying to the development of a model, to its validation or to both. Its checklist is to be used together with the TRIPOD Explanation and Elaboration document.
Title, abstract and introduction
Identify the study as developing a prediction model, validating one, or both, and name the target population and the outcome to be predicted. Summarise in the abstract the objectives, study design, setting, participants, sample size, predictors, outcome, statistical analysis, results and conclusions. Explain the medical context, including whether the model is diagnostic or prognostic, and the rationale for developing or validating it, with references to existing models. Specify the objectives, including whether the study describes the development of the model, its validation or both.
Methods
Describe the study design or source of data, such as a randomized trial, a cohort or registry data, separately for the development and the validation data if applicable. Give the key study dates, including the start and end of accrual and, if applicable, the end of follow-up. Describe the setting, such as primary care, secondary care or the general population, with the number and location of centres, then the eligibility criteria and, if relevant, the treatments received. Define clearly the outcome and all the predictors, including how and when they were measured, report any masking of their assessment, and explain how the sample size was arrived at.
Explain how missing data were handled, such as by complete-case analysis, single imputation or multiple imputation, with details of any imputation method. For development, describe how predictors were handled and selected, the type of model, how it was built and the method of internal validation. Specify the performance measures and, if relevant, those used to compare multiple models. Give details of how any risk groups were created and, for validation, of how the predictions were calculated.
Results
Describe the flow of participants, with the numbers with and without the outcome and, if applicable, a summary of the follow-up time; a diagram may be helpful. Describe the participants' characteristics, such as basic demographics, clinical features and available predictors, with the number of participants with missing data for predictors and outcome. For development, give the number of participants and outcome events in each analysis and, if it was assessed, the unadjusted association between each candidate predictor and the outcome. For validation, set out how the setting, eligibility, predictors and outcome differ from the development data, and compare the distribution of important variables with it.
For development, present the full model needed to make a prediction for an individual, with all regression coefficients and the intercept or the baseline survival at a given time point, and explain how to use it. Report the performance measures with confidence intervals. Describe any updating the validation led to, such as recalibration, and report the updated model's specification and performance.
Discussion and other information
Discuss the limitations, such as a nonrepresentative sample, few events per predictor and missing data. For validation, discuss the results with reference to the performance in the development data and in any other validation data. Give an overall interpretation of the results, considering the objectives, limitations, results from similar studies and other relevant evidence. Discuss the potential clinical use of the model and its implications for future research. Say what supplementary resources are available, such as the protocol, a web calculator and the datasets, and give the source of funding and the funders' role.
Predicted growth read against a treated child
Data from one pediatric registry of patients with growth disorders and short stature were used to develop a number of prediction models, which predict what the response to treatment should be and how a treated child should grow.
The prediction is shown on a graph of the child's growth, with lines for the normal growth of a child, the child's own growth below those norms before treatment and, once growth hormone is started, the range the child should grow within. Growth above or below that range is some sort of red light for the treating physician that something is not going well: either the dose is too high or the patient is not complying.