Clinical decision rule

We will build a clinical prediction model to identify the most relevant predictors of success of rehabilitation and supportive care programs. Program success will be defined as having successfully improved quality of life, physical or psychosocial functioning, or reduced fatigue.

Multivariable backwards logistic regression analysis will be conducted on pooled data to build the prediction model. The variables with the highest p-values will be removed one by one, based on the Wald test, until all remaining variables have a significant pre-determined p-value. Potential predictors include sociodemographic, clinical and personal characteristics at baseline. Relevant mediators and moderators identified in step 2 will also be taken into account when building the prediction model. Subsequently, the predictors included in the model will be checked for interaction with treatment by introducing interaction terms into the model, and evaluating their contribution to the model. We will calculate the probabilities of success for the different categories of the predictors interacting with treatment, accordingly. The performance of the regression model will be evaluated using the Hosmer-Lemeshow goodness-of-fit test. The discriminative ability of the regression model will be evaluated by the area under the receiver operating characteristics (ROC) curve and its 95% confidence interval. In the fourth year, we will validate the prediction model. Internal validation of the model will be determined by a bootstrapping procedure with 200 replications. In each replication, a random sample from the original dataset is drawn with replacement. We will multiply the regression coefficients by the shrinkage factor derived from the bootstrapping procedures to quantify the amount of optimism and to correct for overfitting if necessary. For external validation we will use data from randomized controlled trials in the POLARIS database that are not used for model building (e.g., from the expanded database).

The clinical prediction model will be translated into an easy-to-use clinical decision rule that assists patients and clinicians in making the most objective, evidence-based and well-considered choice for optimal rehabilitation and supportive care program to improve the quality of life. This model will guide treatment choice and makes it possible to predict which patient benefit most from a specific treatment.