Skip to main content

Table 2 Is the prediction model ready for implementation?

From: Evaluating the impact of prediction models: lessons learned, challenges, and recommendations

1. Assess the current state of scientific evidence
 A prediction model should at least have been validated once to assess its predictive performance in new patients or in a new setting. Subsequent diagnostic and therapeutic steps should also have a valid scientific base.
2. Verify the predictive performance of the prediction model in the new setting
 Local practice, medical care, and patient population may not be similar to the setting in which the prediction model was derived. Consider possible differences between the two settings.
3. Tailor the prediction model to optimize the predictive performance in the new setting
 An insufficient predictive performance in the new setting requires a model update. Even simple adjustments may overcome poor performance in the new setting.
4. Develop a real-time strategy to handle missing predictor values when using the model
 Multivariable imputation is preferred over simply omitting predictors. Other predictors of the model, additional patient information, and information about the local clinical process may be used to estimate missing predictor values.