Skip to main content

Table 3 How to present the model predictions?

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

Facilitators: features that increase the ease of use of a prediction model
 F.1 Add a decision recommendation to the predicted probabilities
  Directive prediction tools may be easier for physicians to use in their decision making than assistive prediction tools that provide only predicted probabilities without decision recommendations.
 F.2 Automatic calculation and presentation of the model’s probability within the physician’s workflow
  Minimizing manual predictor value entry and integrating the estimation of the model’s probability in the electronic patient record will facilitate the ease of use of a prediction model for care providers.
 F.3 Provide the reasoning or research evidence behind the predicted probability
  Enhances face value, acceptation and belief in the model, and thus the willingness to use the model’s probabilities to guide decision making.
Barriers that may decrease the ease of use of a prediction model
 B.1 A predicted probability may be difficult to use in decision making, especially without corresponding recommendations
  Weighing the numerical probabilities with other available information will require more cognitive effort from physicians when the probabilities are presented without a corresponding recommendation on subsequent treatment or additional diagnostic testing.
 B.2 When the targeted physicians use an intuitive rather than analytical process of decision making
  When an existing decision-making process is mostly intuitive, it may require more cognitive effort to use probabilistic knowledge in decision making.
 B.3 When the predicted outcome is not a main concern for the physicians
  Physicians will not prioritize their time and efforts to use a prediction model in their decisions, when they consider other problems or outcomes to be more important.
 B.4 A prediction model does not weigh the benefits and risks of treatment or additional diagnostics regarding the patient’s (co)morbidity
  When a physician has more sources of information about the benefits and risks of subsequent treatment decisions, she/he will still have to weigh the model’s predicted probability from the model against this information, which is often perceived as cumbersome.