Prediction Rules and Modeling

Prediction Models

Clinical prediction models may provide the evidence-based input for shared decision making, by providing estimates of the individual probabilities of risks and benefits. Clinical prediction models are sometimes also referred to as prognostic models or nomograms. Clinical prediction models combine a number of characteristics (e.g., related to the patient, the disease, or treatment) to predict a diagnostic or prognostic outcome. Typically, between 2 and 20 predictors are considered. The number of publications with clinical prediction models has increased steeply in recent years in various medical fields.

Applications of Prediction Models

Prediction models are valuable for medical practice and for research purposes. In public health, models may help target preventive interventions to subjects at relatively high risk of having or developing a disease. In clinical practice, prediction models may inform patients and their treating physicians on the probability of a diagnosis or a prognostic outcome. Prognostic estimates may, for example, be useful to assist in planning of the remaining lifetime in terminal disease or give hope for recovery if a good prognosis is expected after an acute event such as a stroke. Classification of a patient according to his or her risk may also be useful for communication among physicians, for research purposes, and for benchmarking.

Prediction models may also assist medical decision making, for example, as part of a decision support system. In the diagnostic workup, predictions can be useful to estimate the probability that a disease is present. When the probability is relatively high, treatment is indicated; if the probability is very low, no treatment is indicated. For intermediate probabilities of disease, further diagnostic testing is necessary. In therapeutic decision making, treatment should only be given to those most likely to benefit from the treatment. Prognostic predictions may support the weighing of harms versus individual benefits. If risks of a poor outcome are relatively low, the maximum benefit will also be relatively low. Any harm, such as a side effect of treatment, may then readily outweigh any benefits. The claim of prediction models is that [Page 894]better decisions can be made with a model than without and that their predictions are sometimes better than those made by physicians.

In research, prediction models may assist in the design and analysis of randomized trials. Adjustment for baseline risk in the analysis of a trial results in higher statistical power for the detection of a treatment effect. Models are also useful to control for confounding variables in observational research, either in traditional regression analysis or with modern approaches such as propensity scores.

...