A simple method to adjust clinical prediction models to local circumstances

Area of Science:

• Clinical epidemiology
• Biostatistics
• Health informatics

Background:

• Clinical prediction models are crucial for estimating disease risk.
• Poor performance of existing models often leads to new model development.
• Model updating using new data presents an alternative to developing new models.

Purpose of the Study:

• To demonstrate a simple updating method for clinical prediction models.
• To evaluate the performance of an updated model in a new patient cohort.
• To assess the utility of model updating when prediction accuracy is suboptimal.

Main Methods:

• A multivariable logistic regression model predicting severe postoperative pain was developed.
• The model's predictive performance was assessed in a separate cohort of 1,035 surgical patients.
• Model calibration was improved by adjusting the intercept based on new data.

Main Results:

• The original model's predicted risks were systematically higher than observed in the validation cohort (62% vs. 36% incidence).
• The updated model demonstrated improved calibration, aligning predicted risks more closely with observed outcomes.
• Model updating enhanced the accuracy of risk estimates in the new patient population.

Conclusions:

• Updating clinical prediction models with new data is a viable strategy to improve performance.
• A simple intercept adjustment can effectively update models for new patient populations.
• Updated models provide more reliable risk estimates, enhancing clinical decision-making.