Statistical primer: using prognostic models to predict the future: what cardiothoracic surgery can learn from...

Area of Science:

• Medical Prognostication
• Statistical Modeling
• Clinical Decision Support

Background:

• Prognostic models are integral to medical assessment, particularly in cardiothoracic surgery, with tools like EuroSCORE being widely adopted.
• Despite their utility, these predictive models face challenges and potential misapplication in clinical practice.
• Understanding the limitations of prognostic models is crucial for accurate patient assessment and treatment planning.

Purpose of the Study:

• To illustrate common issues and pitfalls in the application and interpretation of prognostic models.
• To use a non-medical example, Strictly Come Dancing, to demonstrate the complexities of predictive modeling.
• To emphasize the need for caution when utilizing prognostic models in cardiothoracic surgery and other medical fields.

Main Methods:

• A multivariable prognostic model was developed using data from 19 series of Strictly Come Dancing.
• The model aimed to prospectively predict the outcomes of the 20th series of the show.
• Model performance was evaluated using R-squared and Spearman's rank correlation coefficients.

Main Results:

• An initial model based on demographic data alone showed limited predictive value (R2=0.25, Spearman's=0.22).
• Incorporating early judges' scores significantly improved the model's predictive power (R2=0.40, Spearman's=0.30).
• The findings underscore the importance of including relevant performance metrics beyond basic demographics.

Conclusions:

• Prognostic models require careful and judicious use by researchers and clinicians.
• Extrapolating conclusions solely from demographic data in prognostic models can lead to inaccuracies.
• Models must adequately capture essential prognostic information to be reliable in clinical decision-making.