Prognostic models estimate future outcomes or risks associated with a particular disease for an individual patient. These models use various patient characteristics (like age and smoking status) and clinical data (like blood pressure) to generate predictions about a patient's prognosis. These predictions are used by healthcare professionals in treatment planning, particularly for chronic diseases where predicting the course of the illness can help guide interventions and improve patient outcomes.
One such chronic disease for which many prognostic models have been developed is chronic obstructive pulmonary disease (COPD), a progressive lung disease affecting the ability to breathe. Provided a diagnosis is made early, patients can live for 10 to 20 years. Models have been developed to predict the risk of death up to 10 years after diagnosis, meaning that patients from 10 to 20 years ago are included in the model. However, survival both in the disease area and in the general population may have improved since then. This means that by the time the model is published it may already be, or soon become, out of date.
We will evaluate three existing COPD prognostic models and assess how accurate their predictions are in the current UK population. If the predictions are out of date we will evaluate methods that can be used to update these models to provide more accurate predictions for new patients.
Prognostic models are statistical algorithms that predict the risk of a future event occurring in a particular disease or health condition. They contain multiple predictors such as patient characteristics and clinical data, and are usually developed using logistic or survival regression. The predictions are used by healthcare professionals in treatment prescriptions and disease management, particularly for chronic diseases like cancer and diabetes where predicting the course of the illness can guide interventions and improve patient outcomes.
One such chronic disease for which many prognostic models have been developed is chronic obstructive pulmonary disease (COPD), a progressive lung disease that is particularly prevalent in smokers. The focus of many of these models is mortality at different time points after diagnosis. Models producing longer-term predictions require longer follow-up to estimate the risk of the outcome, meaning that patients diagnosed many years ago have to be included. However, if survival outcomes have improved, the risk for these patients may be different to more recently diagnosed patients, which leads to model miscalibration.
Using recent data we will temporally validate three prognostic models for COPD mortality to evaluate the drift of these models over time, assess the need for recalibration, and if required, recalibrate them for the current UK population.
Period analysis can be used for time-to-event prognostic modelling by using more recent cohorts, but this decreases the sample size which can lead to overfitting. Instead, we will evaluate newly developed methods including temporal recalibration and weighted regression. In temporal recalibration, the predictor effects from the full cohort are preserved while baseline mortality is updated based on a period analysis, giving up-to-date recalibrated predictions that utilise the power of the full sample size. A weighted regression approach accounts for the case-mix to investigate model recalibration by re-weighting the data according to the observed distribution shift.
Mortality (at 1, 5 and 10 years of follow-up following COPD diagnosis)
Sarah Booth - Chief Investigator - University of Leicester
Jonathan Broomfield - Corresponding Applicant - University of Leicester
David Jenkins - Collaborator - University of Manchester
Haya Elayan - Collaborator - University of Manchester
Mark Rutherford - Collaborator - University of Leicester
HES Admitted Patient Care;ONS Death Registration Data;Patient Level Index of Multiple Deprivation