Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Bayesian Prevalence Incidence Cure model for estimating survival using Electronic Health Records with incomplete baseline diagnoses

Matilda Pitt, Robert J. B. Goudie

TL;DR

This work tackles the challenge of estimating baseline prevalence, time-to-incidence, and cure fractions from Electronic Health Records when baseline diagnoses may be missing and not all individuals experience the event. It introduces a Bayesian three-component Prevalence-Incidence-Cure (PIC) model that combines a PI-style mixture with a latent cure fraction, models component membership with covariates via multinomial logistic regression, and models incidence times with a Weibull distribution. Through simulations and a Diabetic Macular Oedema application, the PIC framework demonstrates reduced bias in survival estimates and reveals covariate-driven effects on prevalence, incidence, and cure that PI models can miss, especially when a nonzero cure fraction is present. The method provides a principled, priors-aware Bayesian approach for real-world epidemiological analyses in EHR data, with potential applicability to screening and chronic disease settings where missing baseline information and cure-like outcomes are common.

Abstract

Retrospective cohorts can be extracted from Electronic Health Records (EHR) to study prevalence, time until disease or event occurrence and cure proportion in real world scenarios. However, EHR are collected for patient care rather than research, so typically have complexities, such as patients with missing baseline disease status. Prevalence-Incidence (PI) models, which use a two-component mixture model to account for this missing data, have been proposed. However, PI models are biased in settings in which some individuals will never experience the endpoint (they are 'cured'). To address this, we propose a Prevalence Incidence Cure (PIC) model, a 3 component mixture model that combines the PI model framework with a cure model. Our PIC model enables estimation of the prevalence, time-to-incidence, and the cure proportion, and allows for covariates to affect these. We adopt a Bayesian inference approach, and focus on the interpretability of the prior. We show in a simulation study that the PIC model has smaller bias than a PI model for the survival probability; and compare inference under vague, informative and misspecified priors. We illustrate our model using a dataset of 1964 patients undergoing treatment for Diabetic Macular Oedema, demonstrating improved fit under the PIC model.

A Bayesian Prevalence Incidence Cure model for estimating survival using Electronic Health Records with incomplete baseline diagnoses

TL;DR

This work tackles the challenge of estimating baseline prevalence, time-to-incidence, and cure fractions from Electronic Health Records when baseline diagnoses may be missing and not all individuals experience the event. It introduces a Bayesian three-component Prevalence-Incidence-Cure (PIC) model that combines a PI-style mixture with a latent cure fraction, models component membership with covariates via multinomial logistic regression, and models incidence times with a Weibull distribution. Through simulations and a Diabetic Macular Oedema application, the PIC framework demonstrates reduced bias in survival estimates and reveals covariate-driven effects on prevalence, incidence, and cure that PI models can miss, especially when a nonzero cure fraction is present. The method provides a principled, priors-aware Bayesian approach for real-world epidemiological analyses in EHR data, with potential applicability to screening and chronic disease settings where missing baseline information and cure-like outcomes are common.

Abstract

Retrospective cohorts can be extracted from Electronic Health Records (EHR) to study prevalence, time until disease or event occurrence and cure proportion in real world scenarios. However, EHR are collected for patient care rather than research, so typically have complexities, such as patients with missing baseline disease status. Prevalence-Incidence (PI) models, which use a two-component mixture model to account for this missing data, have been proposed. However, PI models are biased in settings in which some individuals will never experience the endpoint (they are 'cured'). To address this, we propose a Prevalence Incidence Cure (PIC) model, a 3 component mixture model that combines the PI model framework with a cure model. Our PIC model enables estimation of the prevalence, time-to-incidence, and the cure proportion, and allows for covariates to affect these. We adopt a Bayesian inference approach, and focus on the interpretability of the prior. We show in a simulation study that the PIC model has smaller bias than a PI model for the survival probability; and compare inference under vague, informative and misspecified priors. We illustrate our model using a dataset of 1964 patients undergoing treatment for Diabetic Macular Oedema, demonstrating improved fit under the PIC model.
Paper Structure (19 sections, 30 equations, 2 figures, 14 tables)

This paper contains 19 sections, 30 equations, 2 figures, 14 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. PIC setup description
  4. General prevalence-incidence-cure model
  5. Multinomial Logistic-Weibull PIC model
  6. Priors
  7. Mixture model
  8. Incident event time model
  9. Complete prior specification
  10. Model estimation
  11. Simulation study
  12. Data generating process
  13. Comparator method and performance metrics
  14. Prior specification
  15. Simulation results
  16. ...and 4 more sections

Figures (2)

  • Figure 1: Illustrative examples of the disease and observation process for $7$ patients, including the true disease development time $t_i$ and the times of their diagnostic test. The times $l_i$ and $r_i$ of the last negative and first positive test respectively are also shown.
  • Figure 2: Time to a VA value of 70 or above after starting treatment in the DMO application under three different models. The shaded areas represent the 95% credible interval for the PIC model and the 95% confidence interval for the PI and NPMLE model.