Modelling Future Outcomes Using Repeatedly Measured Predictors: Prediction of ICU Admission in Patients with COVID-19

Background:
Repeated biomarker measurements are routinely collected in clinical care and provide dynamic information on disease progression. However, epidemiologic applications that explicitly incorporate repeatedly measured predictors to estimate future binary outcomes remain limited.

Objective:
To demonstrate a two-stage modelling framework for incorporating longitudinal predictors into risk prediction of a future binary outcome, using ICU admission among hospitalized COVID-19 patients as an example.

Methods:
We conducted a retrospective cohort study including 372 patients admitted to King Faisal Specialist Hospital and Research Centre (Riyadh, Saudi Arabia) with PCR-confirmed COVID-19 between March and September 2020. Biomarkers were measured every 72 hours. C-reactive protein (CRP) was selected to illustrate the approach.

A linear mixed-effects model was fitted to repeated log(CRP) measurements to estimate patient-specific intercepts and slopes, accounting for within-patient correlation. Estimated random effects were then included in a logistic regression model to assess their association with ICU admission.

Results:
Increasing and persistently elevated CRP trajectories were strongly associated with ICU admission. For each one-unit increase in the CRP slope, the odds of ICU admission increased by 1.98 (95% CI: 1.52–2.58). Model discrimination was good (AUC = 0.79).

Conclusions:
A two-stage approach provides a practical framework for incorporating repeated predictors into dynamic risk prediction and is broadly applicable to binary outcomes beyond COVID-19.