Hidden Frailty in the ICU: Joint Model Reveals How Death and Discharge Distort Infection Recurrence Estimates

Patients in intensive care units often experience recurrent hospital-acquired infections, and a new methodological study argues that the statistical tools commonly used to count these infections can distort the picture when they treat death and discharge as simple censoring rather than as competing terminal outcomes. The study, published in BMC Medical Research Methodology by Zeytu Gashaw Asfaw, Susanne Weber, Derek Hazard and Martin Wolkewitz, applied a joint frailty model to a publicly available prospective single-center neuro-ICU cohort to show what changes when terminal events are modeled explicitly.

The core problem is one of competing risks. When a patient dies or is discharged, the opportunity to observe further infections ends.

The joint frailty model incorporates two ingredients beyond a standard survival analysis. The first is the frailty term, an unobserved patient-level heterogeneity component capturing the fact that some individuals are intrinsically more prone to infections than others, for reasons not recorded in the data. The second is an association parameter, denoted alpha, which quantifies how strongly the rate of recurrent infections is tied to the hazard of the terminal event. A high positive alpha means patients who accumulate infections faster are also more likely to reach the terminal outcome sooner.

In the first analysis, the researchers combined death and discharge into a single composite terminal event. The model estimated a frailty variance of theta equal to 0.070, indicating individual heterogeneity in infection risk, and a high positive dependency of alpha equal to 4.40 between recurrent infections and the terminal outcome.

The team then extended the model to treat death and discharge as two distinct terminal outcomes, each with its own association with the recurrent infection process. This separate-outcome specification revealed a substantially larger individual heterogeneity, with a frailty variance of theta equal to 0.286. The association parameters remained high and positive: alpha one, linking recurrent infections to death, was estimated at 6.86, and alpha two, linking recurrent infections to discharge, at 5.81.

The choice of terminal event specification also changed what the covariates appeared to do. In the model with separate terminal outcomes, neither sex nor the Charlson Comorbidity Index showed a statistically significant effect on recurrent infections. Yet under the composite terminal event model, female patients had a significantly lower hazard of recurrent infections compared with males, with a hazard ratio of 0.90, a 95 percent confidence interval of 0.82 to 0.98, and a p value of 0.028.

The covariate impacts varied according to the terminal outcome as well. With separate terminal events, a higher Charlson Comorbidity Index was associated with an increased hazard of death, with a hazard ratio of 1.11, a 95 percent confidence interval of 1.05 to 1.18, and a p value below 0.001. The same index was associated with a decreased hazard of discharge, with a hazard ratio of 0.93, a 95 percent confidence interval of 0.90 to 0.97, and a p value below 0.0001.

To probe robustness, the researchers employed recurrent event analysis approaches based on both total time scales and gap time scales as sensitivity analyses. Total time scales measure time from admission to each infection, while gap time scales reset the clock after each event. All analyses were performed in R version 4.5.1 using the frailtypack package, version 3.7.0, making the workflow reproducible for other research groups working with similar cohort data.

The authors conclude that the study reveals a strong link between recurrent infections and terminal outcomes, with moderate heterogeneity among patients. Crucially, they report that neglecting terminal events leads to an overestimation of the recurrence burden and reduces the observed effects.

The neuro-ICU cohort analyzed here was a single center and publicly available, so the specific parameter estimates reflect that population.

Subject of Research: Joint frailty modeling of recurrent hospital-acquired infections with competing terminal outcomes in a neuro-ICU cohort

Article Title: Joint frailty analysis of recurrent hospital-acquired infections highlights the importance of accounting for competing terminal outcomes

Article References: Asfaw, Z. G., Weber, S., Hazard, D., & Wolkewitz, M. (2026). Joint frailty analysis of recurrent hospital-acquired infections highlights the importance of accounting for competing terminal outcomes. BMC Medical Research Methodology. https://doi.org/10.1186/s12874-026-03073-y

Image Credits: AI Generated

DOI: 10.1186/s12874-026-03073-y

Keywords: hospital-acquired infections, joint frailty model, competing risks, intensive care unit, recurrent events, terminal outcomes, frailty variance, Charlson Comorbidity Index, survival analysis, neuro-ICU cohort, frailtypack, statistical methodology