Identifying distinct clusters of ICU survivors by integrating demographic, pre-admission quality of life, and clinical data: a large prospective cohort study

Abstract

Purpose: ICU patients differ in pre-ICU health status, comorbidities, and diagnosis, forming a heterogeneous population with diverse long-term outcomes. This study explored whether clustering ICU patients by demographic, pre-admission quality of life, and clinical data, rather than by diagnosis, could identify subgroups that are more informative for patient-centered outcomes post-ICU.

Methods: Data from the MONITOR-IC prospective cohort study were used. Demographic, pre-admission quality of life, and clinical data from 2361 adult ICU survivors of six hospitals were used to identify clusters, using the k-prototypes algorithm. Data from five additional hospitals (n = 866) were used for external validation. Self-reported physical, mental, and cognitive functioning, and quality of life one year post-ICU were described.

Results: The four identified clusters differed significantly in long-term physical, mental, and cognitive functioning, and quality of life. Cluster-A (n = 204), characterized by a healthy pre-ICU status, high disease severity, low Glasgow Coma Scale, and long ICU stay, had a relatively high quality of life at one year, despite experiencing a mean decline from baseline. Cluster-B (n = 877), also a healthy group before admission but less severely ill at ICU admission, reported the best outcomes. Cluster-C (n = 632) included younger, mostly female patients with moderate impairments both pre- and one-year post-ICU. Cluster-D (n = 648), characterized by a low education level and poor baseline health, reported impaired outcomes one year post-ICU, although improved compared to their pre-admission status. External validation confirmed the generalizability of these results.

Conclusion: This study identified and externally validated four distinct clusters of ICU patients by integrating both clinical and non-clinical data. These clusters, which differed in long-term physical, mental, and cognitive outcomes, challenge conventional disease-based classification, and support a multidimensional approach to define subgroups of ICU patients.

Trial registration: The MONITOR-IC study was registered at ClinicalTrials.gov: NCT03246334.

Keywords: Cluster analysis; Critical care outcomes; Health status; Quality of life.

Fig. 1

Flowchart

Fig. 2

Standardized mean differences of continuous clustering variables (including demographic, pre-admission quality of life, and clinical data) across clusters. Positive and negative values indicate that the variable is relatively high or low in this cluster compared to other clusters. Thus, cluster-A has a low Glasgow Coma Scale and high APACHE IV score compared to other clusters, whereas both clusters A and B have fewer pre-ICU impairments and higher pre-ICU quality of life than clusters C and D. Cluster-C includes younger patients with lower disease severity, while cluster-D stands out due to more pre-ICU impairments and lower pre-ICU quality of life compared to other clusters

Fig. 3

Characteristics of the four identified cluster across all clustering variables (including demographic, pre-admission quality of life, and clinical data), illustrated with boxplots for continuous variables and stacked bar charts for categorical variables. This figure illustrates the low prevalence of comorbidities in cluster-A, and higher prevalence of COPD and diabetes in cluster-D. It also highlights the differences in income source (e.g., more retired patients in clusters B and D; more unemployed patients in clusters C and D), differences in gender (e.g., predominantly female gender in cluster-C), and differences in admission diagnosis (e.g., more cardiac arrest and trauma in cluster-A). Differences in pre-ICU frailty, fatigue, and quality of life between the clusters are also visible

Fig. 4

Quality of life scores of the four identified cluster before admission and one year post-ICU admission, illustrated with boxplots