# Time-series clustering analysis reveals distinct patterns of cytomegalovirus viremia in critically ill adults

## TL;DR

A study used time-series clustering to identify three distinct patterns of CMV viremia in ICU patients, which varied in viral load, treatment, and survival outcomes.

## Contribution

This study introduces a novel application of time-series clustering to classify CMV viremia trajectories in critically ill adults.

## Key findings

- Three distinct CMV viremia trajectory clusters were identified with different viral load patterns and treatment responses.
- Cluster 1 showed high initial viremia and rapid decline with treatment, while cluster 2 had increasing viral load despite partial treatment.
- Survival rates varied significantly between clusters, with cluster 2 and 3 showing higher 1-year survival compared to cluster 1.

## Abstract

Critically ill patients are at increased risk for cytomegalovirus (CMV) reactivation, which is associated with poorer clinical outcomes. However, little is known about the longitudinal viremia trajectories in this population.

This retrospective single-center study was conducted in a medical ICU and included patients with CMV viremia ≥ 1000 International Units CMV–DNA per milliliter whole blood (IU/mL) treated between March 2014 and April 2021. Time-series clustering was applied to identify subgroups of patients with similar longitudinal viremia trajectories.

82 patients were included in the final analysis. Time-series clustering identified three distinct clusters: (1) patients with initial high viremia (median 46,700 IU/mL), 94% receiving treatment and showing subsequent steep reduction of viremia; (2) patients with moderate viremia (median 2720 IU/mL) and subsequent increase in viral load, treated in 52%; and (3) patients with moderate viremia (median 3120 IU/mL), 63% receiving treatment and showing stable viral load in follow-up measurements. No significant differences were identified between the clusters with respect to patient characteristics, including underlying immunosuppression. With respect to disease severity, the Acute Physiology and Chronic Health Evaluation II (APACHE-II) score was highest in cluster 3 and among patients without follow-up CMV–DNA measurements (P = 0.029), while the Sequential Organ Failure Assessment (SOFA) score demonstrated a similar directional trend without reaching statistical significance. Survival differed significantly between the clusters in the Kaplan–Meier estimate (p = 0.008); however, absolute 1-year survival was low across all clusters (cluster 1: 0%, cluster 2: 33%, cluster 3: 32%, patients without follow-up CMV measurement: 14%; p = 0.062). Probable CMV pneumonia with respiratory symptoms and CMV–DNA detection in bronchoalveolar lavage fluid was the most common disease manifestation (cluster 1: 35%; cluster 2: 28%; cluster 3: 7.5%; patients without follow-up CMV measurement: 23%; p = 0.040).

In this hypothesis-generating study, time-series clustering analysis identified three subgroups with distinct longitudinal viremia trajectories which significantly differed in viral load, treatment decisions and survival over time. The diagnostic and therapeutic relevance of longitudinal CMV viremia trajectories and the optimal CMV–DNA threshold for treatment initiation in ICU patients remain undefined and might differ from other cohorts.

The online version contains supplementary material available at 10.1186/s40635-026-00866-9.