Dynamic biomarker trajectories in the first 72 h after infarct-related cardiac arrest: a novel approach to early risk stratification

TL;DR

This study introduces a new method using dynamic biomarker patterns in the first 72 hours after a heart attack-related cardiac arrest to better predict patient outcomes.

Contribution

The novel approach uses serial biomarker trajectories and machine learning to improve early risk stratification after cardiac arrest.

Findings

Survivors and non-survivors showed distinct biomarker patterns by day three post-arrest.

Dynamic biomarker changes over 72 hours independently predicted mortality.

Machine learning techniques like t-SNE revealed outcome-related separations in patient profiles.

Abstract

Central Illustration. Trajectory of biomarker levels following infarct-related cardiac arrest. (A) Radar plots illustrate the relationship between biomarker levels and in-hospital outcomes from day one to day three following infarct-related cardiac arrest. The dashed black line represents the overall cohort mean (set to 0), with each concentric grey ring denoting 0.1 standard deviation increments. Biomarker values for survivors are shown as a green line, while those for non-survivors are shown in red. Deviations above the cohort mean extend beyond the dashed line, while below-average values fall within it, allowing visual comparison of multivariate biomarker profiles over time. (B) The t-distributed stochastic neighbor embedding (t-SNE) algorithm is a dimensionality reduction technique that transforms complex, high-dimensional biomarker data into a two-dimensional space for visual interpretation. By estimating the pairwise similarity of patients based on biomarker patterns, t-SNE projects them so that clinically similar profiles are positioned closer together, while dissimilar profiles are spaced further apart. In this plot, each dot represents an individual patient, colored by hospital outcome (green: survival, red: death). Clustering was performed using biomarker values from day 1 and day 3, revealing distinct patterns and outcome-related separations across the study population.

Cardiac arrest caused by acute myocardial infarction (AMI) is associated with high mortality. Although risk stratification scores exist, they rely primarily on static variables obtained at admission, which do not capture the dynamic pathophysiology of the post-resuscitation phase. This study aimed to evaluate the prognostic value of serial biomarker trajectories during the first 72 h after AMI-induced cardiac arrest.

In this single-center cohort study, 181 patients with AMI-induced cardiac arrest between 2018 and 2024 were analyzed. Routinely measured laboratory biomarkers were assessed over the first three days in the intensive care unit (ICU). Multivariable logistic regression models adjusted for key clinical covariates were used to evaluate associations between biomarker trajectories and in-hospital mortality. Secondary analyses included t-distributed stochastic neighbor embedding cluster (machine learning), radar, Sankey and trend plots to visualize biomarker patterns in survivors and non-survivors.

Of the 181 patients, 65.2% survived to hospital discharge. Survivors and non-survivors showed overlapping biomarker profiles on day one, with clearer separation emerging by day three. Non-survivors demonstrated progressive multi-organ dysfunction, including elevated levels of creatinine, potassium, creatine kinase, lactate, neuron-specific enolase, leukocytes and persistent coagulopathy, while survivors showed restoration of physiological homeostasis. Several biomarkers and their dynamic changes over 72 h independently predicted mortality. Cluster, radar, Sankey and trend plot analyses supported the concept of diverging physiological trajectories between survivors and non-survivors over time.

In patients who survive the initial critical phase after cardiac arrest, early prognostication remains limited due to evolving clinical trajectories. Admission biomarkers alone are insufficient for making definitive decisions. The post-resuscitation period represents a critical “second hit” characterized by systemic inflammation and organ dysfunction. Integrating serial biomarker trends into dynamic risk models, such as with machine learning, offers a more individualized and accurate approach to post-cardiac arrest prognostication and care.

Figures