63
Summer months also presented elevated risk, especially during periods of extreme heat. These patterns 
underscore that cardiac arrest risk is influenced not only by individual weather events but also by 
broader seasonal shifts that combine environmental exposures with behavioral and physiological 
changes.
In addition to environmental drivers, the model highlighted the important role of socioeconomic and 
demographic factors in shaping OHCA risk. The proportion of individuals living below the poverty level 
emerged as one of the strongest non-environmental predictors, followed by education level (high school 
attainment), racial composition, and median age. These variables likely reflect differences in baseline 
health status, access to care, chronic disease burden, and resilience to environmental stressors. Notably, 
these factors remained consistently predictive across regions of varying population size and composition, 
demonstrating that social vulnerability and environmental exposure jointly influence OHCA incidence. 
This reinforces the importance of incorporating both environmental and socioeconomic context into 
predictive models to better identify high-risk populations and guide targeted prevention and response 
strategies.
Unlike traditional statistical approaches, the machine learning model can continuously improve 
prediction accuracy through iterative learning. The model was able to generate reliable forecasts of 
daily OHCA incidence up to seven days in 
advance across national, state, and local 
levels, as well as in regions not included 
in initial training. Figure 1 presents the 
testing results for same-day, 3-day-ahead, 
and 7-day-ahead predictions in external 
settings. The light blue lines indicate the 
observed daily incidence per 100,000 of out-
of-hospital cardiac arrests in the registry-
participating areas. The yellow lines indicate 
the predicted daily incidence per 100,000 
by the XGBoost gradient boosting model 
using predictors selected by invariant causal 
prediction. This type of predictive capability 
is particularly influential for public health 
and emergency response systems, as it 
enables early warning of high-risk periods, 
more efficient allocation of EMS resources, 
and the potential for targeted community 
interventions during periods of elevated risk.
Figure 1. Observed versus predicted incidence of out-of-
hospital cardiac arrest in the invariant causal prediction 
model at varying time intervals. 
From Nakashima T. et al., Development and evaluation of a machine learning 
model predicting out-of-hospital cardiac arrest using environmental factors, npj 
Digital Medicine (2025). Used with Permission. Licensed under CC BY NC ND 4.0. 
https://pubmed.ncbi.nlm.nih.gov/41430486/

View this content as a flipbook by clicking here.