Impact of a high-profile sudden cardiac death on emergency department utilization: a multi-campus interrupted time series and exploratory phenotyping study

Figures

• 

  Fig. 1. Interrupted time series analysis of daily emergency department volumes across main and internal validation cohorts. Notes: longitudinal trajectories of daily emergency department (ED) visit volumes are analyzed across a 30-day window centered on the high-profile sudden cardiac death (SCD) event (March 24, 2026). Cohort organization: panels (A, C, E) represent the main cohort; panels (B, D, F) represent the internal validation cohort. Rows correspond to total ED volume (top), cardiovascular department (middle), and respiratory department (bottom; internal control). Visualization: solid black lines indicate segmented linear regression (ITS) trends. Dashed colored lines represent LOESS smoothing (fraction = 0.6) to capture non-linear underlying patterns. Circles denote daily observed visit counts. Analytical features: the red vertical dotted line marks the SCD event. Δ%) labels quantify the percentage change in mean daily volume during the post-event period relative to the 15-day baseline. Standardization: for each row, Y-axes are synchronized between cohorts to ensure direct comparability. Notably, the cardiovascular (row 2) and respiratory (row 3) panels utilize a common Y-axis…

• 

  Fig. 2. Temporal and demographic heterogeneity of the post-event emergency department surge. Notes: (A) Average daily ED visits stratified by operational shifts in the main cohort. Error bars represent 95% confidence intervals. (B) Heatmap illustrating the standardized percentage change in visit volumes across cohorts and shifts, highlighting a consistent nocturnal escalation. (C) Forest plot depicting the interrupted time series (ITS) estimated level change (representing absolute additional daily visits) across specific patient strata. Points denote point estimates, and horizontal lines represent 95% confidence intervals. Note that the significant intervention effect is concentrated in the <45 years and 45–65 years groups, while the >65 years group shows no significant change. (D) A 24-hour hourly distribution curve comparing average visits per hour before and after the event. The shaded gray area represents the night shift duration (20:00–08:00), visualizing the sustained and nocturnal-heavy nature of the event-associated surge

• 

  Fig. 3. Age demographic evolution of ED presentations following the event. Notes: combined kernel density estimation (KDE) plot (top) and horizontal boxplots (bottom) illustrating the age demographics of the main cohort. The blue distribution represents the 15-day baseline period, and the red distribution denotes the 15-day post-event period. Vertical dashed lines and internal boxplot markers indicate the respective median ages. The black arrow highlights the discrete yet statistically significant leftward shift toward a younger patient profile (Welch’s t-test, p < 0.001). The high degree of overlap between the two periods, combined with the significant p-value, reflects the nuanced impact of media-driven health-related concerns on the internet-engaged population

• 

  Fig. 4. Machine learning attribution and cross- cohort consistency of the media-induced surge population. Notes: (A, B) SHAP summary beeswarm plots illustrating global feature attribution in the main and internal validation cohorts, respectively. Features are ranked by their mean absolute SHAP values. Each point represents an individual patient; color indicates the feature value (red: high; blue: low), and the position on the x-axis indicates the impact on the event-driven surge. Age and nighttime presentation (NP) consistently emerged as the primary drivers across both cohorts. (C) SHAP dependence plot demonstrating the non-linear relationship between patient age and surge impact. A sharp escalation in risk is observed for individuals younger than 45 years (red dashed line), an effect further exacerbated during nocturnal hours (color-coded by NP). (D) Inter- cohort feature consistency scatter plot. The x-axis and y-axis represent the mean absolute SHAP values for each specific feature in the main and internal validation cohorts, respectively. The Pearson correlation coefficient (r = 0.845, p < 0.001) was calculated specifically based on these paired mean SHAP values across all included…