A machine learning based approach to identify carotid subclinical atherosclerosis endotypes

Abstract

Aims: To define endotypes of carotid subclinical atherosclerosis.

Methods and results: We integrated demographic, clinical, and molecular data (n = 124) with ultrasonographic carotid measurements from study participants in the IMPROVE cohort (n = 3340). We applied a neural network algorithm and hierarchical clustering to identify carotid atherosclerosis endotypes. A measure of carotid subclinical atherosclerosis, the c-IMTmean-max, was used to extract atherosclerosis-related features and SHapley Additive exPlanations (SHAP) to reveal endotypes. The association of endotypes with carotid ultrasonographic measurements at baseline, after 30 months, and with the 3-year atherosclerotic cardiovascular disease (ASCVD) risk was estimated by linear (β, SE) and Cox [hazard ratio (HR), 95% confidence interval (CI)] regression models. Crude estimates were adjusted by common cardiovascular risk factors, and baseline ultrasonographic measures. Improvement in ASCVD risk prediction was evaluated by C-statistic and by net reclassification improvement with reference to SCORE2, c-IMTmean-max, and presence of carotid plaques. An ensemble stacking model was used to predict endotypes in an independent validation cohort, the PIVUS (n = 1061). We identified four endotypes able to differentiate carotid atherosclerosis risk profiles from mild (endotype 1) to severe (endotype 4). SHAP identified endotype-shared variables (age, biological sex, and systolic blood pressure) and endotype-specific biomarkers. In the IMPROVE, as compared to endotype 1, endotype 4 associated with the thickest c-IMT at baseline (β, SE) 0.36 (0.014), the highest number of plaques 1.65 (0.075), the fastest c-IMT progression 0.06 (0.013), and the highest ASCVD risk (HR, 95% CI) (1.95, 1.18-3.23). Baseline and progression measures of carotid subclinical atherosclerosis and ASCVD risk were associated with the predicted endotypes in the PIVUS. Endotypes consistently improved measures of ASCVD risk discrimination and reclassification in both study populations.

Conclusions: We report four replicable subclinical carotid atherosclerosis-endotypes associated with progression of atherosclerosis and ASCVD risk in two independent populations. Our approach based on endotypes can be applied for precision medicine in ASCVD prevention.

Keywords: ASCVD; Artificial intelligence; Atherosclerosis; Biological markers; Endotype; Progression of atherosclerosis.

Graphical Abstract

Figure 1

Scatter plot showing the top 10 endotype-specific variables and their impact on endotype prediction. Z-score_sample is the standardized value (scaling to mean of 0 and standard deviation of 1) for each variable. The mean of each variable (Z-score_sample) equals to zero. Z-score_sample lower than zero are represented in light blue and Z-score_sample over the zero in red. The higher the Z score (red dots), the higher the value of original variable. SHAP value: SHapley Additive exPlanations value. It measures the predicted contribution of a variable to one of the four endotypes for each individual. SHAP value of 0 means that a variable does not contribute to predict the endotype. A positive SHAP value indicates that a variable is very likely a predictor of the endotype, while a negative SHAP value indicates in that individual, the variable is not a predictor of the endotype. SHAP_endotype: SHAP_endotype is the mean of |SHAP value| (absolute value). It measures the overall contribution for prediction for a variable to one of the four endotypes (reported with different colours in the Supplementary material online, Figure S8). The higher the log10 SHAP_endotype, the higher the likelihood that the prediction model is accurate. In general, the stacking model predicts endotypes 1 and 4 more reliably as compared to endotypes 2 and 3. In each subplot of endotype, the variables’ importance is ranked from left to right. We randomly sampled 300 individuals in the derived dataset to estimate their SHAP value.

Figure 2

Relative concentrations of biomarkers assigned to endotypes. Biomarkers are grouped into modules of co-expressed protein estimated by the weighted co-expression network analysis to define both protein co-expression modules and relative concentration of each biomarker in the four endotypes. Circulating levels of biomarkers were Z-standardized. Light blue represents the values below the standardized mean, white represents the standardized mean, and red represents the values above standardized mean.

Figure 3

Distribution of ultrasonographic measures across the four endotypes in the derived dataset. Mean value is reported inside a square for each violin plot, while the minimal, 25% quantile, median, 75% quantile, and maximal values are presented at the right side of each violin plot. Violin plots: top-left panel shows the distribution among the four endotypes of the c-IMT_mean-max at baseline; top-middle panel shows the distribution among the four endotypes of the number of plaques at baseline; top-right panel shows the distribution the four endotypes of the area of plaque (including bulb) at baseline; the bottom-right panel shows the c-IMT_{fastest-progr} after 30 months follow-up; and the bottom-middle panel shows the area of plaque (including bulb) progression.

Figure 4

Atherosclerosis-related cardiovascular disease and cardiac events survival curves. The subplots in the upper panel display ASCVD (left), cardiac event (middle), and cerebrovascular event (right) in IMPROVE. The subplots in the bottom panel display myocardial infarction (MI) or ischaemic stroke (IS) (left), myocardial infarction (middle), and ischaemic stroke (right) in PIVUS. Red, endotype 1; light blue, endotype 2; green, endotype 3; dark blue, endotype 4.