Fig. 2: Marginal plots of the top 10 important variables for predicting moderate to severe AKI (at least AKI stage 2) in 48 h. | Nature Communications

Fig. 2: Marginal plots of the top 10 important variables for predicting moderate to severe AKI (at least AKI stage 2) in 48 h.

From: Cross-site transportability of an explainable artificial intelligence model for acute kidney injury prediction

Fig. 2

Each panel demonstrates marginal effects of each of the most impactful features ranked among the top 10 for predicting moderate-to-severe AKI in 48 h with SCr and BUN in the model. The x-axis gives the raw values of each feature and the y-axis gives the logarithmic of estimate odds ratio (i.e., the SHAP value) for moderate-to-severe AKI in 48 h when a feature takes certain value. Each dot represents the average SHAP value over bootstrapped samples with the error bars suggesting a 95% bootstrapped interval. The ‘shaded area’ represents the 95% confidence band for the LOWESS smoother extrapolating across all the dots. The full interactive dashboard can be found at: https://sxinger.shinyapps.io/AKI_shap_dashbd/.

Back to article page