Predicting mortality in hemodialysis patients using machine learning analysis

FIGURE 1:. Cohort selection flow chart. The comparison of mortality models was performed on nine different cohorts that are represented in the dashed boxes. On the left, the number of patients included to evaluate the prediction of mortality at 6 months, 1 year and 2 years. For each mortality prediction period there were three separate analyses according to the minimal period after the first HD session used for the collection of baseline data (input variables: 30, 60 or 90 days).

FIGURE 2:. Prediction of mortality by random forest analysis. The dashed line represents the value of the AUC of the mortality prediction ROC curve obtained by the random forest regression model. Each dot shows the influence of each variable on the AUC value that is obtained if the effect of the specified variable is turned off. This is achieved by randomly changing the values of the variable in the test set. Each graph represents the result of mortality prediction (at 6 months, 1 year and 2 years) and for each mortality prediction period there were three separate analyses according to the minimal period of days after the first HD session used for the collection of the baseline data (input variables 30, 60 and 90 days). (A) Prediction of mortality at 6 months and a 30-day period after the first HD for baseline data collection. (B) Prediction of mortality at 6 months and a 60-day period after the first HD for baseline data collection. (C) Prediction of mortality at 6 months and a 90-day period after the first HD for baseline data collection. (D) Prediction of mortality at 1 year and a 30-day period after the first HD for baseline data collection. (E) Prediction of mortality at 1 year and a…

FIGURE 3:. Influence of variables on mortality by logistic regression. The dashed line represents the value of the AUC of the mortality prediction ROC curve obtained by the logistic regression model. Each dot shows the AUC value obtained if the effect of the specified variable is removed. Each graph represents the result of mortality prediction at 6 months, 1 year and 2 years and for each mortality prediction period there were three separate analyses according to the minimal period of days after the first HD session used for the collection of baseline data (input variables 30, 60 and 90 days). (A) Prediction of mortality at 6 months and a 30-day period after the first HD for baseline data collection. (B) Prediction of mortality at 6 months and a 60-day period after the first HD for baseline data collection. (C) Prediction of mortality at 6 months and a 90-day period after the first HD for baseline data collection. (D) Prediction of mortality at 1 year and a 30-day period after the first HD for baseline data collection. (E) Prediction of mortality at 1 year and a 60-day period after the first HD for baseline data collection. (F) Prediction of mortality at 1 year and a 90-day period after the…