Constructing a random forest model based on clinical characteristics of uremic patients treated with hemodialysis to screen the risk factors affecting anxiety

Objective To construct a random forest prediction model based on clinical features of uremic patients, and assess its value in predicting anxiety among regular maintenance hemodialysis (MHD) patients.

Methods A total of 162 uremic patients treated with MHD were selected as the research subjects. According to the score of the hospital anxiety and depression scale (HADS-14), the patients were divided into the anxiety group (41 cases) and the non-anxiety group (121 cases). The baseline data and clinical characteristics were compared between two groups; Binary logistic regression was used to analyze the influencing factors of anxiety in patients, and the R (R4.1.0) software package and randomForest program package were applied to construct the random forest model. The receiver operating characteristic (ROC) curve was drawn to analyze the risk factors of the random forest model for predicting anxiety in uremic patients after MHD treatment.

Results The HADS-A score of uremic patients after MHD treatment was (5.46 ± 2.68) points. The proportions of body mass index (BMI) ＜ 18.50 kg/m², pruritus, hemoglobin (Hb) ＜ 90 g/L, and Pittsburgh sleep quality index (PSQI) score ≥11 points in the anxiety group were higher than those in non-anxiety group (P ＜ 0.01). The results of multivariate logistic regression analysis showed that the BMI ＜ 18.50 kg/m², pruritus, Hb ＜ 90 g/L, and PSQI score ≥11 points were the risk factors of anxiety in uremic patients after MHD treatment (OR = 3.654, 3.147, 5.287, 5.343, P ＜ 0.01). The %IncMse scoring was adopted, and the feature importance was ranked. Among them, the top three importance scores were Hb, PSQI score and BMI, respectively. The %IncMse × 10^–2 were 23.781%, 22.881% and 22.084%, respectively. P = 0.01 and R² = 0.220 were found in the random forest model. The AUC of the random forest model established based on BMI, pruritus, Hb and PSQI score for predicting anxiety in uremic patients after MHD treatment was 0.757, the sensitivity was 0.911, the specificity was 0.672, and the Youden index was 0.583, which showed good predictive efficacy.

Conclusions The BMI, pruritus, Hb, and PSQI scores are the important influencing factors of anxiety in patients with uremia after MHD treatment. The decision tree model constructed based on these factors has high predictive value.