Identification of risk factors for attempted suicide by self-poisoning and a nomogram to predict self-poisoning suicide

Abstract

Purpose: Suicide is a global concern, especially among young people. Suicide prediction models have the potential to make it easier to identify patients who are at a high risk of suicide, but they have very little predictive power when there is a positive value for suicide mortality. Therefore, the aim of the study is to uncover potential risk factors associated with suicide by self-poisoning and further to provide a trustworthy nomogram to predict self-poisoning suicide among poisoned patients.

Methods: This study prospectively enrolled 237 patients who were treated for poisoning at the Fifth Medical Center of PLA General Hospital (Beijing) between May 2021 and May 2022. Patient's basic characteristics, daily activities, mental health status, and history of psychological illnesses were gathered to examine their predictive power for self-poisoning suicide. On developing a prediction model, patients were split 8:2 into a training (n = 196) group and a validation (n = 41) group at random via computer. The training group worked on model development, while the validation group worked on model validation. In this study, the Hosmer and Lemeshow test, accuracy, and area under the curve were the primary evaluation criteria. Shapley Additive exPlanations (SHAP) was determined to evaluate feature importance. To make the prediction model easy for researchers to utilize, it was presented in nomogram format. Two risk groups of patients were identified based on the ideal cut-off value.

Results: Of all poisoned patients, 64.6% committed suicide by self-poisoning. With regard to self-poisoning attempted suicide, multivariate analysis demonstrated that female gender, smoking, generalized anxiety disorder-7 (GAD-7), and beck hopelessness scale-20 (BHS-20) were significant risk factors, whereas married status, relatively higher education level, a sedentary time of 1-3 h per day, higher sport frequency per week, higher monthly income were significant protective features. The nomogram contained each of the aforementioned nine features. In the training group, the area under curve (AUC) of the nomogram was up to 0.938 (0.904-0.972), whereas in the validation group, it reached a maximum of 0.974 (0.937-1.000). Corresponding accuracy rates were up to 0.883 and 0.927, respectively, and the P-values for the Hosmer and Lemeshow test were 0.178 and 0.346, respectively. SHAP demonstrated that the top three most important features were BHS-20, GAD-7, and marital status. Based on the best cut-off value of the nomogram (40%), patients in the high-risk group had a nearly six-time larger likelihood of committing suicide by self-poisoning than patients in the low-risk group (88.68 vs. 15.38%, P < 0.001). The dynamic nomogram was made available at the following address: https://xiaobo.shinyapps.io/Nomogramselfpoisoningsuicide/.

Conclusions: This study proposes a prediction model to stratify patients at a high risk of suicide by self-poisoning and to guide individual preventive strategies. Patients in the high-risk group require further mental health counseling to alleviate anxiety and hopelessness, healthy lifestyle like quitting smoking and exercising more, and restriction of access to poison and psychiatric drugs.

Keywords: mental health; nomogram; prediction model; self-poisoning; suicide.

Figure 1

Patient's flowchart and creation of a nomogram.

Figure 2

A nomogram to predict risk of self-poisoning. The red dot in each characteristic in the supplied case showed the case's current condition. For instance, the red dot was placed at the “No” box because the case was not a smoker. To determine a unique score for the feature, a line was drawn upward from the box to the score axis. All nine features were combined to create the final score (−2.76). We were able to determine the patients' final suicide risk (91.7%) by drawing a line downward to the projected risk axis.

Figure 3

Area under the curve (AUC) for the nomogram. (A) The training group. (B) The validation group. The light blue indicates the AUC, and its value and the optimal cut-off were provided.

Figure 4

Probability curve and discrimination slope for the nomogram. (A) Probability curve for the nomogram in the training group. (B) Discrimination slope for the nomogram in the training group. (C) Probability curve for the nomogram in the validation group. (D) Discrimination slope for the nomogram in the validation group. The light blue indicates patients without suicide. The light red indicates patients with self-poisoning. A large separation between patients with and without suicide was observed in both training and validation groups.

Figure 5

Calibrating evaluation and clinical usefulness of the nomogram. (A) Calibration curve in the training group; (B) calibration curve in the validation group; (C) decision curve analysis in the training group; (D) decision curve analysis in the validation group.

Figure 6

Feature importance analysis. (A) Bees warm plot; (B) bar plot based on SHAP value.