Suicide detection in Chile: proposing a predictive model for suicide risk in a clinical sample of patients with mood disorders

Abstract

Objective: To analyze suicidal behavior and build a predictive model for suicide risk using data mining (DM) analysis. Methods: A study of 707 Chilean mental health patients (with and without suicide risk) was carried out across three healthcare centers in the Metropolitan Region of Santiago, Chile. Three hundred fortythree variables were studied using five questionnaires. DM and machine-learning tools were used via the support vector machine technique. Results: The model selected 22 variables that, depending on the circumstances in which they all occur, define whether a person belongs in a suicide risk zone (accuracy = 0.78, sensitivity = 0.77, and specificity = 0.79). Being in a suicide risk zone means patients are more vulnerable to suicide attempts or are thinking about suicide. The interrelationship between these variables is highly nonlinear, and it is interesting to note the particular ways in which they are configured for each case. The model shows that the variables of a suicide risk zone are related to individual unrest, personal satisfaction, and reasons for living, particularly those related to beliefs in one's own capacities and coping abilities. Conclusion: These variables can be used to create an assessment tool and enables us to identify individual risk and protective factors. This may also contribute to therapeutic intervention by strengthening feelings of personal well-being and reasons for staying alive. Our results prompted the design of a new clinical tool, which is fast and easy to use and aids in evaluating the trajectory of suicide risk at a given moment.

Acute Mental Discomfort Associated with Suicide Behavior in a Clinical Sample of Patients with Affective Disorders: Ascertaining Critical Variables Using Artificial Intelligence Tools

“…There are also reasons to carry on living even when a person is going through a painful, demanding, or overwhelming situation. These findings are consistent with those published in the literature with regard to reasons that might be powerful protectors against suicide ( 1 , 59 ) and might principally be associated with concern for family and a confidence in one’s ability to face problems ( 24 , 30 , 60 , 61 ).…”

Acute Mental Discomfort Associated with Suicide Behavior in a Clinical Sample of Patients with Affective Disorders: Ascertaining Critical Variables Using Artificial Intelligence Tools

“…There are also reasons to carry on living even when a person is going through a painful, demanding, or overwhelming situation. These findings are consistent with those published in the literature with regard to reasons that might be powerful protectors against suicide ( 1 , 59 ) and might principally be associated with concern for family and a confidence in one’s ability to face problems ( 24 , 30 , 60 , 61 ).…”

Explainable artificial intelligence models for predicting risk of suicide using health administrative data in Quebec

“…The weights in the LR model were used to identify the parameters and attributes that are globally or locally significant for accurate prediction. The important features found with absolute coefficients were consistent with previous research on their associations with suicide risk at the individual, programmatic and community level [60][61][62][63][64]. Considering the 10 most significant associations of in Fig 3, a first set of individual variables pertain to mental disorders and substance use disorders are well established risk factor for suicide [63][64][65][66][67].…”

Machine learning and the prediction of suicide in psychiatric populations: a systematic review

“…In line with this result, other ML studies highlighted the importance of socio-occupational status and well-being [ 56 , 63 , 65 , 87 , 93 ]. Similarly, non-psychiatric health issues have been reported among the features able to predict suicide [ 38 , 56 , 94 ]; moreover, one study reported the use of commonly prescribed opioids (e.g., Fentanyl) as a relevant feature in the prediction [ 57 ].…”

“… Pestian et al, [ 112 ] 130 suicidal patients, 126 non-suicidal patients with mental illness, and 123 controls (Not specified) Mixed diagnoses (Not specified) Not specified SVM (LOOCV) Linguistic and Acoustic Features extracted from open-ended questions Suicide attempts Suicidal vs HC ROC: 0.92 Suicidal vs non-suicidal Patients: ROC: 0.82 Suicidal vs All: ROC: 0.87 By combining linguistic and acoustic characteristics, subjects could be classified into one of the three groups. Barros et al, [ 65 ] 707 mental health patients (564/143) Mixed diagnoses (MDD 53% BD 18% Anxiety 10% Adjustment 10% Dysthymia 1% Others 8%) Not specified CART, k-nearest neighbor, RF, AdaBoost, NN multilayer perceptron, SVM (10-folds CV) 343 sociodemographic and clinical variables Suicide risk (current suicidal behavior – attempts o ideation) SVM Acc: 0.78 Sens: 0.77 Spec: 0.79 CART Acc: 0.72 Sens: 0.71 Spec: 0.74 RF Acc: 0.78 Sens: 0.78 Spec: 0.77 AdaBoost Acc: 0.76 Sens: 0.75 Spec: 0.76 KNN Acc: 0.73 Sens: 0.74 Spec: 0.73 The model shows that the variables of a suicide risk zone are related to individual unrest, personal satisfaction, and reasons for living, particularly related to beliefs in one’s own capacities and coping abilities. Cook et al, [ 45 ] 1453 self-harming patients (944/509) Self-harm (Not specified) Not specified NLP-Based Machine Learning (linear classifier) Open-ended question from a medical app Suicidal ideation Sens: 0.56 Spec: 0.57 PPV: 0.61 The top ten words associated with suicidal ideation were conté (I told), monotona (monotony), Equasim (Ritalin), acosado (harassed), trabajamos (we work), raza (race), aseos (restrooms), resfriado (congested/sick), pronuncio (I pronounce), and rechaza (rejects).…”

“…Moreover, 14 studies predicted suicide ideation alone [ 45 – 55 ] or in combination with suicide attempts [ 56 – 60 ]. Finally, other studies predicted self-harm [ 61 – 64 ], suicide risk [ 38 , 55 , 65 – 70 ], the number of suicide attempts [ 71 ], and the presence of a familiar history of suicide [ 72 ].…”

Acute Mental Discomfort Associated with Suicide Behavior in a Clinical Sample of Patients with Affective Disorders: Ascertaining Critical Variables Using Artificial Intelligence Tools

“…There are also reasons to carry on living even when a person is going through a painful, demanding, or overwhelming situation. These findings are consistent with those published in the literature with regard to reasons that might be powerful protectors against suicide ( 1 , 59 ) and might principally be associated with concern for family and a confidence in one’s ability to face problems ( 24 , 30 , 60 , 61 ).…”

Explainable artificial intelligence models for predicting risk of suicide using health administrative data in Quebec

“…The weights in the LR model were used to identify the parameters and attributes that are globally or locally significant for accurate prediction. The important features found with absolute coefficients were consistent with previous research on their associations with suicide risk at the individual, programmatic and community level [60][61][62][63][64]. Considering the 10 most significant associations of in Fig 3, a first set of individual variables pertain to mental disorders and substance use disorders are well established risk factor for suicide [63][64][65][66][67].…”

Machine learning and the prediction of suicide in psychiatric populations: a systematic review

“…In line with this result, other ML studies highlighted the importance of socio-occupational status and well-being [ 56 , 63 , 65 , 87 , 93 ]. Similarly, non-psychiatric health issues have been reported among the features able to predict suicide [ 38 , 56 , 94 ]; moreover, one study reported the use of commonly prescribed opioids (e.g., Fentanyl) as a relevant feature in the prediction [ 57 ].…”

“… Pestian et al, [ 112 ] 130 suicidal patients, 126 non-suicidal patients with mental illness, and 123 controls (Not specified) Mixed diagnoses (Not specified) Not specified SVM (LOOCV) Linguistic and Acoustic Features extracted from open-ended questions Suicide attempts Suicidal vs HC ROC: 0.92 Suicidal vs non-suicidal Patients: ROC: 0.82 Suicidal vs All: ROC: 0.87 By combining linguistic and acoustic characteristics, subjects could be classified into one of the three groups. Barros et al, [ 65 ] 707 mental health patients (564/143) Mixed diagnoses (MDD 53% BD 18% Anxiety 10% Adjustment 10% Dysthymia 1% Others 8%) Not specified CART, k-nearest neighbor, RF, AdaBoost, NN multilayer perceptron, SVM (10-folds CV) 343 sociodemographic and clinical variables Suicide risk (current suicidal behavior – attempts o ideation) SVM Acc: 0.78 Sens: 0.77 Spec: 0.79 CART Acc: 0.72 Sens: 0.71 Spec: 0.74 RF Acc: 0.78 Sens: 0.78 Spec: 0.77 AdaBoost Acc: 0.76 Sens: 0.75 Spec: 0.76 KNN Acc: 0.73 Sens: 0.74 Spec: 0.73 The model shows that the variables of a suicide risk zone are related to individual unrest, personal satisfaction, and reasons for living, particularly related to beliefs in one’s own capacities and coping abilities. Cook et al, [ 45 ] 1453 self-harming patients (944/509) Self-harm (Not specified) Not specified NLP-Based Machine Learning (linear classifier) Open-ended question from a medical app Suicidal ideation Sens: 0.56 Spec: 0.57 PPV: 0.61 The top ten words associated with suicidal ideation were conté (I told), monotona (monotony), Equasim (Ritalin), acosado (harassed), trabajamos (we work), raza (race), aseos (restrooms), resfriado (congested/sick), pronuncio (I pronounce), and rechaza (rejects).…”

“…Moreover, 14 studies predicted suicide ideation alone [ 45 – 55 ] or in combination with suicide attempts [ 56 – 60 ]. Finally, other studies predicted self-harm [ 61 – 64 ], suicide risk [ 38 , 55 , 65 – 70 ], the number of suicide attempts [ 71 ], and the presence of a familiar history of suicide [ 72 ].…”

Acute Mental Discomfort Associated with Suicide Behavior in a Clinical Sample of Patients with Affective Disorders: Ascertaining Critical Variables Using Artificial Intelligence Tools

“…There are also reasons to carry on living even when a person is going through a painful, demanding, or overwhelming situation. These findings are consistent with those published in the literature with regard to reasons that might be powerful protectors against suicide ( 1 , 59 ) and might principally be associated with concern for family and a confidence in one’s ability to face problems ( 24 , 30 , 60 , 61 ).…”

Explainable artificial intelligence models for predicting risk of suicide using health administrative data in Quebec

“…The weights in the LR model were used to identify the parameters and attributes that are globally or locally significant for accurate prediction. The important features found with absolute coefficients were consistent with previous research on their associations with suicide risk at the individual, programmatic and community level [60][61][62][63][64]. Considering the 10 most significant associations of in Fig 3, a first set of individual variables pertain to mental disorders and substance use disorders are well established risk factor for suicide [63][64][65][66][67].…”

Machine learning and the prediction of suicide in psychiatric populations: a systematic review

“…In line with this result, other ML studies highlighted the importance of socio-occupational status and well-being [ 56 , 63 , 65 , 87 , 93 ]. Similarly, non-psychiatric health issues have been reported among the features able to predict suicide [ 38 , 56 , 94 ]; moreover, one study reported the use of commonly prescribed opioids (e.g., Fentanyl) as a relevant feature in the prediction [ 57 ].…”

“… Pestian et al, [ 112 ] 130 suicidal patients, 126 non-suicidal patients with mental illness, and 123 controls (Not specified) Mixed diagnoses (Not specified) Not specified SVM (LOOCV) Linguistic and Acoustic Features extracted from open-ended questions Suicide attempts Suicidal vs HC ROC: 0.92 Suicidal vs non-suicidal Patients: ROC: 0.82 Suicidal vs All: ROC: 0.87 By combining linguistic and acoustic characteristics, subjects could be classified into one of the three groups. Barros et al, [ 65 ] 707 mental health patients (564/143) Mixed diagnoses (MDD 53% BD 18% Anxiety 10% Adjustment 10% Dysthymia 1% Others 8%) Not specified CART, k-nearest neighbor, RF, AdaBoost, NN multilayer perceptron, SVM (10-folds CV) 343 sociodemographic and clinical variables Suicide risk (current suicidal behavior – attempts o ideation) SVM Acc: 0.78 Sens: 0.77 Spec: 0.79 CART Acc: 0.72 Sens: 0.71 Spec: 0.74 RF Acc: 0.78 Sens: 0.78 Spec: 0.77 AdaBoost Acc: 0.76 Sens: 0.75 Spec: 0.76 KNN Acc: 0.73 Sens: 0.74 Spec: 0.73 The model shows that the variables of a suicide risk zone are related to individual unrest, personal satisfaction, and reasons for living, particularly related to beliefs in one’s own capacities and coping abilities. Cook et al, [ 45 ] 1453 self-harming patients (944/509) Self-harm (Not specified) Not specified NLP-Based Machine Learning (linear classifier) Open-ended question from a medical app Suicidal ideation Sens: 0.56 Spec: 0.57 PPV: 0.61 The top ten words associated with suicidal ideation were conté (I told), monotona (monotony), Equasim (Ritalin), acosado (harassed), trabajamos (we work), raza (race), aseos (restrooms), resfriado (congested/sick), pronuncio (I pronounce), and rechaza (rejects).…”

“…Moreover, 14 studies predicted suicide ideation alone [ 45 – 55 ] or in combination with suicide attempts [ 56 – 60 ]. Finally, other studies predicted self-harm [ 61 – 64 ], suicide risk [ 38 , 55 , 65 – 70 ], the number of suicide attempts [ 71 ], and the presence of a familiar history of suicide [ 72 ].…”