Research Article

Novel Use of Natural Language Processing (NLP) to Predict Suicidal Ideation and Psychiatric Symptoms in a Text-Based Mental Health Intervention in Madrid

Table 1

Baseline characteristics by suicidal ideation status and average GHQ-12 ≥ 4 (,453).

Mean (SD) or %SuicidalityGHQ-12
Never suicidalEver suicidal valueAvg GHQ < 4Avg GHQ ≥ 4 value

Age (cont)40.0 (13.8)41.6 (13.9)<0.00140.9 (0.5)40.2 (0.5)0.316
Percent female59.369.3<0.00159.769.7<0.001
Average nightly sleep (hours) 7.3 (1.6)6.9 (1.9)<0.0017.4 (1.4)6.8 (2.0)<0.001
Self-rated sleep quality (0–100)72.4 (24.1)55.0 (25.1)<0.00173.6 (21.2)52.8 (26.2)<0.001
Self-rated anger rarely (0–100)83.8 (16.3)67.0 (23.5)<0.00183.1 (15.6)66.6 (24.3)<0.001
Self-rated changes in appetite (0–100)55.0 (18.8)50.0 (23.0)<0.00156.2 (17.8)48.7 (23.5)<0.001
Medication adherence (0–100)76.5 (36.7)75.2 (35.1)0.00577.1 (35.4)74.5 (36.0)0.012
Average WHO-561.6 (21.1)38.3 (20.0)<0.00163.3 (18.9)35.0 (18.6)<0.001

Significance was assessed using 2-sample -tests for continuous normal variables, Wilcoxon rank-sum (Mann-Whitney) tests for continuous skewed variables, and Pearson chi-square test for binary variables.