Subgrouping suicidal ideations: An ecological momentary assessment study in psychiatric inpatients

**Background.** Suicidal ideations (SI) are amongst the strongest predictors for suicide attempts, yet reliable prediction models for suicide risk are still lacking. One challenge is that SI may vary when indexed over time. This could be due to SI subgroups that might hold crucial information for suicide risk prediction. We aimed to expand on prior approaches that averaged across the SI trajectories and instead use an approach that respects the temporal nature of SI. **Methods.** First, we applied longitudinal clustering to ecological momentary assessment SI data (5 assessments/day over 28 days) of 51 psychiatric patients (61% female, mean age = 35.26, *SD* = 12.54). Specifically, we used KmlShape, an algorithm that takes the SI raw scores and measurement occasion index as input. Second, we regressed each subgroup on established clinical SI risk factors (i.e., history of suicidal thoughts and behaviors, hopelessness, diagnosis of depression, diagnosis of anxiety disorder, and history of abuse). **Results.** We identified four subgroups with distinct SI patterns: (1) “Episodic intensity SI” (high mean, high variability), (2) “Consistent low average SI” (lowest mean, lowest variability), (3) “Chronic moderate average SI” (low mean, low variability), and (4) “Intermittent high SI” (highest mean, highest variability). Further, the subgroups were meaningfully associated with clinical characteristics, i.e., the least severe SI subgroup (“Consistent low average SI”) entailed the least hopeless individuals (beta = -0.95, 95% CI = -1.04, -0.86), and the most severe SI subgroup (“Intermittent high SI”) the most hopeless (beta = 0.84, 95% CI = 0.72, 0.95). **Conclusion.** Applying longitudinal clustering to EMA collected from patients with SI allows to identification of valid and reliable SI subgroups with more distinct clinical characteristics, an important step towards a better understanding of SI and a basis for improving prediction and prevention. **Trial Registration.** 10DL12_183251.