アイテム詳細

要旨

Background: Psychiatric conditions are ubiquitously heterogeneous, making it critical to find subtypes with differing characteristics. Suicidal ideation involves at least two subtypes, with distinguishable cognitive and behavioural deficits, such as stress-responsivity and impulsivity. We examined whether computational modelling can separate these and/or other subtypes in relevant clinical populations from their performance in aversive Go/NoGo tasks. Methods: We analysed Go/NoGo choices and reaction times from two clinical suicidality groups comprising 129 veterans and 46 individuals with MDD, all assessed for suicidality, impulsivity, and anxiety. The behavioural data were analysed with a range of reinforcement learning models, hierarchically fitted to identify potential subgroups in the populations. Results: Hierarchical modelling arranged the veterans and MDD populations into 3 and 2 subgroups, respectively. We found distinct clinical correlations characterizing each group after controlling for age and sex. Higher scoring groups in both populations exhibited low to moderate correlations of lifetime suicidal ideation and cognitive control (in the veteran population) with performance biases and learning rates. Lower scoring groups showed low to moderate correlations of performance measures with impulsivity, anxiety, and hopelessness in both populations. Conclusions: The computational subgrouping of clinical populations based on behavioural patterns is promising as a way of separating informative subtypes of suicidality. Further analyses are required to establish their clinical utility, but the identification of subgroups with poorer/biased performance throughout the task, which correlated with impulsivity, is consistent with a stress-sensitive phenotype of suicidal ideation.