(PS9-B37) Initial Calibration of a Multidimensional Assessment of Suicide Risk in Time-intensive Longitudinal Designs: Item-bank Development and Calibration

Suicide is a leading cause of death globally (WHO, 2021) with rates continuing to increase (Hedegaard et al., 2018). Recent research emphasizes the identification of time-varying risk factors capable of informing proximal suicidal thought and behavior (STB) detection (i.e., days or hours prior to STBs). Ecological momentary assessment (EMA) designs have become an increasingly popular method for studying STBs and STB risk. However, the burdensome nature of these designs restricts the number and length of instruments feasible to administer. This limits the direct comparison of suicide theories (e.g., Joiner, 2005; Klonsky & May, 2015; O’Connor 2018) and testing complex interactive effects. A potential solution is computerized adaptive testing (CAT), which facilitates increased efficiency while preserving measurement precision. As a tradeoff, CAT requires pre-calibrated item-banks to facilitate their adaptive nature. The current study aims to 1) assemble a preliminary item-bank for STB risk factors in EMA designs, 2) evaluate test dimensionality and 3) calibrate item parameters of the resulting item-bank.

Participants N = 2,498 were recruited through Prime Panels (Chandler et al., 2019). Respondents failing automated bot detection, or more than 1 attention checks were excluded from analyses to safeguard data quality. The final calibration sample resulted N = 1759 adults (957 female; 439 Black; 427 Latinx); 36.33% reported lifetime history of STBs. STB risk constructs were identified through literature review and consultation with experts in the field. Constructs included: thwarted belongingness (TB), perceived burdensomeness (PB), hopelessness, loneliness, defeat, humiliation, entrapment, perceived stress, anger, impulsivity, aggression, and distress tolerance. Items (J = 199) were modified from existing instruments to share a common number of response options and target recent experiences (i.e., past day).

Exploratory item factor analysis (EIFA) was conducted on STB risk variables using pairwise complete data. Parallel analyses and scree plots were used to explore test dimensionality; 12 factors were suggested. Examination of factor loadings showed that perceived stress items did not load onto a common factor. Stress items were removed from the item bank (J = 189) and EIFAs were conducted again. Results supported 12 factors corresponding to TB, PB, loneliness, psychological pain, defeat, humiliation, entrapment, anger, aggression, impulsivity, distress tolerance, and a method factor for reverse keyed items. This method factor was present in all EIFAs. Inter-factor correlations for the final STB risk model ranged from low (r < .01) to high (r = .73). Following EIFA, the item parameters were calibrated using a multidimensional graded response model.

Although preliminary, findings supported a 12-dimensional measure of STB risk variables focused on proximal (i.e., past day) experiences. The recovered factor structure is largely in line with hypotheses including both key theoretical variables and trait-level vulnerabilities for STBs. Future work will develop a CAT item selection algorithm for EMA and seek to validate this instrument in micro-longitudinal designs for clinical populations.