Psychiatric, Sociodemographic, and Community Belonging Correlates of Suicidal Ideation in Canada: Findings from the 2022 Mental Health and Access to Care Survey

Abstract

Objective To estimate the prevalence of suicidal ideation among Canadians, identify psychiatric and sociodemographic correlates of past-12-month ideation and its association with community belonging, and compare the out-of-fold predictive performance of logistic regression and random forest models. Method We analyzed the nationally representative 2022 Mental Health and Access to Care Survey (analytic n = 8,080; weighted population = 25,828,546). Survey-weighted logistic and multinomial regression used the master weight with 1,000 Fay-adjusted bootstrap replicate weights. Five-fold nested cross-validation compared a sociodemographic-plus-psychiatric logistic model, the same model with belonging added, and a random forest, using out-of-fold discrimination, average precision, Brier scores, and grouped permutation importance. Results Past-12-month suicidal ideation was reported by 3.4% (95% CI, 2.9% to 3.9%) and lifetime-only ideation by 7.4% (95% confidence interval (CI), 6.7% to 8.2%). Independent correlates of past-12-month ideation were major depressive episode (MDE) (adjusted odds ratio (aOR) = 5.80; 95% CI, 3.61 to 9.33), very weak community belonging (aOR = 3.64; 95% CI, 1.88 to 7.04), generalized anxiety disorder (aOR = 2.65; 95% CI, 1.56 to 4.50), bipolar disorder (aOR = 2.43; 95% CI, 1.27 to 4.66), and separation/divorce (aOR = 2.25; 95% CI, 1.22 to 4.17). Adjusted probability rose from 2.0% at very strong to 6.0% at very weak belonging. No interaction was supported. Adding belonging did not improve discrimination (area under the curve (AUC), 0.823 vs. 0.823; difference = 0.000; 95% CI, −0.011 to 0.011); the random forest AUC was 0.843 (difference = 0.020; 95% CI, 0.004 to 0.038). Conclusions Psychiatric morbidity, separation/divorce, younger age, and very weak community belonging were associated with past-12-month suicidal ideation. Community belonging was independently associated but did not improve cross-validated discrimination, whereas MDE was the most robust correlate.