Abstract
Background
Adolescent suicide attempts remain difficult to predict, in part because research has concentrated on a narrow set of risk factors. We examined whether self-reported plain-water consumption is associated with past-year suicide attempt among U.S. adolescents, whether that association is graded across levels of intake, and whether a zero-intake response adds information beyond established risk factors.
Methods
We analyzed the 2017, 2019 and 2021 cycles of the U.S. national Youth Risk Behavior Survey (n = 36,779). Missing data, including the plain-water item, were multiply imputed (40 datasets) and pooled by Rubin’s rules, accounting for the survey design. The association was estimated unadjusted, then adjusted for demographics, each of six risk domains, and all domains together. Least absolute shrinkage and selection operator (LASSO) models, tuned under two metrics and two penalty rules, provided variable selection and held-out performance; odds ratios come from unpenalized models.
Results
Reporting no plain water was uncommon (3.6%) and associated with past-year suicide attempt (survey-weighted odds ratio 2.14, 95% CI 1.70–2.71), in each cycle. The association was confined to the zero category: the adjusted odds ratio for none was 2.55 (1.91–3.39) against 3 times/day, whereas among those reporting any plain water there was no gradient (1.01, 0.97–1.06). It remained associated after adjustment for each domain separately (1.45–2.01) but attenuated to 1.34 (0.97–1.85) under simultaneous adjustment. LASSO retained it in every model at the discrimination optimum and, under the one-standard-error rule, in five of six domains but not the full model. Adding it changed held-out discrimination by at most 0.0011.
Conclusions
Reporting no plain-water consumption in the past seven days was consistently associated with higher odds of a past-year suicide attempt among U.S. adolescents. However, the association was confined to the zero-water category, attenuated after simultaneous adjustment for all risk domains, and did not improve prediction. These findings suggest that the zero-water response may function as a behavioral marker of broader risk rather than as an independent risk factor.