Disentangling Alcohol from a Single Latent Dimension of Polysubstance Involvement at Treatment Entry

Background: Polysubstance use is common among people entering alcohol and other drug (AOD) treatment, yet whether co-use reflects a shared dimension or multiple substance-specific patterns has rarely been formally tested in clinical samples.Methods: Using WHO-ASSIST severity bands for 11 substances, we fitted confirmatory factor analyses (ordinal indicators, WLSMV, theta parameterisation) in 18,361 adults entering residential and community AOD treatment in Australia.Findings: An 11-indicator single-factor model fit poorly (CFI = .906, RMSEA = .088); alcohol showed a near-zero loading (λ = −.01). Excluding alcohol improved fit (CFI = .958, RMSEA = .064), and adding three correlated residuals (methamphetamine–GHB, opioids–sedatives, cocaine–MDMA) yielded excellent fit (CFI = .994, RMSEA = .025). The model was invariant across gender and age.Conclusions: Non-alcohol substance involvement was organised along a single latent dimension, with three residual pairings capturing additional clinically interpretable covariation. Alcohol operated independently of this structure. Rather than reflecting multiple substance-specific processes, polysubstance involvement cohered along a single shared dimension that held across gender and age. These findings support moving beyond single-drug classification toward pattern-based assessment of polysubstance involvement, with alcohol evaluated as a structurally distinct dimension.