A Bayesian bivariate analytics framework for joint modeling of psoriasis disability and quality of life

Psoriasis imposes substantial functional and psychosocial burden, yet the Psoriasis Disability Index (PDI) and Dermatology Life Quality Index (DLQI) are conventionally analyzed separately, without explicitly modeling their residual dependence. This cross-sectional study jointly modeled PDI and DLQI in 150 Thai psoriasis patients using Bayesian bivariate regression fitted via Hamiltonian Monte Carlo. All 11 candidate predictors were considered without significance-based screening, with inclusion evidence assessed using Bayesian model averaging. Age showed the strongest inclusion evidence for both outcomes ( BF incl = 126.7060 for PDI; 8.1461 for DLQI). Full-model estimates indicated decreases of 0.3379 PDI and 0.2420 DLQI points per year of age. PASI was positively associated with both outcomes, with moderate inclusion evidence for PDI (3.0339) and weak evidence for DLQI (1.9533). Smoking showed moderate inclusion evidence for PDI (8.1246) and weak evidence for DLQI (1.7242). Standardized contrasts did not establish cross-outcome differences for smoking or comorbidity. Residual correlation remained substantial ( ρ = 0.7632 , 95% CrI: 0.6891–0.8249), compared with a marginal correlation of 0.8070. The freely correlated model achieved higher estimated joint predictive accuracy than the model fixing ρ = 0 ( Δ ELPD LOO = 65.5 ; SE=9.0). However, posterior predictive checks identified discrepancies in DLQI distributional shape and approximately 7% of replicated scores outside permissible ranges. Joint modeling quantified residual dependence and cross-outcome uncertainty, while bounded-score alternatives and external validation remain necessary before clinical application.