OBJECTIVES: Outcomes research in a number of chronic diseases usually involves multiple risk factors and interactions. Furthermore, several correlated outcomes, like incident disease and its subsequent survival patterns, might also be desirable to investigate in a single study. Conventional statistical methodologies cannot effectively demonstrate such complicated causal pathways. Thus, we aim to examine the feasibility of applying generalized structural equation model (GSEM) to build an integrative analytic framework.

METHODS: The data were from a 41-year longitudinal incident and survival study of 54 pre-rheumatoid arthritis (pre-RA) cases and 4:1 matched controls, which was nested in the community-based CLUE cohort. In order to investigate risk factors of RA onset and long-term survival, a GSEM was developed with observed variables including baseline age, gender, cigarette smoking, education, family history of RA, C-reactive protein, soluble interleukin-2 receptor alpha, soluble tumor necrosis factor alpha-receptor 1, interleukin-1 beta, tumor necrosis factor-alpha, isotype-specific rheumatoid factors, luteinizing hormone, cortisol, prolactin, 17-hydroxyprogesterone, 17-hydroxypregnenolone, dehydroepiandrosterone, androstenedione, and estradiol. The GSEM also included four latent variables (androgenic-anabolic, C21-OH steroids, inflammatory cytokines, and immunoreactive proteins) and one multilevel variable (matching age and sex for the outcome of RA). GSEMs with four numerical integration methods were performed in Stata 15.

RESULTS: The GSEM had a convergence problem when the default method of mean-and-variance adaptive Gauss-Hermite quadrature was performed. However, the method of nonadaptive Gauss-Hermite quadrature made the GSEM converge very well. Survival time was associated with RA as well as age, education completed, cigarette smoking, immunoreactive proteins, prolactin, and androgenic-anabolic steroids. Aside from the matching variables, the risk factors of RA included cigarette smoking, family history of RA, and isotype rheumatoid factors.

CONCLUSIONS: It is feasible to use a GSEM in longitudinal incident and survival data, which serves to advance our understanding of risk factors and complex casual pathways.