Background: Illness behaviors—or affective, cognitive, and behavioral responses to symptoms of illness—predict patient outcomes, including symptom exacerbation and functional recovery, and they account for a large proportion of U.S. healthcare costs. Although priorcross-sectional work has examined illness behaviors like symptom reporting in isolation, the measurement of illness behavior using a longitudinal, multi-indicator approach has yet to be explored.

Aim: We evaluated illness behavior as a latent, developmental construct in the Swedish Adoption/Twin Study of Aging (SATSA).

Method: Participants were up to 1,886 individuals (from 1,223 twin pairs) ages 29 to 102 years (Mage baseline = 62.32 years; SD =13.69; 59% Female). Illness behavior indicators included somatic complaints, non-prescription medication use, pain-related disability and perceived illness complications. The psychomotor retardation subscale of the CES-D was used to index somatic complaints, and medication use was a simple composite of 9 dichotomous items on participants’ use of non-prescription medications, such as over-the-counter analgesics, in the previous month. Pain-related disability included a simple composite of three dichotomous items on the presence of neck,back, or shoulder pain that prevented participants from performing daily tasks or activities. Perceived illness disability was a composite of difference scores, calculated from subtracting a physician panel’s objective ratings of disability for each of 35 medical conditions (on a 3-pointscale; 1= Little or no disability; 3= Severe disability) from participants’ self-ratings of how much each of the same endorsed medical conditions interfered with their daily lives (on the same 3-point scale; 1= Not at all; 3= A lot). Positive composite scores reflected higher perceived disability relative to what was expected from the objective ratings, whereas a composite score of zero reflected “accuracy” or agreement in perceived illness complications. Confirmatory Factor Analysis (CFA) was used to evaluate invariance in the loadings of these four indicatorson a latent illness behavior factor across four questionnaire waves (1987-2004).

Findings: Confirmatory factor analyses revealed moderate factor loadings of the four indicators (standardized loadings ranged from .49 to .52, all ps < .0001). Also, practical fit indices from the nested model comparisons suggested strong factorial invariance in the loadings across time (CFI = .96; TLI = .95, RMSEA= .03, 90% CI: [.026, .035]).

Conclusion: Illness behavior as a latent, multi-indicator construct represents a promising focus for longitudinal work on behavior change and maintenance.