Does allostatic load predict incidental coronary events differently among sexes?

Background: One measure to quantify the degree of dysregulation is allostatic load (AL). Typically, AL incorporates information on diverse biomarkers and is associated with health outcomes such as cardiovascular diseases or the incidence of coronary events (C-E).

Aims: This study investigates the predictive performance of different AL scoring methods on the incidence of coronary events (C-E). This study also elaborates sex differences in the baseline risks of C-E and the AL associated risks of C-E.

Design: Longitudinal data analysis of the Heinz Nixdorf Recall Study (Risk Factors, Evaluation of Coronary Calcification, and Lifestyle) of 4327 participants free of C-E at study baseline aged 45-75. The data contains over 13 biomarkers measuring AL.

Methods: After conducting multiple imputations on missing values on AL for 826 participants, the analysis sample consisted of N = 4327 participants. We applied the two most commonly used methods of AL scoring AL (count-based and Z-score) and a recently developed logistic regression weighting method (LRM) approach. Cox regression was used to predict the incidence of C-E for each AL score. Results were estimated without (M0) and with (M1) covariate adjustment, and in a final model (M2), with an interaction between AL and sex.

Results: We found no violation of the proportional hazard assumption and significant differences in the survival curves between the sexes for C-E (Log-rank test: prob. > Chi = 0.000). In M0, all AL-scoring methods predicted C-E significantly, with the LRM based AL-score having the best performance (hazard ratio = 3.133; CI: [2.630, 3.732]; Somer's D = 0.717). After covariate inclusion, differences between the scoring methods levelled, though the count-based method and LRM performed better than the Z-scoring method. The interaction analysis in M2 showed a significant multiplicative interaction for the count-based method (1.254; [1.066, 1.475]) and for the LRM (1.746; [1.132, 2.692]). The additive relative excess risk due to interaction (RERI) measure was negative for the count-based method (RERI = -1.967; [-3.778; -0.156]) and the LRM (RERI = -1.909 [-3.910; 0.091]), indicating subadditivity.

Conclusion: AL scores are suitable for predicting C-E. Differences between the AL-scoring algorithms were only present after including interactions. We value the count-based method as suitable for clinical practice since its calculation is relatively simple, and performance was among the best. Interaction analysis revealed that despite strong sex differences in baseline C-E, the effect of AL is more pronounced for females at high levels of AL; thus, females could benefit more from a potential intervention on AL. We suggest further investigation of sex differences concerning the mediation by physiological and psychological intermediates.