Allostatic Load as a Complex Clinical Construct: A Case-Based Computational Modeling Approach

Abstract

Allostatic load (AL) is a complex clinical construct, providing a unique window into the cumulative impact of stress. However, due to its inherent complexity, AL presents two major measurement challenges to conventional statistical modeling (the field's dominant methodology): it is comprised of a complex causal network of bioallostatic systems, represented by an even larger set of dynamic biomarkers; and, it is situated within a web of antecedent socioecological systems, linking AL to differences in health outcomes and disparities. To address these challenges, we employed case-based computational modeling (CBM), which allowed us to make four advances: (1) we developed a multisystem, 7-factor (20 biomarker) model of AL's network of allostatic systems; (2) used it to create a catalog of nine different clinical AL profiles (causal pathways); (3) linked each clinical profile to a typology of 23 health outcomes; and (4) explored our results (post hoc) as a function of gender, a key socioecological factor. In terms of highlights, (a) the Healthy clinical profile had few health risks; (b) the pro-inflammatory profile linked to high blood pressure and diabetes; (c) Low Stress Hormones linked to heart disease, TIA/Stroke, diabetes, and circulation problems; and (d) high stress hormones linked to heart disease and high blood pressure. Post hoc analyses also found that males were overrepresented on the High Blood Pressure (61.2%), Metabolic Syndrome (63.2%), High Stress Hormones (66.4%), and High Blood Sugar (57.1%); while females were overrepresented on the Healthy (81.9%), Low Stress Hormones (66.3%), and Low Stress Antagonists (stress buffers) (95.4%) profiles.

Keywords: allostatic load; artificial neural nets; case-based modeling; complexity theory; computational modeling; health risk outcomes.

FIGURE 1

U-Matrix and Components Maps for Nine Allostatic Load Profiles: Map A and Map B are graphic representations of the cluster solution arrived at by the Self-Organizing Map (SOM) Neural Net, referred to as the U-Matrix. In terms of the information, they provide, Map A is a three-dimensional (topographical) u-matrix: for it, the SOM adds hexagons to the original 15×11 map to allow for visual inspection of the degree of similarity among neighboring map units; the dark blue areas indicate neighborhoods of cases that are highly similar; in turn, bright yellow and red areas, as in the lower right comer of the map, indicate highly defined cluster boundaries. Map B is a two-dimensional version of Map A that allows for visual inspection of how the SOM clustered the individual cases. Cases on this version of the u-matrix (as well as Map A) were labeled according to their k-means cluster membership (The nine cluster solution shown in Table 2) to see if the SOM would arrive at a similar solution. Map C is a graphic representation of the relative influence that the seven factors (shown in Table 1) had on the SOM cluster solution. The SOM generates a mini-map for the seven factors, each of which can be overlaid across maps A and B. Each of these mini-maps can then be inspected visually to examine what its rates are across the different neighborhoods (clusters of cases). Dark blue areas indicate the lowest rates for a factor; and the bright red areas indicate the highest rates for a factor. For example, looking at the mini-map for Factor 6 (Blood Sugar), its rates are extremely low across most of the map, except for the lower right comer, which is where (looking at Map A and Map B) the SOM placed Cluster 6.

FIGURE 2

Clinical health risk outcomes for nine allostatic load profiles. This figure displays the differences between observed and expected frequencies for each self-reported medical condition. Each of the radii represents a self-reported medical condition, labeled at the top of their respective radius. The case clusters are circumscribed around the 23 points of each circle based on the average frequency on a particular self-reported medical condition. The resulting profile (which constitutes each Cluster’s health risk profile) is in red. Score higher than 0 (the green circle) indicate a greater observed value than expected, whereas scores below 0 indicate a smaller than observed value than expected. For those scores higher than 20, the corresponding medical condition is labeled in red. The three healthy to marginally healthy profiles are at the top, outlined in orange.