From work stress to disease: A computational model

Abstract

In modern society, work stress is highly prevalent. Problematically, work stress can cause disease. To help understand the causal relationship between work stress and disease, we present a computational model of this relationship. That is, drawing from allostatic load theory, we captured the link between work stress and disease in a set of mathematical formulas. With simulation studies, we then examined our model's ability to reproduce key findings from previous empirical research. Specifically, results from Study 1 suggested that our model could accurately reproduce established findings on daily fluctuations in cortisol levels (both on the group level and the individual level). Results from Study 2 suggested that our model could accurately reproduce established findings on the relationship between work stress and cardiovascular disease. Finally, results from Study 3 yielded new predictions about the relationship between workweek configurations (i.e., how working hours are distributed over days) and the subsequent development of disease. Together, our studies suggest a new, computational approach to studying the causal link between work stress and disease. We suggest that this approach is fruitful, as it aids the development of falsifiable theory, and as it opens up new ways of generating predictions about why and when work stress is (un)healthy.

Fig 1. Schematic illustration of our computational model.

In essence, the model explains how activation of the HPA axis, due to work stress and circadian inputs, elevates circulating cortisol levels. In turn, the cortisol response burdens the physiological system, in a fully reversible way (allostatic strain). Decay of cortisol and allostatic strain is assumed to be proportional to their respective current levels, as indicated by the feedback loops. When allostatic strain exceeds a threshold ε_L, it causes permanent damage (allostatic load). Allostatic load is cumulative and non-reversible (hence the integration symbol). When allostatic load exceeds a threshold ε_D it ultimately causes disease. This chain of events is described in detail in the main text. HPA axis = Hypothalamus Pituitary Adrenal-axis.

Fig 2. The count of days (y-axis) that a simulated person received a specific number of impulses (x-axis).

Four random, representative, simulated people are shown. Note: the total amount of simulated days is 200.

Fig 3. Distribution of all impulses over two consecutive days (48 hours).

Two representative days are shown for each of four random, representative, simulated people. The plots illustrate that the number of impulses varies from day to day. This is because the number of impulses in a given day results from a random draw from a Poisson distribution, based on the person’s overall mean (see main text).

Fig 4. Correlations between average daily work impulses and average daily night impulses.

Scenario I, uncorrelated and Scenario II, correlated.

Fig 5. Empirical vs. simulated aggregated day-time cortisol time courses.

a) Empirical data. Cortisol time courses based on data from 19,000 people, replotted from [37] with permission. b) Simulation results. Note: in both panels, data are shown until 16 hours after awakening.

Fig 6. Empirical vs. simulated aggregated night-time cortisol time courses.

a) Empirical data. Night-time cortisol time courses (n = 15; error bars reflect 95% confidence interval around the mean), replotted from [39] with permission. b) Simulation results. Note: in (b), the 95% confidence interval around the average is too small to be discernable.

Fig 7. Empirical vs. simulated 24-hour individual cortisol time courses.

a) Empirical data. 24-Hour cortisol time courses of five random, representative healthy individuals from [26], replotted with permission. b) Simulation results, showing five random, representative individuals based on Scenario I. c) Simulation results, showing five random, representative individuals based on Scenario II.

Fig 8. Illustration of variation of model parameters.

Sample of seven days showing variation on all model variables from five random simulated people.

Fig 9. Simulation results of relative risk of becoming diseased at various time points in the simulation.

Scenario I (a; no correlation between work and night impulses) Scenario II (b; correlation between work and night impulses). The odds ratios are calculated from the standardized averages of work impulses (i.e., the effect sizes represent one SD difference in the average number work impulses). Odd ratios greater than 100 are presented as 100.

Fig 10. New predictions based on our model.

The plot represents the relative risk of developing disease, as a function of different workweek configurations. All predictions are relative to a standard working week (i.e., configuration #1: 5 working days of 8 hours, Monday to Friday). See Table 2 for an explanation of all workweek configurations that we examined.