Food insecurity as a driver of obesity in humans: The insurance hypothesis

Abstract

Integrative explanations of why obesity is more prevalent in some sectors of the human population than others are lacking. Here, we outline and evaluate one candidate explanation, the insurance hypothesis (IH). The IH is rooted in adaptive evolutionary thinking: The function of storing fat is to provide a buffer against shortfall in the food supply. Thus, individuals should store more fat when they receive cues that access to food is uncertain. Applied to humans, this implies that an important proximate driver of obesity should be food insecurity rather than food abundance per se. We integrate several distinct lines of theory and evidence that bear on this hypothesis. We present a theoretical model that shows it is optimal to store more fat when food access is uncertain, and we review the experimental literature from non-human animals showing that fat reserves increase when access to food is restricted. We provide a meta-analysis of 125 epidemiological studies of the association between perceived food insecurity and high body weight in humans. There is a robust positive association, but it is restricted to adult women in high-income countries. We explore why this could be in light of the IH and our theoretical model. We conclude that although the IH alone cannot explain the distribution of obesity in the human population, it may represent a very important component of a pluralistic explanation. We also discuss insights it may offer into the developmental origins of obesity, dieting-induced weight gain, and anorexia nervosa.

Keywords: Obesity; behavioural ecology; eating disorders; food insecurity; hunger-obesity paradox; meta-analysis; overweight; weight regulation.

Figure 1

Output from the model described in section 3 (for details, see Online Appendix A). A. The assumed probability of survival against current level of fat reserves. B. The optimal amount to eat for different levels of fat reserves and four values of the food security parameter p. C. Fat reserves over 20 time periods for individuals who begin with reserves of 1 unit, follow the optimal eating policy for their level of food security, and find food in every period. D. Mean fat reserves over 40 time periods for simulated individuals who find food each period with probability p and follow the optimal eating policy for their level of food security. Points have been jittered in the horizontal dimension to make them more visible.

Figure 2

Estimated log odds ratios for high vs. normal body weight in food insecure vs. food secure individuals, plus their 95% confidence intervals, from the data set overall (line 1), and from various subsets of the data. Zero represents no association. The high body weight outcome varies from association to association (e.g. obesity, overweight), as does the exact specification of the food insecurity variable. For details see text and Online Appendix B.

Figure 3

Modifications to the model from section 3 to explore potential explanations for sex differences. See Online Appendix A, sections 3.2-3.4 for full details. A. Three different locations for the cliff-edge below which starvation becomes likely (controlled by parameter w). B. Steady-state target levels of fat reserves at different values of p for the different cliff-edge locations shown in panel A. C. Three different shapes of the left-hand part of the survival function (controlled by parameter x). D. Steady-state target levels of fat reserves at different values of p for the different shapes shown in panel C. E. Three different slopes of the right-hand part of the survival function, the cost of carrying each additional unit of weight (controlled by parameter y). F. Steady-state target levels of fat reserves at different values of p for the slopes shown in panel E.

Figure 4

Mean level of reserves over 40 days for simulated individuals experiencing different levels of food security p, for two different values of the maximum energy available from food per period, N. When N is small, p becomes a poor predictor of body weight as the variability between individuals at the same level of p becomes greater. Points have been jittered in the horizontal dimension to make individual data points more visible.