Modeling the dynamics of human starvation: incorporating lessons from famines past and present

Abstract

Famines are crises characterized by starvation and malnutrition-related mortality. But representative nutrition and mortality data is often impossible to collect under common famine conditions, impeding real-time estimations and near-term projections of famine severity and therefore allocation of humanitarian assistance. As a supplement to survey data, systems models may be able to project individual outcomes based on more readily available data, to guide food assistance rations, and to predict trajectories of population-level malnutrition and starvation mortality. We provide an overview of systems dynamics modeling, discuss how it has been used to understand energy and nutrient dynamics in the human body, and discuss possible applications of these models, as well as the critical research needs to apply systems dynamics models to real-world crises. We then highlight complicating aspects of famine systems that should be considered as these models are applied. The literature includes several models that could be adapted to help identify critical areas for nutrition research on the biological progression of starvation under famine conditions. By incorporating additional physiological detail, these models could be parameterized to inform humanitarian public health responses to food crises, but improved data sharing will be crucial to model development.

Supplementary Information

The online version contains supplementary material available at 10.1057/s41271-026-00647-4.

Key messages

• Systems dynamics models for the biological processes of starvation could help inform responses to famine, especially when it is not possible to obtain population-level outcome data on malnutrition.

• However, analyses of recent famines reveal the proximal causes of death are much more complex than a simple lack of sufficient energy, with disease, health history, nutritional, and climate stressors playing major roles.

• Models of starvation during famine must take into account the unique evolution of each famine system and the interacting effects of multiple extreme environmental and dietary stressors.

• Better data sharing practices are necessary to allow access to the data necessary to guide systems model development on starvation.

Introduction

Famines and food crises have always been and continue to be a part of the human experience. Even as we write this article, famine is ongoing in Sudan and the Gaza Strip [1, 2]. Humanitarian aid is critical to limit mortality and suffering during famine, but humanitarian crises are receiving smaller and smaller percentages of the funds required to meet urgent needs, for example less than half of the projected needs for Sudan were funded in 2025 [3]. It is therefore essential to maximize the impact of available resources, which requires knowing where and when needs are likely to be highest, and how best to manage aid allocation. Models built on a framework for understanding consequences of reduced diet quantity and quality on health, nutritional status and mortality under famine conditions are potential tools to quantitatively predict outcomes and to guide the allocation of scarce resources.

A standard approach to research on acute malnutrition is to use statistical models to identify indicators and proxy variables associated with nutrition outcomes in past health crises [4–11]. However, this same evidence reveals that the effects of human diet on human well-being are cumulative and depend on the biological, social, and physical context both at the time of food insecurity and throughout the life of individuals. They are even affected by the nutritional histories of their parents and grandparents [12]. Statistical models also generally start with datasets and build on associations observed among a range of recorded variables. Unfortunately, raw datasets from large populations, especially during famine, are not consistently shared. Systems dynamics models for metabolism and physiology present a promising framework for projecting the long-term consequences of disruptions to food access or quality of diet. Such models could bring together known relationships to model the complex pathways to nutritional outcomes.

Models are simplifications of real-world situations. To simplify the study of human biological systems, studies tend to focus on a single risk factor within normal ranges, controlling for complicating factors, essentially focusing on a single relationship between one factor and an outcome. Most starvation models focus on energy and ignore other critical nutrients with little consideration of complex external factors. Famines are also real-world systems where many normal and unusual risks go to extremes, beyond the ranges that are ethical in experimental study designs. It therefore becomes problematic to empirically model something for which there is little experimental data.

This article explores the potential role of system dynamics models to inform public health responses to famines and other food crises. We review literature on starvation modeling and explore the best understood ways in which a famine system may interact with and modify human biological systems during starvation. Due to limited general understanding of this approach, we first provide general background on system dynamics modeling. Then we review potentially relevant literature on system dynamics models for energy balance and starvation and discuss how this framework can provide insight into responses during famine. By first exploring famine and biological systems independently and then the interactions between them, this article aims to facilitate and encourage the eventual development of system dynamics models that can project trajectories of acute malnutrition and related mortality during nutritional crises, up to and including famine.

What is a system dynamics model?

System dynamics models—sometimes called mathematical models, computational models, or simulation models—use mathematical equations for rates of change over time to project future changes to a system, for example, on the composition and functioning of the human body, as discussed in the present paper.

These models are particularly valuable for understanding systems with positive or negative feedback.

Systems dynamics models can be evaluated based on their ability to accurately predict dynamic changes over time under specific conditions, and a given model may be more accurate under some conditions than others, usually under the range of conditions for which it was originally developed, that is, for the range of conditions represented by the data upon which a model is built. More complicated models can often describe dynamics more accurately and in a wider range of conditions but are harder for researchers to analyze than simpler models, so simple models can be desirable as well. More information on these models and how they work is available in the Supplementary File 1 section "What is a systems dynamics model?".

The complicating context of famines

Although ultimately defined as nutrition emergencies, famines, like war and pandemics, include a multitude of inter-related mortality- and nutrition-related risks which become confounded and conflated with many existing vulnerabilities and risks not usually directly associated with mortality, making it difficult and even meaningless to attribute individual factors to individual deaths. In famine systems, the quality of diets declines well before the energy consumed is reduced. Biological models will need to take into account the differential time-dependent effects of reduced diet quality versus total energy intake. Even though famine is defined as a food and nutrition crisis, the final direct cause of most famine deaths is usually illness leading to malnutrition or illness made possible by malnutrition [13–15]. It is a basic tenet of famine mortality that the malnourished are more susceptible to disease and the ill are more susceptible to malnutrition.

While there are a number of epidemiological system dynamics models that include malnutrition as a factor contributing to disease transmission, such as in cholera, COVID-19 and measles [16–18], we know of no simulation models for the effect of infectious disease on starvation physiology. Starvation during famine is particularly difficult to model because it is a disruption of nearly every aspect of life, often including but not limited to access to food, shelter, clean water, and healthcare [1, 13, 14]. In modeling famine mortality, a much more systemic, dynamic view is needed than a comparative list of factors driving malnutrition, making system dynamics models particularly appropriate.

In the next section, we review existing research on the system dynamics of starvation and related physiological processes. It is crucial, however, to keep in mind the broader context, which is discussed in more detail in the Supplementary File 1 sections on the “Complicating Context of Famines” and “Famine: a collapse of systems that interact with human biological systems.”

Modeling energy balance, weight change, and starvation

System dynamics models have been used extensively in biological research exploring human physiology, nutrition, and metabolism [19, 20], including the relationship of energy balance to body weight, composition, and mortality [21–27], blood glucose and insulin kinetics [28–31], the effect of gut contents on satiety [32], the effect of zinc on body growth in infants [33], the effects of pathogens on digestion [34], and the interactions between malnutrition and hypothermia [35]. In this section we review potentially relevant physiology models and discuss their applicability to predict acute malnutrition and associated mortality outcomes during famines.

While there are abundant systems models related to energy balance for obesity, weight loss, and athletics, there are relatively few models on energy balance in starvation or severe acute malnutrition. We focus here on a small number of systems models explicitly addressing energy balance in long-term starvation [26, 27, 36], alongside several more general systems models of energy balance or metabolic flexibility that may be relevant for famine systems [22–25, 30, 31].

Although the simplest model for energy balance can be expressed as a single ‘stock’ for the amount of energy stored (Fig. 1a), correctly modeling relevant metabolic processes requires understanding the conversion of energy stores from different chemical forms in the body (sugars, fats, and proteins), the location of those stores (in adipose tissue, muscles, organs, etc.), the multiple demands for energy within the body, and variation in the rate of energy expenditure change over time. Models vary in both detail and conditions considered by the model, as well as their purpose. Caloin’s [36] model simulates depletion of fat and protein energy stores during zero caloric intake as a function of adiposity and daily metabolic demand specified by the user. Song and Thomas [26] model depletion of fat, protein, and ketones to estimate maximum survival time with no caloric input, assuming only basal metabolic requirements. In a model intended to test the ‘thrifty gene’ hypothesis for obesity, Speakman and Westerterp [27] similarly track fat, protein, and glycogen, while defining a generic trajectory for increasing and decreasing physical activity depending on phase of starvation. Hall’s models [21, 22] do not focus specifically on starvation, but are the most physiologically comprehensive that we have seen, accounting not only for partitioning of fuel sources in the body, but also the macronutrient composition of the diet, and some other physiological variables.

Fig. 1.

Stocks and flows diagrams for basic energy balance and energy conversions within the body. Rectangles represent ‘stocks’ or forms in which the energy may be stored. The arrows annotated with circles and triangles represent the ‘flows’ or rates at which energy moves or is converted between forms. The cloud images indicate sources and destinations outside of the system (details that are not modeled explicitly). a) The most basic model of energy storage in the body states that the change in energy stored over a period of time is equal to the amount of energy consumed minus the energy expended in that time interval. b) The main stocks of energy in the body are blood glucose (rapidly depleted, and therefore not shown), fat, glycogen, and protein (diagram adapted from Hall [23]). The relative importance of flows between these stocks changes as reserves are depleted when the body starves

All of these models are based on well-established scientific understanding of energy stores in the body. The body’s most readily available energy source is glucose circulating in the bloodstream, but this is a relatively small pool of energy at any moment in time. In healthy bodies, glucose is depleted or converted to other chemical forms within hours, after which the body shifts to other energy sources [26, 27]. Because blood sugar levels can fluctuate rapidly and the body seeks to maintain homeostasis through multiple mechanisms, systems models of long-term starvation typically disregard glucose as a stock of energy in the body, focusing instead on energy stored as fat, glycogen, and protein [21–23, 26, 27]. Figure 1b—adapted from Hall’s review of models for energy balance and metabolic regulation in humans [23]—diagrams the key stocks and flows used in this framework.

Although these energy storage tissues are central to the mechanisms of energy balance, it is generally not practical to measure them directly outside of a lab setting. Instead, total body weight is measured, and measurements such as upper arm circumference are used as a proxy to estimate body mass [37, 38], and skinfold thickness may be used to estimate fat mass [39]. Mathematical conversions are required to translate energy storage pools into these real-world measurements. The energy density of fat, sugars, and proteins allows conversion between the caloric energy stored and the masses of these molecules. Lean body weight includes both extracellular water (which can vary significantly over short periods) and skeletal mass (which is typically assumed to be constant). Simpler models may assume constant extracellular water, or water as a constant percentage of tissue mass, depending on the type of tissue [26, 27]. A more sophisticated framework, such as Hall’s model for human energy balance, weight gain, and metabolic adaptation can track extracellular water as a separate stock, including the role of sodium concentration in water retention [22]. Hall’s initial 2006 model, which described energy metabolism during semi-starvation and refeeding, did not model changes in extracellular water, but incorporated laboratory data on extracellular water of semi-starved volunteers in a classic starvation experiment conducted in Minnesota in the 1940s to evaluate model predictions against real-world data from that experiment [21, 40]. Unlike the human starvation model by Song and Thomas, the two energy balance models by Hall explicitly include stocks of nitrogen in the body (in the form of amino acids), including the role of physical activity in maintaining lean body mass (preventing muscle wasting), a consideration omitted from many other models [21, 22, 26].

The suitability of models for understanding famine conditions depends on the scenarios they are designed for and simplifying assumptions made. Models of total starvation (zero caloric intake)—which have been used to estimate the maximum survival time as a function of starting body composition when no food is provided [26, 27, 36]—are useful for scientific understanding, but are unlikely to reflect real-world famine conditions, in which some limited food is almost always available. Models of partial starvation, such as Hall presents, may therefore be more suitable [21, 22]. Another factor is the realism of assumptions about energy expenditure. For instance, Song and Thomas focus only on the basal metabolic energy needed to maintain body function, assuming no energy needs for physical activity or thermoregulation, whereas Speakman and Westerterp assume physiologically adaptive changes in activity levels [26, 27].

Finally, it should be noted that almost all models are based on data from lab-based studies on adult volunteers, most of them are healthy or have the same health issues within a model. Few models consider energy balance in growing children. One exception is the human growth model by Rahmandad, which calculates food energy needs from birth to old age and simulates the long-term effects of malnutrition on child height [25].

While system dynamics models can provide unique insights into how changes in one factor affect other factors and pathways to a given outcome, their construction depends on access to large volumes of raw historic data. During a famine, agencies primarily collect data for operational needs, like the design of interventions. Nevertheless, this data has enormous scientific value, including in the development of models, which may change the course of future famines. Unfortunately, raw data are rarely shared outside of the organizations collecting the data. The lack of robust reference data is a major challenge in model development.

Overall, existing models of energy balance tend to focus on caloric or micronutrient deficiency in isolation, whereas causes of death in real-world famines are much more complicated.

How starvation models could help with humanitarian response

The systems dynamics models described here are illustrative and necessarily incomplete. They focus on the most basic energy stores and related metabolic processes, not taking into account other nutrients that are necessary for energy metabolism and other competing functions. Even these fairly simple models may provide insight into public health responses to food crises, but it is important to be aware of model limitations, and eventually to extend these models with additional factors known to complicate famine systems. The process of model development must necessarily be iterative.

Preliminary models should assemble the current state of knowledge on energy and key nutrient balance in as much detail as possible, under realistic conditions and populations. These initial models would have the immediate benefit of clarifying the timing of physiological shifts as well as the relationships between processes that have otherwise been studied in isolation. They would also highlight gaps where processes or relationships are ill-defined and facilitate the prioritization of gaps for further research. Existing models, such as Hall’s model discussed above, could provide a starting point on which to build [21–23].

A model estimating the various rates of energy and nutrient storage, depletion, and replenishment would support protocols that maximize lean tissue mass accumulation or other complementary micronutrient stores, if that proves advantageous to long-term health. For example, one study showed that increasing the sulfur amino acids in therapeutic foods increased the speed of recovery among children with kwashiorkor (edematous malnutrition) [41]. Such a model could also facilitate the design of humanitarian food assistance packages with nutrient compositions ideal to simultaneously maximize the rates of replenishment of multiple critical nutrient stores.

For public health purposes, it would be especially valuable to design a sufficiently complete model to project the trajectory of a population into famine based on knowledge of general diets and conditions with less reliance on nutrition outcome surveys. Currently, humanitarians employ a menu of metrics on a range of food security indicators that can be collected by telephone that include the consumption of food from different food groups, available foods in an area, people’s access to various foods, as well as livelihood and coping strategies that may impact energy expenditure and the body’s ability to utilize available food. Information is also usually available on environmental factors such as air temperatures, water quality and available sanitation—all of which would affect energy and nutrient requirements. While availability of representative data on the current nutritional status and mortality of a population cannot be collected remotely, these other measures on factors which can affect the requirements of energy provide a powerful starting point for assessing priority regions for humanitarian aid. There are particular challenges, however, in projecting the timing of when a crisis will become acute or when it will have likely crossed certain thresholds in the absence of representative nutrition or mortality outcome data. Incorporation of sophisticated physiological models could nevertheless help support projections in the absence of outcome data.

Such models could have provided clarity, for example, on challenges that global experts have encountered in projecting nutritional trajectories for residents of Gaza during the ongoing siege by Israel. The siege began shortly after Hamas attacks on Israel in October 2023, and in the years that have followed, food permitted into the Gaza Strip has oscillated between sufficiency and extreme deficit. Prior to the conflict and siege, the population had been food secure and healthy, with replete physical reserves, and few individuals had ever fasted beyond the religiously mandated daylight-only fasts during Ramadan. Scientific understanding of typical trajectories is based on African populations whose physical reserves are generally more limited, but who frequently experience food deficits and whose bodies are therefore more ‘thrifty’ during times of dearth. Experts therefore struggled to project Gazans’ descent into undernutrition using the standard ratios of body mass index (BMI), age, sex and caloric intake to weight lost, which had been established through experience with populations whose nutritional history and environmental exposures were very different. Missing from the analysis were the complex interplay of body status, diet history, and metabolic shifts, but also the different rates at which a body depletes reserves and the time and nutrients to replenish those stores.

In this type of situation, a physiological model could provide a critical supplement to existing tools. Conversely, lessons learned in Gaza provide a cautionary message on the dangers of oversimplification—whether qualitative or quantitative. A systems dynamics model could be used to simulate physiological outcomes for different individuals during a food shortage (Fig. 2), helping to plan for different 'what if' scenarios, but such a model must be able to correctly account for relevant contextual details. In the case of Gaza, a physiological model that did not account for diet history would merely have reinforced the incorrect assumptions.

Fig. 2.

Simulated BMI during zero food intake for a 20-year-old man (blue) and a 60-year-old woman (orange), each 1.7 m tall, and with an initial BMI of 30. Trajectories produced using an implementation (available in Supplementary File 2) of the starvation model from Speakman and Westerterp [27]. According to the model, when the two individuals are deprived of food, it would take the younger man just 51 days to reach the 18.5 BMI threshold (dashed red line) for underweight status, whereas the older woman would continue at healthy weight for nearly 80 days

It is also crucial to keep in mind that a physiological simulation describes a unique trajectory for each individual, and no individual will represent the full range of experiences for a population in crisis. Sen’s entitlements framework describes how, in a famine system, hunger is not experienced by all segments of the population in the same way: there are winners and losers depending on social status, access to resources, interpersonal relationships, and other factors [42]. To simulate starvation during a famine, we must account for unequal distribution of food resources. Models would need to be able to adjust to multiple combinations of environmental factors affecting the body to represent different experiences within a population.

Conclusions

Traditional statistical models for famine and food crises identify the relative association of contributing factors and proxy variables in helping to identify and predict the occurrence of famine based on datasets of past experiences but struggle to show the paths through which those factors affect the outcome, limiting their utility to project the trajectories of malnutrition in famine-affected populations. System dynamics models may provide a more appropriate tool to study and project potential outcomes related to starvation in the complex famine environment. However, current models require extensive adaptation to account for mismatches between the conditions typically modeled and realistic famine conditions. Development of models for the interaction among disease, nutrient deficiency, and time-varying physiological stressors such as exertion and exposure with long-term energy imbalance would be a particularly fruitful area for future research. Although it may not be possible to capture the full interacting dynamics of malnutrition, disease, and stressors during crisis, incorporation of a limited number of highly influential factors into existing starvation models could yield insights, complementing research on the effects of malnutrition on disease transmission. There may be challenges in obtaining data to assess the validity of such models, however. Collection of laboratory data on such extreme stressors is clearly unethical, and biometric data of people in ongoing crises tend to be limited. A closer analysis of existing humanitarian datasets on body condition during crises would provide useful insights into the development and testing of such models.

Regardless of the difficulties, the iterative development of a system dynamics models of human energy and key micronutrient balance would serve multiple purposes. The initial basic models, based on existing knowledge, would elucidate relationships between biological processes or associations that are often examined in isolation of the larger context or other interacting processes. The initial models would also identify gaps in our understanding of these systems, indicating priorities for future research, and multiplying the impact of that research by placing its results within a larger dynamic system. As the model evolves and becomes more complete, it will become more useful, for example, to project trajectories that risk famine, or to prioritize non-food interventions that could have an outsized impact on nutritional status. Such a detailed understanding of the biological processes may also contribute to more effective treatment of the malnourished and the design of better food assistance packages. While a complete dynamic systems model may not be realistic in the short run, even initial, more basic models will still provide benefits, and in time, the dream may become a reality.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

A.M.T. and M.F. wrote the main manuscript. A.M.T. prepared Figs. 1 and 2 and M.F. prepared Supplementary Table S1. Both authors reviewed the manuscript.

Data availability

We generated the BMI graph depicted in Fig. 2 using our own implementation of the model described in Speakman and Westerterp’s 2013 paper [21], using StochSD modeling software. We have attached our model implementation as a supplementary.ssd file (Supplementary File 2). We implemented the model using StochSD version 2022.01.02. Readers wishing to test the model can download StochSD from the official website (https://stochsd.sourceforge.io/), or use the online version of the software available at the same site.

Declarations

Conflict of interest

The authors declare no competing interests.