Effect of Meat Price on Race and Gender Disparities in Obesity, Mortality and Quality of Life in the US: A Model-Based Analysis

Introduction There are large differences in the burden and health implications of obesity by race and gender in the US. It is unclear to what extent policies modifying caloric consumption change the distribution of the burden of obesity and related health outcomes. Meat is a large component of the American diet. We investigate how changing meat prices (that may result from policies or from exogenous factors that reduce supply) might impact the burden of obesity by race and gender. Methods We construct a microsimulation model that evaluates the 15-year body-mass index (BMI) and mortality impact of changes in meat price (5, 10, 25, and 50% increase) in the US adult population stratified by age, gender, race, and BMI. Results Under each price change evaluated, relative to the status quo, white males, black males, and black females are expected to realize more dramatic reduction in 2030 obesity prevalence than white females. Life expectancy gains are also projected to differ by subpopulation, with black males far less likely to benefit from an increase in meat prices than other groups. Conclusions Changing meat prices has considerable potential to affect population health differently by race and gender. In designing interventions that alter the price of foods to consumers, it is not sufficient to assess health effects based solely on the population as a whole, since differential effects across subpopulations may be substantial.

population growth (equal to the death rates [2]) and did not allow the birth rate to vary over the course of the simulation with the size of the model population.

Baseline Food Consumption
Average consumption patterns (proportional contribution of each food group to the total energy intake) for each race-gender were identified according to NHANES food recall data, adjusting for survey weighting. Dietary recall is biased toward under-reporting consumption, which makes it a poor tool for assessing total energy intake. Furthermore, this trend toward under-reporting tends to be more pronounced among women than men. [3] In light of these biases, we did not to use dietary recall data to estimate total caloric consumption, and instead use it to estimate relative consumption across the various food groups. We believe that the likelihood of bias in relative consumption across food groups would be diminished and unlikely to vary notably by demographic.
Foods were classified into one of the following groups: red meat, white meat, seafood, grain, dairy, eggs, fruit, vegetables, legumes/nuts/seeds, fats, sweets, sugarsweetened beverages. Meats that could not be classified as red meat, white meat, or seafood were assumed to have the same breakdown (red meat/white meat/seafood) as meats that could be readily classified.

Effect of Price Change on Consumption
Price elasticities of demand and cross price elasticities were drawn from the study performed by Basu et al., [4] in which elasticities were determined according to the quadratic almost ideal demand system [5], combining NHANES consumption data with USDA price history data. This source was selected due to the comprehensive inclusion of own-and cross-price elasticity estimates for all food groups represented in the NHANES 24-hour recall data. There are many other sources for data on food price elasticity, though they tend to be less comprehensive. Andreyeva et al. performed a systematic review of own-price elasticity of demand for several foods. The own-price elasticity values for white meat and red meat estimated by this review were somewhat higher than those used in our base case, but well within the range explored in our sensitivity analysis. [6] We acknowledge that elasticity estimates are accurate within some suitably small range of price changes, and that for larger price changes, predictions for change in consumption, may be inaccurate.

Time Dynamics of Consumption and Weight Change
The time scale over which weight change occurs in response to a change in consumption is long. Within the first year following a shift in consumption, approximately 50% of the weight change towards equilibrium occurs, however it takes approximately three years to get 95% of the way to a new equilibrium weight. [7] We modeled the combined effect of change in consumption due to the price change and change in consumption due to the exogenous tendency to gradually increase consumption as additive. (The impact of the former is only directly observed over approximately the first three years of the simulation, since it is a one-time change.) While it is conceivable that the tendency to gradually increase one's consumption may be affected by a shift in price, we assumed that this is not the case. However, in our sensitivity analysis we test different assumptions about the tendency to increase consumption over time.
The equations we used to determine weight in response to a change in caloric intake assume that current weight is at equilibrium prior to the change in intake. While this constraint is amenable to a one-time shift in consumption, it presents a challenge for assessing the impact of multiple changes in intake over relatively short period of time (e.g., tendency for gradual increase in intake). Therefore, we choose to consider the tendency for individuals to gradually increase their intake over time not on a yearly basis, but instead assumed these gradual increases take place every three years (and that the increase in consumption is three times as large as it would have been if increased yearly).
(The wavy nature of the prevalence curves in Fig C through E is an artifact of the three-year time span over which we model gradual change in consumption in order to allow weight to equilibrate before assessing the impact of another small change in consumption.) The equation relating change in consumption to weight change accounted for individuals' physical activity level. For simplicity, all members of the population were assumed to have the same physical activity level: a ratio of 1.5 total energy expenditure to resting metabolic rate. However, evidence suggests that some groups are more physically active than others. [8] Specifically, men tend to be more physically active than women and younger adults tend to be more active than older adults, but difference between races is far less notable.

Mortality
BMI-dependent risk of death, given gender, race, smoker status, and disease status was based on data from the American Cancer Society's Cancer Prevention Study II. [9] For each race-gender subpopulation, we determined mortality risk ratios according to BMI by weighting the risk ratios given smoker and disease status according to estimated smoking and disease prevalence for the group as reported in the American Cancer Society's Cancer Prevention Study II at baseline. Unlike in the case of the other race-gender groups, the number of black males with BMIs greater or equal to 40 kg/m 2 in this study was too small to warrant assessment of their risk ratio for mortality, so risk ratio for mortality was assessed for black males greater than or equal to 35 kg/m 2 as opposed to separately for 35.0-39.9 kg/m 2 and greater than or equal to 40.0 kg/m 2 . For each race-gender group, given the proportion of the population in each BMI category, and the mortality risk ratios for each BMI category, we used the CDC's overall mortality rate for the group to derive the mortality rates by race, gender, and BMI range (shown in Fig B).
There is considerable debate about whether the relationship between extreme BMIs (particularly low BMIs) and heightened mortality rates is causal. [9][10][11][12][13][14][15] Illness and smoking status are potential confounders. We performed sensitivity analysis to test the impact of explicitly accounting for differing risk of mortality for smokers versus nonsmokers by assigning smoker status and using differing risk ratios for mortality as identified by Patel. [9] We also evaluated sensitivity to the prevalence of illness in the population by testing the extremes: a population that has no disease, and a population in which every individual suffers from disease, applying the differing risk ratios for mortality associated with each.

Obesity Prevalence
An increase in the price of meat reduced obesity prevalence (BMI ≥ 30 kg/m 2 ) relative to the status quo in each race-gender group, as shown in Fig C. During the first three years, weight for individuals (and thus the trajectory of obesity prevalence) accounts for reduction in consumption in response to the price increase combined with the tendency to gradually increase consumption. The degree to which a given price increase impacts prevalence over this period depends on the extent to which the decrease in consumption corresponding to the price increase (which varies with the proportion of a group's calories derived from meat) is counterbalanced by the tendency to gradually increase consumption over this period. Following this initial period, the change in obesity prevalence rises steadily in accordance with the rate of gradual increase in caloric intake (which also varies by race-gender group). Overweight prevalence (BMI ≥ 25 kg/m 2 ) over 15 years, as depicted in Fig D, exhibits trends between subpopulations analogous to that for obesity.
Trends in underweight prevalence (BMI < 18.5 kg/m 2 ) vary by subpopulation, as shown in

Quality of Life
In addition to comparing QALYs lived relative to the status quo over all members of each gender-race group (Fig F1), we assessed this on the basis of initial BMI, as shown in Fig F2 and Fig F3. Among the initially overweight, all race-gender groups experienced an increase in QALYs with increasing price. While black males who were not initially overweight suffered with respect to QALYs, non-overweight white males still received a QALY benefit from increasing meat price. Though these white males were not initially overweight, many of them were on the verge of becoming overweight, so the tendency to gradually increase caloric intake resulted in many of them becoming overweight. Thus the meat price increase posed a barrier to them becoming overweight and enabled them to enjoy the higher utility associated with being of normal weight.

Impact of BMI on Risk Ratio for Mortality (Fig G through K)
There is significant uncertainty around the impact that BMI has on risk of mortality-especially for blacks, given their small samples sizes in the study from which this data is drawn. For this reason, we perform the following several sensitivity analyses.
(Note that these what-if analyses around the impact of BMI on mortality would not impact obesity prevalence, so we present only the effects on life years lived.)

I.
Assume black risk ratios of mortality by BMI are equal to that for whites (Fig G) II. Reduction in black male risk ratios of mortality for BMI in 15-18.5 kg/m 2 and 18.5-20 kg/m 2 ranges to that of the 20-22.5 kg/m 2 BMI range (Fig H) III. Reduction in black male risk ratios of mortality for BMI in 18.5-20 kg/m 2 range to that of the 20-22.5 kg/m 2 BMI range (Fig I) IV. Assume black male risk ratio of mortality for BMI greater than or equal to 40 kg/m 2 is equal to that for whites ( Fig J) In each of these analyses, we find that black males, who in the base case received by far the least mortality benefit, are projected to enjoy substantial increases in life years with increasing meat price, relative to the base case. In analyses III and IV, black males are projected to achieve mortality benefit comparable to that of white women. In analysis I, both black males and black females benefit to an even greater extent than white males.
Additionally, in the base case, the relationship between BMI and risk of mortality is  Given uncertainty about whether the observed trend toward gradual increase in consumption year over year will continue at a rate consistent with historical data (the status quo assumption), we assess how our results change under two alternative scenarios: (1) the increase in daily consumption proceeds at a rate of half that observed historically, and (2) complete cessation of the trend to eat more each year. Reduction in the tendency to gradually increase consumption produced less steep increases in obesity prevalence following the first three years (during which prevalence is strongly influenced by the price increase). This resulted in greater reductions in obesity prevalence over 15 years (Fig R) than in the base case in which we assumed continuation of the tendency to gradually increase consumption over time (Fig 2). However, the relative difference in obesity prevalence 15 years out under various meat price increases compared to no change in meat price (Fig S) did not differ in these what-if analyses relative to the base case (Fig 3). In comparing life years lived relative to the status quo under a given meat price increase (Fig   T), the results for reduced/eliminated tendency to gradually increase consumption were not notably different than in the base case for white males, white females, and black females. However, for black males, reduction in the tendency to increase consumption results in lower life expectancy.

Explicit Modeling of Smoker Status (Fig U and V)
In this analysis, instead of arriving at an average risk ratio for mortality by BMI for individuals of each race-gender group by performing a weighted average of those for smokers and nonsmokers, we assign each individual in the model a smoker status, such that their risk of mortality by BMI accounts for this in addition to their race and gender.
This does not impact our results.
Degree of Prevalence Disease (Fig W) Patel's assessment of risk ratio for mortality by BMI for each race-gender differentiates between those who, at the initiation of the American Cancer Society's Cancer Prevention Study II had one or more of the following conditions: cancer, heart disease, stroke, respiratory disease, current sickness, or weight loss of at least 10 lbs. in the previous year. [9] Those with any of these conditions are identified by Patel  Effect of Substitution/Complements (Fig X and Y) Cross price elasticity data suggests that an increase in meat price would result in reduction in the consumptions of many other types of food in addition to meats. If we account for changes in consumption of non-meats in addition to meats, the effect of meat price increase on obesity prevalence and life years lived is more dramatic than in the base case, with further reductions in obesity prevalence relative to no meat price change ( Fig   S24). This results in additional life years lived relative to the base case for white males, white females, and black females. Black males, however receive less mortality benefit than in the base case, and are harmed if meat prices increase by 50%.
Price Elasticity of Demand for Meats (Fig Z and AA) We assess the impact of meat price elasticity of demand values over a wide range by executing our model with elasticity of demand for red meats, white meats, and seafood at (1) half and (2) double their base case values. Halving the elasticities resulted in less dramatic reductions in obesity prevalence relative to the status quo, while doubling it yielded drastic reductions in prevalence (Fig Z). Halved elasticity of demand yielded smaller increases in life years relative to the base case (Fig AA). The extreme reductions in obesity prevalence assuming doubled elasticity would be less beneficial for black males than the base case (and posed harm relative to the status quo for the largest price increases examined), but benefited the other subpopulations.
Price Increase Confined to Specific Meat Types (Fig BB through EE) Our base analysis assumes price increase across all types of meat. While it is likely that, due to substitution effects, the price of all meats would rise in response to a sustained increase in the price of one type of meat (e.g., red meats), here we evaluate the effect of a price increase confined to red meat alone (Fig BB and CC) and seafood alone (Fig DD and   EE). In both of these cases the results are similar to the base case, but scaled to reflect the smaller portion of diet made up by red meat alone or seafood alone.
Price Decrease (Fig FF and GG) We also assessed the impact of a 5%, 10%, 25%, or 50% reduction in meat price.