RESEARCH AND PRACTICE

The Association Between Food Prices and the Blood Glucose Level of US Adults With Type 2 Diabetes Tobenna D. Anekwe, ScD, and Ilya Rahkovsky, PhD

The American Diabetes Association recommends that people with diabetes follow a diet that includes carbohydrates from fruits, vegetables, whole grains, legumes, and low-fat milk; contains 2 or more servings of fish per week (excluding commercially fried fish fillets); limits saturated fat to less than 7% of total calories and dietary cholesterol to less than 200 milligrams per day; and minimizes trans fat.1 However, adhering to such a diet is difficult for many people with diabetes. Dietary modification is only temporary for most patients with diabetes and the cost of a diabetes-healthy diet has been identified as a significant barrier to maintaining the diet,2 particularly for lowincome patients.3 In light of these obstacles, it is important to understand how the prices of healthy and unhealthy foods affect the diet and health outcomes of people with diabetes. We examined the association between food prices and blood sugar biomarkers among adults with type 2 diabetes. We focused on people with diabetes because elevated blood glucose in this group is associated with a host of medical complications.1 We know of at least 1 other study that investigated how blood glucose levels respond to food prices: Rashad4 found that higher prices for low-glycemic-index foods (orange juice and bananas—foods that do not spike blood sugar levels much after consumption) were positively associated with blood sugar, whereas higher prices for high-glycemicindex foods (bread and ice cream) were negatively associated with blood sugar among people without diabetes, although these associations were not statistically significant. Other research has found that food prices are linked to a number of health outcomes. In their review of the empirical literature, Powell and Chaloupka5 concluded that food prices are associated with body mass index, especially among low-socioeconomic-status populations and people at risk for overweight or obesity. Meyerhoefer and Liebtag6 found that the price

Objectives. We estimated the association between the price of healthy and less-healthy food groups and blood sugar among US adults with type 2 diabetes. Methods. We linked 1999–2006 National Health and Nutrition Examination Survey health information to food prices contained in the Quarterly Food-atHome Price Database. We regressed blood sugar levels on food prices from the previous calendar quarter, controlling for market region and a range of other covariates. We also examined whether the association between food prices and blood sugar varies among different income groups. Results. The prices of produce and low-fat dairy foods were associated with blood sugar levels of people with type 2 diabetes. Specifically, higher prices for produce and low-fat dairy foods were associated with higher levels of glycated hemoglobin and fasting plasma glucose 3 months later. Food prices had a greater association with blood sugar for low-income people than for higherincome people, and in the expected direction. Conclusions. Higher prices of healthy foods were associated with increased blood sugar among people with type 2 diabetes. The association was especially pronounced among low-income people with type 2 diabetes. (Am J Public Health. 2014;104:678–685. doi:10.2105/AJPH.2013.301661)

of low-carbohydrate foods is associated with a higher likelihood of having a diabetes diagnosis and a higher level of medical expenditures among people with type 2 diabetes. Rahkovsky and Gregory7 demonstrated that the prices of vegetables, processed foods, whole milk, and whole grains are associated with blood cholesterol levels. A number of studies have found that higher prices for fast food and soda and lower prices for vegetables are associated with lower weight for children.8---11 Understanding the relationship between food prices and dietary health is policyrelevant. In recent years, politicians and public health professionals have suggested that taxing foods high in calories, saturated fat, or added sugar, and subsidizing healthier foods, may improve dietary quality and health outcomes for the general population.12---15

METHODS Several tests monitor blood glucose levels.16 One common test is the fasting plasma glucose

678 | Research and Practice | Peer Reviewed | Anekwe and Rahkovsky

(FPG) test, which is performed after a fast of at least 8 hours. Levels of 70 to 99 milligrams per deciliter (mg/dL) are considered to be in the normal range, whereas an FPG of 100 to 125 mg/dL indicates prediabetes, and 126 mg/dL and higher indicates diabetes.17 Another test is the glycated hemoglobin (GHb) test, or HbA1c, which “has become the preferred standard for assessing glycemic control.”18(pS92) The GHb test measures the percentage GHb, which is an indicator of a person’s average blood glucose levels during the previous 2 to 3 months.18 Fasting is not needed for this test. Levels lower than 5.7% are considered to be in the normal range, levels of 5.7% to 6.4% indicate prediabetes, and levels of 6.5% and higher indicate diabetes.19 The FPG and GHb test results do not necessarily agree, because they measure blood glucose levels over different time scales. The FPG test reflects glucose levels at a single point in time, whereas GHb measures average glucose level over the previous 2 to 3 months.20 For a discussion of the advantages and disadvantages of these tests, see Sacks.20

American Journal of Public Health | April 2014, Vol 104, No. 4

RESEARCH AND PRACTICE

Data To create our data set, we combined data from the National Health and Nutrition Examination Survey (NHANES) and the Quarterly Food-at-Home Price Database (QFAHPD) by using National Center for Health Statistics (NCHS) restricted data on calendar quarter and market region as a link. NHANES is a nationally representative survey conducted by NCHS to assess the health and nutritional status of the US population. We used NHANES’s demographic, socioeconomic, dietary, and health-related data, including respondents’ GHb and FPG test results. Our data set used 4 two-year waves of NHANES: 1999---2000, 2001---2002, 2003---2004, and 2005---2006. We did not include the years after 2006 because they do not include price information for random-weight items (i.e., products, such as many fresh fruits and vegetables, that are sold by weight and generally do not have Universal Product Codes [UPCs]). The QFAHPD is a database that tracks nominal food prices over time in the United States. The database is constructed from 1999---2006 Nielsen Homescan consumer surveys, and includes the mean price of 52 food categories for every annual quarter for 35 market regions that cover the contiguous United States (Appendix A, available as a supplement to the online version of this article at http://www.ajph.org). Food prices are presented as the average price per 100 grams for each food category for each quarter in each market region. Prices in QFAHPD were estimated by using Homescan’s household weights to make the sample nationally representative. Details about QFAHPD can be found in Todd et al.21 We dropped 2 of the 52 food groups— whole-grain mixes and frozen whole grains— because prices for these food groups are missing for most observations, as consumers rarely purchase these products. We then collapsed the remaining 50 food groups into 10 groups: produce, lean protein, oils and raw nuts, regular meat, low-fat meat, processed foods, regular dairy, low-fat dairy, packaged refined grains, and packaged whole grains. (See Appendix B, available as a supplement to the online version of this article at http://www.ajph.org, for the composition of these 10 food groups.) We

calculated the price for each of these 10 food groups as the weighted average of the price of the foods within each group, for each quarter and market region combination. We weighted food prices by “expenditure weights,” or the percentage of dollars spent on that food within its larger food group in a given quarter and market region. We used previous-quarter (rather than current-quarter) prices for all analyses because GHb reflects blood sugar levels over the past few months and FPG levels, rather than changing immediately, take roughly 3 weeks to change in response to dietary changes.22,23 We created 2 different samples of people with diabetes, based on either the GHb or the FPG definition. The GHb sample included people with GHb of 6.5% or higher and people who were told by a clinician that they had diabetes. The FPG sample included people with FPG of 126 mg/dL or higher and people who were told by a clinician that they had diabetes. We dropped from the analysis individuals younger than 20 years because they had not been asked certain questions in the NHANES interview (e.g., questions about marital status, cholesterol, and cardiovascular health). We also excluded individuals with type 1 diabetes because this condition is caused by genetics and autoimmune reactions rather than by diet.24 The NHANES survey does not explicitly identify the type of diabetes an individual has, so we followed Koro et al.25 and assumed that people had type 1 diabetes if they had been diagnosed with diabetes before age 30 years and had started taking insulin within 1 year of diagnosis.

Statistical Analysis Our goal was to estimate the relationship between food prices and diabetes biomarkers (i.e., GHb and FPG) among adults with type 2 diabetes. We assumed that the causal pathway from food prices to diabetes biomarkers runs through food purchases and subsequent food consumption. Our regression model was DiabetesBiomarkeritj ¼ b  Priceitj þ ð1Þ d  Demographicsitj þ c  Healthitj þ u  PhysicalActivityitj þ lj þ st þ ei where DiabetesBiomarkeritj is the blood sugar outcome for person i in annual quarter t living in market region j, and s is the time trend

April 2014, Vol 104, No. 4 | American Journal of Public Health

measured in annual quarters and l is the market---region fixed effect. We included the time trend and market---region fixed effect to control for correlations that time and geography each have with food prices and diabetes biomarkers. Priceitj is a vector of food prices, and b is the quantity we are estimating (i.e., the association of food prices with diabetes biomarkers). Healthitj is a vector of the following variables: body mass index, whether the respondent had a history of diabetes, whether the respondent’s family members had a history of diabetes, and whether the respondent had ever had his or her blood cholesterol checked. PhysicalActivityitj is a vector of whether and how many minutes the respondent walked or biked during the previous 30 days, and whether the respondent engaged in any vigorous physical activity for 10 minutes or more during the previous 30 days. We applied sample weights in all regression analyses and descriptive statistics so that the sample represents the national population of people with diabetes. Per NHANES guidance,26 we used mobile examination center sample weights for the GHb sample, and we used fasting subsample mobile examination center weights for the FPG sample. To restrict our analysis to the subpopulation of people with type 2 diabetes, we used the subpopulation option in Stata (“subpop”). We also accounted for survey stratification to correctly estimate standard errors. For purposes of comparison, we report estimates from a regression that does not control for market---region fixed effects. We conducted analyses with Stata version 11 (StataCorp LP, College Station, TX). We measured statistical significance at a P level of less than .05. Variables we controlled for were individual and family history of diabetes, body mass index, gender, age, race/ethnicity, household size, education, physical activity, alcohol consumption, meals per week outside the home, health insurance status, poverty---income ratio, and an indicator for whether an individual had ever had blood cholesterol checked. These variables are likely to affect blood sugar levels. We expected positive coefficients on the prices of healthy foods and nutrients—that is, products high in dietary fiber, lean protein, and whole grains—because higher prices should reduce consumption of these beneficial foods

Anekwe and Rahkovsky | Peer Reviewed | Research and Practice | 679

RESEARCH AND PRACTICE

and increase blood sugar levels. By contrast, we expected negative coefficients on the prices of less-healthy foods and nutrients—products rich in saturated fats and refined grains—because, as prices increase, blood sugar levels should decrease because of lower consumption of these foods. We also compared the association between food prices and blood sugar for low-income versus higher-income populations. For this, we used equation 1, except that now we interacted the food-price variable Priceitj with a dummy variable that indicated whether the respondent was higher-income (above 185% of the poverty line), low-income (131% to 185% of the poverty line), or lowest-income (at or below 130% of the poverty line). “Higher-income” was the baseline group. We chose these cutpoints because individuals at or below 185% of the poverty level qualify for the Women, Infants, and Children food assistance program, and individuals at or below 130% of the poverty line additionally qualify for the Supplemental Nutritional Assistance Program (SNAP; formerly known as the food stamp program) and for free school meals. These thresholds therefore identify populations that are relevant to US food assistance programs. We expected people with lower income to have a higher price elasticity of blood sugar (i.e., have blood sugar levels that are more responsive to changes in food prices).27

RESULTS Table 1 presents descriptive statistics for the GHb and FPG samples. The 2 samples are different because the GHb and FPG tests are not identical; also, some people had been dropped from the FPG sample because they had not fasted.28---31 Despite the different composition of the 2 samples, their descriptive statistics were quite similar. The median and mean percentage differences between means across all variables were only 2.1% and 5.6%, respectively (Table 1).

Average Effects of Food Prices Table 2 presents results from fixed-effects and ordinary least square models estimating the association of food prices with GHb and FPG among adults with type 2 diabetes who were aged 20 years and older. The fixed-effects results differed from the ordinary least square

results, which suggests the importance of controlling for market region in this model. For this reason, our preferred model is fixed-effects. Higher prices for produce were associated with higher GHb levels one quarter (3 months) later. The coefficient for produce indicates that a 1-standard-deviation increase in produce prices (2.2 cents per 100 grams) was associated with an increase in GHb of 0.29% points (calculated as the effect size of 13.16 times the standard deviation of 0.022 dollars per 100 grams), which is 3.9% of the mean GHb level of 7.45%. Results for FPG (Table 2) show positive and significant coefficients for the produce and low-fat dairy prices. The coefficient for produce indicates that a 1-standard-deviation increase in produce prices (2.2 cents per 100 grams) was associated with an increase in FPG of 20.5 mg/dL, or 12.6% of mean FPG (162.13 mg/dL). For low-fat dairy items, a 1-standard-deviation increase in price (3.2 cents per 100 grams) was associated with an increase in FPG of 9.4 mg/dL, or 5.8% of mean FPG. Table 3 presents results from a fixed-effects model with the produce category disaggregated. The GHb results reveal that none of the produce coefficients were statistically significant. However, FPG results indicate that the price of fruit was positively and significantly associated with FPG levels, and that fruit is was driving the effect of produce at least in the case of the FPG outcome. Table 4 shows results with the low-fat dairy category disaggregated. Low-fat cheese had a positive and significant association with both GHb and FPG, but low-fat milk and low-fat yogurt and other dairy were not significantly associated with either biomarker.

Effect of Food Prices in Low-Income vs Higher-Income Households Appendix C (available as a supplement to the online version of this article at http://www. ajph.org) presents results from the interaction model that compares low- and higher-income people living with type 2 diabetes. We found evidence supporting our hypothesis that people in different income groups respond differently to changes in the price of several food groups, namely, lean protein, oils and nuts, processed foods, low-fat dairy, and packaged refined grains. A price increase for healthy food groups

680 | Research and Practice | Peer Reviewed | Anekwe and Rahkovsky

(lean protein and low-fat dairy) was associated with a relative increase in blood sugar among the poorer groups, and a price increase for less-healthy foods (processed foods and packaged refined grains) was associated with a relative decrease in blood sugar among the poorer groups. These results indicate that low-income groups had higher price elasticity of blood sugar (i.e., their blood sugar was more responsive to changes in food prices) than the higher-income group. Only in the case of oils and nuts (which are part of a diabetes-healthy diet32---34) did we find an unexpected result: a price increase was associated with a relative decrease in blood sugar levels among the group with income at 131% to 185% of the poverty level compared with the higher-income group. This finding might be explained if lower-income people were more likely than high-income people to use oils in less-healthy ways, such as deep frying.35

Robustness Check To test the strength of our results, we reanalyzed the data by using an alternative definition of food prices. Here, we defined prices as the price of nutrients (e.g., dollars per gram of fat purchased) rather than the price per 100 grams of a particular food group. To calculate nutrient prices, we combined Nielsen Homescan 1999---2006 data with nutrition information (for random-weight items) from the US Department of Agriculture Standard Reference database as well as nutrition information (for UPC items) from the Gladson UPC database. We included nutrients that are most relevant to the health of people with diabetes: calories, carbohydrates, fiber, saturated fat, and sugar. We calculated nutrient prices as dollars per calorie and as dollars per gram of carbohydrates, fiber, saturated fat, or sugar, for each combination of annual quarter and market region. We also weighted nutrient prices by using the Homescan household statistical weights to make them nationally representative. The results are shown in Appendix D (available as a supplement to the online version of this article at http://www.ajph.org), and they broadly support the conclusions from our main analysis. We found that the price of calories was negatively associated with GHb, and the price of fiber was positively associated with

American Journal of Public Health | April 2014, Vol 104, No. 4

RESEARCH AND PRACTICE

TABLE 1—Weighted Descriptive Statistics for Adults With Diabetes: National Health and Nutrition Examination Survey Linked to Food Prices Contained in the Quarterly Food-at-Home Price Database, United States, 1999–2006 Variable

GHb (n = 1582),a Mean (SD)

FPG (n = 841),b Mean (SD)

Absolute Value of the Percentage Difference Between Means,c %

Glucose measure GHb, %

7.448 (1.734)

FPG, mg/dL Commodity price

160.577 (58.642)

Produce, $/100 g

0.223 (0.022)

0.223 (0.021)

0.097

Lean protein (eggs, poultry, fish), $/100 g

0.545 (0.090)

0.546 (0.086)

0.22

Oils and raw nuts, $/100 g

0.729 (0.094)

0.727 (0.091)

0.23

Regular red meat, $/100 g

0.654 (0.068)

0.655 (0.065)

0.046

Low-fat red meat, $/100 g

0.787 (0.074)

0.784 (0.069)

0.41

Processed foods, $/100 g

0.504 (0.042)

0.505 (0.042)

0.27

Regular dairy, $/100 g Low-fat dairy, $/100 g

0.532 (0.080) 0.211 (0.032)

0.534 (0.075) 0.210 (0.031)

0.30 0.57

Refined grains, $/100 g

0.352 (0.028)

0.353 (0.027)

0.27

Packaged whole grains, $/100 g

0.479 (0.029)

0.479 (0.029)

0.095

Calories, $/calorie

0.002 (0.00003)

0.002 (0.00003)

0.71

Carbohydrates, $/g

0.015 (0.0003)

0.015 (0.0003)

0.47

Fiber, $/g

0.273 (0.006)

0.272 (0.005)

0.67

Saturated fat, $/g

0.171 (0.005)

0.169 (0.004)

1.41

Sugar, $/g Diagnosed with diabetes

0.036 (0.002) 0.800 (0.376)

0.035 (0.001) 0.694 (0.426)

1.71 14.15

BMI, kg/m2

32.647 (7.013)

32.678 (7.342)

0.095

Female

0.512 (0.470)

0.480 (0.462)

6.46

Age, y

59.109 (13.206)

58.744 (13.022)

0.62

Mexican American

0.081 (0.256)

0.072 (0.240)

11.18

Other Hispanic

0.069 (0.238)

0.057 (0.214)

19.12

Non-Hispanic White Non-Hispanic Black

0.617 (0.457) 0.168 (0.352)

0.660 (0.438) 0.156 (0.336)

6.83 7.75

2.644 (1.389)

2.568 (1.290)

2.92

Race/ethnicity

Household size Education < 9th grade

0.142 (0.328)

0.132 (0.313)

6.97

9th–11th grade

0.175 (0.357)

0.153 (0.333)

13.10

High-school graduate

0.254 (0.410)

0.281 (0.416)

9.94

Some college

0.266 (0.416)

0.272 (0.412)

2.06

0.163 (0.347)

0.161 (0.340)

1.25

Mother

0.228 (0.395)

0.218 (0.382)

4.30

Father

0.128 (0.314)

0.148 (0.329)

14.55

‡ college graduate Family history of diabetes diagnosis

Maternal grandmother

0.074 (0.246)

0.097 (0.274)

27.23

Maternal grandfather

0.041 (0.187)

0.036 (0.173)

12.62

Paternal grandmother

0.049 (0.204)

0.059 (0.217)

17.42

Paternal grandfather

0.027 (0.151)

0.026 (0.148)

0.61

Brother Sister

0.122 (0.308) 0.142 (0.328)

0.113 (0.293) 0.139 (0.320)

7.59 2.21 Continued

April 2014, Vol 104, No. 4 | American Journal of Public Health

Anekwe and Rahkovsky | Peer Reviewed | Research and Practice | 681

RESEARCH AND PRACTICE

TABLE 1—Continued Exercise Walked or biked over past 30 d Minutes walked or biked on those days Any vigorous physical activity for ‡ 10 min over past 30 d

0.187 (0.365)

0.204 (0.372)

8.61

45.709 (49.709)

37.325 (29.372)

20.19

0.185 (0.363)

0.197 (0.366)

6.41

How often drank alcohol over past 12 mo

3.065 (20.074)

3.414 (20.620)

10.77

Meals/wk from outside the home Health insurance

2.423 (2.952) 0.879 (0.306)

2.454 (2.850) 0.887 (0.293)

1.27 0.83

Poverty–income ratio

3.307 (4.108)

3.464 (4.732)

4.66

Ever had blood cholesterol checked

0.876 (0.310)

0.882 (0.298)

0.73

Median of the absolute values

2.06

Mean of the absolute values

5.55

Note. BMI = body mass index (calculated as weight in kilograms divided by height in meters squared); FPG = fasting plasma glucose; GHb = glycated hemoglobin. Means and standard deviations were estimated by using sample weights to represent the diabetic population of the United States. a The GHb sample includes people whose GHb reading was in the diabetic range (i.e., ‡ 6.5%) and people who were told by a clinician that they had diabetes. b The FPG sample includes people whose FPG reading was in the diabetic range (i.e., ‡ 126 mg/dL) and people who were told by a clinician that they had diabetes. c Calculated as the absolute value of 100*(mean 1 – mean 2)/(the average of mean 1 and mean 2).

–1284.71 times the standard deviation of $0.00003 per calorie). Because the price of calories was highly correlated with some of the other nutrient prices, we removed it to reduce collinearity, and then reestimated the model. In

GHb, although this latter association did not reach statistical significance. A 1-standarddeviation increase in the price of calories was associated with a small decrease in GHb, of 0.04% points (calculated as the effect size of

this reestimated model, the price of saturated fat was negatively associated with GHb. A 1-standard-deviation increase in the price of saturated fat was associated with a small decrease in GHb of 0.002% points. Finally, we

TABLE 2—Effect of Previous-Quarter Food Group Prices on Diabetes Biomarkers: Fixed-Effects and Ordinary Least Squares Models: National Health and Nutrition Examination Survey Linked to Food Prices Contained in the Quarterly Food-at-Home Price Database, United States, 1999–2006 GHb (n = 1582) Variable Produce Lean protein

FPG (n = 841)

FE (95% CI)

OLS (95% CI)

FE (95% CI)

OLS (95% CI)

13.16* (0.24, 26.07)

11.83 (–0.37, 24.02)

932.95** (251.82, 1614.09)

711.95* (104.46, 1319.45)

1.52 (–1.35, 4.39)

–0.19 (–2.31, 1.93)

72.58 (–82.90, 228.06)

–42.98 (–144.08, 58.13)

Oils and raw nuts Regular red meat

–0.88 (–2.41, 0.65) –5.29 (–11.79, 1.20)

–0.50 (–1.74, 0.74) –1.03 (–4.58, 2.51)

–24.99 (–82.65, 32.68) 16.76 (–259.78, 293.31)

–20.91 (–76.52, 34.70) –16.18 (–216.77, 184.42)

Low-fat red meat

–1.56 (–3.96, 0.85)

–1.52 (–3.05, 0.019)

–90.29 (–194.76, 14.18)

Processed foods

3.61 (–3.38, 10.59)

3.91* (0.41, 7.40)

28.45 (–316.04, 372.93)

Regular dairy

1.96 (–3.56, 7.47)

–1.03 (–3.88, 1.83)

Low-fat dairy

3.31 (–2.24, 8.85)

–0.06 (–4.43, 4.30)

292.99* (33.15, 552.83)

Packaged refined grains

4.93 (–3.61, 13.46)

–1.03 (–7.04, 4.98)

–131.18 (–536.62, 274.25)

–74.67 (–407.26, 257.92)

Packaged whole grains

1.30 (–3.60, 6.20)

0.98 (–2.67, 4.64)

39.34 (–191.44, 270.13)

83.38 (–121.92, 288.68)

0.225

0.203

R2

180.37 (–47.43, 408.16)

–27.74 (–116.37, 60.89) –31.14 (–230.43, 168.15)

0.251

–101.99 (–266.59, 62.62) 249.45* (47.39, 451.51)

0.193

Note. CI = confidence interval; FE = fixed effects; FPG = fasting plasma glucose; GHb = glycated hemoglobin; OLS = ordinary least squares. The FE regressions controlled for Quarterly Food-at-Home Price Database–market-region fixed effects. The FE and OLS regressions included the following covariates: a linear time trend measured in quarters, age, age-squared, body mass index category (defined as weight in kilograms divided by height in meters squared), gender, race/ethnicity, education, poverty–income ratio, household size, whether an individual walked or biked in the past 30 days and for how long on average over those days of activity, whether an individual was engaged in vigorous physical activity for ‡ 10 minutes in the past 30 days, number of alcoholic drinks per year, number of times eating out per week, indicators of whether an individual refused to provide information about restaurants and alcohol consumption, health insurance status, ever had blood cholesterol checked, ever told by clinician that one has diabetes, whether one takes diabetic pills to lower blood sugar, and whether various blood relatives, living or deceased, had been told by clinician that they had diabetes (these relatives are mother, father, sister, brother, and grandmothers and grandfathers, both maternal and paternal). *P < .05; **P < .01.

682 | Research and Practice | Peer Reviewed | Anekwe and Rahkovsky

American Journal of Public Health | April 2014, Vol 104, No. 4

RESEARCH AND PRACTICE

TABLE 3—Effect of Previous-Quarter Food Group Prices on Diabetes Biomarkers—Produce, Disaggregated: National Health and Nutrition Examination Survey Linked to Food Prices Contained in the Quarterly Food-at-Home Price Database, United States, 1999–2006 Variable

GHb (n = 1582), Effect (95% CI)

FPG (n = 841), Effect (95% CI)

Fruit

6.66 (–4.47, 17.79)

765.95** (271.31, 1260.59)

Dark green vegetables

0.79 (–6.03, 7.60)

–16.95 (–306.87, 272.98)

Orange vegetables

3.03 (–3.40, 9.45)

180.97 (–136.78, 498.72)

–8.00 (–20.64, 4.64) –2.53 (–9.48, 4.42)

–459.42 (–1072.21, 153.37) 177.72 (–131.68, 487.11)

Starchy vegetables Other nutrient–dense vegetables Other watery vegetables

3.08 (–1.67, 7.83)

Legumes

4.51 (–8.88, 17.91)

Lean protein

1.71 (–1.64, 5.07)

15.75 (–206.30, 237.80) –157.26 (–669.41, 354.89) 65.47 (–103.37, 234.31)

Oils and raw nuts

–1.09 (–2.86, 0.67)

–13.47 (–84.45, 57.50)

Regular red meat

–3.67 (–10.63, 3.29)

–15.56 (–303.54, 272.42)

Low–fat red meat

–1.01 (–3.22, 1.19)

–83.81 (–184.91, 17.30)

Processed foods Regular dairy

4.57 (–2.22, 11.35) 1.04 (–5.04, 7.13)

20.28 (–326.47, 367.02) 199.08 (–34.12, 432.27)

Low–fat dairy

2.79 (–2.71, 8.30)

Packaged refined grains

6.71 (–1.32, 14.75)

–51.35 (–403.64, 300.94)

Packaged whole grains

0.67 (–4.23, 5.57)

–54.51 (–313.16, 204.14)

0.229

0.256

R2

300.50* (16.63, 584.37)

Note. FPG = fasting plasma glucose; GHb = glycated hemoglobin. These regressions controlled for a linear time trend measured in quarters, and Quarterly Food-at-Home Price Database–market-region fixed effects, as well as the following covariates: age, age-squared, body mass index category (defined as weight in kilograms divided by the square of height in meters), gender, race/ethnicity, education, poverty–income ratio, household size, whether an individual walked or biked in the past 30 days and for how long on average over those days of activity, whether an individual was engaged in vigorous physical activity for ‡ 10 minutes in the past 30 days, number of alcoholic drinks per year, number of times eating out per week, indicators of whether an individual refused to provide information about restaurants and alcohol consumption, health insurance status, ever had blood cholesterol checked, ever told by clinician that one has diabetes, whether one takes diabetic pills to lower blood sugar, and whether various blood relatives, living or deceased, had been told by clinician that they had diabetes (these relatives are mother, father, sister, brother, and grandmothers and grandfathers, both maternal and paternal). *P < .05; **P < .01.

reestimated both of those regressions using FPG as the outcome variable, and found that, when we included the price of calories in the model, the price of sugar was negatively associated with FPG. A 1-standard-deviation increase in the price of sugar was associated with a small decrease in FPG of 0.99 mg/dL. When we excluded the price of calories from the model, both the price of saturated fat and the price of sugar were negatively associated with FPG. A 1-standard-deviation increase in the prices of saturated fat and sugar were respectively correlated with a decrease in FPG of 0.08 mg/dL and 0.98 mg/dL.

DISCUSSION We found that food prices were associated with the blood sugar of people with type 2

diabetes. Our main results were (1) the price of produce 3 months previously was positively associated with blood sugar levels (as measured by GHb and FPG) and (2) the price of low-fat dairy items 3 months previously was positively associated with blood sugar levels (as measured by FPG). Our alternative specification (using nutrient prices instead of food-group prices) supports the main results. The price of calories was negatively associated with GHb, the price of sugar was negatively associated with FPG, and the price of saturated fat was negatively associated with both GHb and FPG. (These effects were smaller than those from the main specification, perhaps because the nutrient prices have a smaller standard deviation relative to mean than do the weight-based food prices.) These findings are intuitive because diets low

April 2014, Vol 104, No. 4 | American Journal of Public Health

in saturated fat and sugar support the health of people with diabetes, and low-calorie diets are recommended for overweight people with diabetes.1 We also found that price increases for healthy foods were associated with a relative increase in blood sugar levels for low-income US adults with type 2 diabetes compared with their higher-income counterparts. Likewise, price increases for less-healthy foods were associated with a relative decrease in blood sugar levels for the low-income group compared with their higher-income counterparts. These findings suggest that low-income US adults with type 2 diabetes benefit more (in terms of blood sugar) from low prices of healthy food than their higher-income counterparts. We should note that throughout the time period examined in this study, SNAP benefits could not be used to buy the following foods and beverages: beer, wine, liquor, and hot foods. Thus, SNAP participants would have had to pay out of pocket for items such as alcoholic beverages and “hot deli foods,” perhaps making any price increases for those items even more of a disincentive against buying them. (In this study, “hot deli foods” were classified as part of the processed foods category.)

Limitations Our study has a number of limitations. Our data did not allow us to disaggregate unsweetened beverages (e.g., diet sodas) from sweetened beverages (e.g., regular sodas), so we were unable to examine associations for those 2 groups separately, which would have been useful because unsweetened beverages presumably do not have an impact on blood sugar levels in the same way that sweetened beverages do. Another limitation is that we were unable to use data from years after 2006 because random-weight foods (e.g., many fruits and vegetables) were not available in QFAHPD in those later years. Finally, we were unable to estimate associations for the prices of all foods, as our data set included prices for food at home but not for food away from home. One limitation for this line of research is the disconnect between health and economic surveys. For example, NHANES has a wealth of health information but very little about respondents’ food purchases or the food prices

Anekwe and Rahkovsky | Peer Reviewed | Research and Practice | 683

RESEARCH AND PRACTICE

can be ordered at http://www.ajph.org by clicking the “Reprints” link. This article was accepted August 29, 2013. Note. The views expressed herein are those of the authors and may not be attributed to the Economic Research Service or the US Department of Agriculture.

TABLE 4—Effect of Previous-Quarter Food Group Prices on Diabetes Biomarkers—Low-Fat Dairy, Disaggregated: National Health and Nutrition Examination Survey Linked to Food Prices Contained in the Quarterly Food-at-Home Price Database, United States, 1999–2006 Variable Produce Lean protein

GHb (n = 1576), Effect (95% CI)

FPG (n = 840), Effect (95% CI)

15.68* (2.58, 28.79)

1263.80** (623.04, 1904.56)

1.83 (–1.07, 4.73)

52.76 (–101.83, 207.36)

Oils and raw nuts Regular red meat

–1.14 (–2.30, 0.021) –5.33 (–11.80, 1.13)

–22.24 (–65.88, 21.39) 132.39 (–165.34, 430.12)

Low-fat red meat

–1.11 (–3.77, 1.56)

Processed foods

–96.87 (–206.32, 12.57)

4.22 (–2.20, 10.65)

–26.91 (–341.72, 287.89)

Low-fat milk

–4.04 (–31.68, 23.60)

–991.96 (–2447.62, 463.70)

Low-fat cheese

0.60* (0.046, 1.16)

58.63** (29.35, 87.92)

Low-fat yogurt and other dairy

–5.08 (–16.43, 6.26)

–109.25 (–648.14, 429.64)

Regular milk

1.60 (–18.47, 21.67)

Regular cheese Regular yogurt and other dairy

1.58 (–1.98, 5.13) 1.05 (–4.84, 6.94)

Solid fats

–7.70** (–12.49, –2.92)

586.07 (–425.32, 1597.45) 25.84 (–146.59, 198.27) 1.39 (–266.19, 268.97) –100.06 (–333.28, 133.16)

Packaged refined grains

7.02 (–1.15, 15.19)

–206.17 (–573.78, 161.43)

Packaged whole grains

1.49 (–3.79, 6.76)

137.63 (–145.84, 421.10)

R2

0.232

0.258

Note. FPG = fasting plasma glucose; GHb = glycated hemoglobin. These regressions controlled for a linear time trend measured in quarters, and Quarterly Food-at-Home Price Database–market–region fixed effects, as well as the following covariates: age, age-squared, body mass index category, gender, race/ethnicity, education, poverty–income ratio, household size, whether an individual walked or biked in the past 30 days and for how long on average over those days of activity, whether an individual was engaged in vigorous physical activity for ‡ 10 minutes in the past 30 days, number of alcoholic drinks per year, number of times eating out per week, indicators of whether an individual refused to provide information about restaurants and alcohol consumption, health insurance status, ever had blood cholesterol checked, ever told by clinician that one has diabetes, whether one takes diabetic pills to lower blood sugar, and whether various blood relatives, living or deceased, had been told by clinician that they had diabetes (these relatives are mother, father, sister, brother, and grandmothers and grandfathers, both maternal and paternal). *P < .05; **P < .01.

respondents faced. On the other hand, Nielsen Homescan has a plethora of economic information but until recently has had scant health information. It would be ideal to combine the advantages of the 2 data sets into 1. Fortunately, recent enhancements to Nielsen Homescan and the Information Resources Inc Consumer Panel data set provide new opportunities for research. In 2007, Nielsen added a health survey that asks Homescan respondents about their chronic health conditions. Information Resources Inc later added its own health survey plus a survey of the prescription drugs that respondents take. We believe that these data sets provide new possibilities for research on the relationship between food prices and dietary health.

Conclusions Our results complement conclusions from a recent study7 that also used QFAHPD and

NHANES data and found that, for the population of US adults aged 50 years and older, the price of vegetables was significantly and positively associated with blood cholesterol and the price of regular milk was significantly and negatively associated with blood cholesterol. Taken together, that study and the present study underscore the importance of the affordability of healthy diets for mitigating the risk of chronic diseases among US adults, a topic of continuing interest in the research and policymaking communities.36---38 j

I. Rahkovsky originated the idea and performed the statistical analysis. T. Anekwe wrote most of the article, prepared the tables, and performed most of the reviewing and editing. Together, the authors created the analysis plan, discussed the econometric issues, and approved the final version.

Acknowledgments We thank Jean Buzby, PhD, Karen Hamrick, PhD, and Vicki Burt, ScM, RN, for helpful comments on earlier drafts of this article.

Human Participant Protection Institutional review board approval was not needed because we used only secondary data to conduct this study.

References 1. American Diabetes Association. Nutrition recommendations and interventions for diabetes. Diabetes Care. 2008;31(suppl 1):S61---S78. 2. Vijan S, Stuart NS, Fitzgerald JT, et al. Barriers to following dietary recommendations in type 2 diabetes. Diabet Med. 2005;22(1):32---38. 3. Horowitz CR, Williams L, Bickell NA. A community-centered approach to diabetes in East Harlem. J Gen Intern Med. 2003;18(7):542---548. 4. Rashad I. Obesity and diabetes: the roles that prices and policies play. Adv Health Econ Health Serv Res. 2007;17:113---128. 5. Powell LM, Chaloupka FJ. Food prices and obesity: evidence and policy implications for taxes and subsidies. Milbank Q. 2009;87(1):229---257. 6. Meyerhoefer CD, Leibtag ES. A spoonful of sugar helps the medicine go down: the relationship between food prices and medical expenditures on diabetes. Am J Agricultural Econ. 2010;92(5):1271---1282. 7. Rahkovsky I, Gregory CA. Food prices and blood cholesterol. Economics Hum Biol. 2013;11(1):95---107. 8. Auld MC, Powell LM. Economics of food energy density and adolescent body weight. Economica. 2009;76(304):719---740. 9. Powell LM, Bao Y. Food prices, access to food outlets and child weight. Economics Hum Biol. 2009;7(1): 64---72. 10. Sturm R, Powell LM, Chriqui JF, Chaloupka FJ. Soda taxes, soft drink consumption, and children’s body mass index. Health Aff (Millwood). 2010;29(5):1052---1058.

About the Authors The authors are with the Economic Research Service, US Department of Agriculture, Washington, DC. Correspondence should be sent to Tobenna Anekwe, Economic Research Service, US Department of Agriculture, 1400 Independence Ave SW, Washington, DC 20250-1800 (e-mail: [email protected]). Reprints

684 | Research and Practice | Peer Reviewed | Anekwe and Rahkovsky

Contributors

11. Wendt M, Todd JE. The Effect of Food and Beverage Prices on Children’s Weights. Washington, DC: US Department of Agriculture, Economic Research Service; 2011. 12. Brownell KD, Farley T, Willett WC, et al. The public health and economic benefits of taxing sugar-sweetened beverages. N Engl J Med. 2009;361(16):1599---1605.

American Journal of Public Health | April 2014, Vol 104, No. 4

RESEARCH AND PRACTICE

13. Cash SB, Sunding DL, Zilberman D. Fat taxes and thin subsidies: prices, diet, and health outcomes. Acta Agriculturae Scandinavica, 2005;2(3-4):167---174. 14. Chouinard HH, Davis DE, LaFrance JT, Perloff JM. Fat taxes: big money for small change. Forum Health Econ Policy. 2007;10(2):1---28. 15. Jacobson MF, Brownell KD. Small taxes on soft drinks and snack foods to promote health. Am J Public Health. 2000;90(6):854---857. 16. American Diabetes Association. Executive summary: standards of medical care in diabetes—2012. Diabetes Care. 2012;35(suppl 1):S4---S10. 17. National Library of Medicine. Glucose test—blood. 2012. Available at: http://www.nlm.nih.gov/medlineplus/ ency/article/003482.htm. Accessed September 13, 2012. 18. American Diabetes Association. Tests of glycemia in diabetes. Diabetes Care. 2004;27(suppl 1):s91---s93. 19. American Diabetes Association. Diagnosis and classification of diabetes mellitus. Diabetes Care. 2012; 35(suppl 1):S64---S71. 20. Sacks DB. A1C versus glucose testing: a comparison. Diabetes Care. 2011;34(2):518---523. 21. Todd JE, Mancino L, Leibtag E, Tripodo C. Methodology Behind the Quarterly Food-at-Home Price Database. Washington, DC: US Department of Agriculture, Economic Research Service; 2010. 22. Rasmussen OW, Thomsen C, Hansen KW, Vesterlund M, Winther E, Hermansen K. Effects on blood pressure, glucose, and lipid levels of highmonounsaturated fat diet compared with a highcarbohydrate diet in NIDDM subjects. Diabetes Care. 1993;16(12):1565---1571.

nhanes/nhanes1999-2000/LAB10AM.htm. Accessed September 17, 2012. 31. National Health and Nutrition Examination Survey. 2003---2004 data documentation, codebook, and frequencies. 2009. Available at: http://www.cdc.gov/nchs/ nhanes/nhanes2003-2004/L10AM_C.htm. Accessed September 17, 2012. 32. Jenkins DJA, Kendall CWC, Banach MS, et al. Nuts as a replacement for carbohydrates in the diabetic diet. Diabetes Care. 2011;34(8):1706---1711. 33. Jenkins DJA, Hu FB, Tapsell LC, Josse AR, Kendall CWC. Possible benefit of nuts in type 2 diabetes. J Nutr. 2008;138(9):1752S---1756S. 34. American Diabetes Association. Fat and diabetes. 2012. Available at: http://www.diabetes.org/food-andfitness/food/what-can-i-eat/fat-and-diabetes.html. Accessed September 18, 2012. 35. Gittelsohn J, Franceschini MCT, Rasooly IR, et al. Understanding the food environment in a low-income urban setting: implications for food store interventions. J Hunger Environment Nutr. 2008;2(2-3):33---50. 36. Cassady D, Jetter KM, Culp J. Is price a barrier to eating more fruits and vegetables for low-income families? J Am Diet Assoc. 2007;107(11):1909---1915. 37. Waterlander WE, Steenhuis IH, de Boer MR, Schuit AJ, Seidell JC. The effects of a 25% discount on fruits and vegetables: results of a randomized trial in a threedimensional web-based supermarket. Int J Behav Nutr Phys Act. 2012;9:11. 38. Leone AF, Rigby S, Betterley C, et al. Store type and demographic influence on the availability and price of healthful foods, Leon County, Florida, 2008. Prev Chronic Dis. 2011;8(6):A140.

23. Bosse MC, Davis SC, Puhl SM, et al. Effects of Zone diet macronutrient proportions on blood lipids, blood glucose, body composition, and treadmill exercise performance. Nutr Res. 2004;24(7):521---530. 24. American Diabetes Association. Diagnosis and classification of diabetes mellitus. Diabetes Care. 2004;27 (suppl 1):s5---s10. 25. Koro CE, Bowlin SJ, Bourgeois N, Fedder DO. Glycemic control from 1988 to 2000 among U.S. adults diagnosed with type 2 diabetes. Diabetes Care. 2004; 27(1):17---20. 26. National Health and Nutrition Examination Survey. Continuous NHANES Web tutorial: tips STATA. 2010. Available at: http://www.cdc.gov/nchs/tutorials/ nhanes/stata_tips.htm. Accessed June 10, 2012. 27. Andreyeva T, Long MW, Brownell KD. The impact of food prices on consumption: a systematic review of research on the price elasticity of demand for food. Am J Public Health. 2010;100(2):216---222. 28. National Health and Nutrition Examination Survey. 2001---2002 data documentation, codebook, and frequencies. 2007. Available at: http://www.cdc.gov/nchs/ nhanes/nhanes2001-2002/L10AM_B.htm. Accessed September 17, 2012. 29. National Health and Nutrition Examination Survey. 2005---2006 data documentation, codebook, and frequencies. 2008. Available at: http://www.cdc.gov/nchs/ nhanes/nhanes2005-2006/GLU_D.htm. Accessed September 17, 2012. 30. National Health and Nutrition Examination Survey. 1999---2000 data documentation, codebook, and frequencies. 2009. Available at: http://www.cdc.gov/nchs/

April 2014, Vol 104, No. 4 | American Journal of Public Health

Anekwe and Rahkovsky | Peer Reviewed | Research and Practice | 685