The study area consisted of the city of St Louis, Mo, and the eastern part of St Louis County, Missouri, the area between the Missouri River on the east and Interstate 270 (the outer belt of the St Louis area) on the west. This area is considered by many residents to comprise the urban area of St Louis. This area includes 233 square miles, 220 census tracts, and 912,323 people (25).

The audit tools developed for this study build upon previous work (20) and reflect current dietary intake guidelines established by the U.S. Department of Agriculture (USDA) (26). The recommendations include eating a variety of fruits and vegetables each day and selecting from all five vegetable subgroups (dark green, orange, legumes, starchy vegetables, and other vegetables). In addition, the recommendations suggest eating lean, low-fat, or fat-free meat, poultry, and dairy products. The audit tools were designed to assess the extent to which the environment — supermarkets and fast food restaurants — provided individuals with a selection of foods that would enable them to follow these recommendations.

Supermarket audit tool

The supermarket audits were structured to determine the extent to which the selection of foods available in each supermarket enabled individuals to meet USDA recommendations. The fruit and vegetable section of the audit tool was created as a checklist that included each item identified by the USDA's Continuing Survey of Food Intakes by Individuals (CSFII) as currently being consumed by adults living in urban midwestern cities (78 total fruits and vegetables). The checklist provided a place for the auditor to indicate whether each item was available in a fresh, frozen, or canned form in each store.

We used the USDA's Agriculture Handbook 8 (27) to develop the audit tool for assessing the availability of lean, low-fat, and fat-free meat, poultry, and dairy products. Similar to the fruit and vegetable section of the audit tool, this section was created as a checklist for the auditor to indicate whether each type of lean beef, skinless chicken, or low-fat or reduced-fat cheese was available in each store. The audit tool also allowed the recorder to identify the availability of fat-free, 1/2%, 1%, 1 1/2%, 2%, and whole milk.

Fast food restaurant audit tool

The fast food restaurant audit tool assessed the extent to which the menu options at each fast food restaurant provided the opportunity for individuals to meet the recommended dietary intake based on the availability and preparation of the foods (e.g., broiled or baked rather than fried). Fast food restaurants were defined as restaurants where customers place orders at the counter (i.e., with no waitress or waiter available for service). Twenty-six major fast food restaurant chains were audited including hamburger, sandwich, Mexican, chicken, and pizza chains. All fast food restaurants in this study provided menus through a corporate Web site. A checklist was created based on the menu of each chain (24). Audit tools were designed for each restaurant chain so that all items included on the corporate menu that would enable customers to meet one of the recommended eating patterns was placed on a checklist. For example, the items on the corporate menu (and thus the checklist) that were identified as providing the opportunity to eat fruits and vegetables may have included a garden salad, a chef salad, or a taco salad, depending on the chain. Each auditor used the corporation-specific checklist to assess which of the food options that met the recommended dietary guidelines were carried by local branches or franchises.

Conducting audits

Between 2003 and 2004, audits were conducted in person at all stores that were identified by the 2000 business census as either supermarkets or major-chain grocery stores and had addresses that were geographically located within the study area (N = 81). Each store was audited by two research staff: one observer (who visually noted all the items) and one recorder (who recorded the items on a standard data sheet). Each auditor participated in a half-day training session and followed an auditor for another half day. Using the 78-item fruits and vegetable checklist, the auditor recorded whether each store carried that fruit or vegetable and whether it was available fresh, frozen, or canned. Similarly, each auditor looked for all available meat, poultry, and dairy options. The auditors checked the checklist for each item the supermarket carried; they did not count the number of each item the store had. (For example, they noted whether there were peaches but not the number of peaches.) This process was chosen because variations in the actual number of peaches (or other items) might reflect purchasing and stocking patterns within the store rather than the availability of an item.

A two-stage process was used to assess the selection of options that met dietary intake recommendations at local fast food restaurants. The first phase entailed stratifying the study area into census tracts by three racial and three poverty groups. Each fast food restaurant was placed within a stratum based on its location. A random sample of two of each type of fast food chain within each census tract was then audited by telephone. These audits were conducted by auditors who participated in a half-day training session. Once the manager or supervisor of the fast food restaurant agreed to an audit, the auditor went through the checklist of items based on the corporate menu and asked the manager or supervisor to indicate whether each item was available at the branch or franchise.

The data were reviewed after 130 fast food restaurants, or approximately half the sample, had been audited. It was then determined that there were few differences in the availability of healthier options within chains, regardless of geographic location or stratum (race or income). Overall, the variance between restaurants within the same chain was less than 5% for most chains (ranging from 0% to 5%). The lack of variability from our calls, along with the possibility of respondent bias, resulted in a decision to score each fast food restaurant based on its corporate menu. Although the restaurants may have varied more than we ascertained through our sampling and may not have had all of the items from the corporate menu, this method allowed us to give each restaurant its best possible rating.

Availability of healthy choices 

A composite score was created for each supermarket by combining the observed number of different fruits and vegetables available and lean, low-fat, and fat-free meat, poultry, and dairy options. Using the composite score, z scores were calculated for each supermarket. The z scores, based on the mean availability and standard deviation of all items, allowed comparisons among supermarkets. Tertiles were created to indicate high, medium, or low availability of fruits and vegetables and low-fat options based on the distribution of z scores.

Each fast food restaurant audited also received a composite score based on the total number of items available that met dietary guidelines. The z scores were calculated for each restaurant. Based on the distribution of z scores, ratings were divided into tertiles and labeled high, medium, or low potential for meeting dietary intake recommendations.

2000 census data 

To determine racial distribution and poverty rates in the study area, we used 2000 U.S. census data at the census-tract level. Only two racial groups were considered because, according to the U.S. Census Bureau, 95% of the population residing in the study area self-identifies as white or African American. Census tracts were identified as primarily African American if 75% or more of the population in the area self-identified as African American or as primarily white if 75% or more of the population in the area self-identified as white. All other areas were identified as mixed.

The percentage of the population living below the U.S. federal poverty level was measured using 2000 U.S. census data. The poverty rate is a measure that seems to be robust across various diseases and levels of geography; it has a link to possible policy implications; and it is comparable over time (28,29). We grouped poverty rate into three levels: less than 10% of the population living in poverty, 10% to 19.9% of the population living in poverty, and 20% or more of the population living in poverty.

Address matching and spatial clustering 

The street addresses of the fast food restaurants identified were converted to approximate geographic locations and assigned a latitude and longitude. First, all addresses were preprocessed using ZP4 (Semaphore Corp, Pismo Beach, Calif) and then address-matched using ArcView 3.2 (ESRI, Redlands, Calif) with the Redistricting Census 2000 TIGER/Line as the reference files. Unmatched or questionably matched addresses (scoring lower than 85) were recoded using the Internet-based EZ-Locate system (Tele Atlas North America, Inc, Lebanon, NH). Of the 32 unmatched addresses, one was matched to the centroid of the ZIP code, one was matched to the centroid of the ZIP+2 code, one was a near match, and the remaining were matched at the block-face level.

Spatial clustering of supermarkets and fast food restaurants was determined using a spatial scan statistic performed with the software SaTScan (Martin Kulldorff, Boston, Mass) (28,29). The statistic uses a circular window of variable radius that moves across the map. The null hypothesis was that the rate of fast food restaurants or supermarkets (the number of businesses expected per 100,000 population) was the same in all windows. Clusters were defined as areas that had either a lower or higher rate of fast food restaurants or supermarkets than expected. The process of cluster detection was run through 999 Monte Carlo permutations of the data set to identify combinations of clusters of higher and lower rates of fast food restaurants (or supermarkets). The analyses were purely spatial, with a maximum cluster size of 50% of the population size. A Poisson distribution was assumed. The output provided the most likely cluster as well as several secondary clusters. Data associated with these clusters included the location identifiers, search radius and center coordinates, standard rate ratio, and a P value based on the log likelihood ratio test for each cluster. The cluster results were mapped in ArcGIS version 9 (ESRI, Redlands, Calif).

The SaTScan method was run eight times using different data and parameters. We assessed separately the spatial clustering of supermarkets and fast food restaurants regardless of the audit results. Next, we adjusted for the underlying racial distribution and poverty rate based on 2000 census data and determined whether the areas of higher or lower than expected number of supermarkets were still present. We then examined spatial clustering of supermarkets and fast food restaurants that scored in the highest tertile. We also adjusted these results by the racial distribution and poverty rate to determine whether these factors could explain the spatial clustering.

Figure 1 shows the location of the 81 supermarkets in the study area; the area has a rate of 8.9 supermarkets per 100,000 population. Table 2 identifies the number of possible clusters of supermarkets within the study area. The unadjusted analysis showed no clustering of supermarkets; the most likely cluster showed P = .86. Adjusting for racial distribution and poverty rate resulted in similar findings: there was no clustering of supermarkets detected in the study area; the most likely cluster showed P = .48.

Of the 81 supermarkets, 26 were in the highest tertile. Table 2 shows two clusters of supermarkets in the highest tertile, and Figure 2 shows their location. Cluster 1, located in the southern part of the study area, had 23 supermarkets in the highest tertile when only 9.7 had been expected (ratio of observed/expected = 2.4; P = .001). This area included census tracts that were primarily white or racially mixed.

Alternately, Cluster 2 was located in the northeastern part of the study area (Figure 2). Nine supermarkets in the highest tertile were expected to be found, but there were no supermarkets in this area (ratio of observed/expected = 0, P = .003). The 311,491 primarily African American and lower-income people in this area were less likely than expected to have access to supermarkets where healthy choices were available.

Adjusting for the underlying racial distribution and poverty rate for the supermarkets in the highest tertile resulted in no significant clusters (P = .11).

There were 355 fast food restaurants located in the study area with a rate of 39.0 restaurants per 100,000 population (Figure 3).  Table 3 shows that four clusters were identified in the unadjusted analysis; Figure 4 shows their location. In Cluster 1, representing 52 census tracts and 20.6% of the study area population, 73 fast food restaurants were expected, and 31 were observed (ratio of observed/expected = 0.4; P = .001). The other three clusters indicated a higher than expected number of fast food restaurants but were based on a much smaller number of census tracts.

After adjusting for the racial distribution and poverty rate, only two clusters remained (Figure 5). Cluster 1 still showed fewer fast food restaurants than expected (ratio of observed/expected = 0.1; P = .004), but the area of Cluster 1 in the adjusted analysis contained only eight census tracts (Table 3). Cluster 2 remained unchanged in size and location after adjustment for racial distribution and poverty rate (ratio of observed/expected = 3.1; P = .001). Cluster 3 in the unadjusted analysis was no longer statistically significant after adjustment (P = .69). In addition, Cluster 4 in the unadjusted analysis was also no longer significant after adjustment (P = .053).

Of the 355 fast food restaurants, 120 were considered to be in the highest tertile. Two clusters were detected in the unadjusted analysis (Figure 6). Cluster 1 was located in the northern part of the study area and consisted of 75 census tracts and 31.7% of the study area population. The people in this cluster, primarily African American and lower income, were less likely than expected to have fast food restaurants in the highest tertile (ratio of observed/expected = 0.3; P = .001). A second, much smaller cluster was located mainly in the east central part of St Louis County and consisted of 17 census tracts. People in this cluster, primarily white and middle to higher income, were more likely than expected to have access to fast food restaurants that offered healthier food options (ratio of observed/expected = 3.0; P = .01). After adjustment for racial distribution and poverty rate, no clusters were detected.