Introduction

As global eating patterns increasingly shift toward increased consumption of foods prepared away from home [1–3], various strategies are needed to promote healthy food choices in retail environments. To increase transparency and encourage healthy eating, many countries have introduced or implemented nutritional labeling policies such as menu labeling, front-of-package nutrition disclosures, warning labels, and others [4]. Calorie menu labeling, which requires food outlets to post the calorie content of menu offerings, is one such strategy that has been implemented in several countries, including the United States, South Korea, and Saudi Arabia [5]. As other countries consider passage or implementation of similar menu labeling policies, more real-world evidence is needed regarding the effect they have on customer meal choices.

In May 2018, the US Food and Drug Administration required implementation of calorie menu labeling in food establishments with ≥20 locations, which was mandated by the 2010 Affordable Care Act (ACA) [6]. Evidence from rigorous quasi-experimental evaluations of this policy is mixed; smaller studies have generally found no change in the calorie or nutrient content of consumer purchases after labeling [7–13], but studies examining large databases of retail transactions have detected small decreases in meals’ calorie content [14,15]. However, to our knowledge, no studies have examined these associations since nationwide implementation of this policy. Nationwide implementation may have led to improvements in customer meal purchases because of increased attention surrounding the policy and because customers were exposed to labeling at nearly all large chain restaurants [16]. Both are mechanisms that may have enhanced awareness of the policy and encouraged healthier purchasing. Restaurants also may have reformulated their menus to reduce the calorie content of offerings after labeling, which could reduce customer calorie purchases without requiring changes to their behavior [5]. Evidence of reformulation has not been observed in Canada [17] or Australia [18] after implementation of regional labeling laws, but no studies to our knowledge have investigated reformulation after nationwide implementation of menu labeling. Evidence on post-implementation effects of US calorie labeling is thus important to gain a broader understanding of the effect of this policy and to offer guidance to policymakers in countries where it is being considered.

Using sales data from a large fast food franchise in the southern US that voluntarily implemented calorie labeling in April 2017, we previously found that the calorie content of purchased meals declined by 4% after labeling, with an increasing trend that attenuated this association over the year that followed [14]. This study ended 1 month before calorie labeling was implemented nationwide. We have obtained an additional year of sales data from the franchise, which allows us to now report changes in calorie and nutrient content of meals purchased 2 years after franchise calorie labeling (April 2017 to April 2019), including an examination of whether purchases changed specifically after nationwide implementation of labeling (May 2018 to April 2019).

Methods

We leveraged a natural experiment to determine how the nutrient content of purchased meals in fast food restaurants changed after franchise calorie menu labeling and nationwide implementation of calorie labeling. We used data from a 2-year pre-labeling period to predict what meals’ nutrient content would have been in the post-labeling period had labeling not happened; we compared these to the observed nutrient content to estimate the effect of labeling.

Results

Our final sample included 59 restaurants in Louisiana, 41 in Texas, and 4 in Mississippi. On average, each restaurant had sales data available for 198 weeks. Our analytic dataset represented 332 million items purchased and 67 million transactions over the study period (Table 1). In the pre-labeling period, there were about 4.9 items purchased per transaction, and transactions had a mean (SD) of 1,486 calories (152), 63.5 g fat (7.6), 182.5 g carbohydrate (16.8), 50.2 g protein (6.1), 20.6 g saturated fat (2.5), 67.0 g sugar (7.0), 16.5 g fiber (1.7), and 2,841 mg sodium (378).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Transactions, item purchases, and mean (SD) calorie and nutrient content of purchased meals in restaurants overall and by calorie labeling implementation period.

https://doi.org/10.1371/journal.pmed.1003714.t001

Franchise labeling in April 2017 was associated with a level decrease of −54 calories/transaction (95% CI: −67, −42, p < 0.0001) and a trend increase of 3.3 calories/transaction per 4-week period (95% CI: 2.5, 4.1, p < 0.0001) over the subsequent 13 months that was independent of the baseline trend. Nationwide implementation of labeling in May 2018 was associated with another level decrease of −82 calories/transaction (95% CI: −88, −76, p < 0.0001) and a trend decrease of −2.1 calories/transaction per 4-week period (95% CI: −2.9, −1.3, p < 0.0001) (Fig 1, Table 2). The overall change in trend after franchise implementation, inclusive of both the post-franchise and post-nationwide labeling periods, was still positive compared to the baseline trend (i.e., [β_{weekfranchise} = 3.3] + [β_weeknation = −2.1] = 1.2 calories/transaction per 4-week period [95% CI: 0.80, 1.63 from 1,000 bootstrapped samples]). In the last week of the study, the mean transaction contained 73 (95% CI: 65, 81) fewer calories than what would be expected had labeling not been implemented, a 4.7% decrease. We observed generally similar results in sensitivity analyses (S1 Table). However, when using only 1 year of pre-labeling data, the model predicted that by the end of the study, the mean transaction increased by 43 calories (95% CI: 22, 63).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Level and trend changes in mean purchased calories per transaction after franchise labeling and implementation of labeling nationwide.

(A) The graph shows mean calories/transaction across all restaurants (gray dots) and the predicted calories/transaction from the model (red line). The model is also adjusted for seasons and holidays, but these effects are not depicted here. The time surrounding franchise labeling (April 2017) is depicted as dashed vertical lines, and the time surrounding nationwide implementation (May 2018) is depicted as dotted lines. (B) Same as (A) but magnified to see level and trend changes more easily.

https://doi.org/10.1371/journal.pmed.1003714.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Interrupted time series for change in mean calories and mean nutrients purchased per transaction after franchise calorie labeling implementation (April 2017) and after nationwide calorie labeling implementation (May 2018).

https://doi.org/10.1371/journal.pmed.1003714.t002

For other nutrients, we overall observed small level changes and minimal trend changes after each labeling period (Table 2). At the end of the study, the model-based estimates of nutrient content were lower for all nutrients than expected had labeling not been implemented (total fat: −3.0%; carbohydrates: −5.0%; protein: −5.1%; saturated fat: −1.8%; sugar: −7.0%; fiber: −4.4%; sodium: −5.0%). For some nutrients, these decreases offset the overall positive trends observed in the baseline period. For example, when comparing predicted nutrient content at the beginning and end of the study, we found slight improvements for saturated fat (12.6% of calories [April 2015] versus 12.0% of calories [April 2019]) and fiber (10.9 g/1,000 calories versus 11.6 g/1,000 calories). However, for other nutrients, the decrease at the end of the study did not offset positive baseline trends. For example, we did not find improvements for sugar (18.0% of calories [April 2015] versus 18.3% of calories [April 2019]) or sodium (2,748 mg versus 2,850 mg) at the end of the study.

When we ran models that assumed an intervention only at the time of franchise labeling (ignoring nationwide implementation), level and trend changes were different than in the main analyses (S2 Table). However, these models suggested end-of-study decreases in calorie and nutrient purchases that were very similar to results from our main analyses (e.g., 74-calorie/transaction decrease [95% CI: 67, 82]) for all nutrients except sodium (larger change of −208 mg/transaction, 95% CI: −224, −190).

We observed a stronger trend increase in calories/transaction after franchise labeling in lower-income census tracts than in higher-income census tracts (Table 3). By the end of the study, transactions in higher-income census tracts had on average 95 fewer calories (95% CI: 85, 107) than expected had labeling not occurred, but those in lower-income tracts had on average only 49 fewer calories (95% CI: 39, 59) than expected. This pattern was consistent across all nutrients.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Interrupted time series for change in mean calories and nutrients purchased per transaction after restaurant calorie labeling implementation (April 2017) and after nationwide calorie labeling implementation (May 2018) by median income of restaurant census tracts¹.

https://doi.org/10.1371/journal.pmed.1003714.t003

We observed minimal reformulation of calories and nutrient content of the 100 top-selling items that remained on the menu in the pre- and post-labeling periods (S3 Table). Items that were newly introduced in 2018 did not have lower nutrient content than those introduced in 2016 to 2017 (i.e., pre-labeling). We also did not find statistically significant lower nutrient content of items newly introduced in 2019 (versus 2016 to 2017), though the point estimates were much stronger. For example, items newly introduced in 2019 had on average 108.5 fewer calories (95% CI: −16.9, 233.8, p = 0.09) than those newly introduced in 2016 to 2017. We also found that the median calorie content of all items offered at the restaurants decreased over time (from 290 calories in 2015 to 235 calories in 2019); nutrient content did not appear to change appreciably over time (S4 Table). Lastly, when exploring mechanisms for changing calories over time, we observed slight level increases in calories purchased per item after each labeling period, but these were followed by trend decreases that largely attenuated these associations (S5 Table). We observed small decreases in mean items/transaction after franchise labeling (β_franchise = −0.3 items/transaction, 95% CI: −0.3, −0.2, p < 0.0001) and after nationwide labeling (β_nation = −0.4 items/transaction, 95% CI: −0.5, −0.4, p < 0.0001), neither of which was followed by a trend change.

Discussion

Data from this large natural experiment support an association of calorie labeling with small decreases in calorie and nutrient content of purchased fast food meals 2 years after initial franchise implementation and 1 year after nationwide implementation. By the end of the study, the average transaction contained an estimated 73 fewer calories (−4.7%) than expected had labeling not occurred. Labeling was also associated with overall small decreases in nutrient content of meals purchased that were consistent with calorie results. These results support modest improvements in purchased meal quality in response to calorie menu labeling.

Our previous analysis, which examined only franchise labeling over 1 year [14], found a similarly small level decrease and post-implementation trend increase in calories/transaction after labeling (small differences were apparent because the current study included 1 additional month of data after franchise labeling but before nationwide labeling). The present analysis indicates longer-term reductions in calories/transaction following nationwide implementation, albeit with an estimated increased trend of 1.2 calories/transaction per 4-week period. The reason for this increasing trend is unknown, but it may be explained if some customers stopped responding to the labels over time. Alternatively, the franchise could have engaged in other business practices to counteract labeling (e.g., marketing), though we could not examine this with our data. If the association persists over time, even the small reduction we observed could have an important population-level impact on energy balance and health [14,28].

The small decreases we observed in meals’ nutrient content (i.e., besides calories) are likely due to the fact that customers purchased overall less food after labeling. Because most of the nutrients we analyzed contribute to calories, reduced purchases of food overall would lead to reductions in nutrient content that were similar to the reduction in calories, which is what we observed (i.e., 3% to 5% decreases for calories, total fat, carbohydrates, and protein). Additionally, our exploratory analysis found that labeling was associated with decreases in purchased items/transaction but not with decreases in purchased calories/item. This finding is consistent with a previous study of labeling in a coffee shop chain, which found a 6% decrease in calories purchased post-labeling that was mostly driven by customers buying fewer items [15].

Although labeling was associated with reduced absolute nutrient content, the nutritional composition of meals (i.e., nutrient content relative to calorie content), which is an independent predictor of chronic disease risk [29], did not consistently change for the better. For example, at the end of the study, the predicted sugar content (18.3% of calories) and sodium content (2,850 mg) were slightly higher than at the beginning of the study. Even nutrients that did improve by the end of the study (e.g., saturated fat: 12.0%; fiber: 11.6g/1,000 calories) still had overall lower nutritional quality than US dietary recommendations. The 2020–2025 Dietary Guidelines for Americans recommends that adults consume <10% calories from saturated fat, about 14 g fiber/1,000 calories, and <2,300 mg sodium per day for reducing chronic disease risk [30]. It also recommends that adults consume <10% calories from added sugar, though we only had data on total sugar. Although we do not know whether purchased meals exceeded these limits for individuals (due to lack of individual-level data), the nutrient composition of meals suggests overall low quality, even after labeling.

We observed little evidence of menu reformulation among the top-selling items after labeling and found that the distribution of nutrient content for menu items was similar over time, suggesting an overall small role of changes in the nutrient content of restaurant offerings. It has been hypothesized that menu labeling would encourage restaurants to reduce the calorie content of their menu offerings, which would lower the calorie content of purchased meals without requiring changes to customer behavior. Our results that find minimal changes in the calorie content of continuously offered items are consistent with findings from other studies that have assessed post-labeling menu reformulation [17,18]. This has not yet been formally analyzed across all major restaurant chains in the US since nationwide implementation. At the same time, we observed that items newly introduced in 2019 had on average 109 fewer calories than items newly introduced in 2016 or 2017, though the 95% CI was wide and crossed the null due to small sample size. It is possible that labeling did not cause restaurants to reformulate their standard offerings but did cause them to introduce more lower-calorie items, which could improve customer meal choice. This should be analyzed in the future in a larger sample of restaurant chains.

Our stratified results by median household income of restaurant census tracts revealed a potential disparity in the effect of calorie labeling on customer purchase quality. By the end of the study period, the post-labeling decrease in calorie content in higher-income neighborhoods (−95 calories/transaction) was nearly twice that of the decrease in lower-income neighborhoods (−49 calories/transaction). Given that low socioeconomic status is associated with poorer diet quality and increased risk of obesity in the US and other high-income countries [31–33], this disparity could potentially worsen existing inequities in diet quality and nutrition-related disease. Because we did not have individual-level data, we cannot confirm whether there is a true disparity between individuals of different socioeconomic status. However, previous studies of individuals have found that those with higher incomes are more likely to see and use calorie menu labels than those with lower incomes [34–36]. Based on prior studies, other labeling systems (e.g., traffic light labels and warning labels) may be more effective than nutrient disclosures [37,38], particularly among low- and middle-income populations [39,40].

Most previous quasi-experimental studies in real-world settings in the US have not detected associations between calorie labeling and calorie content of meal purchases [7–12]. However, most of these studies were not powered to detect very small differences in calories purchased, took place in different areas of the US, and examined purchases over shorter periods of time. One previous study of similar size that used data from a large coffee shop chain found a 6% decrease in calories purchased after New York City’s 2008 implementation of calorie labeling, compared to cities that did not require labeling then [15]. Previous studies that examined changes in nutrient content have also been too small to detect the minor changes we observed [9,10,13]. Studies evaluating the effect of labeling on food purchases outside of the US are more limited, perhaps because other countries have implemented labeling more recently or done so on a voluntary basis [5]. Results of studies in Canada and Australia suggest improved customer meal choice in response to labeling [41–43]; these have either assessed many types of menu labeling interventions (i.e., not just calorie information) or have been conducted in laboratory settings, where results may not generalize to real-world settings.

This study has limitations. First, we did not have a control group, which limited our ability to adjust for time-varying confounders [44]; however, we adjusted for some population-level confounders (e.g., Hurricane Harvey). Moreover, ITS analyses are generally robust to individual-level confounders assuming the population remains stable over time [26]. Second, we did not know how many people were included as a part of each transaction. Because many transactions probably included meals for >1 person, the associations we observed are likely to be smaller for individuals. Third, if, labeling did not change the meal purchased but rather some post-purchasing behavior (e.g., if some people ate less of the meal), we might have underestimated the effect of labeling on individuals’ health. Fourth, given that all restaurants were located in southern states in the US, our results may not generalize to other parts of the US or to other countries. Fifth, like all statistical models, our models may have been misspecified (e.g., not adjusting for seasonality correctly), which could lead to bias. However, most sensitivity analyses that tested our modeling assumptions yielded similar results with the overall same conclusion. The sensitivity analyses that assumed level and trend changes only at the time of franchise labeling found different level and trend changes, but the predicted change at the end of the study was very similar to our main analysis. This implies robustness of our main finding that labeling was associated with decreased nutrient content of meals 2 years after franchise labeling. Another sensitivity analysis that used just 1 year of pre-implementation data suggested weaker associations after franchise labeling level than our main analysis. Although it is difficult to know which model better predicts the baseline trend, we preferred our main model because the additional data points could provide more stable estimation of the baseline trend. Our 2-year baseline main analysis also may have adjusted for seasonality better than the 1-year baseline analysis because it covered pre-labeling data from 2 of each season, instead of 1 each in the sensitivity analysis.

To the best of our knowledge, this study is the first to examine calorie labeling and restaurant purchases after nationwide implementation in the US. Among 104 restaurants, calorie labeling was associated with a small decrease in mean calories/transaction 2 years after initial franchise labeling and nearly 1 year after implementation of labeling nationwide. The absolute nutrient content of purchased meals decreased in a consistent way; however, the nutritional composition of meals was similar to pre-labeling levels and exceeded dietary recommendations. Further research should examine this policy in chain full-service restaurants and other food establishments where labels are required on prepared foods (e.g., supermarkets). In fast food settings, additional interventions and policies are likely necessary to further reduce the calorie content and improve the nutritional quality of meals.

Supporting information