Abstract

Smoking is the leading cause of preventable morbidity and mortality in the United States, accounting for more than 480 000 deaths each year, $133 billion in medical costs, and $156 billion in lost productivity.1,2 Although smoking rates have declined, nearly 1 in 4 US adults still use tobacco, and 16.8% of persons in the United States smoke cigarettes.3,4 Reducing tobacco use continues to be a public health priority.

Accumulating evidence suggests that restricting access to and limiting opportunities to smoke tobacco reduces tobacco use.5–7 Prohibiting smoking in public and private locations, increasing financial costs through taxation, and raising the tobacco purchasing age are all linked to decreases in smoking prevalence.5,7 Limiting the number of locations where tobacco can be purchased is also effective. When San Francisco, California, and Boston, Massachusetts, prohibited tobacco sales in pharmacies, cigarette purchasing declined.6

On September 3, 2014, after changing its name to CVS Health, CVS removed tobacco products from its more than 7800 retail stores in 47 states nationwide and began a high-profile media and advertising campaign both in and outside of its stores to encourage smoking cessation.8 Little is known about whether such a decision by a large retail pharmacy chain led to reductions in tobacco use, or if consumers simply switched their purchasing to alternative retailers. We sought to examine the effect of CVS’s decision on cigarette purchasing by evaluating 2 complementary data sources. With household-level data, we compared the effect of CVS Health’s decision to stop selling tobacco on cigarette purchasing by consumers who previously purchased cigarettes at CVS versus those who purchased at other retail outlets. We used state-level data to assess whether the CVS decision led to population-level reductions in cigarette purchasing, comparing states with substantial CVS retail market share to states with no CVS retail presence. We hypothesized that restricting access to tobacco would reduce cigarette purchasing at the household and population levels.

METHODS

Data were provided by IRi Worldwide, a firm that collects retail purchasing data by using various methods to understand and track consumers’ purchasing behaviors for specific retail goods.9 To assess changes in cigarette purchasing associated with tobacco removal, we used monthly data for unique households participating in IRi’s Household Panel Survey, a nationally representative, opt-in panel in which household members regularly scan and report all retail purchases and corresponding purchasing locations. All purchases are recorded, regardless of retailer. Data included unique households’ demographic data (female or male head of household’s age, race, education, marital status, occupation, and Hispanic ethnicity; family size; presence of children; income; rented or owned residence) and monthly cigarette pack purchasing volume, frequency, and location. IRi tracks households’ participation monthly and flags those households who do not report any purchases of any retail good within a given month. We used this participation flag to determine households’ study eligibility and, when appropriate, to censor those households that were no longer contributing information to the panel. All households were de-identified.

Because many readers may be unfamiliar with such companies and data collection, it may be useful to think of an analogous situation: the Nielsen Company’s monitoring of television viewership, in which households are asked to report on their television program viewing on a monthly basis to understand shows’ popularity, what the characteristics are of people who are watching, and other data.10

To compare state-level purchasing of tobacco products before and after CVS tobacco removal in states where CVS Pharmacies are located versus states with no CVS Pharmacies, IRi provided monthly point-of-sale cigarette purchasing data from all drug, food, big box, dollar, convenience, and gas station retailers in 26 states: Alabama, Arizona, Arkansas, Colorado, Florida, Georgia, Illinois, Indiana, Kentucky, Louisiana, Maryland, Michigan, Missouri, Nevada, New York, North Carolina, Ohio, Oklahoma, Oregon, Pennsylvania, South Carolina, Tennessee, Vermont, Virginia, Washington, and Wisconsin. Only 26 states’ data were available because of competitive restrictions. To protect retailers’ competitive information, IRi only releases data if there are more than 2 retailers reflected in the relevant household purchase behaviors (in this case, cigarettes) at all levels of geography (e.g., local market, county, state).

Statistical Analysis

Household-level studies.

In the cohort study, descriptive statistics depict household characteristics and cigarette purchasing frequency (number of cigarette purchase transactions during the baseline period) and volume (number of cigarette packs during the baseline period for all and ≥ 3 pack purchasers). We calculated unadjusted risks (number of households that stopped cigarette purchasing divided by total households) and rates (number of households that stopped cigarette purchasing divided by household-months). We calculated household-months as the sum of the months contributed by each household before having the outcome or being censored. We evaluated the unadjusted and adjusted hazards of stopping cigarette purchases by using Cox proportional hazards models and generalized estimating equations to account for repeated measurements across households.13,14 Adjusted models included all household demographic characteristics. We also adjusted for baseline cigarette purchasing frequency and purchasing volume, as we anticipated that these might differ substantially between CVS-exclusive and other purchasers.

In the cross-sectional study, we calculated the same descriptive statistics and unadjusted risks as described previously. In a multivariable logistic regression, we compared changes in stopping cigarette purchasing between the tobacco removal period and the comparison period occurring 1 year before for CVS-exclusive versus mixed purchasers, adjusting for all covariates described previously, including purchasing frequency and volume. Because of the rarity of the outcome, the resulting odds ratio approximated a risk ratio.

State-level studies.

For each interrupted time series, linear regression models allowed for an immediate-level change in September 2014 at tobacco removal and a postremoval slope change to assess longer-term trend effects. All models used generalized estimating equations to account for repeated measurements.13,14 We adjusted for known seasonality in cigarette purchasing (e.g., New Year’s resolutions to quit smoking) with quarterly indicators and for state-specific, time-varying tobacco control and health care program changes that might influence cigarette purchasing: cigarette taxes,15 Medicaid expansion (or not) that expanded access to smoking cessation resources,16 and the yearly ratio of a state’s tobacco control spending to the Centers for Disease Control and Prevention’s recommended spending for that state.17,18 Using only the statistically significant parameter result (slope change), we quantified how many fewer packs were purchased per smoker over the 8-month follow-up by summing the results across the 8 months:

We then extrapolated this result to calculate how many fewer packs were purchased in states with CVS market share of greater than or equal to 15% over the 8 months after removal by multiplying the pack purchasing changes per adult smoker in the 8 months by the total estimated adult smokers in each of the states with CVS market share greater than or equal to 15%, and then summed all these states’ results. We then calculated the percentage change in pack purchases in these states over the 8 months after removal by comparing how many fewer packs were purchased to the estimated total pack purchases in these states had tobacco removal from CVS not occurred.

RESULTS

Results from the household-level studies are described first, followed by those from the state-level studies.

DISCUSSION

To our knowledge, this is the first study to evaluate the impact of a large retail company’s decision to end cigarette sales on unique household and population-level cigarette purchasing. Our findings suggest that the decision to eliminate cigarette sales at 1 retail pharmacy chain had a meaningful effect on cigarette purchasing behavior. Although some consumers of tobacco products at CVS certainly altered their cigarette purchasing locations when cigarettes became unavailable at CVS, our findings demonstrate that other CVS tobacco consumers purchased less tobacco.

In survival analyses, households that had purchased cigarettes exclusively at CVS Pharmacy were 38% more likely to stop buying cigarettes after CVS stopped tobacco sales, and those CVS Pharmacy consumers who bought more cigarettes, “3-or-more-pack purchasers,” were more than twice as likely to stop buying cigarettes, likely reflecting the greater disruption in their tobacco use and purchasing behaviors when CVS removed tobacco. Separate difference-in-differences analyses corroborated these findings. Similarly, at the population level, in the 13 states with greater than or equal to 15% CVS Pharmacy retail market share, consumers purchased 95 million fewer packs of cigarettes over the 8 months subsequent to tobacco removal (equivalent to 5 fewer packs per smoker), representing a 1% reduction in sales in these states. In another population-level analysis, we observed a dose–response relationship between CVS Pharmacy retail market share and reductions in cigarette purchasing that confirmed the directionality and magnitude of the intervention versus control state findings.

We used numerous approaches and data sources, as each had limitations. We characterized households as having stopped purchasing cigarettes, but we were unable to confirm smokers’ actual quitting efforts or intentions.19 We were similarly unable to determine whether single or multiple smokers resided within a household, so our analyses may underestimate effects. Recognizing disparities in baseline characteristics between exposure groups, we explicitly adjusted for cigarette pack purchasing volume, purchasing frequency, and demographics in analyses. Although survey households may not have been generalizable to the broader population of smokers who did not opt in to a household survey or to residents of a particular state, the CVS decision was associated with a reduction in cigarette purchasing among CVS-exclusive purchasers, and this reduction suggests that the decision was effective in tobacco harm reduction.

Several other issues deserve mention. The retail point-of-sale data were more generalizable, but did not allow for the identification of unique individuals or households and their longitudinal cigarette purchasing patterns. However, we estimated the number of adult smokers in a state to determine the denominator, and looked at a comprehensive sample of retailers selling tobacco to assess overall cigarette sales. Moreover, the September 4 tobacco removal date occurred a full month before the timeline communicated to the public, reducing the likelihood that people switched their purchasing behaviors right before the removal date (e.g., stockpiling cigarette purchases from CVS or leaving abruptly, which would have biased the analyses).

We were unaware of any specific tobacco-cessation efforts that occurred at the same time, but we cannot exclude the possibility. In a related vein, 2015 Behavioral Risk Factor Surveillance System data were not available, so we used 2014 data for both 2014 and 2015 smoking prevalence. To address threats to the validity, analyses included both historical and concurrent controls, to take into account the underlying decline in smoking prevalence in the United States,3 the rising popularity of electronic nicotine-delivery devices, and other trends.

Finally, we were only able to include 26 states in our analyses. Because excluded states might differ from included states in important factors (e.g., greater cigarette purchasing because of higher nicotine consumption), we cannot ensure the generalizability of the study to all 50 states. However, our analyses relied on IRi data, widely regarded as among the best available, highly accurate data to understand consumer purchasing.20 Taken together, we are reassured that multiple study designs, data sources, and analytic approaches produced consistent results, both in direction and magnitude.

Overall, our results are consistent with those from the broader tobacco literature. Multiple studies examining the impact of restricting tobacco access have observed reductions in tobacco use similar or greater in magnitude to those we observed. In 2011, New York City implemented a sales ban on flavored noncigarette tobacco products (e.g., cigars and smokeless tobacco). Compared with sales in 10 proximal counties, the ban was associated with a 10.2% greater absolute decrease (P < .01) in flavored noncigarette tobacco sales.21 Raising the tobacco purchasing age from 18 to 21 years in 1 suburban Massachusetts community was associated with a 6 percentage point decline in smoking prevalence (from 13% in 2006 to 7% in 2010) among high-school youths, compared with a 3 percentage point decline in 16 surrounding communities.8

In this study, we found that a large, nationwide retailer’s decision to remove tobacco from its stores meaningfully reduced household-level and population-level cigarette purchasing and, thus, presumably, consumption. Our findings have implications for public health professionals seeking to further reduce tobacco use. Currently, many large retailers in the United States sell cigarettes, including multiple pharmacy chains.22,23 Our research indicates that if these retailers, particularly pharmacies, stop selling cigarettes, there would be an overall reduction in tobacco consumption, and, as a result, a substantial health benefit. The findings from this study highlight another strategy that can reduce tobacco use and improve the nation’s health.

ACKNOWLEDGMENTS

HUMAN PARTICIPANT PROTECTION

Footnotes

REFERENCES