To our knowledge this is the first study that identified individuals who switched from smoking cigarettes to using ST (moist smokeless tobacco, dip, chewing tobacco, or snus) or quit tobacco from US healthcare administrative claims data to assess the potential impact of smoking behavior change on short-term healthcare costs. Findings from this study demonstrate that, while significant challenges exist as discussed below in detail, it is feasible to use individual-level administrative claims data to assess the real-world impact of transitioning from cigarette smoking to lower risk NCP.
Among male AWS with COPD, greater reductions in healthcare costs were seen in the 12 months after quitting all tobacco products compared to patients continuing cigarette smoking after assuming a modest burden of two comorbidities and exacerbations. These results are consistent with the literature that smoking cessation reduces healthcare costs among patients with COPD (27, 28). A similar trend of greater healthcare cost reduction was observed among AWS who quit smoking and switched to ST in the SW cohort, but the difference was marginally statistically significant (p=0.08), possibly due to the small number of patients in the ST cohort and reduction in statistical power after the introduction of covariates.
A strength of this study is the use of the DiD model where the individual serves as their own control, which accounts for time-invariant characteristics, both observed and unobserved. In addition, we adjusted for arguably two of the most relevant variables, namely, comorbidity and COPD exacerbation, to account for potential differences in health conditions across the cohorts. It is not surprising that the total number of comorbidities and COPD exacerbations were associated with an increase in direct healthcare costs. Further, after accounting for the total number of comorbidities and COPD exacerbations, the estimated total costs for COPD patients were attenuated rather dramatically. However, the estimated reduction in direct healthcare costs became more robust for COPD patients who changed their combustible smoking behavior compared to those who continued to smoke. This result cannot rule out regression to the mean for sicker COPD patients who changed their smoking behavior compared to healthier COPD patients who continued smoking as an explanation. Similar to quitting smoking, switching from cigarette smoking to ST can relieve COPD symptoms, which leads to reduction in healthcare costs among COPD patients, as ST use is associated with lower risks compared to cigarette smoking (29, 30). Despite the plausibility underlying this observation, these results should not be interpreted as definitive evidence for causal relationships because of the observational nature of the study, the relatively small sample size of the SW cohort, and the limitations discussed below. Studies with more robust sample sizes and appropriate longitudinal study design will be needed to further explore the impact of switching to ST on direct healthcare costs.
While a study assessing the impact of switching to a range of NCP would have more meaningful public health implications, this study focused on ST only because healthcare claim codes did not exist for modern NCP like ENDS and nicotine pouches. Some limitations of this study are inherent in any retrospective analysis using healthcare claims data. First, this study was limited to only those individuals with commercial health coverage or private Medicare coverage. Consequently, results of this analysis may not be generalizable to AWS with other insurance or without health insurance coverage. Second, the potential for misclassification of smoking and ST use status, covariates, or study outcomes was present as patients were identified through administrative claims data as opposed to medical records. As with any claims databases, the MarketScan® Databases rely on administrative claims data for clinical detail. These data are subject to data coding limitations and data entry error. Observable smoking behavior change is limited to the duration of a patient’s follow-up period and is reliant upon diligent coding by the healthcare provider. In addition, due to incomplete capture of tobacco use information, it is safe to assume that some patients without any evidence of tobacco use in their claims data actually used tobacco products, which likely attenuated the effect size observed. Third, not accounting for smoking duration and intensity, which can significantly impact both the incidence and exacerbation of COPD, is another limitation of this study because such information is not well documented in claims data. Fourth, important covariates that are strongly associated with tobacco product use behaviors and smoking-related diseases (e.g., lifestyle factors, BMI, blood pressure, high cholesterol, etc.) are not captured or are limited in the claims database and not accounted for in the study. Systematic differences between AWS who continue to smoke and those who quit or transitioned to ST may contribute to the differences in healthcare costs among the three cohorts (see additional discussions below). Fifth, despite the large size of the Marketscan® Databases, the SW cohort is relatively small, which limited the statistical power of the analysis and the interpretation of the study findings. Finally, we only had two time points for each individual. Therefore, we could not assess whether the parallel trends assumption of DiD models was met or not. We accounted for potential factors that could lead to the violation of the parallel trend assumption in the Adjusted Model and conducted an additional sensitivity analysis given that COPD exacerbations were associated with the smoking behaviors of interest (i.e. quitting and switching). Future studies with longitudinal data with at least two time points before the index date will be able to better assess the assumption. Nonetheless, we provided a “proof of concept” in this study and laid a foundation for future studies.
The paucity of NCP use history documentation in healthcare claims records is a major challenge for RWE studies assessing the impact of transitioning away from cigarette smoking. Among tobacco use behaviors documented in claims records, the overwhelming majority are related to cigarette smoking, primarily through ICD diagnosis codes, which have been reported to have perfect specificity in identifying AWS (31). However, since ICD codes are used for the diagnoses of diseases rather than document tobacco use behavior per se, they have relatively low sensitivity (0.32) in identifying AWS (31). Combing natural language processing (NLP) of unstructured clinical notes from electronic health records (EHR) with ICD codes has been reported to substantially increase the sensitivity (0.82) of identifying AWS using EHR (31). Therefore, adding EHR as an additional data source and using ICD codes in combination with NLP extraction of smoking status from unstructured clinical notes will substantially enhance the identification of adults who currently smoke cigarettes as well as adults who previously smoked and quit smoking than using claims data alone. NLP extraction of information on smoking duration and intensity from clinical notes in EHR may further enhance the design of future studies. Not having standardized codes for NCP other than ST precluded the investigation of modern NCP. Nonetheless, considerable increases in the documentation of ENDS use in unstructured clinical notes over the last 10 years has been reported (32-35), with some health systems adding specific fields for capturing ENDS use in their EHR (36, 37). While existing evidence suggests ENDS use is still substantially under-documented in EHR (33, 35, 38, 39), NLP extraction of modern NCP use data from clinical notes in EHR will facilitate assessment on the impact of switching to modern NCP. Creating standardized ICD codes for modern NCP like ENDS and nicotine pouches, broader implementation of SNOMED CT (which already include ENDS-related codes) in EHR systems, in combination with more diligent documentation of tobacco product use in medical records will greatly facilitate the assessment of the public health impact associated with modern NCP.
Dual- and poly-product use is common among AWS on their transitioning journey to NCP. For example, two studies among patients with ENDS use documentation in different EHR databases both reported that over half of the patients smoked cigarettes concurrently (34, 35). A hybrid study design linking detailed tobacco use history information including duration, intensity, and dual/poly product use collected directly from consumers through questionnaires to existing records in EHR and/or healthcare claims represents a good option to further enhance the design of future RWE studies on NCP.
Another major challenge for RWE studies on transition from cigarette smoking the NCP is mitigating confounding by baseline health-related attributes that are differentially associated with various tobacco use behavior changes. Studies on patients with COPD have counterintuitively reported better outcomes for current smokers including lower odds of COPD exacerbation than former smokers (40, 41). Higher healthcare utilization and cost around the time of smoking cessation has also been reported in the literature (25, 26). Previous research has shown that recent quitters are more likely to utilize more healthcare services within a year than current smokers (23). These results have been interpreted as evidence that recent diagnoses of major diseases often leads AWS to quit smoking, which would likely attenuate the beneficial effects of quitting smoking among recent quitters when assessed cross-sectionally (24) and confound studies assessing the impact of smoking behavior change for patients with symptomatic COPD (42). At baseline, the QT and SW cohorts in our study also had higher healthcare utilization rates, higher overall healthcare costs, and higher DCI scores than the CS cohort (Table 2), indicating higher comorbidities. In addition, higher proportions of patients in the QT and SW cohorts had at least one COPD exacerbation in the quarter immediately before the index dates than the CS cohort (Figure 5). Higher numbers of baseline comorbidities and exacerbations have been found to be associated with increases in baseline direct healthcare costs (21), which is consistent with data for the QT and SW cohorts in our study. Combined, these observations strongly suggest that some patients in the QT and SW cohorts in this study changed their smoking behavior after COPD exacerbation or new disease diagnosis (22), which would bias the results of the analysis by violating the parallel trends assumption for DiD models. Therefore, it is not surprising that analyses accounting for comorbidities and COPD exacerbations resulted in more robust estimated reductions in direct healthcare costs in the QT and SW cohorts compared to the CS cohort. In this study, while we showed that DiD models may be useful to control for time-invariant variables, careful consideration should be given to potential time-varying confounders in future studies.
In line with the recent FDA guidance on RWE (43), we identified six areas to enhance data relevance and reliability as well as methodologies for collecting and analyzing RWD that would allow for more robust assessments of the real-world clinical and healthcare cost impact of modern NCP that have been determined to be “appropriate for the protection of the public health” (44) and authorized to be marketed in the United States by the FDA: (1) more diligent screening and documentation of comprehensive tobacco use history information including duration, intensity, and dual/multiple product use by healthcare providers in medical records; (2) creation and adoption of new standardized codes for documenting use of modern NCP like ENDS, nicotine pouches and heated tobacco products; (3) leveraging artificial intelligence and/or natural language processing to maximize the utilization of unstructured data in EHR; (4) combining data from both EHR and claims databases to enable adjustment for important demographic attributes associated with cigarette smoking; (5) using a hybrid study design by collecting comprehensive tobacco use history directly from consumers to overcome the inherent limitations of medical records due to practitioners’ resource constraints, and subsequently linking that information to medical records; and (6) leveraging longitudinal data to better control for the “sick quitter” phenomenon which would likely attenuate the observed beneficial effects among AWS who quit smoking or switched to tobacco products at the lower end of the risk continuum.