Relating Onset of Health Conditions to Changes in Tobacco/Nicotine Use — Analyses based on the U.S. PATH Study *

In the past smokers who were discovered to have smoking-related diseases have been encouraged to quit smoking, with such advice not always being followed. There is now the possibility of switching to new products such as e-cigarettes. If these products prove to be safer alternatives this could be a useful option for improving health outcomes in diseased populations.

There is an extensive literature on the extent to which use of, and changes in the use of tobacco products affects the subsequent risk of cancer, chronic obstructive pulmonary disease (COPD), cardiovascular disease (CVD) and other diseases (1). However, there is far less literature available on how onset of disease might affect subsequent use of tobacco/nicotine products. Smokers with smoking-related disease may often be advised to reduce their consumption or quit, but the extent to which smokers may actually do so is not well established. Many of these studies (e.g., GOLČIĆ et al. (2) and GUMMERSON et al. (3) for cancer, BRUMMETT et al. (4) and KIM et al. (5) for heart disease, and HIRAI et al. (6) and HIRAYAMA et al. (7) for COPD or respiratory disease) determine changes in smoking habits only in a group with a defined disease or group of diseases, and do not collect comparable data in a healthy control group. Even, where they do (e.g., PALMER et al. (8) and WESTMAAS et al. (9) for cancer, KWON et al. (10) for diabetes and hypertension, and LIU et al. (11) and JONES (12) for COPD or respiratory disease) they tend to limit attention to quitting, not considering the possibility that those with a diagnosis with a poor outcome might increase their smoking, or re-initiate it (possibly because they wish to enjoy their final days while they still can), or switch to other tobacco/nicotine products, such as e-cigarettes. Furthermore, most studies are relatively short-term, not allowing useful information to be obtained on survival or other health changes, few, (e.g., (13–15)) reporting on any health outcome.

Here we report detailed analyses investigating the effects of onset of a range of diseases on aspects of established cigarette smoking and e-cigarette use. These are based on adult data from those who were adults (aged 18+ years) in Waves 1 to 4 of the Population Assessment of Tobacco and Health (PATH) study (16, 17), a nationally representative longitudinal cohort study in the U.S. supported by federal funds. The publicly available databases (https://doi.org/ 10.3886/ICPSR36231.v18) provide extensive information on tobacco/nicotine product use, health status and other factors at each follow-up Wave. Wave 1, conducted in September 2013 to December 2014, involved 45,971 individuals, 13,651 adolescents aged 12–17 years and 32,320 adults aged 18+ years. Wave 2 was conducted in October 2014 to October 2015, and Wave 3 in October 2015 to October 2016. By Wave 4, conducted in December 2016 to January 2018, there were data on 33,821 adults. 6,065 were from a new cohort, and of the remaining 27,756, 5,131 were adolesecents at Wave 1 and were thus excluded, leaving 22,625. Of these 20,594 had cigarette smoking at both waves, 19,576 had e-cigarette data at both waves, and 18,018 had information on both types of tobacco/nicotine product use at both waves.

The main analytical method involved studying changes in established tobacco/nicotine use between Waves 1 and 4 according to whether or not onset of the condition occurred in Waves 2 or 3. The individuals included in each analysis were adults (aged 18+ years) at Wave 1, who at Wave 1 had never been told by a doctor or health professional that they had the health condition or certain related conditions, and had established smoking data available at Waves 1 and 4. For each combination of health condition and change in tobacco/nicotine use considered, the main analyses took account of the population weighting at Wave 1 and adjusted for sex (male, female) and detailed age group (18–24, 25–34, 35–44, 45–54, 55–64, 65–74, 75+ years) as at Wave 1, with estimates of the relationship between onset of the condition and change in established tobacco/nicotine use (expressed as an odds ratio (OR) with 95% confidence interval, with a positive value indicating that the change was more likely in those with onset of the condition) calculated by weighted logistic regression for the whole population and also separately by broad age group (18–34, 35–54, 55+ years). Additional analyses tested, by including a suitable interaction term in the model, whether the association varied by sex, or by broad age group. Also Sensitivity analysis 1 excluded those who had ever used smoked tobacco/nicotine products other than cigarettes or e-cigarettes (traditional cigars, filtered cigars, pipes, hookahs, smokeless tobacco, snus and dissolvable products), while Sensitivity analysis 2 adjusted not only for sex, and detailed age group, but also for total household income (split into five levels, with those with missing values being considered as an extra level), census region (Northeast, Midwest, South, West), and use of any of the tobacco/nicotine products excluded in Sensitivity analysis 1.

Initial analysis considered 17 different changes in to-bacco/nicotine habits between Waves 1 and 4. Changes 1–3 related to initiation of smoking (never smokers at Wave 1, but current or former smokers at Wave 4), quitting smoking (current smoking at Wave 1, but not currently smoking at Wave 4), and re-initiation of smoking (former smokers at Wave 1, but current smokers at Wave 4), with changes 4–6 relating to initiation, quitting and re-initiation of e-cigarettes smoking similarly defined. Changes 7–9 relate to initiation, quitting and re-initiation of smoking in those who had never used cigarettes, while changes 10–12 relate to initiation, quitting and re-initiation of e-cigarettes in those who had never smoked. The final five changes relate to current dual use of cigarettes and e-cigarettes - change 13 to current cigarette-only smokers becoming dual users, change 14 to current e-cigarette-only users becoming dual users, and changes 15–17 relating to dual users quitting, becoming current users of cigarettes only, or becoming current users of e-cigarettes only. Preliminary analysis showed that eight of these changes (5, 6, 11, 12, 14–17) occurred relatively infrequently (see Supplementary File 1 which gives the number of changes for each health condition and definition of tobacco/nicotine use change) and attention in this paper is restricted to the following nine changes:

Initial analysis considered the onset of 18 health conditions, but it was decided to limit attention to nine of them for the reasons explained in Table 1.

All analyses were limited to those without the condition of interest at Wave 1, with analyses of some conditions also excluding those with other conditions, as also shown in the Table 1. Some additional analyses were also carried out, as described below.

The first two additional analyses considered the effect of alternative exclusions. Additional analysis A excluded those who at Wave 1 had any of the individual conditions studied, while Additional analysis B excluded those who at Wave 1 had any major smoking-related condition, and also, where appropriate, the particular condition considered.

The next set of additional analyses attempted to account for reverse causation, avoiding the possibility that, in the main analyses, some of the changes in tobacco/nicotine use between Waves 1 and 4 might have occurred before onset of the health condition studied. Additional analysis C was restricted to those who had unchanged tobacco/nicotine habits at Waves 1 and 2, and related changes in tobacco/nicotine use between Waves 1/2 and Wave 4 according to whether onset of the condition occurred in Wave 2. Additional analysis D was restricted to those who had unchanged tobacco/nicotine habits at Waves 1, 2 and 3, and related changes in tobacco/nicotine use between Waves 1/2/3 and Wave 4 according to whether onset of the condition occurred in Wave 3.

The final set of additional analyses studied changes in the number of cigarettes smoked. Three versions of it were conducted. Additional analysis E restricted attention to those smoking cigarettes at both Waves 1 and 4. Additional analysis F restricted attention to those smoking cigarettes at Wave 1 but counted quitting at Wave 4 as a reduction in amount smoked. Additional analysis G also included nonsmokers (never and former smokers) at Wave 1, counting both quitting and relapse as changes in amount smoked. In these analyses, correction was made for clearly erroneous values for the daily amount smoked for some individuals in the data reported in the PATH study using the methodology given previously in the publication which first described the problem (18). Also, for those reporting someday rather than everyday smoking, the amount smoked was divided by 3 if smokers last smoked three or more days ago, by 2 if they last smoked yesterday or the day before yesterday, and was left unchanged if they smoked in the past hour or sometime today.

In order to conduct the analyses described above, the relevant data were transferred to a ROELEE database, and analysed using the ROELEE program (Release 59, Build 49). All these analyses could also be run using the GLM Package and the Step Function from the R Program (https://www.r-project.org/).

We have carried out extensive analyses of the relationship of quitting smoking to onset of various health conditions. Our main analyses adjusted for sex and broad age groups, while other analyses controlled for additional variables, excluded users of other tobacco/nicotine products, restricted attention to those who had never used e-cigarettes, or attempted to adjust for reverse causation. None of these analyses showed any strong tendency for onset of a health condition to predict quitting. Even in the analyses excluding other tobacco/nicotine product users, ORs never exceeded 1.86 and significant increases were only seen for four of the nine health conditions with sufficient data for analysis - any major respiratory condition (OR 1.86, 95% CI 1.22–2.83), high blood pressure (1.84, 1.20–2.83), high cholesterol (1.75, 1.17–2.62) and any major smoking-related condition (1.73, 1.18–2.56). Given the lack of significantly reduced ORs for any condition in any analysis, the results seem, however, to be consistent with a tendency for onset of health conditions to lead to some increase in quitting.

Other nationally representative U.S. studies in which quit rates have been compared in those with and without specific health conditions have produced rather variable results. Thus a large study based on the 2018 Behavioral Risk Factor Surveillance System (11) found that adults with COPD had similar adjusted quit rates to those without COPD, despite making more quit attempts, while a smaller study based on the 1998 to 2013 Surveillance, Epidemiology and End Results study (19) found no difference in quit rates between bladder cancer cases and controls. Another study, based on the 2009 Behavioral Risk Factor Surveillance System (20), reported that the quitting rate for those without cancer (57%) was intermediate between the rate for those with tobacco-related cancer (55%) and for those with other cancers (62%). Clearer evidence that onset of a health condition is associated with increased quitting came from three studies. Based on the Health Professionals follow-up study (21) men diagnosed with cancer were more likely than matched men without cancer to quit smoking within two years (OR 2.5, 95% CI 2.0–3.0) or within four years (1.6, 1.3–2.0), though ORs varied considerably by cancer type. Based on the Cancer Prevention Study II Nutrition Cohort, WESTMAAS et al. (9) found that two-year quit rates were higher among smokers diagnosed with cancer than among smokers not diagnosed with cancer (31.3% vs 19.5%) with a similar difference observed after four years (43.0% vs 33.8%), the cancers considered excluding those of the lung, head and neck, oesophagus and any metastatic cancer. In an analysis based on waves of the National Health and Aging Trends study in 2011 and 2012 (22) the odds of smoking cessation were noted to increase with a new diagnosis of chronic illness since the 2011 wave.

Other changes in tobacco/nicotine use that we analyzed have rarely, if ever, been considered in other publications. However, it is interesting to note that though there is little evidence that onset of a health condition affects the probability that former cigarette smokers would re-initiate current cigarette smoking (a single marginally significant positive relationship for diabetes, one of the nine health conditions studied, being unconvincing and requiring independent replication), there is more evidence that onset of a health condition affects the probability that never cigarette smokers would initiate cigarette smoking. Here all the ORs in the main analysis exceeded 1.0 and three (any cardiovascular condition, high blood pressure and any major smoking-related condition) were significant (at p < 0.05). However, none of the ORs were very large, the highest being 1.65 (95% CI 1.31–2.08), and again some independent confirmation would be valuable. Despite onset of a health condition apparently encouraging some never or former smokers to become current smokers, the results in Table 5 show some tendency for amount smoked by Wave 1 cigarette smokers to reduce between Waves 1 and 4 following onset of a health condition, though the reduction is smaller when the analyses consider the whole Wave 1 population.

Our clearest finding is that onset of a health condition consistently predicts initiation of e-cigarettes, with all the ORs in the main analysis significantly above 1, the largest values being 2.86 (95% CI 2.10–3.90) for any major respiratory condition, and 2.51 (1.52–4.14) for cancer. Though controlling for more variables, excluding users of tobacco/nicotine products other than cigarettes, and restricting attention to those who had never smoked cigarettes, made some ORs become non-significant they all remained positive. We are unaware that this finding has been reported previously. How much onset of a health condition predicts initiation of e-cigarettes may depend on their availability at the time, and public perception of their risks compared to those from cigarettes.

Our study has a number of strengths, including use of a nationally representative population which has widely been used in research, and the inclusion of detailed weighted analyses allowing for study of the effect of bias and confounding. It does, however, have some weaknesses. The sample size of 22,526 adults with smoking data at Waves 1 and 4 is not that large, and does not allow precise estimates of the OR relating some health conditions to some changes in tobacco/nicotine use. Also, disease diagnosis is not based on independent medical confirmation, severity of disease is not taken into account, and some of the disease categories combine different diseases, possibly with different effects on changes in tobacco/nicotine use. Furthermore, the analyses relating onset of health conditions to quitting do not determine whether the quitting is long-term. The fact that errors have been discovered in the reporting of daily amount smoked (see methods) might suggest that other undiscovered errors are possible. Some older participants who stated that they have never smoked cigarettes at Wave 1 might not have bothered to mention past smoking, so that reports of current or past smoking at Wave 4, possibly affected by recent onset of a smoking-related health condition, might not necessarily imply that they had actually taken up smoking after Wave 1. Bias from this source would have the most effect on our analyses on initiation of cigarette smoking. Furthermore, details are not available on the nature and extent of any advice given to modify smoking habits. This would vary by the disease diagnosed, by U.S. state and by hospital within state. Detailed study of subsequent health effects following changes in tobacco/nicotine use after onset of health conditions is beyond the scope of this paper, though we note that the effect of changes in smoking habits (and of the advice given to modify smoking habits) is likely to vary substantially by disease, changes being likely to have the least effect in diseases with a very poor prognosis when diagnosed.

Based on analyses relating changes in tobacco/nicotine use between Waves 1 and 4 of the PATH study to onset of various health conditions in Waves 2 or 3, the clearest association noted was that for all nine health conditions studied, rates of initiation of e-cigarettes were higher in those where onset occurred than in those where it did not. The increase was consistently seen in the main and various sensitivity analyses with ORs often statistically significant at (p < 0.05). Positive associations were also often seen with quitting smoking, though these were weaker than for initiating e-cigarettes and rarely significant. Positive associations were also evident with initiating cigarette smoking, but these are subject to bias.

This work was supported by Philip Morris Products S.A.

Both authors are long-term consultants to various tobacco companies.

We thank Yvonne Cooper for secretarial assistance and Dr. A. Van der Plas for helpful comments.