Associations between Electronic Cigarettes, Smokeless Tobacco, and Age-related Macular Degeneration in the 2017 United States National Health Interview Survey

I read with interest the findings from the recent study by Ochoa et al on associations of cigarette-alternative tobacco products with age-related macular degeneration (AMD). ¹ Based on data from the 2017 United States National Health Interview Survey (NHIS), the authors reported that adults who ever used smokeless tobacco had higher odds of developing AMD compared to never users (adjusted odds ratio [aOR] 1.49 [95% CI, 1.02 to 2.18]). They also raised concerns about the risk of AMD with e-cigarette use, though this association did not reach statistical significance (aOR 1.08 [95% CI, 0.73 to 1.60]).

While these results may initially appear concerning, the NHIS 2017 dataset and authors’ analyses contain serious limitations that may impact the validity of the findings. These limitations in the study by Ochoa et al are not uncommon in epidemiologic analyses of data from cross-sectional national surveys, and include the following:

1. Exposure is defined overly broadly. Ochoa et al define smokeless tobacco and e-cigarette exposure as ever use, ie, even one-time use. This level of exposure may have negligible direct health effects: the majority of ever use may be transient, ² and even for cigarette smoking – which is much higher on the continuum of risk from nicotine products than either smokeless tobacco or e-cigarette use ² – it may take heavy quantities and/or long durations of use to pose clinically relevant risks.^3,4

2. The number of cases is insufficient. A measure such as current use (which is also available in the NHIS 2017 dataset) may be a more reasonable exposure measure. However, critically, simple tabulations of the NHIS 2017 data show only 7 cases of AMD among participants reporting current smokeless tobacco use and 8 cases among those reporting current e-cigarette use, and nearly all those with AMD reported a history of cigarette smoking (see point 3 below). Ochoa et al do not directly report the number of cases, which is necessary to evaluate the reliability of results. Given the rule of thumb that at least 10 cases per covariate are needed to reliably estimate associations in a logistic regression model, ⁵ the available data may be inadequate to reliably estimate an association between smokeless tobacco or e-cigarette use and AMD.

3. Analyses did not rule out cumulative and/or lingering effects of prior smoking. Cigarette smoking is a well-documented risk factor for AMD ⁶ and all but 1 participant with AMD reported either current or former cigarette smoking. Ochoa et al’s analyses were adjusted for the number of cigarette packs smoked per day, but in addition to the small case number limiting the adequacy of adjustment (as noted above in point 2), this adjustment may be insufficient to rule out the contribution of cumulative cigarette smoking to AMD.

   At a minimum, it is necessary to adjust for pack-years, duration of smoking, and (for those who quit) time since quitting to factor out the cumulative and/or lingering effects of cigarette smoking on health outcomes. ⁷ Without such detailed adjustment for cumulative smoking exposure, any cases of AMD may be more likely explained by cumulative smoking exposure rather than ever use of smokeless tobacco or e-cigarettes.

   Moreover, it is not clear how Ochoa et al handled former cigarette smoking; presumably, the covariate of packs per day is only among participants who were current smokers, leaving it unclear whether people who formerly smoked were removed from the analysis altogether or misclassified as never smokers. Since AMD risk could be elevated from former smoking, either infraction invalidates the authors’ models.

   Alternatively, the confounding-by-smoking-history issue can be avoided entirely by examining smokeless tobacco or e-cigarette use among never smokers. However, this further limits the number of cases (see point 2 above), since most people who use smokeless tobacco or e-cigarettes either currently smoke or formerly smoked cigarettes.

4. Temporality of the association is unclear and possibly backwards. Being a cross-sectional survey, the NHIS does not contain information on which occurred first, the smokeless tobacco or e-cigarette trial, or AMD diagnosis. In fact, a prior methodology report showed that for many health outcomes, the diagnosis of AMD occurs decades before e-cigarette use initiation, especially for conditions with onset more common at older ages (eg, chronic obstructive pulmonary disease). ⁸ Indeed, AMD by definition is age-related, and Ochoa et al report the mean age of the participants with AMD as 70.9 years. However, e-cigarette use is rare among adults ages 65 years and older ⁹ – in other words, there is a large separation between the population with the exposure and the population with the outcome. Additionally, as e-cigarettes are a relatively newer product, it is even more likely that the AMD diagnosis occurred well before e-cigarette use; Ochoa et al note this limitation, implying that it underestimates the possible association between e-cigarette use and AMD. Smokeless tobacco use appears to be less age-dependent, though the relevant temporality remains unclear, especially considering that most smokeless tobacco users also smoked cigarettes.

5. Interpretation assumes a one-way causal relationship. The NHIS data are observational and cross-sectional, and thus cannot be used to infer causality. Not only do Ochoa et al omit this as a limitation, but they appear to draw causal conclusions, eg, concluding that these findings “identified smokeless tobacco use as a novel factor associated with AMD among a nationally representative sample, suggesting its avoidance may reduce the risk of developing AMD.”

As discussed above, the unclear and likely reverse temporality, small case number, and unaccounted-for risk from prior and/or cumulative smoking history also preclude causal conclusions. In addition, the possibility of reverse-causality is not considered, eg, a “sick-quitter effect” whereby health complications from cigarette smoking motivate someone to switch to a reduced-risk product such as smokeless tobacco or e-cigarettes. ⁸

In summary, the study by Ochoa et al, with its above-described limitations, does not necessarily provide reliable evidence that ever use of smokeless tobacco and/or e-cigarettes poses a risk of AMD. The exposure measure of ever use has negligible risks in and of itself. Since smokeless tobacco and e-cigarette use overlap heavily with cigarette smoking, any apparent association is at least partly attributable to cumulative smoking history. Case number is insufficient for the statistical adjustment necessary to examine whether there is a unique effect of smokeless tobacco or e-cigarette use. The association between smokeless tobacco use and AMD is interpreted as causal; however, the data likely contain reverse-directionality and possibly reverse-causality. Many of these flaws could be attributed to limitations in the NHIS data. However, these flaws may be enough to render analyses of this type inconclusive: the “signal,” if present at all, is too weak in the face of the “noise” of confounding and reverse-directionality.

Arielle Selya, PhD
Senior Scientist, Pinney Associates, Bethesda, MD, USA