Smoking is associated with impaired verbal learning and memory performance in women more than men

Abstract

Vascular contributions to cognitive impairment and dementia (VCID) include structural and functional blood vessel injuries linked to poor neurocognitive outcomes. Smoking might indirectly increase the likelihood of cognitive impairment by exacerbating vascular disease risks. Sex disparities in VCID have been reported, however, few studies have assessed the sex-specific relationships between smoking and memory performance and with contradictory results. We investigated the associations between sex, smoking, and cardiovascular disease with verbal learning and memory function. Using MindCrowd, an observational web-based cohort of ~ 70,000 people aged 18-85, we investigated whether sex modifies the relationship between smoking and cardiovascular disease with verbal memory performance. We found significant interactions in that smoking is associated with verbal learning performance more in women and cardiovascular disease more in men across a wide age range. These results suggest that smoking and cardiovascular disease may impact verbal learning and memory throughout adulthood differently for men and women.

Figure 1

Participants who self-report current smoker status perform worse on paired-associates learning (PAL) compared to non-smokers across 18–85 year olds. Linear regression fit (line fill ± 95% confidence interval (CI)) of the PAL total number of correct from 18 to 85 years old. Lines were split by Smoker versus Non-smoker. (F(30, 81,670) = 560. 58, p = 0e+00, N = 81, 700).

Figure 2

The main effect of self-reported smoking status on PAL performance across 18–85 year olds separately between men and women. Simple effects analyses revealed that women’s memory is negatively impacted by smoking (β =  − 1.01 word pairs, std error = 0.14, p = 7.83e−13; (B)) whereas men’s is not significantly impacted (β =  − 0.27 word pairs, std error = 0.18, p = 0.135: (A)). Shown are the linear regression fit lines (± 95% confidence interval (CI)) of the PAL total number of correct from 18 to 85 years old.

Figure 3

The main effect of a self-reported cardiovascular disease (CVD) composite score on PAL performance across 18–85 year olds. CVD composite was a sum score of self-reported heart disease, hypertension, diabetes, and stroke. Individuals were treated as groups based on their composite score (0 as the control group compared to 1, 2, and 3+). Linear regression fit (line fill ± 95% confidence interval (CI)) of the PAL total number of correct from 18 to 85 years old (F(31, 81,669) = 542.88, p = 0e+00).

Figure 4

The main effect of a self-reported CVD composite score on PAL performance across 18–85 year olds separately between men and women. Simple effects analyses revealed the negative impact of cardiovascular disease on memory had slightly larger effect sizes in men compared to women in the 3 group (men: β =  − 1.80 word pairs, std error = 0.41, p = 1.3e−5; women: β =  − 1.27 word pairs, std error = 0.35, p = 0.0003).

Figure 5

The propensity score matching main effect of self-reported smoking on PAL performance across all ages separately between men and women. We conducted propensity score matching (PSM) analysis matching smokers and non-smokers for various health and lifestyle factors from self-report including: age, race, ethnicity, marital status, handedness, education level, number of daily medications, history of diabetes, seizures, cancer, stroke, hypertension, heart disease, family history of Alzheimer disease, drug abuse, loss of consciousness, and dizziness. The dependent variable was the total number of correct word pairs entered across the three trials of PAL tests (range of 0–36). When collapsed across age, PSM suggested there is no effect of smoking on memory in males (β =  − 0.21 (− 0.62 to 0.19, 95% credible interval)) and a negative effect in females (β =  − 0.54 ( − 0.93 to −0.14, 95% credible interval)).

Figure 6

(A) We conducted 1000 down-sample linear regression models of the MindCrowd cohort between the ages of 18 and 85 years for the interaction effect (β, y axis) of sex × smoking on paired associate learning (PAL) for each indicated total sample size (x axis). For each analysis, we had an equal amount of smokers and non-smokers and women and men. The horizontal red line indicates the effect size estimated by the total study sample. Green filled circles indicate an individual down-sampled comparison that resulted in a statistically significant association (p < 0.05), black dots are non-significant comparisons. Red arrow highlights that at samples sizes approximating 5000 one could potentially produce a significant beta value with the opposite sign from the largest sampled model. (B) is the same data displayed to easily see the positive relationship between significant betas and sample size.

Figure 7

Artificial error introduction suggests the present smoking results are likely not due to self-report error. A Monte Carlo simulation was used to determine the effect of introducing artificial error (in addition to any real self-report error already in the data). Error was introduced in 1% increments (x-axis) by randomizing smoker status. Each box represents 10,000 model simulations, and plotted are the p-values for each simulation. The red dashed line represents a significance level of α = 0.05. At 0% simulated error, we report our measured p-value. In the whole cohort and women specifically (A,B), we measure a significant effect in > 90% of simulations after 10% additional self-report error is introduced. For the sex × smoking interaction term, 75% of simulations are statistically significant after 10% additional self-report error is introduced.

Figure 8

We examined the smoking effect on PAL through the use of permutation testing. This was performed one million times per model. The smoking data label for every participant was randomly assigned and the t-statistic for the main effect of smoking in the whole cohort (A), women only (B), and men only was re-calculated (C). We also conducted permutation tests on the interaction between sex × smoking on PAL (D). Black dashed line indicates the full model results statistic. Results from permutation testing indicate the present results are likely not due to chance.

Figure 9

A visual representation of US smoking rates by age derived from data from the National Health Interview Study (NHIS) for years 2015–2018 and MindCrowd smoking rates per age for men and women. The NHIS is is one of the major household survey-based data collection programs of the National Center for Health Statistics (NCHS) which is part of the Centers for Disease Control and Prevention (CDC). These data demonstrate that men typically have higher smoking rates compared to women of the same age and that MindCrowd, while demonstrating lower rates of smoking overall, follows the same trend as observed by the NHIS.