Association Between Smoking Behavior and Cognitive Functioning in Patients With Psychosis, Siblings, and Healthy Control Subjects: Results From a Prospective 6-Year Follow-Up Study

This study was performed within the naturalistic multicenter Genetic Risk and Outcome of Psychosis (GROUP) cohort study. The full sample consisted of 1,119 patients with a diagnosis within the nonaffective psychotic spectrum, 1,059 unaffected siblings, and 586 unrelated healthy control subjects. The study procedures and inclusion and exclusion criteria for participants have been described in detail elsewhere (17). Patients diagnosed as having a nonaffective psychotic disorder according to DSM-IV criteria were recruited by clinicians from four university study sites and their surrounding mental health care facilities in the Netherlands and Belgium (18). Trained investigators conducted clinical interviews with patients and applied rating instruments. All patients, unaffected siblings, and healthy control subjects were invited to take part in a baseline assessment and in follow-up assessments 3 and 6 years after enrollment. We analyzed all individuals for whom complete data were available for the baseline measurement of smoking in the past 12 months. The study was approved by the Medical Ethics Committee of the Academic Medical Center of Utrecht. Written informed consent was obtained from participants before they were enrolled in the study. Release 5.00 of the GROUP database was used for the analyses.

Student t tests, Pearson chi-square tests, and one-way analysis of variance were used to compare baseline differences in demographic characteristics and clinical outcomes between smokers and nonsmokers in patients, unaffected siblings, and healthy control subjects. To illustrate the baseline difference in cognitive performance, one cognitive score per domain was standardized to a z-score, using the values of the nonsmoking control subjects as reference. Next, we used R, version 3.3.2 (26), and the lme4 package (27) to perform linear mixed-effects analyses of the relationship between cognitive measures and smoking status over a period of 6 years. Visual inspection of residual plots revealed no obvious deviations from homoscedasticity or normality, with the exception of the accuracy score on the Continuous Performance Test. Linear mixed models were applied to the raw scores and raw change scores for each cognitive domain. In all analyses, p values were calculated by the Kenward-Roger approach, using the pbkrtest package in R (28), which has been shown to produce the most acceptable type I error rates in mixed-effects models (29). Given the seven outcomes in five independent cognitive tasks, we used Bonferroni correction to minimize the risk of type I errors, and thus the two-tailed significance threshold was set at 0.007 (0.05/7).

To answer the multi-cross-sectional research questions, we entered into the model, as fixed effects, smoking status during the past 12 months, time, and all confounders selected a priori. As random effects, we added intercepts for subjects and by-subject random slopes for the effect of time. Each variable was added in a forward approach, and the Akaike information criterion was used to compare model fit. A lower Akaike information criterion estimate for the model after adding the covariates indicates a better model fit. Post hoc linear mixed-model analyses were conducted if a significant difference was found between smokers and nonsmokers for a specific task. To specifically investigate the association of cognition with smoking severity, we then entered the number of cigarettes per day instead of smoking status and subsequently entered the same fixed and random effects as stated above.

To answer the research questions concerning the effect of change in smoking behavior, we determined change in smoking status at 3-year follow-up compared with baseline and at 6-year follow-up compared with 3-year follow-up. In a similar fashion, a change score was calculated for each period for number of cigarettes per day and raw scores for each cognitive task. Next, we used linear mixed models with the same a priori covariates to explore differences in change scores of cognitive performance between subjects who quit smoking, those who started smoking, and those whose smoking status did not change. The same set of fixed and random effects as used to answer the first research question were entered. If a significant result was found, we conducted post hoc analyses to examine the relationship between change in number of cigarettes per day and cognitive performance with the same set of fixed and random effects.

In total, we analyzed data from 1,094 patients with a nonaffective psychosis, 1,047 siblings, and 579 control subjects at baseline. The demographic and clinical characteristics of smokers and nonsmokers are presented in Table 1. A total of 717 (65.5%) patients were diagnosed as having schizophrenia (DSM-IV codes 295.1, 295.2, 295.3, 295.6, and 295.9). Baseline smoking rates were significantly higher in patients compared with siblings and compared with healthy control subjects (66.6%, 38.3%, and 25.2%, respectively; χ²=312.49, p<0.001). Patients who smoked also used significantly more cigarettes per day (mean=17.8, SD=8.6) than did smoking siblings (mean=13.1, SD=9.1) or control subjects (mean=12.0, SD=7.7) (F=60.28, p<0.001). Smoking patients, siblings, and control subjects showed significant differences in various demographic characteristics and in the presence and severity of psychopathology compared with nonsmokers (Table 1). Unadjusted baseline cognitive scores were lowest in the patient group and highest in the nonsmoking control group (Figure 1).

To explore the overall association between smoking status and cognitive outcomes over 6 years, we used data from 916 patients, 947 siblings, and 552 control subjects (Table 2; see also section 1 in the online supplement). Data on cognitive performance scores were missing for a maximum of 179 patients, 100 siblings, and 27 control subjects. In patients, mixed-effects analyses revealed a significant negative association between smoking and score on the digit-symbol coding task (estimate=−2.38, SE=0.84, p=0.005). Post hoc analysis (see section 2 in the online supplement) revealed a similar negative association between number of cigarettes per day and score on the digit-symbol coding task (estimate=−0.10, SE=0.03, p=0.002). In siblings, significant negative associations were found between smoking and score on the arithmetic task and the block design task (Table 2). Post hoc analyses also showed negative associations between number of cigarettes per day and score on the block design task (estimate=−0.13, SE=0.04, p=0.002). In control subjects, we found a significant negative association between smoking and score on the digit-symbol coding task (Table 2). Finally, analyses yielded a negative association between number of cigarettes per day and score on the digit-symbol coding task (estimate=−0.25, SE=0.07, p=0.001). We did not find significant associations between smoking status and performance on the Auditory Verbal Learning Test and the Continuous Performance Test in any of the three groups.

Overall, patients, siblings, and control subjects showed significantly better performance on most cognitive tasks during follow-up (see sections 3 and 4 in the online supplement). We found that smoking behavior remained unchanged in most individuals from one assessment to the next (patients, 89.6%; siblings, 86.1%; control subjects, 90.7%). Over time, however, more individuals quit smoking (patients, 6.7%; siblings, 7.8%; control subjects, 5.4%) than started smoking (patients, 3.7%; siblings, 6.1%; control subjects, 3.9%).

We found a significant positive association between smoking cessation and change scores on the digit-symbol coding task (estimate=4.90, SE=1.73, p=0.005), indicating that processing speed improved in patients who quit smoking compared with patients who did not change their smoking behavior (see section 3 in the online supplement). This association was not found in siblings and control subjects. No significant associations were found between smoking initiation and changes in cognitive functions in any of the groups (see section 3 in the online supplement). In the patient group, post hoc analyses revealed a significant negative association between change in number of cigarettes per day and change in score on the digit-symbol coding task (estimate=−0.18, SE=0.05, p=0.001).

Previous studies on smoking and cognition in patients with psychotic disorders have reported contradictory findings. Our results are in line with the largest cross-sectional study so far, which reported reduced functioning in smoking patients compared with nonsmoking patients with psychotic disorders, based on a cognitive composite score in a total of 400 patients (14). In two other cross-sectional studies (12, 30) that included patients with first-episode psychosis (N=304 and N=159), no differences were found between smokers and nonsmokers on specific cognitive tasks. One small study (13) found an advantage of smoking (N=32) compared with nonsmoking (N=15) for verbal memory outcomes in patients with schizophrenia. However, research in patients with psychotic disorders has varied in sample characteristics, with differences in illness duration, sample size, and adjustment for covariates. Our study also adds relevant findings from analyses of number of cigarettes smoked per day, showing a dose-response relationship between number of cigarettes and impairments in processing speed in patients and control subjects, and in reasoning and problem solving in siblings.

Because cognitive tasks that assess working memory and problem solving also impose a time limit, processing speed could represent a core cognitive symptom that is negatively associated with smoking. This hypothesis is consistent with the already mentioned negative association between smoking and a cognitive composite score in patients with psychotic disorders (14) and with previous studies showing that overall cognitive performance in patients with schizophrenia can be explained by one component or a single factor based on five domains (31, 32). Although we did not test this in the present study, the lower performance in processing speed in smoking patients could represent worse cognitive performance overall. Finally, in studies in the general population, chronic cigarette smoking has repeatedly been found to be associated with lower cognitive functioning in multiple domains beyond processing speed (16, 33). Large studies, including the present study, point to a negative association between smoking and cognitive functioning for both patients and individuals without psychosis.

Lower cognitive performance in smokers probably reflects a multifactorial etiology, with a large number of directly cytotoxic compounds from cigarettes (e.g., carbon monoxide, polycyclic aromatic hydrocarbons) causing adverse effects in the brain (16). Despite the large body of evidence regarding smoking and general health problems, studies concurrently assessing cognitive, neurobiological, and genetic factors in relation to chronic smoking are still scarce. A review of this issue in the general population (16) that summarized a limited number of studies stated that chronic smoking is related to globally decreased brain perfusion, an abnormal increase in global brain atrophy, and structural and biochemical abnormalities in the anterior frontal region, subcortical nuclei, and commissural white matter.

A study on the short-term effects of nonsmoking on cognitive performance in 26 patients with schizophrenia (34) found no significant effect on any cognitive task after 3 weeks of smoking abstinence. To our knowledge, the present study is the first to find a significant positive association between smoking cessation and processing speed in patients with psychotic disorders. We have not found any studies on the association between sustained tobacco abstinence and cognitive performance in patients with psychosis. A meta-analysis of abstinence from various kinds of substances, but not tobacco, found that abstinence generally results in partial cognitive recovery (35). There are currently no published studies in patients or the general population that evaluated whether this positive association can be explained by neurobiological processes, such as brain plasticity (e.g., capacity to change in response to smoking cessation) and perfusion rates. Nevertheless, the accumulating evidence concerning the neurotoxic effect of persistent smoking is substantial. In the general population, it has been found repeatedly that former smokers showed less cognitive decline than those who continued smoking in later life (11, 36). Moreover, more lifetime years of smoking has been found to be related to poorer performance on measures of cognitive efficiency, processing speed, and visuospatial skills (33). Research relating measures of cognitive function to neurobiological integrity is therefore needed to understand whether or to what extent the harm of smoking is reversible in patients with psychotic disorders.

The main strengths of this study are the large sample size, the presence of two comparison groups, and the prospective long-term nature of the study, with follow-up assessments after 3 and 6 years. The study also has some limitations. First, given the observational design, no definite causal conclusions can be drawn, since reverse causation cannot be ruled out (e.g., patients who show improvement in cognitive performance also stop smoking). Second, and related to the first limitation, this study focused on spontaneous changes in smoking status and their relation to cognitive functions. Lower cognitive functioning is a known vulnerability for relapse of substance abuse (37), and patients who quit smoking could therefore represent a subgroup that had a larger increase in cognitive performance than those who did not change their smoking behavior. Third, information about exposure to smoking was limited. Observation points within the GROUP study have a 3-year interval, and participants were interviewed regarding their smoking behavior only over the 12 months before each assessment. Smoking history (i.e., number of pack-years) was not assessed, and this barred correction for length of exposure. This is of interest because a greater number of lifetime years of smoking has been found to be related to poorer performance on measures of cognitive efficiency, processing speed, and visuospatial skills (36). Fourth, no data were available on blood levels of cotinine or carbon monoxide. However, in a systematic review and meta-analysis about the validity of self-reported smoking behavior (38), interviewer-administered questionnaires, such as the CIDI-SAM, which was used in this study, were shown to produce accurate data. Fifth, GROUP patients represent a relatively high-functioning cohort, probably with higher cognitive performance than average patients with psychosis, which limits the generalizability of our findings. Finally, a minor limitation was the use of accuracy scores from the Continuous Performance Test to assess sustained vigilance. This measure showed ceiling effects, not reacting to transformation. By also including the mean reaction time scores from the Continuous Performance Test, we aimed to bypass assessment of nondiscriminating values.

We found substantially higher smoking rates in patients with psychotic disorders compared with unaffected siblings and control subjects, and lower cognitive performance in several domains in smokers compared with nonsmokers, irrespective of their illness or vulnerability status. Two main hypotheses have been proposed to explain the high smoking rates in patients with schizophrenia: the self-medication hypothesis and the shared-vulnerability hypothesis (7). The self-medication hypothesis postulates that patients often smoke to alleviate clinical symptoms, such as cognitive impairments. This theory is criticized by researchers who believe that shared genetic and shared environmental factors together with neurological deficits make individuals with schizophrenia more vulnerable to tobacco use and nicotine dependence (39, 40). Our prospective results on the cognitive effects of changes in smoking status in patients with nonaffective psychosis do not support the self-medication hypothesis. This is illustrated by our finding that smoking in patients with psychotic disorders is associated with impairment in domains similar to those observed in smoking nonpatients (after correcting for crucial confounders such as variation in psychopathology) and the fact that patients who quit smoking showed a significant improvement in processing speed. Moreover, our finding that smoking is more prevalent in both patients and unaffected siblings of patients (i.e., siblings without illness-related confounders) compared with healthy control subjects points to the presence of shared genetic or environmental factors that increase the risk of both smoking and psychosis. Still, patients often report that they smoke because of a need for stimulation and activation (41). We argue that more psychoeducation about the potential long-term negative effects of smoking on cognitive performance is necessary. The results of our study further underline the importance of smoking cessation treatment to increase cognitive, mental, and somatic health in patients with psychosis.