Within-person decline in pregnancy smoking is observable prior to pregnancy awareness: Evidence across two independent observational cohorts

Abstract

Decreased consumption of nicotine and other drugs during pregnancy appears to be a cross-species phenomenon from which mechanism(s) capable of interrupting addictive processes could be elucidated. Whether pregnancy influences smoking behaviour independent of women's knowledge of the pregnancy, however, has not been considered. Using repeated measures analysis of variance (ANOVA), we estimated within-person change in mean cigarettes/day smoked across the estimated date of conception but prior to individually reported dates of pregnancy recognition using longitudinal smoking data from two independent observational cohorts, the Growing Up Healthy (GUH, n = 271) and Midwest Infant Development Studies (MIDS, n = 145). Participants smoked an average of half a pack/day in the month immediately before conception (M (SD) = 12(8.1) and 9.5(6.7) cigarettes/day in GUH and MIDS, respectively). We observed within-person declines in smoking after conception, both before (M_GUH = −0.9; 95% CI −1.6, −0.2; p = 0.01; M_MIDS = −1.1; 95% CI −1.9, −0.3; p = 0.01) and after (M_GUH = −4.8; 95% CI −5.5, −4.1; p < 0.001; M_MIDS = −3.3; 95% CI −4.4, −2.5; p < 0.001) women were aware of having conceived, even when women who had quit and women who were planning to conceive were excluded from analyses. Pregnancy may interrupt smoking-related processes via mechanisms not previously considered. Plausible candidates and directions for future research are discussed.

1 ∣. INTRODUCTION

Progress on the neurobiological understanding of addiction has not translated to proportional gains in addiction treatment.¹ One barrier may be the nearly exclusive focus on the minority of users of a given substance who develop a substance use disorder at the expense of also understanding why the majority of users do not.² This is particularly true in research on substance use during pregnancy. The intense focus on prenatal drug exposure as a leading preventable risk factor for health problems in offspring, with prenatal tobacco being most studied,^3,4 has overshadowed a remarkable protective phenomenon—the apparent protective effect of pregnancy on an individual's smoking and substance use behaviour.^5-7 Here, dichotomous characterisation (yes/no) of smoking during pregnancy obscures the prevalence and significance of within-person decline in smoking frequency across pregnancy⁸ evident among smokers who quit and those who do not.^5,9-11 Thus, while pregnancy is commonly viewed as a window of opportunity to intervene to change behaviour, we propose that pregnancy is also an opportunity to understand how behaviour changes spontaneously, without intervention.¹²

Spontaneous reduction and cessation of smoking during pregnancy is widely assumed to reflect motivation to minimise foetal exposure.¹³ As such, pregnancy's protective effect on substance use is typically conceptualised as being psychological in nature (i.e. involving maternal cognitive and affective processes).^5,12,14-19 Indeed, we recently linked smokers' capacity to perceive others' somatic pain with likelihood of quitting smoking (as confirmed by smoking biomarkers) once a pregnancy is recognised.²⁰ Whether pregnancy influences smoking behaviour prior to, and/or independent of, a deliberate intention to protect foetal well-being, however, is less clear, although plausible. Changes in olfactory and gustatory sensations, nausea, and vomiting have been suspected to influence alcohol consumption during pregnancy,²¹ and there is increasing recognition of interactions between reproductive hormones and smoking.²² Less known is evidence that pregnancy's protective effect on substance use is not unique to humans.

Pregnant rats exhibited reduced nicotine self-administration relative to their nonpregnant female counterparts in one prior study.²³ This finding and gestational decreases in voluntary alcohol consumption in a number of rodent species (Figure 1, top) and one nonhuman primate species^24-26 support the potential benefit of a bidirectional translational approach to explicating putative biological mechanisms underlying pregnancy's protective effect on addictive processes.²⁷ The current study takes the first step using the natural experimental design shown at the bottom of Figure 1.

FIGURE 1.

TOP: Cross-species conservation of pregnancy's protective effect on substance use implicates mechanism(s) other than concerns about intrauterine exposure. BOTTOM: In humans, the temporal discrepancy between the onset of pregnancy (conception) and recognition of pregnancy enables inference of putative biological influences on smoking independent of psychological processes (deliberate harm reduction) occurring once the pregnancy is recognised.

In humans, any protective biological effect of pregnancy on smoking, if present, overlaps with psychological processes related to concern about foetal exposure to smoking. An exception is the interval of variable length between the estimated date of conception and the date when women first become aware of the pregnancy, (Figure 1, bottom) particularly when the pregnancy is unplanned. Thus, to probe the possibility that putative biological influences of early pregnancy attenuate smoking, we exploited this temporal discordance between the onset and recognition of pregnancy. Change in smoking during this interval before women are aware of the pregnancy would implicate processes other than women's intention to protect the foetus.

Using this design, we quantified mean within-person change in smoking level (cigarettes per day) beginning immediately before conception through the date the pregnancy was recognised, as reported by pregnant individuals during second trimester time-line follow back interviews, the gold standard method of retrospective assessment of patterns of substance use over time.²⁸ We hypothesised that conception would be associated with a significant within-person decrease in mean cigarettes per day prior to the date when the pregnancy was recognised. Demonstrating this would support the likelihood that pregnancy's protective effect on substance use is conserved across species and thus the validity of an animal model to examine specific biological mechanism(s).

2 ∣. METHODS

This study involves secondary analysis of existing de-identified data (Figure 2) originally developed for the Midwest Infant Development Study (MIDS) and the Growing Up Healthy Study (GUH). Including these two cohorts into our study served to minimise selection bias that may be unique to individual studies. Institutional Review Board approval of all procedures was obtained prior to the conduct of current and original studies.

FIGURE 2.

Derivation of analytic samples from original cohorts

MIDS is a prospective study of prenatal tobacco exposure and offspring developmental outcomes.²⁹ Pregnant women were recruited from multiple obstetric clinics located in two mid-sized Midwestern cities. Smokers were oversampled (75%), while women younger than 18 years old, delivering before 35 weeks, reporting illegal drug use (including marijuana) or binge drinking (> 3 drinks/occasion) were excluded.^28,30 Data from 463 MIDS mothers who were smoking at the time of conception were screened for inclusion in the integrated sample (Figure 2).

Testing our hypothesis that conception is associated with a significant decline in use prior to pregnancy recognition necessitated unique data points of smoking frequency at the following intervals: (a) before conception, (b) between conception and pregnancy recognition (which varies widely between participants) and (c) after the pregnancy was recognised (again, different for each participant). Because smoking was ascertained at monthly intervals in MIDS (every 28 days) rather than at daily intervals in as done in GUH, women who recognised the pregnancy less than 28 days after the last menstrual period (LMP) were excluded since a separate data point for smoking between conception and pregnancy recognition was unavailable (n = 81; 33%). Women missing data on average daily cigarettes (n = 14; 6%) were also excluded.

Thus, the final analytic sample consisted of 145 women who reported smoking at the time of conception. Maternal sociodemographic factors and pregnancy characteristics strongly associated with smoking during pregnancy were examined across cohorts. These included age at baseline, educational attainment, self-reported race/ethnicity, parity, marital/partner status, and partner smoking status.^15,31-35

GUH is a longitudinal study of N = 415 adult (≥18 years of age) female smokers and nonsmokers recruited at ≤ 20 weeks of gestation from an urban Northeastern hospital-based obstetric clinic (2006–2009) aimed at understanding psychobiological and family processes associated with elevated risk for conduct disorder in children of smokers.¹¹ Smokers were oversampled 2:1 with demographically matched nonsmokers, while women younger than 18 years old, more than 20 weeks pregnant at the time of enrollment, reporting any illegal drug use except marijuana, or endorsing binge drinking (> 3 drinks/occasion) were excluded. For the current study, data from 286 GUH mothers who reported smoking at the time of conception were screened for inclusion in the integrated sample (Figure 2). Fourteen individuals (5%) reported recognising their pregnancy before their estimated date of conception and were thus excluded from study analyses. Individuals with missing data on average daily number of cigarettes during pregnancy were also excluded (n = 1; 0.01%). The final analytic sample consisted of 271 women (Figure 2).

In the analytic sample, participants' mean age (SD) was 24.6(5.1) and 24.5 (5.0) years in MIDS and GUH, respectively. Gestational age at pregnancy recognition in MIDS and GUH was 6.2 (2.2) and 6.1 (2.4) weeks, respectively. Just under half (45%) of participants were nulliparous in both cohorts. GUH was more racially and ethnically diverse; 33.2% were non-Hispanic white, 39.5% non-Hispanic Black and 18.1% Hispanic participants versus 75.9% non-Hispanic white participants in MIDS. Additionally, 38.5% of women in GUH reported some college education versus 51% in MIDS. Mean smoking immediately before pregnancy was M_GUH = 12.0 cigarettes/day (SD = 8.1) versus M_MIDS = 9.5 cigarettes per day (SD = 6.7).

2.1 ∣. Assessments of smoking

Smoking was assessed using calendar-based time-line follow back (TLFB) interviews,²⁸ considered the gold standard for retrospective reporting of substance use. TLFB interviews differ substantially from standard questionnaires. Rather than querying respondents about substance use relative to arbitrary benchmarks in time (i.e., past 12 months, past 30 days), TLFB interviews mimic the way in which memories are spontaneously structured and recalled—that is, in relation to personally salient events, identified by respondents. Respondents are instructed to begin by marking as many of these salient events as they can remember over the particular interval of interest on a calendar, then fill in the number of cigarettes smoked on each day in relation to these events as they are spontaneously recalled in the order they are recalled.²⁸

In MIDS, women filled in the mean number of cigarettes per day smoked for each month, beginning with the month prior to the LMP, followed by the month starting with their LMP, then the next 9 months of pregnancy. Final monthly windows including determination of the date of conception were assigned later during data processing, after all available information on gestational age (ultrasound, delivery) was available. This approach may have minimised the impact of social desirability bias on reporting of pregnancy smoking in the earliest weeks of gestation. In GUH, participants marked the number of cigarettes smoked on each day of the calendar, beginning 3 months prior to the estimated date of conception.

2.2 ∣. Key pregnancy dates

As shown in Figure 1, the temporal discordance between the estimated date of conception and the date women recognised the pregnancy was the key interval during which time any within-person change in smoking was used to infer a ‘putative biological effect’ of pregnancy on smoking frequency, separate from a putative ‘psychological effect’ associated with knowledge of the presence of a developing foetus. We estimated the date of conception as 14 days after the first day of the women's LMP for both cohorts, mimicking current obstetric practice. In cases when the date of the LMP is uncertain, the date of conception is estimated from fundal height measured during ultrasound exam or from clinical examination.³⁶ In MIDS, LMP date was self-reported and presumed to be accurate. We then derived date of conception as 14 days after the LMP, based on the biological average date of ovulation as 2 weeks after the start of the menstrual period. In GUH, women were asked to provide their estimated due date, which may have been based on LMP or may have been updated based on the ultrasound. The date of conception was derived by counting back 38 weeks from estimated due dates reported by participants. The date when women recognised the pregnancy was ascertained in both cohorts.

2.3 ∣. Statistical analysis

Coordinated analysis is a type of Integrative Data Analysis in which longitudinal cohorts have nonidentical data collection protocols and key variable measurement, and thus, direct data harmonisation is not possible.^37,38 In coordinated analysis, data from the included cohorts are analysed separately to account for their unique features. Meta-analysis is then used to create summary effects. In our study, the level of detail on smoking measurement varied by cohort, and this harmonisation of average daily smoking was impractical. Therefore, we used a coordinated data analysis approach.

In this study, we estimated mean daily cigarettes smoked in five time periods: (1) before pregnancy, (2) conception of the pregnancy prior to recognition of the pregnancy, and the first (3), second (4), and third (5) months after pregnancy recognition. Visual representations of these time periods are shown in Figure 3 graphs. All months were considered as 28-day intervals. In GUH, the daily smoking ascertainment allowed for exact inclusion of each of these intervals (i.e. the conception to recognition period included only the days between estimated conception and recognition), and the first month after pregnancy recognition started within the date of pregnancy recognition. In MIDS, average daily cigarettes were reported in months, starting with the month prior to the LMP. Accordingly, the conception to recognition period included the month(s) between LMP and recognition month, and the month of recognition was considered the first month after pregnancy recognition. In both studies, the conception to recognition period varied by individual, although most women recognised during the second month of the pregnancy. As a sensitivity analysis, we limited the samples to only those women who recognised during the second month of pregnancy, with no qualitative change in findings.

FIGURE 3.

TOP: Estimated effects of pregnancy onset (conception) and pregnancy recognition on smoking, operationalised as mean within-individual change in cigarettes per day from coordinated and meta-analysis. BOTTOM: Graphical depiction of mean within-individual change in smoking by cohort

Within each study, a repeated measures analysis of variance (ANOVA) was used to detect overall differences in smoking, in mean cigarettes/day, across each of five time periods of interest with a compound symmetry correlation structure assumed.³⁹ Least squared means of cigarettes smoked per day in each interval were also calculated to ensure correct variance estimation across multiple time points. No additional predictors were considered in the main statistical models. Random effects meta-analysis with a DerSimonian–Laird estimator was used to create a combined summary estimate of the differences between the mean daily cigarettes measures between five time periods of interest.⁴⁰ Q statistics and associated p-values were used to test homogeneity of the effects between cohorts. All analyses were conducted in SAS version 9.4. Statistical significance was set a priori at 2-sided p < .05.

3 ∣. RESULTS

Mean cigarettes/day, graphically depicted in the top row of Figure 3, declined between conception and pregnancy recognition by about one cigarette/day in both cohorts (M_MIDS = −1.1; 95% confidence interval −1.9, −0.3; p = 0.01; M_GUH = −0.9; 95% confidence interval −1.6, −0.2; p = 0.01). Smoking declined further and more robustly during the 2 months after the pregnancy was recognised (M_MIDS = −3.3 cigarettes/day; 95% confidence interval −4.4, −2.5; p < 0.001); M_GUH = −4.8 cigarettes/day, 95% confidence interval −5.5, −4.1; p < 0.001) before plateauing at a mean of 4.0 cigarettes/day (SD = 7.5) in MIDS and a mean of 4.6 cigarettes/day (SD = 7.1) in GUH. By the third month after the pregnancy was recognised, the percentage of women who had stopped smoking altogether was 42% in MIDS and 22% in GUH.

To rule out the possibility that observed declines in mean smoking levels were driven by an increasing proportion of women who had stopped smoking altogether over time, we included two sensitivity analyses. First, we examined the subset of women in each cohort who were still smoking any amount at each interval thereby excluding those who had quit (Figure 4, top row). Next, we examined the subset of women who smoked any amount at every interval examined in this study and throughout the remainder of pregnancy (Figure 4, middle row). Patterns of decline in smoking among these subsets of ‘nonquitters’ and ‘persistent smokers’ were remarkably similar and even more robust relative to patterns shown for the total sample. Finally, to rule out the possibility that postconception-prerecognition decline was driven by deliberate systematic reductions in smoking by women who had planned to conceive, we included an additional sensitivity analysis to examine results stratified by whether the pregnancy was reported as planned versus unplanned (Figure 4, bottom).

FIGURE 4.

Sensitivity analyses of: (a) persistent pregnancy smokers who smoked throughout the entire pregnancy (TOP); (b) nonquitters, that is, women still smoking at each interval (n's shown along x-axis) (MIDDLE); and (c) unplanned pregnancies only (BOTTOM)

Meta-analytic, averaged associations are shown in Figure 3. Declines in smoking that occurred from preconception through the second month of pregnancy awareness were all statistically significant. Overall, mean within-individual changes in smoking were similar across cohorts, indicative of low heterogeneity between studies and evidenced by nonsignificant values for the Q statistic. An exception was between-cohort changes in smoking during the first month of pregnancy awareness. Here, women in GUH demonstrated a steeper decline in mean cigarettes/day relative to those in MIDS (Q = 7.43; p = 0.006).

4 ∣. DISCUSSION

Two prior within-person analyses of Swedish medical, criminal and pharmacy registry data supported a striking protective effect of pregnancy on risk for illicit drug and alcohol abuse.^6,7 The current study extends these findings by employing a within-person analysis of data from two independent observational cohorts to estimate the protective effect of pregnancy on cigarette smoking. We additionally estimated the effect of being pregnant on smoking, separate from the effect of being aware of the pregnancy.⁴¹ Recognition of the pregnancy, which varied widely with respect to gestational age, was associated with a significant within-person decline in smoking as expected (Figure 3). Importantly, within-person reduction in smoking was evident regardless of whether smokers did or did not quit smoking before delivery (Figure 4). However, this is the first study to our knowledge to show a postconception decline before the pregnancy was recognised in two independent cohorts, recruited during different years and from different regions of the United States.

We considered that an apparent decline in mean cigarettes/day could be driven by zero values among women who had already quit smoking. However, sensitivity analyses including only women who were still smoking at any particular interval (Figure 4, top) and only women who never quit smoking at any time during the pregnancy (Figure 4, middle) ruled out this possibility. We also considered that observed postconception-prerecognition declines in smoking could have been driven by deliberate reductions in smoking frequency over time made by women who were hoping to conceive. If this were true, declines would only be observed in planned pregnancies. Again, as shown at the bottom of Figure 4, this possibility was also not supported by sensitivity analysis of only unplanned pregnancies in GUH (pregnancy planning was not assessed in MIDS). We and others have previously viewed pregnancy-associated reduction in smoking as evidence of the power of investment in offspring to interrupt addictive processes.^5-7,18-20 While this may be true, results of the current study indicate that other protective processes may be involved that exert their effects independent of the awareness of having conceived.

Changes in olfactory and gustatory sensations, including an aversion to the taste or smell of cigarettes during the first trimester, for example, could contribute to a reduction or suspension of smoking.^42,43 Interestingly, an inverse relationship between smoking and ‘morning sickness’ (i.e. nausea and vomiting of pregnancy) has previously been observed in many but not all studies.^21,44 While smoking has been viewed as (paradoxically) protective against nausea and vomiting during pregnancy and also postoperatively,⁴⁵ results of the current study suggest this relationship could be bidirectional.

Hormonal factors related and unrelated to nausea and vomiting are also plausible mechanisms of observed postconception-prerecognition declines in smoking⁴⁶ with progesterone being most studied.^47,48 Less studied is human chorionic gonadotropin (hCG), the hormone assayed by pregnancy tests, which rises exponentially between the fifth and tenth weeks of gestation. This rise in serum hCG concentration coincides with the postconception-prerecognition declines in smoking observed in this study and the onset of nausea and vomiting for most pregnant individuals.^44,49 In fact, an inverse dose-dependent relationship has previously been observed between frequency of pregnancy smoking and serum hCG concentration, although directionality is again unclear.⁵⁰ Repeated prospective assessments of smoking, nausea and vomiting and hormonal processes during pregnancy would be needed to elucidate directionality of these relationships.

The slope of decline in postconception smoking, while statistically significant before pregnancy recognition, was much steeper after recognition, before plateauing by 3 months after pregnancy recognition (Figures 3 and 4). Interpretation of these different rates of change in smoking is challenging, however, since putative biological and psychological processes were not assessed in this study, and importantly, overlap once the pregnancy is recognised. For example, we cannot discern the extent to which the reduction in smoking of about four cigarettes/day observed in the month after pregnancy recognition is attributable to pregnancy recognition alone or reflects additive effects of pregnancy recognition and ongoing biological processes that began at conception. As gestational decreases in nicotine self-administration have been reported in pregnant rats relative to their nonpregnant counterparts,²³ animal models could be extremely useful for isolating and examining biological processes directly in the absence of the confounding sociocultural framework present in humans.

Replication of our findings in other cohorts and across other is also recommended to support a novel line of bidirectional translational research aimed at elucidating mechanism(s) of gestation that interrupt substance-related processes. This approach of deconstructing pregnancy's protective effect on addictive processes would provide a complementary strategy to the conventional disease-based investigations. When we consider the profound positive impact of pregnancy on women's smoking^5,51 and substance use^6,7,31 and the urgent need for more effective and acceptable treatment for substance use disorders, the potential impact of the proposed line of research is clear.

Findings also have implications for understanding the impact of prenatal drug exposures on offspring development by underscoring the need to account for the substantial within-person variability in smoking across gestation. There is early evidence that failure to account for within-person variability in maternal smoking during pregnancy (and foetal exposure) leads to substantial error in estimates of early developmental risk.⁹ Thus, the quantification of within-person changes in the use, and co-use, of other substances across the perinatal period is a promising and impactful area for future research.

Results should be considered within several study limitations. First, the key outcome was derived from information on smoking occurring before women were enrolled. This necessitated the use of data that were retrospectively recalled, albeit using gold standard calendar-based interview methods. Prospective assessment of women before conception may also be problematic, however, since women who are trying to conceive may continue to reduce smoking thereby confounding the effect of interest. Accuracy would be optimal if cigarettes per day were assessed prospectively in a sample of nonpregnant women who were not planning nor wanting to become pregnant (and thus not reducing their smoking), but then inadvertently happened to conceive anyway and continue in the study. Such a study is of course impractical. Nonetheless, replication of our findings using data from larger population-based samples in which the number of unplanned pregnancies is much larger is strongly recommended to confirm these results.

Finally, granularity of smoking data (i.e. number of observations) used in our coordinated analysis was limited to monthly reports of mean cigarettes per day, the finest granularity available in both cohorts. This may have resulted in some error in classification in cases where conception or recognition of the pregnancy occurred in the middle of a given month. To overcome this limitation, an important next step is to model trajectories of smoking using daily smoking observations, available only in the GUH cohort. Extending our design to examining marijuana use in GUH is also possible and recommended to identify between-substance differences.

5 ∣. CONCLUSION

Investment in offspring is likely to be a powerful psychological motivation to reduce or suspend smoking during pregnancy. Declines in pregnancy smoking occurring before pregnancy recognition found in this study, however, indicate that other processes should be examined as candidate targets for intervention. Given pregnancy-associated declines in nicotine and other drug consumption in other mammalian species, we propose employing both preclinical and clinical studies to isolate candidate mechanism(s) and confirm ecological validity of preclinical findings, respectively, in an iterative bidirectional manner. This strategy to expedite progress is ideally matched to the urgency and potential impact of this research.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.