Comparison of Nicotine Dependence Indicators in Predicting Quitting among Pregnant Smokers

Abstract

Research in the general population of smokers indicates that across various measures of nicotine dependence, time to first cigarette (TTFC) is the strongest single-item predictor of quitting success. Whether those findings generalize to pregnant smokers is unclear. To investigate this matter, we compared TTFC to cigarettes per day (CPD) and the Heaviness of Smoking Index (HSI) in predicting late-pregnancy abstinence among 289 pregnant women enrolled in four smoking-cessation trials assessing the efficacy of financial incentives. Logistic regression was used to compare predictors with model fit measured using the c statistic (range = 0.5 [poor prediction] to 1.0 [perfect prediction]). In simple regressions, model fit was comparable across the three measures although strongest for CPD alone (c = 0.70, 0.68, 0.66 for CPD, HSI, and TTFC, respectively). In a stepwise multiple regression, treatment entered first (c = 0.67), then CPD (c = 0.77), quit attempts pre-pregnancy (c = .81), TTFC (c = .82), and quit attempts during pregnancy (c = .83). We saw no evidence supporting TTFC as the optimal predictor of quitting among pregnant smokers. Instead, the evidence supported using CPD and TTFC together or CPD alone if using only a single predictor.

Introduction

Understanding the relationship between nicotine dependence and success in quitting smoking is important to effective tobacco control and regulatory efforts. In the tobacco control area, such measures can be used to assign participants to more or less intensive interventions depending on the severity of one’s nicotine dependence level (e.g., Kalman et al., 2011). In the tobacco regulatory science area, these measures can be used to index differences in the dependence potential between different products as is being done, for example, in studies investigating the effects of reducing the nicotine content in cigarettes to very low levels (e.g., Benowitz et al., 2012; Benowitz et al., 2015; Hatsukami et al., 2015).

Using data collected in four randomized controlled cessation trials and one international epidemiological study in the general population of smokers, Baker et al (2007) reported that Fagerstrom Test for Nicotine Dependence (FTND; Heatherton, Kozlowski, Frecker, & Fagerstrom, 1991) scores were the strongest predictor of short- and longer-term quitting relative to other dependence assessment instruments (i.e., Heaviness of Smoking Index [HSI], Nicotine Dependence Syndrome Scale [NDSS], Wisconsin Inventory of Smoking Dependence Motives [WISDM]). Further analyses indicated that the variance in the relationship between FTND scores and quitting success was accounted for by two items, Cigarettes per Day (CPD) and Time to First Cigarette (TTFC), with the latter item being the much stronger predictor of the two items. The HSI, which combines categorical versions of CPD and TTFC, also predicted quitting in the clinical trials in the Baker et al. report, but further analyses showed that to be attributable to the TTFC item.

While the Baker et al. (2007) report and others (e.g., Fagerstrom, 2003) provide compelling evidence that TTFC is a stronger predictor of quitting than CPD in the general population of smokers, whether those findings are generalizable to pregnant smokers is unclear and is the overarching focus of the present report. Clearly there is a great deal of interest in cigarette smoking and use of other tobacco and nicotine delivery products in pregnant smokers due to the substantial adverse short- and longer-term impacts on the developing fetus (Bakker & Jaddoe, 2011; Cnattingius, 2004; Dietz et al., 2010). There are also well-substantiated differences in the smoking of pregnant women compared to the general population that warrant caution against assuming that observations regarding the effects of tobacco and nicotine in one group can be readily generalized to the other. At a very basic level, for example, nicotine metabolism is accelerated among pregnant compared to non-pregnant users (Dempsey, Jacob, & Benowitz, 2002). As another example at the clinical level, nicotine replacement therapy is a staple of best practices for increasing smoking cessation rates among non-pregnant smokers (U.S. Public Health Service, 2008) whereas its efficacy in pregnant women is equivocal at best (e.g., Coleman et al., 2012).

The evidence from the literature on predictors of quitting success among pregnant smokers also suggests caution in assuming that TTFC is superior to CPD. In at least one study on spontaneous quitting among pregnant smokers that we are aware of, CPD and TTFC each independently predicted spontaneous quitting (Ockene et al., 2002). In at least three other studies on spontaneous quitting, pre-pregnancy CPD significantly predicted spontaneous quitting, but TTFC was not examined in these studies (Cnattingius et al., 1989, Severson et al., 1995; White et al., 2015). In a prior report by our group on predictors of quitting among pregnant smokers enrolled in the same voucher-based incentives trials that are the focus of the present study, CPD predicted late-pregnancy smoking status (Lopez et al., 2015). While TTFC was collected in those trials as part of the FTND, it was not examined as a potential predictor in Lopez et al. (2015), or in an earlier report that examined predictors of quit success among the women from three of four previous cessation trials as well as predictors of spontaneous quitting (Higgins et al., 2009).

In a trial conducted by other investigators also examining the effectiveness of financial incentives among pregnant smokers, the HSI did not predict quitting success, but no information on the validity of CPD and TTFC as single predictors was provided in this report (Ierfino et al., 2015). Lastly, in a trial investigating whether adding a brief smoking-cessation intervention to usual midwife care increased quitting among women still smoking and those who recently quit spontaneously, TTFC predicted smoking status at three and six months post-intervention, but CPD does not appear to have been investigated in this study (Hajek et al., 2001).

While the above studies document important predictive relationships between CPD and TTFC with spontaneous and late-pregnancy quitting success among pregnant women, we know of only one study that included CPD and TTFC as independent predictors, and that study suggested that each accounted for significant independent variance in predicting quitting rather than TTFC being the dominant predictor. To our knowledge, the present study represents the first systematic comparison of the relative predictive validity of CPD, TTFC, and the HSI in predicting quitting success among pregnant smokers.

Methods

Participants

Participants (N = 289) were from four previous randomized controlled clinical trials that were all conducted in the same university-based outpatient research clinic (Heil et al., 2008; Higgins et al., 2004; Higgins unpublished data reported in Higgins et al., 2012 review; Higgins et al., 2014). This same data set has been used in prior studies of predictors of abstinence among pregnant cigarette smokers (Higgins et al., 2009; Lopez et al., 2015; White et al., 2014), although none focused on comparisons of dependence indicators or included the TTFC or HSI measures.

Participants in these trials were recruited from local obstetrical clinics and the Federal Special Supplemental Nutrition Program for Women, Infants, and Children office located in the Burlington, Vermont area. Trial inclusion criteria were (a) self-reported smoking in past week that was biochemically verified using urine cotinine toxicology testing (≥ 80 ng/ml, enzyme immunoassay test [Microgenics Corporation, Fremont, CA] run on a Roche Cobas Mira analyzer), (b) not planning to leave the area within 6 months of delivering, and (c) English speaking. The above cut point for determining smoking status was based on previous research indicating nearly perfect agreement (≥ 98%) between classifications based on enzyme immunoassay testing (EMIT) using a cut point of approximately 80 nl/ml (79–87 ng/ml) and gas chromatography ([GC], Higgins et al., 2007). This cut point has been used in all of our previous trials to determine smoking status at intake and at subsequent routine visits and formal assessments, and was used for all participants in the present study. Exclusion criteria were (a) incarceration, (b) having previously participated in an incentives-based smoking cessation trial or living with a trial participant, (c) currently enrolled in opioid substitution therapy, (d) current use of psychomotor stimulant or antipsychotic medications, (e) greater than 25 weeks gestation, and (f) living in a group home.

Participants completed study intake assessments for each trial at which they reported the number of cigarettes smoked per day over each of the 7 days preceding the interview, from which we derived an average CPD for each participant. They also completed the Fagerstrom Test for Nicotine Dependence (Heatherton et al., 1991), which includes an item asking “how soon after waking do you smoke your first cigarette?” with the following four response options and associated # of points: ≥ 1 hr of rising = 0 points, 31–59 mins = 1 point, 6–30 mins = 2 points, and ≤ 5 mins = 3 points. The HSI comprises both the CPD and TTFC items from the FTND. The CPD item on the HSI is scored on a 0–3 scale, where smoking ≤ 10 cigarettes per day = 0 points, 11–20 CPD= 1 point, 21–30 CPD = 2 points, and ≥ 31 CPD = 3 points. TTFC is scored on the 0–3 scale described above, thus total HSI scores range from 0–6. The HSI was not originally used in our previous trials but we were able to calculate participant scores on this scale based on their responses to the CPD and TTFC items on the FTND.

Treatment Conditions

Treatment conditions have been described in detail previously (Higgins et al., 2004; Heil et al., 2008). In brief, participants were randomly assigned to one of the following two treatment conditions: (a) Contingent vouchers condition wherein participants earned vouchers exchangeable for retail items contingent on providing cotinine-negative urine toxicology results. Participants earned $6.25 for the first cotinine-negative sample, and $1.25 for each consecutive negative sample up to a maximum value of $45. Positive samples or missed sample visits reset the voucher value at $6.25, but two consecutive negative samples restored the value to its pre-reset level. (b) Noncontingent vouchers (control) condition wherein vouchers were received for attending scheduled clinic visits and submitting urine tests, independent of smoking status. Participants in both conditions also received usual care for smoking cessation via their obstetrical clinics (e.g., discussing the advantages of quitting during pregnancy), as well as brief counseling provided by the study staff.

Statistical Methods

Baseline sociodemographics, smoking characteristics, and psychiatric symptoms were compared between smokers and abstainers at a late-pregnancy assessment (i.e., ≥ 28 weeks gestation) using t tests for continuous measures and chi-square tests for categorical variables. The correlations between CPD, TTFC, and the HSI were examined using Pearson’s r. Associations between CPD, TTFC, and the HSI with smoking status were examined using a two-step logistic regression modeling sequence: In Step 1, we examined associations between (a) CPD alone, (b) TTFC alone, and (c) HSI alone with late-pregnancy smoking status. In Step 2, forward stepwise logistic regression modeling was used to examine whether CPD, TTFC, and the HSI independently predicted late-pregnancy smoking status in a single model when adjusting for treatment and other potential predictors of smoking status. Treatment condition and all baseline characteristics associated with late pregnancy smoking status at p < .25 were included in the regression in Step 2, along with CPD, TTFC and the HSI, with the criterion for entry and retention in the final model set at p < .05. To determine whether associations between CPD, TTFC, the HSI, and smoking status were moderated by treatment condition, logistic regression analyses were also conducted using models that examined main effects and interactions with treatment condition.

The ability of the different models to discriminate smoking status was measured using the c statistic (Hosmer, Lemeshow, & Sturdivant, 2013). For Step 2 in the above regression analysis, which incorporated more than one item, the c statistic was calculated as successive variables were added and therefore reflects the predictive ability of the model taking into account the most recently added variable and those that entered previously. The c statistic ranges from 0.5 (indicating very poor prediction) to 1.0 (indicating perfect prediction). Values between 0.7–0.79 are considered acceptable, values between 0.8–0.9 are considered excellent, and values > 0.9 are considered outstanding.

For the purpose of representing a meaningful change in smoking, odds-ratios for CPD were calculated to represent a 5-cigarette difference. TTFC was analyzed as four categories (those who smoked within ≤ 5 mins of rising, 6–30 mins of rising, 31–59 mins of rising, and ≥ 1 hr of rising), with ≥ 1 hr functioning as the reference category.

Results

Study Participants

Participants in this study were young, socioeconomically disadvantaged women, with the majority being less than 25 years of age, having completed 12 or fewer years of education, without full-time employment, unmarried, and without private health insurance (Table 1). In univariate analyses, those who were able to quit by the late-pregnancy assessment were on average more educated, smoked fewer CPD before pregnancy and at baseline, had their first cigarette later upon rising, started smoking at an older age, were more likely to report having attempted to quit smoking before and during the current pregnancy, and were more likely to report that smoking would harm their baby.

Table 1.

Participant Characteristics and Correlations with Late-pregnancy Smoking Status (N = 289).

Characteristic	Overall (N = 289)	Abstinent (n = 74)	Smoker (n = 215)	p value
Demographics
Age (years)	24.0 ± 5.1	24.4 ± 4.8	23.9 ± 5.2	.55
% Caucasian	94	96	93	.38
Education				.02*
% > 12 years of education	18	18	18
% = 12 years of education	52	64	48
% < 12 years of education	30	18	34
Weeks pregnant at baseline	10.7 ± 5.1	10.3 ± 5.34	10.9 ± 5.0	.45
% Primagravida	56	61	54	.29
% Married	18	13	19	.28
% With private insurance	20	20	20	.96
% Working for pay outside of home	49	57	46	.11
Smoking Characteristics
Cigarettes per day pre-pregnancy	19.6 ± 9.0	16.4 ± 7.6	20.8 ± 9.1	<.001***
Cigarettes per day at baseline	9.2 ± 6.6	6.2 ± 4.5	10.2 ± 6.9	<.001***
Time to first cigarette after waking up				<.001***
% ≤5 minutes	16	9	18
% 6–30 minutes	29	16	33
% 31–59 minutes	17	15	18
% ≥1 hour	38	59	30
Age first started smoking cigarettes	14.5 ± 2.8	15.3 ± 2.3	14.3 ± 2.9	<.01**
% Living with another smoker	79	78	79	.97
% With no smoking allowed in home	45	51	42	.18
% With none or few friends/family who Smoke	21	23	20	.65
% Attempted to quit pre-pregnancy	67	84	61	<.001***
Number of quit attempts during Pregnancy	0.8 ± 2.1	1.6 ± 3.5	0.6 ± 1.2	.01**
Minnesota Nicotine Withdrawal Scale Score	1.6 ± 0.8	1.6 ± 0.8	1.6 ± 0.9	.78
% Endorsing that smoking will greatly harm baby	84	92	82	.04*
Psychiatric Symptoms
Stress rating	5.6 ± 2.6	5.8 ± 2.4	5.5 ± 2.7	.36
Beck Depression Inventory	10.6 ± 7.2	10.3 ± 7.0	10.7 ± 7.3	.70
% History of depressive symptoms	39	38	39	.80
Treatment condition				<.001***
Non-Contingent	42% (122)	18% (13)	51% (109)
Contingent	58% (167)	82% (61)	49% (106)

Note: M ± SD or %

^*significant correlation with smoking status (p < .05)

^**(p < .01)

^***(p < .001)

Correlations Between Dependence Indicators

CPD and TTFC were moderately correlated (r = 0.49, p < .0001), and as expected HSI scores were strongly correlated with both CPD (r = .74, p < .001) and TTFC (r = .93, p < .0001).

Logistic Regression Models

Results of the regression analyses examining CPD, TTFC and the HSI are shown in Table 2. In Step 1, where each predictor was considered alone in simple logistic regressions, each significantly predicted late-pregnancy smoking status. The predictive ability of CPD alone was in the lower end of acceptable range for model fit (c = 0.70) while HSI and TTFC fell slightly below that level (c = .68 and .66, respectively).

Table 2.

Effects of CPD, TTFC, and HSI on Late-pregnancy Smoking Abstinence (N = 289).

Step	Effect	OR	95% CI	p-value	ca
1a	CPD	0.48	0.35 – 0.67	<.001***	0.70
1b	TTFC			<.001***	0.66
	≤ 5 min vs ≥ 1 hr	0.26	0.11 – 0.65
	6–30 min vs ≥ 1 hr	0.25	0.12 – 0.51
	31–59 min vs ≥ 1 hr	0.42	0.19 – 0.90
1c	HSI	0.59	0.47 – 0.74	<.001	0.68
2	Treatment	6.56	3.09 – 13.93	<.001***	0.67
	CPD (per 5 cig increase)	0.54	0.37 – 0.80	.002**	0.77
	Attempted to quit pre-pregnancy	3.55	1.65 – 7.63	.001**	0.81
	TTFC			.02*	0.82
	≤ 5 min vs ≥ 1 hr	0.65	0.21 – 1.99
	6–30 min vs ≥ 1 hr	0.34	0.14 – 0.79
	31–59 min vs ≥ 1 hr	0.31	0.12 – 0.75
	Attempted to quit during pregnancy	1.18	1.00 – 1.38	< .05	0.83

^aStep 1 shows the results of simple univariate logistic regression modeling for each dependence indicator showing unadjusted odds ratios (OR) and 95% confidence intervals (CIs). The c statistic refers to the ability of each indicator to discriminate the outcome (late-pregnancy smoking status). Step 2 shows the results of a forward step-wise logistic regression along with adjusted ORs and 95% CIs. The changes in the c statistic in Step 2 reflect improvements in overall model fit as additional significant predictors entered.

CPD= Cigarettes Per Day, TTFC= Time to First Cigarette, HSI= Heaviness of Smoking Index

^*significant correlation with smoking status (p < .05)

^**(p < .01)

^***(p < .001)

In Step 2, where treatment and other baseline measures were considered along with CPD, TTFC, and HSI as potential predictors in a forward stepwise logistic regression model, CPD, having attempted to quit smoking both before and during the current pregnancy, and TTFC were significant independent predictors but the HSI was not. Treatment condition entered the model first with a model fit (c) of 0.67. CPD entered the model second, increasing model fit from c = 0.67 to c = 0.77. Attempting to quit smoking pre-pregnancy entered the model third, increasing fit from c = 0.77 to c = 0.81. TTFC was the fourth predictor to enter the model, increasing fit from c = 0.81 to c = 0.82, and having attempted to quit smoking during the current pregnancy was the fifth and final predictor increasing model fit (c) to 0.83. Neither CPD or TTFC interacted significantly with treatment condition indicating no discernible differences in the model’s ability to predict outcomes in the contingent incentives intervention and noncontingent incentives control conditions (Wald X²(1) = 0.23, p = .63 and Wald X² (3) = 0.53, p = .91, respectively).

Discussion

The present study represents the first explicit comparison of nicotine dependence indicators in pregnant women to use comparable methods (e.g., prospective comparisons across multiple randomized cessation trials) and data analytic techniques (e.g., direct and systematic comparisons of the predictive validity of different indicators) as Baker et al. (2007)’s investigation of this same topic in the general population of smokers. Results of the study indicated that individually, CPD, TTFC, and HSI scores are correlated and comparable predictors, with CPD being a slightly better predictor of late-pregnancy quitting success than HSI or TTFC values in the present study. The evidence indicates that using CPD and TTFC together is better at predicting quitting success than either measure alone, but if one was to use only a single predictor in this population, than CPD alone is best. Certainly the present results offer no evidence that TTFC is a preferred predictor over CPD as has been reported in studies conducted in the general population of smokers (Baker et al., 2007; Fagerstrom, 2003). Importantly, these results provide further caution about assuming that information garnered from the general population of smokers generalizes to pregnant women.

Considering that CPD and TTFC independently predicted quitting success in the present study, it may seem odd that the HSI was not the superior predictor. However, recall that the CPD item on the HSI is scored on a 0–3 scale, where smoking ≤ 10 cigarettes per day = 0 points, 11–20 CPD= 1 point, 21–30 CPD = 2 points, and ≥ 31 CPD = 3 points. Those categories are not an ideal match with the pregnant smoker population as most of those who fail to quit nevertheless reduce their smoking rate usually by almost 10 CPD, which places many in the 0 points category and leaves few in the higher smoking categories. Relative to treating CPD as a continuous variable as was done when examining CPD as an individual predictor in the present study, the HSI method of scoring CPD categorically reduces the variance in this predictor. While that limitation did not preclude HSI scores from significantly predicting quitting success when used alone in the present study, it did appear to render it a slightly weaker predictor than using CPD as a continuous measure alone or in combination with TTFC.

The present findings demonstrating the strong predictive ability of CPD have practical implications for research on smoking cessation treatments. Of course, identifying which smokers have difficulty quitting is a prerequisite to targeting treatment modifications to improve outcomes among difficult-to-treat smokers. Thus, the present results represent a useful initial step towards providing an empirical rationale for parametric manipulations or other treatment modifications directed towards improving treatment outcomes in this population. For example, we are currently conducting a comparative effectiveness trial in which women randomized to a financial incentives condition receive higher-value vouchers if they smoke ≥ 10 cigarettes per day compared to those who smoke < 10 CPD.

In addition to tobacco control implications, the present findings also have implications for tobacco regulatory science (Ashley & Backinger 2012; Ashley, Backinger, van Bemmel, & Neveleff, 2014). The 2009 Family Smoking Prevention and Tobacco Control Act grants the U.S. Food and Drug Administration (FDA) regulatory authority over the manufacture, marketing, and distribution of tobacco products and hence the authority to enact the proposal put forward in Benowitz and Henningfield’s (1994) seminal paper recommending decreasing the nicotine content of cigarettes below a minimum threshold for producing nicotine dependence. Related to those developments, there has been an increase in research focused on the potential for very low nicotine content (VLNC) cigarettes to decrease dependence among current smokers thereby making quitting easier among those who may wish to discontinue smoking (e.g., Benowitz et al., 2015; Donny et al., 2015). The FDA’s interest in VLNC extends to vulnerable populations, including pregnant women. Although response to VLNC’s has been studied in some of the vulnerable populations of interest to the FDA (e.g., smokers with serious mental illness, AhnAllen, Bidwell, & Tidey, 2015; Tidey, Rohsenow, Kaplan, Swift, & Ahnallen, 2013), there is a need for research on VNLC’s among others, including pregnant women. For example, our group is preparing to conduct studies on VLNC’s in pregnant smokers for which sensitive measures of nicotine dependence will be a critically important component. The present findings indicating that CPD may be a somewhat stronger nicotine-dependence predictor of quitting success than TTFC among pregnant smokers, and that using the two measures in combination may be better than either alone, has clear practical implications for such tobacco regulatory studies.

Because the present research used a largely Caucasian sample of women who were enrolled in voucher-based incentives trials, the generalizability of the present results to other samples of pregnant smokers should be done cautiously pending replication in other more diverse samples and among pregnant women receiving alternative cessation treatments. However, the lack of an interaction between baseline levels of nicotine dependence and treatment condition (contingent incentives or non-contingent incentive controls) suggests that the relationships observed should have generality beyond incentive treatments per se. This need for additional research to establish the generality of the present findings to other treatments notwithstanding, the results of the present study have the potential to inform future research with pregnant women that is focused on improving treatment effectiveness or regulatory policy.