Does Impulsiveness Moderate Response to Financial Incentives for Smoking Cessation Among Pregnant and Newly Postpartum Women?

Abstract

We examined whether impulsiveness moderates response to financial incentives for cessation among pregnant smokers. All participants were randomized to either a condition wherein financial incentives were delivered contingent on smoking abstinence or to a control condition wherein incentives were delivered independent of smoking status. The study was conducted in two steps: First, we examined associations between baseline impulsiveness scores and abstinence at late pregnancy and 24-weeks postpartum as part of a planned prospective study of this topic using data from a recently completed, randomized controlled clinical trial (N = 118). Next, to increase statistical power, we conducted a second analysis collapsing results across that recent trial and two prior trials involving the same contingent incentive and control conditions (N = 236). Impulsivity was assessed using a delay discounting (DD) of hypothetical monetary rewards task in all three trials and Barratt Impulsiveness Scale (BIS) in the most recent trial. Neither DD nor BIS predicted antepartum or postpartum smoking status in the single or combined trials. Receiving abstinence-contingent incentives, lower baseline smoking rate (cigs/day), and a history of quit attempts pre-pregnancy predicted greater odds of antepartum abstinence across the single and combined trials. No variable predicted postpartum abstinence across the single and combined trials, although a history of antepartum quit attempts and receiving abstinence-contingent incentives predicted in the single and combined trials, respectively. Overall, this study provides no evidence that impulsiveness as assessed by DD or BIS moderates response to this treatment approach while underscoring a substantial association of smoking rate and prior quit attempts with abstinence across the contingent incentives and control treatment conditions.

Introduction

Impulsiviveness is considered by many to be a risk factor for substance use disorders and an emerging predictor of treatment outcome among those attempting to discontinue substance abuse (Baker et al., 2003; Bickel, Odum, & Madden, 1999; Heil et al., 2006; MacKillop et al., 2007; Madden, Bickel, & Jacobs, 1999; Mitchell, 1999; Vuchinich & Simpson, 1998; Washio et al., 2011). The construct is measured in a variety of ways, including through use of the delay-discounting (DD) of monetary rewards task and Barratt Impulsiveness Scale (BIS, version 11), which are the focus of the present study (Barratt & Patton, 1983; Bickel & Marsh, 2001).

DD is a behavioral-economic task that is used to examine individual differences in the rate at which rewards lose value and aversive events lose salience as delay to their receipt increases (Bickel & Marsh, 2001). DD tasks used with humans often involve hypothetical monetary outcomes (e.g., Bickel, Odum, & Madden, 1999; Madden et al., 1997). Participants make a series of choices between immediately available hypothetical monetary rewards and a delayed, larger amount across different temporal delays. Identifying indifference points across a range of temporal delays and then fitting Mazur’s (1987) hyperbolic equation to the data allows one to quantify the degree to which individuals discount delayed rewards (Johnson & Bickel, 2002). Participants generally prefer smaller, more immediate over the larger, more delayed rewards although there are considerable individual differences in the degree to which that applies. The shape of the discounting function is generally hyperbolic. The BIS is among the most widely used standardized instruments designed to assess individual differences in impulsivity. The instrument includes 30 items that individually contribute to a total score as well as additional factor scales (Patton, Stanford, & Barratt, 1995; Stanford et al., 2009). The instrument has been well validated using other neuropsychological measures of impulsivity in a wide variety of populations (Spinella, 2004; Spinella, 2007).

There is a growing body of evidence demonstrating that DD is associated with a wide-range of health-related risk behaviors (Bickel et al., 2007; Bickel et al., 2012; Bradford, 2010; Davis et al., 2010; Epstein et al., 2010; Rogers et al., 2010). Additionally, greater DD is associated with dependence on cocaine (Heil et al., 2006), opioids (Madden, Bickel, & Jacobs, 1999), alcohol (MacKillop et al., 2007; Vuchinich & Simpson, 1998), and nicotine (Baker et al., 2003; Bickel, Odum, & Madden, 1999; Mitchell, 1999). Greater discounting is also associated with the number of cigarettes smoked per day (Ohmura, Takahashi, & Kitamura, 2005), and smoking relapse in laboratory studies (Dallery & Raiff, 2007; Mueller et al., 2009) as well as outcomes in several clinical trials (MacKillop & Kahler, 2009; Sheffer et al., 2012; Yoon et al., 2007). There is at least one trial in which DD predicted treatment response in cocaine-dependent outpatients receiving the same incentive-based treatment approach that will be investigated in the present study (Washio et al., 2011). More recently, a study was reported examining whether discounting changes with sustained smoking abstinence (Secades-Villa et al., 2014). Smokers and abstainers showed no differences on DD immediately post-treatment; however, abstainers were discounting significantly less compared to smokers at a 12-month follow-up, suggesting that sustained smoking abstinence is associated with reductions in DD.

Congruent with the relationships between greater DD and increased risk for substance use disorders outlined above, higher BIS total scores discriminate between cocaine dependent individuals and controls (Lane et al., 2007) and between ecstasy users and controls (Bond et al., 2004). BIS scores have also been associated with more severe substance use disorders. Turning to treatment response, higher BIS total scores predict treatment dropout among cocaine-dependent (Moeller et al., 2001; Moeller et al., 2002; Patkar et al., 2004; Winhusen et al., 2013) and methamphetamine-dependent (Winhusen et al., 2013) outpatients.

As can be expected, some exceptions or boundary conditions on the positive associations between DD and risk for unhealthy behavior patterns outlined above have been reported as well. In a study of adolescent cannabis abusers, greater DD predicted poorer treatment outcomes in univariate analyses but not after controlling for the potential confounding influence of sociodemographic characteristics in multivariate analyses (Stanger et al., 2012). Bradstreet and colleagues (Bradstreet et al., 2012) reported that DD did not differentiate significantly between pregnant spontaneous quitters and continuing smokers. Spontaneous quitting is a term used to describe a small proportion of women who successfully self-initiate smoking cessation shortly after learning of their pregnancy (Heil et al., 2014; Solomon & Quinn, 2004). By contrast, what might be considered another aspect of impulsivity (e.g., social discounting) significantly differed between quitters and smokers, with quitters showing less social discounting (i.e., selfishness) than smokers as a function of degree of separation (i.e., greater generosity). More recently, White and colleagues (White et al., 2014) conducted a follow-up study to Bradstreet et al. (2012), again examining the ability of DD to predict spontaneous quitting among pregnant smokers. Social discounting was not examined in the follow-up study as it was not assessed in the majority of participants included in this larger study. DD, educational attainment, and smoking rate (cigarettes smoked per day) were each univariate predictors of quitting. In multivariate analyses, educational attainment remained a significant predictor and there was a significant interaction of DD and smoking rate, with DD being a significant predictor at lower but not higher smoking rates. Turning to studies in the general population of smokers, at least one study revealed a gender difference with DD discriminating between male but not female smokers and non-smokers (Jones, Landes, Yi & Bickel, 2009). A longitudinal twin study of discounting from early to late adolescence reported moderate heritability and a significant association with substance use in males but not females and no association with substance use problems in either sex (Isen, Sparks, & Iacono, 2014). Lastly, DD scores at study intake assessment failed to predict treatment outcomes in a study examining an incentive-based intervention designed to reinforce exercise among college undergraduates (Pope, 2013).

To our knowledge, the association between DD, BIS, and response to incentives-based smoking cessation has not been reported. The studies mentioned above (Bradstreet et al., 2012; White et al., 2014) regarding pregnant smokers examined spontaneous quitting and not response to a formal intervention. Examining whether the construct of impulsiveness moderates response to incentive-based interventions is of particular interest as such interventions are designed to offer relatively immediate reinforcement for healthy choices and thus might be expected to be more effective than other treatment approaches with more impulsive individuals who may be unable to tolerate the longer delays typically involved with naturalistic rewards for quitting smoking during pregnancy (e.g., improved birth outcomes). The results from the Washio et al. (2011) study on response to incentive-based interventions among cocaine-dependent outpatients mentioned above are consistent with that position. Washio and colleagues reported a significant association between greater DD and lower rates of abstinence in an incentive-based intervention condition with relatively low reinforcement magnitude but not in a condition with higher magnitude reinforcement. That is, when the magnitude of the incentives was low and thus less effective, DD predicted abstinence levels. However, as the magnitude of the incentive value increased and efficacy increased in parallel, the ability of DD to predict abstinence levels decreased to non-significant levels. Considering that incentive-based interventions are increasingly being used to effectively alter the same health-related behaviors for which impulsivity is reported to be a risk factor (Higgins, Silverman, et al., 2012), knowing whether response to incentive-based interventions may vary across different levels of impulsivity severity is an important question. Finally, knowing as much as possible about moderators of response to this evidence-based treatment for smoking cessation among pregnant and newly postpartum women should be helpful to ongoing efforts to further increase treatment effect size (Higgins et al., 2014; Higgins et al., 2012).

We examined these questions first as a prospective study that was conducted as part of recently reported randomized controlled clinical trial (Higgins Washio, Lopez, et al., 2014). Second, to increase statistical power and reduce the likelihood of false-negative results, we examined these questions using data from that single trial in combination with results from two previously reported trials that also involved abstinence-contingent and non-contingent incentives conditions as well as the DD task although not the BIS (Heil et al., 2008; Higgins et al., 2012).

Methods

Participants

All participants (N = 236) were enrolled in randomized controlled clinical trials from which the smoking cessation and birth outcomes results have been reported previously (Heil et al., 2008; Higgins et al., 2012; Higgins, Lopez et al., 2014). One hundred and forty of these participants contributed data to a prior study on predictors of smoking cessation in pregnant women that did not examine impulsiveness as a potential predictor (Higgins et al., 2009). The present trial was conducted in a university-based outpatient research clinic and approved by the local institutional review board. Participants were recruited from local obstetric clinics and the Federal Special Supplemental Nutrition Program for Women, Infants and Children in the Burlington, Vermont area. To be eligible for inclusion, participants had to report currently smoking at the first prenatal care visit with biochemical verification, reside within the county in which the clinic is located with no plans to leave the area for 6 months following delivery, and speak English. Exclusion criteria included incarceration, previous participation in a trial on incentives for smoking abstinence or living with a trial participant, current opioid substitution therapy, current use of psychotropic medications other than antidepressants, being greater than 25 weeks gestation, and living in a group residence (for more details see Higgins, Washio, Lopez, et al., 2014; Higgins, Washio, et al., 2012).

Assessments

At the study intake assessment, participants completed questionnaires examining socio-demographics, current smoking status/history, smoking environment and motivations, confidence and intentions to quit smoking, and provided breath and urine specimens. They also completed the DD (all trials) and BIS (only recent trial that is examined separately herein) tasks. Abbreviated versions of the intake assessment were administered again at seven subsequent assessments completed antepartum (early- and late-pregnancy) and postpartum (2-, 4-, 8-, 12-, and 24-weeks). Smoking status was biochemically verified at each assessment using urine cotinine testing (cutpoint = ≤ 80 ng/ml; Enzyme Multiplied Immunoassay Technique, Microgenics Corporation, Fremont, CA; a Roche Cobas Mira analyzer, Dade Behring Inc., Deerfield, IL).

The DD task was completed in a quiet room with a staff member present. The DD task used a notebook computer running Microsoft Visual Basic 6.0. The DD program has been described previously (Johnson and Bickel, 2002). Briefly, participants were seated in front of the computer screen, which displayed the following message:

Imagine that you have a choice between waiting (length of time) and then receiving $1,000 or receiving a smaller amount of money right away. Please choose between the two options.

In the instructions, the length of time given was either 1 day, 1 week, 1 month, 6 months, 1 year, 5 years, or 25 years. When participants were ready to begin the task, they clicked on the start button located on the screen, and the DD program was initiated. Participants chose between two different options, always a fixed amount ($1,000) at a fixed delay, or a smaller amount available immediately. The DD program adjusted the value of the smaller reward across trials according to an algorithm wherein different values of the smaller reward were presented until an indifference point was found, in which the value of the smaller, immediate amount was subjectively equivalent to the delayed $1,000 reward (Johnson & Bickel, 2002). Once the indifference point for a given delay was determined, the next delay was introduced until an indifference point was established for each of the 7 delays noted above. Delays were presented in a fixed ascending or descending order for a given participant but determined randomly across participants. Prior to assessment of each new delay, participants were presented again with the instructions listed above.

The BIS is a 30-item self-report instrument designed to measure the construct of impulsiveness was administered at the same assessment intervals noted above for DD (Barratt & Patton, 1983; Patton et al., 1995). A paper and pencil version of the BIS was completed by participants in a quiet office setting with a study staff member present. While it is common to report BIS total scores to represent overall impulsivity severity, BIS scores can also be analyzed in several other ways corresponding to a three-factor structure (Patton et al., 1995; Stanford et al., 2009). However, multiple concerns have been expressed regarding the psychometrics supporting the three-factor structure including in substance abusing populations (e.g., Steinberg, Sharp, Stanford, & Tarp, 2013; Reid, Cyders, Moghaddam, & Fong, 2014). Thus we focus on only the BIS total score in the present study.

Treatment Interventions

All study participants were assigned to an incentive-based smoking-cessation intervention wherein they earned vouchers exchangeable for retail items contingent on abstinence from recent smoking or to a control condition in which vouchers of comparable value were received independent of smoking status. This trial also included a third condition, which was a revised contingent incentives condition that was designed to increase abstinence above levels achieved with the usual incentives condition. Each of these interventions are described below. Note, however, that the outcomes achieved with the usual and revised incentive conditions did not differ and thus were combined for purposes of the present study. Those in the additional two trials included in the larger study sample received only the usual contingent or noncontingent voucher conditions. Additionally, as was noted above, all participants across trials completed the DD task, but only the 118 in the most recent trial (Higgins et al., 2014) completed the BIS. All other study details including recruitment, assessment, and overall treatment condition were consistent across trials..

Usual contingent voucher condition (CV)

Vouchers redeemable for retail items were earned contingent on submitting breath CO specimens ≤ 6 ppm during the initial five days of the cessation effort. Beginning in Week 2, vouchers were delivered contingent on urine-cotinine levels ≤ 80 ng/ml, a criterion that required a longer duration of smoking abstinence than breath CO. Voucher delivery was independent of self-reported smoking status and based exclusively on meeting the biochemical-verification criterion. Vouchers began at $6.25 and escalated by $1.25 per consecutive negative specimen to a maximum of $45.00, where they remained barring positive test results or missed abstinence monitoring visits. Positive test results or missed visits reset the voucher value back to the original low value, but two consecutive negative tests restored the value to the pre-reset level. The incentives intervention was in place from study initiation through 12-weeks postpartum.

Revised contingent voucher condition (RCV)

The same voucher schedule as outlined above was followed in this RCV condition except that potential earnings were rescheduled, moving $300 forward as bonuses that could be earned during Weeks 1–6 by meeting a ≤ 4 ppm breath CO criterion during Week 1, testing cotinine negative at the first urine test on the 2nd Monday of the quit attempt, and thereafter by submitting two cotinine-negative specimens per week through Week 6. More specifically, bonuses earned by reaching a cutoff of ≤ 4 ppm CO during Week 1 started at $18.75 and increased by $3.75 for each successive negative sample reaching a maximum potential bonus of $33.75 for the 5^th consecutive negative specimen meeting the ≤ 4-ppm CO cutoff during Week 1. Women in this condition earned the same incentive as in the CV condition if they met the ≤ 6 ppm CO but not the ≤ 4 ppm cutoff in Week 1. Testing cotinine-negative on the 2nd Monday resulted in an additional bonus of $87.50 above usual CV incentive earnings on that date. Five more bonuses of $15.50 each were available on Thursdays (2^nd test day of each week) during Weeks 2–6 if a woman also tested negative for smoking at the earlier test conducted that same week.

Noncontingent voucher control condition (NCV)

In this condition, vouchers were delivered independent of smoking status. Voucher values were $15.00 per visit antepartum and $20.00 per visit postpartum, values that resulted in payment amounts comparable to average earnings in the CV condition in prior trials (Heil et al., 2008). All else was the same as in the CV and RCV conditions.

Other services

In addition to the interventions, described above, participants in all treatment conditions received usual care for smoking cessation provided through their obstetric clinics, which typically involves provider inquiry regarding smoking status and a discussion of the advantages of quitting during pregnancy. Study staff provided additional cessation counseling to all participants during four visits within two weeks of study entry, at the final antepartum visit, and during three postpartum study visits. For women who quit during pregnancy, brief counseling also occurred during routine smoking-status monitoring visits whenever temptations to smoke were reported. As a counseling guide, we used a printed booklet tailored for pregnant smokers (ACOG, 2001).

Statistical Methods

Regarding abstinence outcomes, we examined late pregnancy and 24-week postpartum 7-day point-prevalence abstinence (i.e., biochemically-verified self-report of no smoking in the 7-days prior to the assessment). Regarding the DD task, Mazur’s (1987) hyperbolic equation was fitted to each subject’s DD data using nonlinear regression. Goodness of fit was evaluated on the basis of model R²s and residual plots. Each subject’s derived discounting parameter (k) was used to compute corresponding log transformed k values to account for the skewed distribution of the discounting score for subsequent analyses as well as ED50 values (1/k). Log k was used in the statistical analyses. ED50 values represent the estimated delay at which the immediate value of the reinforcer was discounted by 50% and provide a more intuitive interpretation of the rate of discounting (Yoon & Higgins, 2008).

The analytic plan was implemented in the following 6 steps for the single trial and multiple trial analyses: (1) Subject characteristics were compared between the two treatment conditions using chi-square tests for categorical measures or t tests for continuous variables. (2) Associations between intake DD log k values, BIS total scores and baseline participant characteristics were conducted using Pearson’s correlation coefficients for continuous measures and t-tests for dichotomous measures. (3) Univariate relationships between baseline characteristics, smoking status (late-pregnancy point-prevalence assessment and 24-week postpartum point-prevalence assessment) and treatment condition, DD and BIS total scores (BIS was included in single trial only) were assessed using logistic regression. Additional models were run to examine the main effects of DD, BIS total scores and treatment group as well as the interaction of both DD and BIS with treatment condition. (4) Next, backward elimination logistic regressions were used to model predictors of smoking status with DD, BIS, treatment condition, DD x treatment and BIS x treatment interaction terms, and all baseline characteristics that were significantly correlated with DD, BIS or smoking status included in the initial step of the modeling procedure. The criterion for retention in the models was set to alpha = 0.05. Separate regressions were conducted to predict 7-day point prevalence abstinence at the late pregnancy and 24-week postpartum assessments. (5) Women were categorized as smokers or abstainers based on their 24-week point-prevalence smoking status, and changes in DD and BIS scores from intake to 24-week postpartum were examined across smoking status groups and within each group using a repeated measures analysis of variance. (6) Steps 1 through 5 were completed in the larger sample (N = 236) to examine the reliability of the findings. All analyses were performed using SAS Version 9 statistical software (SAS Institute, Cary NC). Statistical significance was defined as p < .05.

Results

Participant Characteristics

Participant characteristics for the single trial are shown in Table 1. On average, participants were relatively young and economically disadvantaged, with a mean age of less than 25 years, approximately 19% had less than a high school education, less than 25% had private insurance, and less than 20% were married. There were no significant differences between treatment conditions on any of the baseline socio-demographic or smoking characteristics examined. There were also no significant baseline differences between the two treatment conditions in log k values or BIS total scores. Participant characteristics for the multiple trials are shown in Table 2 and consistent with the single trial there were no significant differences between the two treatment conditions on any of the baseline socio-demographic or smoking characteristics examined. There were also no significant baseline differences in log k values between the two treatment conditions.

Table 1.

Participant Characteristics of Single Trial

	Overall	Contingent (n=79)	Noncontingent (n=39)	p-value
Sociodemographics
Age	24.6 ± 0.5	24.5 ± 0.5	24.7 ± 0.9	.83
% <High School Education	18.8	15.4	25.6	.18
% Caucasian	93.1	94.9	89.5	.28
Weeks Preg at Baseline	10.3 ± 0.4	10.1 ± 0.5	10.7 ± 0.7	.43
% Primagravida	64.1	64.1	64.1	1.00
% Working for Pay	55.1	54.4	56.4	.84
% Private Insurance	23.7	26.6	17.9	.30
% Married	16.1	13.9	20.5	.36
Baseline Delay Discounting
Delayed Discounting	−6.5 ± 0.2	−6.4 ± 0.3	−6.7 ± 0.3	.43
BIS total score	66.5 ± 1.0	66.0 ± 1.2	67.3 ± 1.6	.55
Smoking Characteristics
Age 1st Cigarette	15.5 ± 0.3	15.6 ± 0.4	15.2 ± 0.3	.52
Cigs/day Pre-Preg	18.3 ± 0.8	18.6 ± 0.9	17.8 ± 1.4	.63
Cigs/day at Baseline	8.7 ± 0.6	9.1 ± 0.6	7.8 ± 0.8	.29
NWQ	1.5 ± 0.1	1.5 ± 0.1	1.6 ± 0.1	.67
Quits Attempts During Preg	0.6 ± 0.1	0.7 ± 0.2	0.5 ± 0.2	.33
% History of Quitting Pre-Preg	72.9	78.5	61.5	.05
% Living with Smokers	81.4	83.5	76.9	.38
% No Smoking in Home	61.0	60.8	61.5	.93
% Around None/Few Smokers	27.1	25.3	30.8	.53
Psychiatric Characteristics
Beck Depression Inventory	10.6 ± 0.7	10.4 ± 0.8	10.8 ± 1.1	.79
% History of Depression	41.5	39.2	46.1	.47
Stress Rating (1–10)	5.6 ± 0.2	5.5 ± 0.3	5.8 ± 0.4	.69

Note: Values reported are means ± standard errors unless otherwise specified.

Table 2.

Participant Characteristics of Multiple Trials

	Overall	Contingent (n=137)	Noncontingent (n=99)	p-value
Sociodemographics
Age	24.4 ± 0.3	24.6 ± 0.5	24.1 ± 0.5	.42
% <High School Education	26.2	28.2	23.5	.42
% Caucasian	93.2	93.4	92.9	.87
Weeks Preg at Baseline	10.1 ± 0.3	9.9 ± 0.3	10.5 ± 0.4	.27
% Primagravida	56.2	56.6	55.6	.87
% Working for Pay	49.6	51.1	47.5	.58
% Private Insurance	22.0	25.5	17.2	.13
% Married	16.5	14.6	19.2	.35
Baseline Delay Discounting
Log k	−6.1 ± 0.2	−6.0 ± 0.2	−6.2 ± 0.3	.68
Smoking Characteristics
Age 1st Cigarette	14.7 ± 0.2	14.9 ± 0.3	14.4 ± 0.3	.21
Cigs/day Pre-Preg	18.9 ± 0.5	18.8 ± 0.7	19.0 ± 0.8	.87
Cigs/day at Baseline	9.0 ± 0.4	8.8 ± 0.5	9.2 ± 0.5	.61
NWQ	1.6 ± 0.1	1.6 ± 0.1	1.5 ± 0.1	.54
Quits Attempts During Preg	0.7 ± 0.1	0.8 ± 0.2	0.7 ± 0.2	.78
% History of Quitting Pre-Preg	67.4	69.3	64.6	.45
% Living with Smokers	79.2	78.8	79.8	.86
% No Smoking in Home	48.7	48.2	49.5	.84
% Around None/Few Smokers	21.2	22.6	19.2	.52
Psychiatric Characteristics
Beck Depression Inventory	10.6 ± 0.5	10.4 ± 0.6	10.8 ± 0.7	.68
% History of Depression	36.4	37.2	35.3	.77
Stress Rating (1–10)	5.5 ± 0.2	5.5 ± 0.3	5.6 ± 0.3	.79

Note: Values reported are means ± standard errors unless otherwise specified.

Associations Between Baseline Impulsivity Levels and Other Participant Characteristics

For the single trial, baseline log k was significantly associated with two baseline characteristics (i.e., educational attainment and history of depression) (Table 3). BIS total scores were significantly associated in a positive direction with nicotine withdrawal scores, stress ratings and BDI scores and in a negative direction with age started smoking.

Table 3.

Significant Associations between Delay Discounting, BIS Total Score, and Baseline Characteristics

Single Trial
	Mean log k ± SE	t(df)	p-value
Educational Attainment		−2.24 (112)	.03
less than high school	−5.5 ± 0.5
at least high school	−6.7 ± 0.2
History of Depression		−2.01 (113)	.05
Yes	−6.0 ± 0.3
No	−6.8 ± 0.3
Correlation with BIS total score
Age 1st Cigarette	−0.22		0.02
NWQ	0.27		<.01
Stress Rating (1–10)	0.24		.01
Beck Depression Inventory	0.30		<.01

Multiple Trials
	Mean log k ± SE	t(df)	p-value
Educational Attainment		−2.05 (227)	.04
less than high school	−5.5 ± 0.3
at least high school	−6.2 ± 0.2
Living with Smokers		−2.53 (230)	.01
no	−7.0 ± 0.3
yes	−5.9 ± 0.2
No Smoking in Home		−2.06 (230)	.04
smoking allowed	−5.7 ± 0.3
no smoking allowed	−6.5 ± 0.2

When examined across the multiple trials, baseline log k was significantly associated with three baseline characteristics: educational attainment as in the single trial, as well as living with other smokers, and rules against smoking in the home, which were not significant predictors in the single trial (Table 3). History of depression, which had been a significant predictor in the single trial, was not significant in analyses conducted across the multiple trials.

Univariate Predictors of Smoking Status

Univariate predictors of smoking status at the late-pregnancy and 24-weeks postpartum assessments are shown in Table 4. In the single trial, neither baseline log k nor BIS total scores were significantly associated with smoking status at either of those assessments. Treatment condition was significantly associated with 7-day point prevalence smoking status at the late-pregnancy and 24-week postpartum assessments. Among other potential moderators, late-pregnancy smoking status was significantly associated with three baseline-smoking characteristics (i.e., number of cigarettes smoked per day at baseline, number of quit attempts pre-pregnancy, and number of quit attempts antepartum but prior to treatment entry). Twenty-four week postpartum smoking status was significantly associated with only the number of antepartum quit attempts prior to treatment entry.

Table 4.

Significant Unadjusted Associations of Treatment Condition and Other Participant Characteristics with Point Prevalence Abstinence

Single Trial	Late-Pregnancy			24-weeks Postpartum
	Wald χ2	OR (95% CI)	p-value	Wald χ2	OR (95% CI)	p-value
Treatment C vs. NC	5.69	3.11 (1.22 – 7.91)	.02	1.69	2.40 (0.64 – 8.98)	.19
Cig/Day at Baseline (per 5 cigs)	7.07	0.58 (0.38 – 0.86)	<.01	3.07	0.59 (0.32 – 1.06)	.08
Quits Pre-Pregnancy ≥ 1 vs. 0	3.87	2.70 (1.01 – 7.25)	.05	0.67	1.74 (0.46 – 6.58)	.41
Quits Antepartum	5.51	1.49 (1.07 – 2.09)	.02	4.41	1.39 (1.02 – 1.88)	.04

Multiple Trials	Late-Pregnancy			24-weeks Postpartum
	Wald χ2	OR (95% CI)	p-value	Wald χ2	OR (95% CI)	p-value
Treatment C vs. NC	16.65	4.45 (2.17 – 9.13)	<.001	4.55	3.39 (1.10 – 10.42)	.03
Education < HS vs. ≥ HS	4.69	0.42 (0.19 – 0.92)	.03	0.39	0.70 (0.22 – 2.17)	.53
Age of 1st Cigarette	3.66	1.10 (1.0 – 1.22)	.05	0.96	1.08 (0.93 – 1.25)	.33
Cig/Day Pre-Pregnancy (per 5 cigs)	8.55	0.73 (0.60 – 0.90)	<.01	1.74	0.81 (0.60 – 1.10)	.19
Cig/Day at Baseline (per 5 cigs)	13.35	0.54 (0.39 – 0.75)	<.01	1.13	0.65 (0.40 – 1.05)	.08
Quits Pre-Pregnancy ≥ 1 vs. 0	7.12	2.67 (1.30 – 5.50)	<.01	1.89	2.2 (0.71 – 6.78)	.17
Quits Antepartum	6.13	1.27 (1.05 – 1.52)	.01	0.47	1.06 (0.90 – 1.26)	.49

In analyses conducted across multiple trials, baseline log k was not significantly associated with smoking status at either of those assessments. Treatment condition was again significantly associated with 7-day point prevalence smoking status at the late-pregnancy and 24-week postpartum assessments. Among other potential moderators, late-pregnancy smoking status was significantly associated with educational attainment, which was not significant in the single trial. Late-pregnancy abstinence was also associated with four baseline smoking characteristics that were also significant predictors in the single trial (number of cigarettes per day pre-pregnancy, number of cigarettes smoked per day at baseline, number of quit attempts pre-pregnancy, and number of quit attempts antepartum but prior to treatment entry); it was also associated with one smoking characteristic that was not a predictor in the single trial (age of first cigarette). There were no other significant univariate associations between participant baseline characteristics and late-pregnancy or 24-week postpartum smoking status.

Logistic Regression Analyses Predicting Late-Pregnancy Smoking Status

As described above, backwards elimination logistic regressions were conducted predicting late pregnancy smoking status with DD, BIS (single trial only), treatment condition, DD x treatment and BIS x treatment (single trial only) interaction terms, and all baseline characteristics that were significantly correlated with DD, BIS or smoking status included (Table 5).

Table 5.

Logistic Regressions Predicting Late-Pregnancy Point-Prevalence Abstinence

Single Trial
Final Block	Wald χ2	O.R. (95% C.I.)	p-value
Treatment C vs. NC	6.40	3.77 (1.35 – 10.57)	.01
Cigs/day at Baseline (per 5 cigs)	10.15	0.47 (0.29 – 0.75)	.001
Quits Pre-Pregnancy	5.17	3.72 (1.20 – 11.54)	.02
NWQ	5.28	1.96 (1.10 – 3.49)	.02

Multiple Trials
Final Block	Wald χ2	O.R. (95% C.I.)	p-value
Treatment C vs. NC	14.82	4.45 (2.08 – 9.52)	<.001
Cigs/day at Baseline (per 5 cigs)	15.49	0.47 (0.33 – 0.69)	<.001
Quits Pre-Pregnancy	5.36	2.54 (1.15 – 5.61)	.02
Education less than HS	4.27	0.41 (0.17 – 0.96)	.04

Within the single trial, neither DD nor BIS were retained in the model as main effects or as interaction terms with treatment. Significant predictors that were retained in the model included treatment condition (χ² = 6.40, O.R. = 3.77 (1.35 – 10.57), p =.01), the number of cigarettes smoked per day at baseline (χ² = 10.15, O.R. = 0.47 (0.29 – 0.75), p =.001), a history of quit attempts pre-pregnancy (χ² = 5.17, O.R. = 3.72 (1.20 – 11.54), p =.02), and nicotine withdrawal scores (χ² = 5.28, O.R. = 1.96 (1.10 – 3.49), p =.02). Concordance between the predicted probabilities and late-pregnancy smoking status for this model was C = 0.79, which indicates reasonable concordance (i.e., models are typically considered reasonable when the C-statistic is higher than 0.7 and strong when C exceeds 0.8 (Hosmer & Lemeshow, 2000; Hosmer & Lemeshow, 1989).

The model based on multiple trials did not retain DD as a main effect or interaction term with treatment consistent with results from the single trial. Also consistent with results from the model based on the single trial was retention of treatment condition (χ² = 14.82, O.R. = 4.45 (2.08 – 9.52), p < .001), the number of cigarettes smoked per day at baseline (χ² = 15.49, O.R. = 0.47 (0.33 – 0.69), p < .001), and a history of quit attempts pre-pregnancy (χ² = 5.36, O.R. = 2.54 (1.15 – 5.61), p = .02) as significant predictors in the model. Educational attainment was also a significant predictor in this model based on multiple trials (χ² = 4.27, O.R. = 0.41 (0.17 – 0.96), p = .04), where it had not been in the single trial. Note that nicotine withdrawal, which had been included in the final model based on the single trial, was not a significant predictor in this model based on analyses across multiple trials. Concordance between the predicted probabilities and late-pregnancy smoking status for this model was C = 0.78, which was almost identical with the 0.79 score for the model based on the single trial.

Logistic Regression Analyses Predicting 24-week Postpartum Smoking Status

The number of antepartum quit attempts prior to treatment entry was the only predictor retained in the backwards elimination logistic regression model for the single trial (χ² = 4.41, O.R. = 1.39 (1.02 – 1.88), p = .04; Table 6). The concordance for the model was relatively weak (C = 0.61).

Table 6.

Logistic Regressions Predicting 24-week Postpartum Point-Prevalence Abstinence

Single Trial
Final Block	Wald χ2	O.R. (95% C.I.)	p-value
Quits Antepartum	4.41	1.39 (1.02 – 1.88)	.04

Multiple Trials
Final Block	Wald χ2	O.R. (95% C.I.)	p-value
Treatment C vs. NC	4.55	3.39 (1.05 – 10.42)	.03

Within the analyses based on multiple trials, treatment condition was the only predictor retained in the final backwards elimination logistic regression model (χ² = 4.55, O.R. = 3.39 (1.05 – 10.42), p = .03; Table 6). Log k was not retained as a main effect or an interaction term with treatment. The concordance for the model was again relatively weak (C = 0.63).

Changes in Impulsivity Over Time

There were no significant changes in log k from the intake to 24-week postpartum assessments when examining all participants in the single trial independent of smoking status (F(1,79)=0.08, p =.78). The same was true within each smoking status group. Among women who were non-smokers at 24-weeks postpartum, mean intake log k was −6.70 (median ED50 = 2.7) while mean 24-week postpartum log k was −6.93 (median ED50 = 1.6) (F(1,79) = 0.10, p = .76). Among those who were smokers, mean intake log k was −6.61 (median ED50 = 2.1) and mean 24-weeks postpartum log k was −6.62 (ED50 = 1.3) (F(1,79) = .01, p = .99).

The same analyses were conducted examining changes in DD over time in the multiple trials. There were no significant changes in log k from the intake to 24-week postpartum assessments when examining all participants independent of smoking status (F(1,136)=0.29, p=.59). The same was true within each smoking status group. Among women who were non-smokers at 24-weeks postpartum, mean intake log k was −6.70 (median ED50 = 1.6) while mean 24-week postpartum log k was −7.32 (median ED50 = 3.6) (F(1,136)=0.87, p=.35). Among those who were smokers, mean intake log k was −6.30 (median ED50 = 1.5) and mean 24-weeks postpartum log k was −6.07 (ED50 = 1.0) (F(1,136)=.91, p=.34).

There was a significant change in BIS total scores independent of smoking status in the single trial (F(1,83)=5.56, p=.02). The mean BIS was 66.5 at intake and 63.1 at 24-weeks postpartum. This trend was observed in both smoking status groups. Among women who were non-smokers at 24-weeks postpartum, mean intake BIS was 66.9 while mean 24-week postpartum BIS was 62.4 (F(1,83)=2.82, p=.10). Among those who were smokers, mean intake BIS was −66.1 and mean 24-weeks postpartum BIS was 63.8 (F(1,83)=4.30, p=.04).

Discussion

The primary purpose of the present study was to thoroughly examine whether response to this incentive-based treatment for smoking cessation during pregnancy and early postpartum was moderated by individual differences in impulsiveness. Impulsiveness was operationalized as DD or BIS scores in the single trial two commonly used measures of the impulsivity construct in substance abuse and other health research (Bickel et al., 2012; Reid et al., 2012). In analyses across trials, impulsiveness was operationalized as DD. We chose to examine this question as part of a single prospective, randomized clinical trial to bypass the limitations of retrospective studies. We included the analyses based on multiple trials to reduce the likelihood of false-negative results related to a small sample size and of false positive results by including a check on the generality of study findings across the single trial and this larger data set. We saw no evidence that individual differences in impulsiveness as measured by either DD or BIS was a significant moderator of smoking cessation outcomes in the single prospective trial or when we examined DD across multiple trials. That is, there was no evidence of a significant main effect of either impulsiveness measure or interactions of them with treatment condition in predicting antepartum or postpartum smoking status.

One possible explanation for these negative results is that there may not have been sufficient between-subject variability in discounting to discern a moderating influence. In considering that possibility we compared discounting data from the larger data set in the present study with data from two prior studies that we reported wherein significant associations between DD and drug abstinence were observed (Washio et al., 2011; White et al., 2014). Between-subject variance across the three studies appears to be quite comparable, with overall median log k and interquartile ranges in the present, Washio et al. and White et al studies being −6.19 and 3.23, −5.49 and 3.58, and −6.52 and 3.00, respectively. A more likely explanation for the negative results may be found in pre-pregnancy smoking rates. In the White et al. study on spontaneous quitting among pregnant smokers, DD interacted with pre-pregnancy smoking rate in predicting spontaneous quitting among pregnant women, with greater discounting predicting lower odds of quitting among women with smoking rates less than 10 cigs/day per day but above that rate the odds of quitting were uniformly low across DD levels (White et al., 2014). On average, women in the present study reported smoking almost 19 cigs/day pre-pregnancy, a rate at which DD had minimal association with spontaneous quitting. To the extent that those results with spontaneous quitting extend to treatment-assisted quitting, the relatively high rates of pre-pregnancy smoking rates in the present study likely precluded discerning any significant association between discounting rates and the likelihood of quitting antepartum or postpartum.

Turning to what did predict abstinence from smoking in the present study, the results are strongest for predicting late-pregnancy abstinence. Three variables predicted across the single and multiple trials: being assigned to the incentive-based intervention, having lower baseline smoking rates, and having tried to quit smoking prior to the current pregnancy, with the model fit of these predictors being reasonably strong. Having less than 12 years of education was a predictor in analyses based on multiple trials, but not the single trial. Each of those four predictors is consistent with results from earlier examinations of predictors of treatment response in this population (Higgins et al., 2009). Nicotine withdrawal, by contrast, was a predictor in the single trial in the present study and in an earlier report of predictors of response to this treatment model (Higgins et al., 2009), but its failure to be retained in the final model collapsing across multiple trials in the present study underscores that the strength of its association with quitting is weaker than the others mentioned above and may or may not be included in predictive models depending on the strength of the others in a particular sample.

Regarding 24-week postpartum abstinence, our ability to predict was relatively weak and certainly weaker than late-pregnancy outcomes. Having made one or more quit attempts during the current pregnancy but prior to entering treatment was a positive predictor in the single trial but not across multiple trials and having received the incentives intervention was the single significant predictor in analyses based on multiple trials. Clearly the association of prior quit attempts with success in sustaining abstinence through 24-weeks was not sufficiently robust to be retained in the predictive model based on multiple trials, but, as suggested further below, encouraging women to keep trying to quit before and during pregnancy has no evidence of adverse effects associated with it that we know about and is a prudent course of action with any smoker. The association of the incentives intervention with 24-week abstinence, twelve weeks after incentives were discontinued, is a reliable finding (see Higgins, Washio, Heil, et al., 2014). The absolute abstinence levels are modest (~15%) although certainly superior to controls (~2–3%) and thus whether statistical significance is observed can be expected to vary depending on sample size. We have previously reported other positive postpartum outcomes among women who receive the incentives intervention, including increased breastfeeding duration (Higgins, Higgins, et al., 2010) and lower rates of depressive symptomatology among women at risk for postpartum depression (Lopez et al., in press). An increased likelihood of sustaining abstinence longer term can be considered another important benefit associated with this treatment model.

Because the women included in the present study were all from randomized controlled trials involving common treatment conditions, we can infer a causal relationship of treatment and its effects on antepartum and postpartum smoking status. Of course, we cannot similarly infer causality between the associations of baseline smoking rate, histories of prior quit attempts, or educational attainment with smoking status in the present study. That said, we see little downside to using those associations to further bolster the rationale for encouraging child-bearing aged girls and women who do smoke to reduce their smoking rates as low as possible and to make attempts to quit smoking prior to and during pregnancy and to keep trying despite prior failure. Aside from the influence of treatment, we cannot assume that any of these practices will directly lower the likelihood of smoking during pregnancy, but each is reliably and robustly associated with a decreased likelihood of smoking through a pregnancy (Higgins et al., 2009; Kandel et al., 2009). We know of no evidence of any of these practices being associated with adverse outcomes nor do we envision their inclusion in comprehensive smoking prevention strategies targeting women somehow coopting resources that could be better directed at practices with stronger empirical support (Higgins et al., 2009). Regarding improving outcomes with women who present for treatment with characteristics predicting poor response to this incentives-based treatment model, the utility of higher value incentives and the possibility of combining incentives with nicotine replacement therapy or other pharmacotherapies merit consideration (Higgins, Washio et al., 2012).

In contrast to a prior report indicating that sustained abstinence reduces discounting in a general population of smokers (Secades-Villa et al., 2014), we saw no discernible changes in DD from intake to end of study among abstainers or smokers in the present study. Women in the present study typically entered treatment at approximately 10-weeks gestation and most who responded to the intervention would have quit smoking within the first two weeks of treatment (Higgins et al., 2006) meaning that many would have been abstinent from smoking for almost 1 year at the time of the 24-week postpartum assessment. That is comparable to the period of sustained abstinence in the Secades-Villa et al. (2014) study. Yet this length of abstinence was not associated with a significant reduction in DD values in the present study nor in BIS scores, which decreased over time but independent of smoking status. Important to note is that participants in the Secades-Villa et al. study were assessed 1-year post-treatment compared to only 3-months post-treatment among participants in the present study. Women in the present study also experienced many unique changes associated with pregnancy and early postpartum in addition to smoking status, such as hormonal fluctuations and stress surrounding the pregnancy and postpartum period, which could have contributed to the different results observed. Further research will be necessary to more fully characterize the extent to which sustained smoking abstinence is associated with reductions in DD and individual differences in who may experience that effect.

The present study has several limitations that merit mention, including the exclusive use of women from a small metropolitan area with an almost exclusively Caucasian population, and a sample comprised of women willing to participate in treatment-outcomes studies. How well this incentivized treatment model and results from the present study on predictors of a positive treatment outcome with this approach generalize to more diverse samples and other treatments and settings is largely an unanswered question although controlled trials conducted in at least one other U.S. state (Oregon) were positive (Donatelle et al., 2004) and an effectiveness study in at least one other country (Scotland) resulted in positive outcomes (Radley et al., 2013). Considering the broad generality that has been observed with the use of financial incentives to decrease use of other substances (Lussier et al., 2006) as well as other health-related risk behaviors (Higgins, Silverman, et al., 2012) and because these incentives interventions are based on the fundamental behavioral science principle of reinforcement (Higgins et al., 2004), we are optimistic that this strategy for reducing smoking and associated relationships will have generality to diverse samples and settings. There have been encouraging developments around the use of this approach with opioid-dependent pregnant women, which is a particularly challenging subgroup of smokers for whom effective treatments are sorely needed (Tuten et al., 2012). Another important future challenge in this research effort is getting a larger proportion of women to respond. We hope that increasing understanding about who is and is not currently benefitting from the intervention as was done in the present study will facilitate achievement of that goal.