Differential Mechanisms of Change in Motivational Interviewing versus Health Education for Smoking Cessation Induction

Abstract

Objective:

To determine if Motivational Interviewing (MI) versus health education (HE) elicited different types of client language and whether these differences were associated with outcomes in a randomized clinical trial for cessation induction among people who smoke with low motivation to quit.

Methods:

A secondary data analysis was conducted using data from the MI and HE arms of a trial in which people who smoke (N=202) with low desire to quit were randomly assigned to four sessions of MI, HE or brief advice. Mediation analyses examined two types of client language: change talk (CT) and a novel form of client speech called “learning talk” (LT). Outcomes were assessed at baseline, 3- and 6-months.

Results:

With HE as the reference group, MI resulted in greater change talk (OR=3.0, 95% CI 1.7–5.5) which was associated with better outcomes (average d=0.34, SD=0.13) and HE resulted in greater learning talk (OR = 0.05, 95% CI 0.02–0.10) which was also associated with better outcomes (average d=0.42, SD=0.08). Indirect parallel mediation effects on quit attempts were significant for both MI-CT (OR=1.4, 95% CI 1.1–1.7) and HE-LT (OR=0.4, 95% CI 0.2–0.7).

Conclusions:

MI and HE were both efficacious via different pathways to change, confirming the utility of MI in this RCT as well as highlighting the potential of HE based on the “5 R’s” for smoking cessation. These findings emphasize the value of exploring theorized mechanisms of action of interventions evaluated in RCTs.

Given the costs of randomized clinical trials, it is crucial to extract as much information as possible via secondary analyses of existing RCTs. In particular, there is growing focus on assessing how treatment works (Longabaugh & Magill, 2011; Orford, 2008), including situations where a treatment fails to demonstrate efficacy (Klemperer, Hughes, Callas, & Solomon, 2017b), or to explore how therapies based on different theoretical frameworks show comparable efficacy (Gaume, Heather, Tober, & McCambridge, 2018). Under such situations, there is value in exploring mediation effects (i.e., indirect effect) despite an absence of significant differences due to treatment (O’Rourke & MacKinnon, 2018). Furthermore, it is important to disentangle whether a lack of differential effectiveness across active therapies is due to factors common across treatments or due to different mechanisms that result in equivalent outcomes. Understanding how treatments lead to behavior change is crucial to refining and improving interventions and may also suggest opportunities for matching patients to treatments. This report focuses on mechanisms of change in a randomized clinical trial that evaluated the efficacy of motivational interviewing (MI) compared to a health education (HE) control condition for cessation induction in people who smoke tobacco.

Motivational Interviewing (MI) has demonstrated efficacy for a range of health behavior outcomes, including smoking cessation (Hettema & Hendricks, 2010; Lundahl et al., 2013), where it has been incorporated into practice guidelines (Fiore et al., 2008; US Dept of Health and Human Services, 2020). However, there is marked variability in MI outcomes across studies (Miller & Rose, 2009), and this has spurred a growing body of research examining mediators of MI’s efficacy (Pace et al., 2017). One mechanism that has received considerable attention involves the elicitation of client language in support of behavior change (i.e., change talk) over language that is in opposition to change (i.e., sustain talk). Furthermore, client change talk may be an active mechanism of change across other therapeutic approaches (e.g., cognitive behavioral therapy; Aharonovich, Amrhein, Bisaga, Nunes, & Hasin, 2008) and across other patient conditions (e.g., anxiety; Lombardi, Button, & Westra, 2014). Additionally, MI-consistent therapist behavior leads to increases in change talk (Magill et al., 2018), including in studies of smoking cessation treatment (Boardman, Catley, Grobe, Little, & Ahluwalia, 2006; Catley et al., 2006).

Based on an initial study of commitment language during MI counseling and alcohol use outcomes, MI process researchers have typically examined the frequency of the occurrence of specific types of change talk—statements of desire, reasons, need, ability, commitment, and taking steps to change (Amrhein et al, 2003). Change talk research also encompasses language that expresses opposition to making behavior change, which in MI is referred to as “sustain talk”. Studies examining links between change talk and behavior change have yielded mixed results; however, meta-analyses have consistently indicated that sustain talk or the proportion of change talk to sustain talk are significantly related to outcomes, with sustain talk being a negative predictor (Magill et al., 2014; Pace et al., 2017). Another meta-analysis explored the effect of the different subtypes of change talk in studies of addictive behavior and found that none of the aforementioned subtypes were positive predictors of behavior change outcomes (Magill et al., 2019). However, Magill et al. (2014) also noted heterogeneity in the pooled change talk effect sizes and the study of change talk subtypes was limited by the preponderance of older adolescent or younger adult alcohol users in samples. There is need for additional research with respect to these mechanisms, particularly for smoking cessation interventions.

Because MI had only demonstrated efficacy for smoking cessation when compared to brief advice or other less intensive treatments, the parent trial for this study aimed to compare MI to a matched intensity alternative (HE) to confirm that MI’s efficacy in prior studies was not merely due to greater intervention time. A lower intensity brief advice intervention arm was also included. Also novel to this trial was the focus on participants who were not ready or motivated to quit smoking and a primary outcome of quit attempts. Analysis using the MI Treatment Integrity (MITI) Code (Miller, Yahne, Moyers, Martinez, & Pirritano, 2004) demonstrated that MI was delivered with fidelity and in a manner that was distinctly (and appropriately) different from the HE comparison (Catley et al., 2016). Trial outcome analysis unexpectedly revealed no significant differences between groups in the proportion who made a quit attempt by the 6-month follow up (MI group 52%; HE group 61%, Brief Advice group 45%). Furthermore, HE had significantly higher biochemically verified abstinence rates at 6 months (7.8%) than brief advice (0.0%) (8% risk difference, 95% CI=3%–13%, p=0.003), with the MI group falling in between (2.9% abstinent, 3% risk difference, 95% CI=0%–6%, p=0.079). Across all outcomes there was a consistent pattern in which HE had the best outcome, followed by MI, and then Brief Advice, although the only difference between HE and MI that was significant was motivation to quit (Cohen’s d=0.36, 95% CI=0.12–0.60).

In clinical trials, treatments can fail for different reasons: they can fail to impact the theorized mechanism of action, change in the mechanism of action can fail to impact the outcome, or the control condition can be effective through the same or a different mechanism of action. In the case of MI, it was important to ascertain whether MI elicited client language consistent with MI theory. In this study, the focus was on the elicitation of any language indicating a desire to change (“desire change talk”) because experience conducting sessions suggested this was a clinically meaningful milestone for people who smoke who were all particularly low in motivation to quit [mean motivation on a 0 to 10 point scale was 1.9 (SD = 1.9)]. For counselors, this milestone served as a transition point where the counselor could focus on barriers and facilitators to change rather than whether change was a desirable outcome. For disentangling the effects of the two treatments it was also important to examine the degree to which the elicitation of desire change talk (CT) was specific to MI.

While it is possible that change talk mediated the efficacy of HE, it was also possible that a different mechanism might have been in play with HE. The HE intervention was based on the “5 Rs” framework—Repeatedly discuss Relevant Risks of smoking, Rewards of cessation, and Roadblocks to quitting (AHRQ, 2012; Fiore et al., 2008)—which has previously been found to be effective (Carpenter, Hughes, Solomon, & Callas, 2004), though mechanistic research is lacking (US Dept of Health and Human Services, 2020). However, we hypothesized that HE might work by increasing client learning. The lack of client language measures for health education interventions necessitated the development of a novel coding system that would parallel the coding of change talk. We created a coding system for learning talk (LT) based on an inductive approach, not tied to any specific theory of behavioral change. LT encompassed client statements in response to information provided that indicated interest in gaining more information, realization or gain of knowledge, and the information was being absorbed or reflected upon.

In summary, this report used differential mediation analyses to test for two distinct mechanisms of change across MI vs HE interventions (see Figure 1). It was hypothesized that the two mediators (CT & LT) would be differentially influenced by MI vs HE treatment. Relative to HE, MI was expected to result in more desire CT, which would subsequently be related to tobacco cessation related outcomes (i.e., quit attempts, cessation, medication use, motivation to quit, and confidence to quit). In contrast, HE was expected to result in more LT, which would also be related to these outcomes.

Figure 1.

Parallel mediation - observed variable model. Solid lines denote significant links as assessed by structural equation modeling (package lavaan in R). HE = Health Education. MI = Motivational Interviewing. N=100 (HE) & 102 (MI). CI = bootstrapped confidence interval.

Method

Design and Participants

Data were obtained from a completed randomized clinical trial comparing MI to HE, and both to brief advice for smoking cessation induction in people who smoke with low desire to quit (Catley et al., 2016). Full details of the methods are provided elsewhere (Catley et al., 2012); they complied with the Helsinki Declaration and were reviewed and approved by the Institutional Review Board of the University of Missouri Kansas City (#0978). In this study, only data from the two active intervention arms were used because the brief advice arm did not have comparable dialogue between counselor and client amenable to coding. Adults who smoked cigarettes daily (expired-air carbon monoxide ≥ 7ppm) with no immediate cessation plans along with a low motivation to quit (≤ 6 on 0–10 self-report scale) were eligible to participate. Recruitment took place in a large Midwestern city (2010–2011) via various methods that advertised for “smokers not ready to quit” (for details see Harris et al., 2016). Two HE participants did not have audio-taped recordings of sessions, resulting in analysis of 102 MI and 100 HE participant audio transcripts for these analyses. Mean age for the 202 participants was 46.0 (SD=11.0) years, with 43.6% being female. Most participants were non-white (65.8% Black, 5.0% other/multiple races), non-Hispanic (98.0%), with lower income (75% ≤ $2,000 per month). Average number of years smoked was 29.8 years (SD=11.9) and average cigarettes smoked per day were 16.5 (SD=8.5).

Procedure

During recruitment, participants were told that a desire to quit smoking was not a requirement and that the study was about how healthcare providers should communicate with patients about smoking. Recruits who were highly motivated or ready to quit smoking at baseline were not eligible for the study. After obtaining written consent, baseline measures were collected and participants were computer-randomized to group. Those assigned to active treatments were slated for 4 intervention sessions of either MI or HE, with the first in-person session immediately after baseline assessment. Participants were then scheduled to receive a phone session around week 6, an in-person session around week 12, and a final phone session at week 18, with scheduling adjusted if a participant set a quit date. Follow-up assessments were completed at 3 and 6 months after baseline.

Interventions

Motivational Interviewing.

Counselors were trained (by authors D.C. & K.R.) to deliver MI as described by Miller and Rollnick (2013). Counselors worked to maintain a collaborative, empathic, and autonomy-supportive relationship and targeted the discrepancy between current smoking behaviors and the client’s personal health goals. Counselors endeavored to elicit “change talk” in support of quitting using affirmations, open-ended questions, reflections and summaries while avoiding the encouragement of “sustain talk” (statements opposed to quitting smoking) and confronting or challenging participants about their smoking. Providing unsolicited advice was avoided, however, when requested or judged necessary information was provided in an MI-consistent style (using the “elicit, provide, elicit” strategy).

Health Education.

The HE intervention was designed to be an intensive active intervention comparable to MI. Counselors were trained (by authors D.C., K.G. and K.P.R.) to provide semi-scripted education about the risks of smoking, benefits of quitting, and strategies for cessation. Counselors maintained an educational focus with an “advice-oriented” style and incorporated the “5 R’s” based on the U.S. Clinical Practice Guideline (Fiore et al., 2008). Sessions followed a structured protocol which included assessment of smoking history and personalized recommendations. Counselor-participant interactions were constrained to providing clarifying information about the information provided and scripted brief responses regarding common barriers to quitting that participants might raise.

Assistance to quit.

In both interventions, any participant who expressed a desire to quit smoking during a session was provided with a self-help brochure and assisted to formulate a cessation plan. Counselors maintained the relevant MI or HE communication style and encouraged the use of empirically supported behavioral strategies and pharmacotherapy (varenicline, nicotine patch or lozenge) which was offered without charge. The availability of free medications was not mentioned until the participant committed to quitting by setting a quit date.

Therapists & Intervention Fidelity.

Three female counselors with master’s-level training and experience in MI delivered both interventions to avoid confounding treatment versus therapist effects. Counselors were trained on study-specific treatment protocols and training continued until counselors met fidelity criteria for three consecutive sessions. Throughout the study counselors received regular group supervision of a randomly selected recent audio recording from separate expert clinicians for each of the interventions (weekly for MI, every other week for HE). Supervisors completed study-specific rating scales verifying that procedures and content were completed as required and that the counseling style was adherent to either MI or HE. To verify fidelity, 10% of sessions (38 MI, 27 HE) were assessed by independent expert coders blind to study hypotheses and treatment assignment using the MI Treatment Integrity Code (Miller et al., 2004). Results, presented elsewhere, confirmed that MI was delivered with fidelity and was rated higher on empathy, collaboration, autonomy support, reflection to questions ratio, and lower on giving information than HE (Catley et al., 2016).

Measures

Participant characteristics included self-reported race, ethnicity, gender, age, socioeconomic status, and smoking history. The Severity of Dependence Scale (Grassi, Ferketich, Enea, Culasso, & Nencini, 2014) and the Heaviness of Smoking Index (Etter, Duc, & Perneger, 1999) assessed tobacco dependence.

The first counseling session for each participant was coded for CT and LT in separate passes by independent coders based on audio recordings. Due to resource constraints, only one session was coded for each participant. The first session was selected to maximize the number of participants that could be included and because we expected it to reveal important clinical indicators of the participants likely trajectory in the study. Change talk coding was based on the Motivational Interviewing Skill Code (MISC 2.1), with a focus on “desire” change talk (e.g., “I want to quit”). Desire change talk was the focus because the elicitation of desire change was a therapeutic milestone indicating the counselor could shift from the focus from whether to quit to how to quit. Learning talk was coded using a novel system developed for this project (see Supplemental Materials). Positive occurrences included statements of interest in receiving information or clarification (e.g., “So, is that why it is so hard for people to quit?”), statements indicating realization or surprise (e.g., “ I didn’t know a lot of these things were wrong”), and statements indicating information was being absorbed, or reflected upon (e.g., “I’m just taking in all this information”). As with desire change talk, only learning language that was relevant to the target behavior of smoking cessation was coded as positive. A subset of sessions was coded by multiple raters for reliability analyses. To make it feasible to code all participants in the trial we also reduced the coding burden by coding only the first instance, if present, of desire CT or LT. Therefore, each session was coded based on the occurrence or absence of any CT or LT and analyzed as “yes” or “no” at the level of the participant.

Outcomes aggregated across the 6-month follow-up included self-report of any serious quit attempt for at least 24 hours, any cessation medication use assessed with a self-report checklist of pharmacotherapies, and self-reported 7-day point-prevalence abstinence at last follow-up (each coded yes/no). Supplemental outcome measures included motivation and confidence to quit smoking averaged across the 3-month and 6-month follow-ups. Motivation was assessed with an aggregate of three items: motivation to quit (0, not at all; 10, extremely); motivation to quit in the next 2 weeks (0, not at all; 10, extremely); and the Contemplation Ladder (0, no thought of quitting; 10, taking action to quit; Biener & Abrams, 1991). Confidence to quit was assessed by aggregating two self-report items: confidence to quit and confidence to quit in the next 2 weeks (0, not at all; 10, extremely).

Data Analysis

Descriptive information about this sample has been published previously (Catley et al., 2016). All statistical analyses were conducted using R version 3.6.0 (see Supplemental Materials). Preliminary analyses assessed the reliability of coding for each of the client language measures by calculating the kappa across multiple coders. The effect of treatment assignment on the two client language measures was analyzed with logistic regression. Associations between client language and tobacco quitting outcomes were assessed using generalized linear modeling including both change talk (CT) and learning talk (LT) in the model. Binary outcomes (e.g., quit attempt) were analyzed with the binomial family - logit link whereas continuous outcomes (e.g., motivation to quit) were analyzed with the Gaussian family - identity link.

Mediation analyses focused on determining if there were two distinct pathways to quitting behavior that corresponded to each of the two active treatments (Figure 1). Because both MI and HE were efficacious compared to brief advice, the observed total effect of treatment on outcome was not relevant in these analyses where there was a lack of significant differences between the active treatments (O’Rourke & MacKinnon, 2018). In this context, the analysis was focused on estimating the magnitude and significance of mediation effects, or the efficacy of each treatment transmitted by CT or LT (Borsari et al., 2015; Feingold, MacKinnon, & Capaldi, 2018; Pek & Hoyle, 2016). One path consists of Treatment (MI) leading to CT leading to outcome (MI-CT-O) and the other of Treatment (HE) leading to LT leading to outcome (HE-LT-O). Because both treatments were believed to be active, focus was only on the coefficients associated with the indirect paths that would indicate evidence of distinct and parallel pathways. Given that treatment was coded as 0 or 1 (for HE or MI respectively), it was hypothesized that the indirect path estimates for CT vs LT would be opposing in sign (e.g., less vs greater than 1 for odds ratios, + vs – for proportions). Observed variable structural equation modeling (SEM) analyzed multiple mediators simultaneously, using the lavaan package (version 0.6–5), which accommodates binary endogenous variables (e.g., mediators) using diagonally weighted least squares estimation (Rosseel, 2012). Because the mediators are dichotomous measures, estimation for missing data via maximum likelihood was not available and cases without follow-up values are dropped. Robust confidence intervals for the product of coefficients representing each path were calculated with a bootstrap method using 10,000 samples (Hayes, 2018).

Other approaches to mediation analysis have been recommended in the literature because of limitations with the linear method (Imai, Keele, & Tingley, 2010). Therefore, data were also analyzed using the potential outcomes framework, which uses counterfactual definitions to establish a formal framework for causal inference (e.g., what would have happened if someone in the MI group had actually been in the HE group?). The total treatment effect is then decomposed into direct and indirect effects (i.e., mediated effects; for additional explanation see Imai, Keele, Tingley, & Yamamoto 2011). Because this approach is not tied to a specific statistical model, it can easily accommodate linear and nonlinear relationships and may yield results different from SEM, especially when mediators are dichotomous (Rijnhart, Twisk, Eekhout, & Heymans, 2019). These additional analyses were done, in part, to assess of the robustness of findings across analytic frameworks. These analyses were done using the mediation package in R (version 4.4.7), which accommodates binary endogenous variables and yields a metric scaled in terms of changes in probability (Tingley, Yamamoto, Hirose, Keele, & Imai, 2014). The main statistic of interest is the measure of indirect effects, or Average Causal Mediation Effects (ACME) for CT and LT. Additional sensitivity analyses included controlling for baseline covariates (nicotine dependence, motivation to quit, race), counselor effects, and treating missing outcomes as cessation not attempted.

Results

Client Language

Analyses of CT and LT scorings (Yes=1, No=0) by two coders (n=50 for LT and n=37 for CT) suggested good measurement reliability with Kappas of 0.80 for LT and 0.94 for CT (p<0.001) and intra-class correlations of 0.80 (LT, 95% CI 0.67–0.88) and 0.94 (CT, 95% CI 0.89–0.97). CT occurred in 83 participants (41.1%) and LT occurred in 105 participants (52.0%). The two measures did not appear to be significantly correlated (Χ² = 0.11, p>0.7). For example, for the 83 participants where CT occurred, 41 (49.3%) had no LT and 42 (50.6%) had LT as well.

As expected, treatment type was associated with which type of client language was more likely to be observed during the session (Figure 2A): CT was more likely to occur in the MI sessions relative to HE (53.9% vs 28.0%; OR=3.0, 95% CI 1.7–5.5, p<0.001) while LT was more likely to occur in HE sessions relative to MI (84.0% vs 20.6%; OR = 0.05, 95% CI 0.02–0.10, p<0.001). Note that opposing ranges for the odds ratios is due to the coding of treatment = 1 for MI and 0 for HE, even though both treatments are active; flipping the code for HE results in an equivalent OR=20.0. Thus, the treatments differed robustly in the type of client language elicited, which provides support for two distinct a-paths causally flowing from treatment assignment.

Figure 2.

Frequency of client language occurrence across treatments (2A) and percentage making a quit attempt based on occurrence of change talk X learning talk (2B) with the proportion in Motivational Interviewing (MI) versus Health Education (HE) denoted by the stacked columns.

Examination of the link between the client language and outcomes revealed significant predictive relationships for both CT & LT, even when controlling for the effects of the other language variable. As shown in Table 1, CT was significantly predictive of quit attempts, cessation medication use, and motivation and confidence to quit (all p<0.05). For example, the percentage of participants making a quit attempt was 72% among those with CT versus 51% among those with no CT (RR = 1.44, 95% CI 1.2–1.8, p<0.002). Similarly, LT was predictive of quit attempts, medication use, and motivation and confidence to quit, but LT was also predictive of abstinence. For example, the percentage of participants making a quit attempt was 72% among those with LT versus 49% among those with no LT (RR = 1.48, 95% CI 1.2–1.9, p<.002). Effect sizes across the outcomes averaged d=0.34 (SD=0.13) for CT and d=0.42 (SD=0.08) for LT. For participants where both CT and LT occurred (n=40), the percentage making a quit attempt was 82% (SD=40), compared to 65% (SD=50) in cases where just one of the two occurred, and to 36% (SD=50) in cases when neither occurred (Figure 2B). Using the situation where neither client language occurred as the reference, the odds ratio when just one of the two occurred was 3.3 (95% CI 1.7–6.9) and 8.4 (95% CI 3.2–24.3) when both occurred. There was a lack of a significant CT by LT interaction effect across the outcome measures, suggesting that a simple additive model was sufficient to describe these effects. Thus, subsequent modeling focused on the parallel path effects of each of these variables, without the inclusion of a CT by LT interaction term.

Table 1.

Relationship Between In-Session Client Language and Quitting Outcomes Aggregated Across 3 & 6 Month Follow-Up

	Change Talk (CT)				Learning Talk (LT)
	No	Yes		p	No	Yes		p
	M ± SD	M ± SD	OR	[95% CI]	M ± SD	M ± SD	OR	[95% CI]
Binary Outcomes
Quit Attempts a	0.51 ± 0.048	0.74 ± 0.050	3.00	0.001 [1.55,5.81]	0.49 ± 0.053	0.72 ± 0.046	2.96	0.001 [1.57,5.59]
Cessation Medication Use a	0.24 ± 0.041	0.43 ± 0.056	2.42	0.007 [1.27,4.61]	0.24 ± 0.046	0.39 ± 0.050	2.11	0.025 [1.10,4.06]
Abstinence (Self Report) b	0.09 ± 0.028	0.14 ± 0.040	1.70	0.267 [0.67,4.32]	0.05 ± 0.023	0.18 ± 0.039	4.60	0.008 [1.48,14.33]
				p				p
Continuous Changes Scores			d	[95%CI]			d	[95%CI]
Motivation to Quit	3.09 ± 3.02	3.87 ±2.64	0.28	0.046 [0.01,0.56]	2.66 ± 2.70	4.10 ± 2.90	0.50	0.001 [0.23,0.78]
Confidence to Quit	2.52 ± 3.14	3.63 ±3.23	0.36	0.014 [0.08,0.64]	2.40 ± 3.01	3.50 ± 3.32	0.35	0.013 [0.08,0.63]

Note: Significant client language effects denoted with bold face (p<0.05). M = mean, SD = standard deviation, OR = odds ratio, CI = confidence intervals, d = standardized mean difference = standardized beta coefficient for the multivariate model: outcome ~ change talk + learn talk. CT & LT coded (0,1) and analyzed using generalized linear modeling, function glm(), in the R statistical platform. Measures defined in the text.

^aAny quit attempt made during follow-up, n= 107 & 78 for CT and 88 & 97 for LT.

^bBased only on last follow-up; n=106 & 76 for CT and 87 & 95 for LT

Mediation Model

The focus of the mediation analyses was on the estimation of the effects sizes for the two distinct and parallel indirect paths, one involving MI→CT and the other involving HE→LT (Figure 1). SEM analyses revealed that both indirect pathways mediated by client language were significantly and differentially related to outcomes (Table 2). The path estimates were opposite in sign as expected because the indirect effects via binary mediators (coded 0, 1) were differentially influenced by treatment assignment (coded 0, 1). Indirect path coefficients for CT mediating the effects of MI were significant for quit attempts, self-reported abstinence, motivation, and confidence to quit. Indirect path coefficients for LT mediating the effects of HE were significant for quit attempts, motivation, and confidence to quit. Although not significant, the HE-LT paths were similar for cessation medication use and abstinence (p<0.07). Note that the simple model relating treatment assignment to outcome yields a non-significant path coefficient for quit attempts (p>0.3), consistent with the lack of treatment differences in the parent trial.

Table 2.

SEM Parallel Mediation Analysis of Motivational Interviewing - Change Talk versus Health Education - Learning Talk

	MI → Change Talk → Outcome					HE → Learning Talk → Outcome
Binary Outcomes	ab	se	OR	95% CI	p	ab	se	OR a	95% CI	p
Quit Attempts	0.30	0.11	1.35	[1.14, 1.72]	0.008	−0.89	0.25	0.41	[0.24, 0.65]	0.001
Cessation Medication Use	0.18	0.12	1.19	[0.97, 1.56]	0.139	−0.70	0.38	0.50	[0.22, 1.00]	0.064
Abstinence (Self Report)	0.25	0.11	1.29	[1.8, 1.64]	0.020	−0.53	0.28	0.59	[0.33, 0.97]	0.053
Continuous Change Scores	ab	se	d	95% CI	p	ab	se	d	95% CI	p
Motivation to Quit	0.47	0.21	0.16	[0.02, 0.31]	0.027	−1.24	0.54	−0.43	[−0.78, −0.07]	0.021
Confidence to Quit	0.53	0.26	0.16	[0.02, 0.31]	0.039	−1.50	0.61	−0.47	[−0.82, −0.11]	0.015

Note: ab = product of coefficients. OR = Odds ratios, or e^(ab), characterizing the mediation path in SEM for binary outcomes. d denotes the standardized ab mediation effect size for continuous measures. Boldface denotes significant mediation effect. SEM estimates based on bootstrap with 10000 redraws.

^aOdds ratios were expected to be in the opposite direction (i.e., < 1) because of the coding of HE=0 and MI=1.

Analyses using a counterfactual framework produced similar findings (Figure 3). The Average Causal Mediation Effects, which is analogous to the mediation or indirect effects, were significantly different from zero and opposite in sign for MI-CT versus HE-LT for all of the outcomes (all p<0.05) except for self-reported abstinence, for which the pattern was similar (p<0.10). The pattern of results with respect to mediation effects were robust across reanalysis under different analytical conditions (e.g., including baseline covariates of nicotine dependence and motivation to quit, treating missing as smoking, examination across participants’ race - see Supplemental Materials).

Figure 3.

Average Causal Mediation Effects (ACME) for the Health Education (HE) → Learning Talk → Outcome pathway versus the Motivational Interviewing (MI) → Change Talk → Outcome pathway. Estimates and 95% confidence intervals are plotted and generated from mediation function in R with 10,000 bootstraps. The path estimates were opposite in sign because the indirect effects via binary mediators (coded Yes = 1, No = 0) were differentially influenced by treatment assignment (coded MI = 1, HE = 0).

Discussion

As hypothesized, results were consistent with different theorized causal paths between MI versus HE and cessation outcomes, with change talk as a mediator of MI’s efficacy and learning talk as a mediator of HE’s efficacy. These results indicate both treatments worked but differed in the distribution of in-session client behaviors that were associated with outcomes (i.e., mechanisms). This study contrasted two very different therapeutic interventions associated with distinct counselor behaviors and attempted to equate them on common factors known to be important (e.g., contact time, intensity, counselor). As previously reported, working alliance was also similar between the two active treatments (Grobe et al., 2020). This might suggest that, after equating for general therapeutic factors, details regarding the specific theoretical approach matter. Whereas prior studies have found evidence that CT is related to positive therapeutic outcomes in different treatment types for addictive behavior (Aharonovich et al., 2008; Amrhein, Miller, Yahne, Palmer, & Fulcher, 2003), these results were consistent with the notion that differing treatments may causally influence different change processes. Furthermore, CT and LT appeared as distinct and unrelated processes that could act in an additive fashion to increase the probability of favorable outcomes. Thus, the elicitation of one type of client change process does not exclude the possibility of other active process also playing a role in behavior change.

The demonstrated efficacy of HE addresses concerns raised in the recent Surgeon General’s report (US Dept of Health and Human Services, 2020) whether the “5 R’s” are effective in those who are not motivated to quit smoking and suggest that such interventions may be worth the effort, at least with respect to how they were implemented in this trial. However, it is important to note that HE was constructed as a reasonable alternative active intervention, equating contact time and intensity. Counselors followed a detailed protocol presenting comprehensive information in a nonjudgmental manner with minimal interaction and pre-scripted responses to common objections to quitting (Catley et al., 2012). This may not equate to a less intense or non-manualized application of guidelines (AHRQ, 2012) in clinical practice. Nevertheless, this approach may be more cost effective and easily disseminated than alternatives that require more counselor skill and training (Rasu et al., 2020).

Regardless, this study presents a novel client language mechanism that may tap a distinct aspect of client processing that is also predictive of outcomes. Although the coding system used here was not based on any specific theory of behavior change, knowledge and belief systems are important components of many theories of behavioral change (Webb, Sniehotta, & Michie, 2010). For example, having comprehensive information about medications and disease processes may be crucial in determining adherence to HIV treatment (Fisher, Fisher, Amico, & Harman, 2006). Likewise, the learning of behavioral skills is known to be important in mediating behavioral change (Alexander, Hogen, Jordan, DeVellis, & Carpenter, 2017). Other models, such as the Elaboration Likelihood Model, emphasize the role of effortful central processing of relevant information in determining the effects of persuasive communications (Wilson, 2007). These results suggest that client language indicative of relevant learning mediates the efficacy of educational based interventions. In any case, more research on understanding and improving the efficacy these types of interventions is clearly warranted.

These findings also provide some clarification for MI process research in that change talk was predictive of outcomes. Existing research provides the strongest support for the importance of sustain talk and the proportion of overall change talk to sustain talk (Magill et al., 2014; Pace et al., 2017). A meta-analysis of the effects of subtypes of change talk found only sustain talk and some indicators of the proportion of change talk to sustain talk (e.g., reasons change talk) were related to outcomes (Magill et al., 2019). The significant role of desire CT in this study stands in contrast to those findings but is most likely related to the different clinical context. While we assumed that moving a client from “having no interest” to expressing any interest in quitting was clinically meaningful, the key to our analysis was that the expression of desire change talk proved to be sufficiently uncommon in the first session that participants could be distinguished by whether or not they expressed desire change talk at all. This may have been in part due to selecting participants based on having low desire to quit but also because participants were middle-aged on average and had been smoking for more than 20 years. The meta-analysis of the effects of types of change talk predominantly included studies of adolescents and young adults that were focused on alcohol and drug use (Magill et al., 2019). These participants may be less entrenched in their addictive behavior or more likely to provide “acquiescent” or “dubious” change talk that doesn’t reflect their true intentions (Miller & Rollnick, 2013). These results support these assumptions and suggest the value of continuing to examine desire change talk in addition to other client language (e.g., sustain talk) particularly across diverse demographic groups and target behaviors.

Another implication of the findings is that a combination of treatment approaches that elicits both change talk and learning talk could be even more effective. Our results indicated that sessions with both CT and LT may be associated with better outcomes than sessions with either type of language alone. These results augment findings from the clinical trial in that the significant mediation effect of change talk suggests that MI was efficacious in this trial, with effect sizes comparable with other reports (see Lindson, Thompson, Ferrey, Lambert, & Aveyard, 2019 for a review). However, learning processes may also be important mechanisms and could potentially be used to enhance MI interventions. This is consistent with evidence supporting the efficacy of interventions that combine MI with structured, personalized, feedback (Vader, Walters, Prabhu, Houck, & Field, 2010).

Collectively, these results reinforce the importance of examining null findings from clinical trials, especially in situations where comparative efficacy is being compared across different treatments (Klemperer, Hughes, Callas, & Solomon, 2017a). Previous secondary analyses from this trial suggested that one explanation for the underperformance of MI related to moderation of MI’s efficacy based on participant’s race (Grobe et al., 2020), with race differences being accounted for by other participant characteristics (e.g., preference for a directive approach). Inspection of mediation results across race confirm the idea that the MI-CT path was more pronounced for non-African American participants whereas the HE-LT path was more pronounced for African American participants. However, the study was underpowered to detect moderated mediation and visual inspection suggests that the differential pathways were robust across groups (see Supplemental Materials). Nevertheless, these mediation results provide additional clarification of the null findings from this RCT.

Interpretation of results from these secondary analyses should be placed in context of several issues and limitations discussed in previous reports, including the fact that these were primarily middle-aged African American participants who had no interest in quitting smoking and were lower in socioeconomic status. While we have firm grounds for the causality of treatment impacting client language, care should be taken when interpreting the links between client language and outcome. Such associations, while prospective in nature, may be confounded by some third variable. In other words, the client language coding used here may serve as a proxy for some other important client process variable. For example, both learning and expressing change talk could be pathways to client engagement which could be the ultimate underlying common mechanism of change. However, it should be noted that the occurrence of learning talk and change talk are unlikely to merely be indicators of more client speech. This is because in HE client speech was minimized while in MI client speech was encouraged. If amount of client speech was the mechanism of change MI would have been far more effective than HE. In addition, if activation of speech was a common confounder, we would also have expected the two measures of client language to be at least partially correlated whereas we found them not to be significantly associated. Our study was also limited in its use of simple binary indicators of desire CT and LT and the coding of only the first session. This limited the coding burden, allowing us to include virtually all of the participants in the clinical trial in our analysis.

Despite these limitations, these results reinforce the idea that different therapies may tend to evoke different pathways to change. These different mechanisms are not exclusive to one approach, but rather are more likely under their respective protocols (MI-CT vs HE-LT) and could work in an additive fashion to increase the likelihood of quitting smoking. Future research should continue to focus on determination of shared versus unique factors that are associated with positive outcomes both across and within theoretical frameworks.

Public Health Significance Statement:

Two types of counseling interventions, Motivational Interviewing and Health Education, targeting people who smoke with low motivation to quit were compared to determine how they helped to encourage participants to try to quit. Results suggest that Motivational Interviewing worked by fostering expression of desire to quit (desire change talk) whereas Health Education worked by fostering learning. These different approaches might be effective for different patients or might be more effective when combined.