Detecting Imminent Smoking Lapse Risk: Prospective Lapse Risk Algorithm versus Participant Retrospective Self-Report

Abstract

Improving detection of smoking lapse risk factors could increase smoking cessation rates among socioeconomically disadvantaged adults, who are less likely to quit than the general population. This study used data from a randomized controlled trial that compared the efficacy of two smartphone-based smoking cessation interventions for socioeconomically disadvantaged adults. Daily ecological momentary assessments (EMAs) assessed current smoking lapse risk based on a previously developed algorithm. Participants were instructed to self-initiate EMAs when they were about to lapse and after a lapse. After self-reported lapses, participants were asked questions about the number of hours of awareness of heightened lapse risk prior to the lapse and coping skills that could have prevented the lapse. Overall, 157 participants self-initiated an EMA to report a smoking lapse during the 13-week post-quit study period. Participants reported detecting warning signs prior to 70.06% of lapses; however, only 30% of lapses were anticipated more than two hours in advance. The lapse risk algorithm detected elevated risk in 68.93% of lapses that were preceded by an EMA within 24 hours. A logistic mixed-effects model indicated that on average the algorithm detected heightened lapse risk earlier than participants reported they were aware of heightened lapse risk, AOR=3.34, 95% CI [1.50-7.42]. Participants most frequently endorsed coping with the urge to smoke and stress as skills that would have helped them prevent lapses. EMA-informed algorithms show promise for detecting heightened risk for smoking lapse before participant recognition, an important step for developing effective real-time smoking cessation interventions for socioeconomically disadvantaged adults.

Smoking cigarettes remains the leading cause of preventable death in the United States (U.S. Department of Health and Human Services, 2014, 2020). Although overall rates of smoking have declined over time, adults with lower socioeconomic status (SES) are more likely to smoke (Cornelius et al., 2023; Hiscock et al., 2011) and less likely to successfully quit smoking compared to the general population (Leventhal et al., 2022; U.S. Department of Health and Human Services, 2020, 2024). Disparities in smoking rates may be exacerbated by the sparsity of interventions that are specifically designed to help adults with lower SES to quit smoking (Garg et al., 2022; Kendzor et al., 2010; O’Connell et al., 2022; Pisinger et al., 2022). Further, individuals with low SES experience lapses more frequently during quit attempts (Cambron et al., 2020). Improving the detection and management of lapse risk warning signs in real-time is critically important to addressing disparities in the success of smoking quit attempts in low-SES groups and developing effective interventions for this population.

Smartphones could improve the accessibility of interventions for smoking cessation. Approximately 90% of Americans own a smartphone, and while disparities in smartphone ownership exist, most of those with less education, lower income, and minoritized race/ethnicity, own smartphones (typically more than 80%; Pew Research Center, 2024). Smoking cessation interventions delivered via smartphone applications (apps) offer several advantages over traditional treatment approaches, including increased availability, increased scalability, lower costs, and the ability to deliver tailored content in naturalistic settings (Businelle et al., 2024). Further, there is emerging evidence that smartphone-based interventions can be effective at increasing smoking quit rates, especially when paired with other treatment options (e.g., pharmacotherapy; Guo et al., 2023).

One challenge to predicting and preventing lapses during smoking quit attempts is the dynamic nature in which risk factors influence behavior. Ecological momentary assessment (EMA) is a useful tool for capturing changes in lapse risk factors in real-time (Shiffman et al., 2008). With EMA, participants are typically prompted to complete brief surveys on their smartphone at several timepoints throughout the day. The repeated nature of measurement in EMA studies not only minimizes effects of recall bias in participant self-reports, but also permits researchers to capture important information as it unfolds in real-time across dynamic psychological and environmental contexts (Shiffman et al., 1997; Shiffman et al., 2008; Stone & Shiffman, 1994). For example, Businelle et al. (2016a) used EMA data to create a six-item smoking lapse risk estimator (i.e., urge to smoke, stress, alcohol consumption, interactions with someone smoking, cigarette availability, motivation to quit smoking) that predicted over 80% of smoking lapses within 4 hours of a lapse in adults with low SES that were undergoing a quit attempt.

Despite these advancements in the prediction of smoking lapse, little is known about the urgency in which lapse abatement approaches need to be deployed to prevent a lapse. Smoking lapses can be prevented by engaging in coping strategies at moments of high risk (Ferguson & Shiffman, 2009; Hébert et al., 2018; O’Connell et al., 1998; O’Connell et al., 2007; O’Connell et al., 2008; Shiffman, 1982, 1984; Shiffman et al., 2020). However, it remains unclear which types of coping skills—or combinations of coping skills—are most effective at reducing smoking lapse risk (Bliss et al., 1989; Shiffman et al., 2020) or how long the protective effects of engaging in coping skills may last (Selva Kumar et al., 2022). Only a few studies have evaluated the time course of smoking lapse risk factors. In one study, episodes of heightened urge to smoke were found to result in smoking lapse approximately 11 minutes after onset (Shiffman et al., 1996). In another study, alcohol use was found to increase smoking lapse risk over 24 hours after the end of a drinking episode (Walters et al., 2024). Otherwise, there is limited research on which smoking lapse risk factors pose imminent versus diffuse risk for lapse (cf. Businelle et al., 2016a) and no clear timelines have been established. Real time self-reports from EMAs show promise for identifying areas in which coping skills-based interventions can be improved to better manage smoking lapse risk factors and prevent future lapses (Hébert et al., 2021; Shiffman et al., 1996).

The purpose of this study was to 1) examine the temporal windows in which participants perceived warning signs prior to a lapse using smartphone-based EMAs, 2) compare participant-reported warning signs with algorithm-detected windows of elevated smoking lapse risk, and 3) identify perceived coping strategies that could have increased the odds of maintaining abstinence among individuals who lapsed.

Method

Participants and Procedure

Data from the present study were collected as part of a clinical trial (NCT03740490) that compared the efficacy of the Smart-T (Businelle et al., 2016b; Hébert et al., 2020; Hébert et al., 2018) and QuitGuide (Smokefree.gov, 2016) apps for smoking cessation. Participants were eligible for the study if they: 1) were ≥18 years of age, 2) were willing to quit smoking in the next seven days, 3) smoked ≥ 5 cigarettes per day at the time of enrollment; smoking status was verified by either expired carbon monoxide (CO) levels ≥ 7 ppm (for those who attended the baseline session in person; i.e., pre-COVID-19) or by texting a picture of their pack of cigarettes when requested by staff during the remote enrollment interview (i.e., post COVID-19), 4) had no contraindications to using nicotine replacement therapy (NRT), 5) agreed to complete CO tests and daily EMAs on a study-provided or personal smartphone for 27 weeks, 6) had household income ≤ 200% of the federal poverty guideline, 7) could read English at a 7^th grade reading level or higher (based on a score ≥ 4 on the Rapid Estimate of Adult Literacy in Medicine-Short Form; Arozullah et al., 2007), and 8) agreed to complete the 26-week post-quit follow-up assessments (i.e., qualitive interview and quantitative survey via smartphone).

Participants who were eligible and interested were randomized into one of two smoking cessation intervention groups: Smart-T or QuitGuide. Randomization was stratified to a study group based on race, biological sex, and cigarettes smoked per day. Participants in both groups were provided up to 12 weeks of NRT (i.e., nicotine patch and gum/lozenge) and compensated for completing: a baseline assessment, daily EMAs for 27 weeks, and a follow-up assessment. In addition, the Smart-T and QuitGuide apps provided smoking cessation psychoeducation and on-demand tips (e.g., on how to manage cravings). Both groups were asked to record lapses (cigarettes smoked during the quit attempt). A key feature that differentiated the interventions was that Smart-T provided treatment messages to participants that were tailored to specific, current risk factors. For additional study details, see Hébert et al. (2025). This study was approved by the Institutional Review Board at the University of Oklahoma Health Sciences, and informed consent was obtained from all participants.

Measures

Demographic Characteristics

Demographic and background information was collected during the baseline assessment. Participants reported their race, ethnicity, biological sex, age, average number of cigarettes smoked per day, and annual household income.

Ecological Momentary Assessment (EMA)

EMAs were either prompted based on time (i.e., morning and evening daily diary EMAs, random sampling EMAs) or self-initiated by the participant (event sampling). Participants were prompted to complete EMAs daily for 26 weeks after the scheduled quit date. Specifically, five EMAs were prompted each day for the first 4 weeks (i.e., one morning EMA 30 minutes after the participant’s preset waking time, one evening EMA 60 minutes before sleep time, three random EMAs), three EMAs were prompted each day for weeks 5 through 13 (i.e., one morning EMA 30 minutes after waking, one evening EMA 60 minutes before sleep time, one random EMA), and only the evening EMA was prompted each day for the final 13 weeks. The study smartphone app prompted random EMAs during each participant’s normal waking hours (i.e., the app split each participant’s waking hours into equal epochs based upon the number of random EMAs that were to be scheduled and selected a random time for each random EMA). Participant phones rang/vibrated for 5 minutes to alert them that a prompted EMA was available. Participants were required to begin EMAs during this 5-minute window and they had up to 10 minutes to complete each EMA once initiated.

Participants were instructed to self-initiate an EMA any time they believed they might lapse by clicking a button labeled “I Am About to Slip.” Fifteen minutes after completing the brief participant-initiated EMAs, participants received a prompted follow-up EMA. Participants could also indicate that they already lapsed by pressing a button labeled “I Already Slipped.” In both the “I Am About to Slip” follow-up and “I Already Slipped” assessments, participants could report how many cigarettes they smoked and how long ago they smoked. Smoking lapse was defined as self-reported smoking of at least one cigarette during a self-initiated lapse EMA (binary: lapse = 1, no lapse = 0). Additionally, participants were asked to report their perception of warning signs with one question: “Looking back, how many hours before you smoked did you have warning signs that you might lapse?” and given the response options of “I had NO signs that I would lapse,” “1 hour ago,” “2 hours ago,” “3 hours ago,” “4 hours ago,” “5 to 6 hours ago,” “7 to 8 hours ago,” “9 to 12 hours ago,” “13 to 24 hours ago,” and “More than 24 hours ago.” In another question, participants were asked, “Improving which skills would help you to stay quit in the future?” and instructed to select all responses that applied to the lapse from a list (i.e., coping with urges or cravings to smoke, coping with stress, coping with other people smoking, coping with arguments, coping with frustration or anger, coping with restlessness, coping with depression, learning more about the harms of smoking, and other skills that are not listed).

During all EMAs, participants were asked to indicate if and when they smoked a cigarette (“Today, how long ago did you last smoke a cigarette?” Response options included “0 = I have not smoked today (not even a puff),” “0-15 minutes ago,” “16-30 minutes ago,” “31-45 minutes ago,” “46 minutes to 1 hour ago,” “1 hour and 1 minute to 1 hour and 15 minutes ago,” “1 hour and 16 minutes to 1 hour and 30 minutes ago,” “1 hour and 31 minutes to 1 hour and 45 minutes ago,” “1 hour and 46 minutes to 2 hours ago,” “2 hours and 1 minute to 2 hours and 30 minutes ago,” “2 hours and 31 minutes to 3 hours ago,” “3 hours and 1 minute to 4 hours ago,” to “12 = more than 4 hours ago”), respond to six questions that were combined to create the weighted smoking lapse risk score (i.e., urge to smoke, stress, cigarette availability, proximity of other people who are smoking, motivation to stay quit, recent alcohol consumption; Businelle et al., 2016a), and indicate the odds that they would smoke today. The prompts and response options for the EMA items that informed the lapse risk estimator are shown in Table 1.

Table 1.

Prompted EMA Items and Response Options

Factor - question	Response Options
Mark the response that most applies to you RIGHT NOW.	0	1	2	3	4	5
Urge - I have an urge to smoke.		Strongly Disagree	Disagree	Neutral	Agree	Strongly Agree
Stress - I feel stressed.		Strongly Disagree	Disagree	Neutral	Agree	Strongly Agree
Cigarette availability - Cigarettes are available to me.		Not at All	With Extreme Difficulty	With Difficulty	Fairly Easy	Easily Available
Motivation to quit - I am motivated to AVOID smoking.		Strongly AGREE	AGREE	Neutral	DISAGREE	Strongly DISAGREE
Others smoking – Are you interacting with anyone who is smoking?	No	Yes
Recent alcohol consumption - I drank alcohol within the last hour.	No	Yes
Participant perceived odds of smoking - How likely is it that you will smoke between now and the end of the day?	There is a 0 percent chance	There is a 25 percent chance	There is a 50 percent chance	There is a 75 percent chance	There is a 100 percent chance

Note. For each question, participants were instructed to, “Mark the response the most applies to you RIGHT NOW.”

The lapse risk estimator, embedded at the end of each EMA, calculated a weighted score based on the following equation (Businelle et al., 2016a):

$$\text{Lapse risk score}=\left(\text{Urge}-3\right)*\text{.2}+\left(\text{Stress}-3\right)*\text{.2}+\left(\text{Cigarette availability}-3\right)*\text{.7}+\left(\text{Low cessation motivation}-3\right)*\text{.2}+\left(\text{Interacting with others smoking}\right)\text{+}\left(\text{Alcohol use}\right),$$

in which scores could range from −2.6 to 4.6. Participants were classified as high-risk for smoking lapse if their risk score was greater than or equal to 1.0. Additionally, participants were classified as high risk for lapse if their perceived odds of smoking were greater than 0% (see Table 1). This single question has been previously found to be one of the strongest predictors of smoking lapse (Hébert et al., 2021). Participants in the Smart-T group received a tailored message after each prompted EMA, the content of which was based on the highest rated lapse risk factor (Businelle et al., 2016b; Hébert et al., 2025; Hébert et al., 2020). A hypothetical sequence of study events is depicted in Supplementary Figure 1.

Data Analysis

Participant demographics and smoking lapse characteristics were summarized. The analytic dataset included self-initiated lapse EMAs (i.e., “I Just Slipped” and “I’m About to Slip” follow-up assessments in which participants reported smoking at least one cigarette) and prompted EMAs that prospectively assessed smoking lapse risk. Only data collected during post-quit study weeks 1 through 13 were included in the present analyses because during weeks 14 through 26 only one EMA was prompted per day. The key outcome variables for this study were: (1) self-reported smoking lapse, (2) perceived warning signs, (3) algorithm-detected lapse risk, and (4) perceived coping skills.

Logistic regression was used to determine if group assignment (Smart-T or QuitGuide) was associated with self-reported lapse characteristics. To determine whether group assignment was associated with warning sign detection, a two-level (with lapses nested within participants) generalized linear mixed-effects regression model (GLMM) with a logit link function was fit to the warning sign data (binary: 0 = no warning sign, 1 = any warning sign). Another GLMM was fit to test whether heightened lapse risk was detected by prospective EMAs before the timepoint in which participants retrospectively reported detecting warning signs. For each self-reported smoking lapse, a new binary outcome variable was created. Lapses were coded as 1 if at least one prompted EMA detected “high-risk” for smoking lapse before the participant detected lapse warning signs (up to 24 hours prior to lapse), and coded as 0 if the participant detected lapse warning signs before or within the same time window in which a prompted EMA detected “high risk” for smoking lapse (see Supplementary Figure 1 for further explanation). Smoking lapses that were not preceded by any prompted EMAs within 24 hours were excluded from analysis. The GLMM was fit in two stages. First, the data were fit to an intercept-only model to estimate the odds that the lapse risk algorithm detected heightened lapse risk before participants perceived the warning signs. Second, treatment group, study day, age, race, sex, and baseline cigarettes per day were added to the model. To increase the likelihood that lapse assessments were only counted during a quit attempt (and not a period of relapse), self-reported lapses were excluded from analyses if they occurred less than 24 hours after a previously reported lapse. All analyses were performed using R (R Core Team, 2024) and GLMMs were run with the lme4 and glmmTMB packages (Bates et al., 2015; Brooks et al., 2017). For all inferential tests, alpha was set at .05.

Results

Across the 13-week post-quit period, participants self-initiated 464 “I Already Slipped” assessments and 997 “I am About to Slip” assessments. A prompted “I am About to Slip” follow-up assessment was delivered 15 minutes after the “I am About to Slip” assessment; however, only 556 (55.77%) of these follow-up assessments were completed. Notably, participants indicated a smoking lapse on only 13.67% of the completed “I am About to Slip” follow-up EMAs (76/556). Thus, a total of 540 self-initiated smoking lapses were reported. Overall, smoking lapse EMAs were self-initiated by 41.94% (n = 91/217) of QuitGuide participants, OR = 0.72, 95% CI [0.55-0.94], and 31.43% (n = 66/210) of Smart-T participants. Compared to QuitGuide, Smart-T participants were significantly less likely to self-initiate a lapse EMA, OR = 0.63, 95% CI [0.43-0.94]. Demographic characteristics for the subsample of participants who self-initiated a lapse EMA (n = 157) are shown in Table 2. On average, Smart-T participants reported 3.55 lapses (SD = 4.87; Range: 1-28) and QuitGuide participants reported 3.36 lapses (SD = 4.25; Range: 1-25). However, the distribution of lapse EMA frequencies was positively skewed: 40.76% of participants only self-initiated a smoking lapse EMA once. The median (IQR) number of self-initiated lapse EMAs reported by Smart-T and QuitGuide participants was 2.0 (1.0-3.75) and 2.0 (1.0-4.0), respectively.

Table 2.

Characteristics of Participants that Self-Initiated a Lapse EMA (n=157)

	mean or N	SD or %
Age, mean (SD)	52.45	10.61
Race, N (%)
American Indian/Alaska Native	8	5.10
Black or African American	39	24.84
More than one race	6	3.82
White	104	66.24
Biological Sex, N (%)
Female	119	75.80
Male	38	24.20
Household Income, N (%) *
<30K	116	75.32
30K to <60K	33	21.43
≥60K	5	3.25
Education, N (%)
High school or greater	137	87.26
Less than high school	20	12.74
Cigarettes per day, mean (SD)	17.11	8.19

^*Three participants did not indicate their annual household income

Anticipating Lapse Risk

Figure 1 shows the percentage of lapses in which participants perceived warning signs across timepoints leading up to the lapse. Overall, participants reported that they perceived warning signs prior to 70.06% of all lapses (Smart-T: 74.67%; QuitGuide: 66.44%). Participants were more likely to perceive warning signs in the hours preceding the lapse compared to more distal timepoints. Warning signs were detected more than 1 hour in advance in less than half of all lapses (Smart-T: 48.03%; QuitGuide: 42.47%) and more than 4 hours in advance in less than a quarter of all lapses (Smart-T: 22.27%; QuitGuide: 9.25%).

Figure 1. Number of Hours Perceived Warning Signs Preceded Self-Reported Smoking Lapses.

Note. Percentage of lapse EMAs (y-axis) as a function of the number of hours perceived warning signs preceded a lapse (x-axis). Unfilled bars show data from the Smart-T group and filled bars show data from the QuitGuide group.

Participants did not detect any warning signs of imminent smoking lapse in 29.94% of all self-initiated lapse reports (Smart-T: 25.33%; QuitGuide: 33.56%). A GLMM was fit to compare the likelihood of experiencing a lapse with or without warning signs between groups. For the QuitGuide group, participants were more than three times as likely to lapse with warning signs than without, OR = 3.21, 95% CI [1.77-5.32]. The odds of detecting a warning sign prior to lapse were an additional 1.23 times higher in the Smart-T group compared to the QuitGuide group, but this difference was not statistically significant, 95% CI [0.53-2.87].

Lapse Risk Algorithm

Figure 2 shows the cumulative percentage of lapses detected by the lapse risk algorithm (unfilled symbols) and participant perceived warning signs (filled symbols; replotted from Figure 1) at timepoints leading up to a lapse. The cumulative lapse detection rate for the algorithm is calculated as the number of lapses preceded by at least one “high-risk” EMA divided by the number of lapses in which at least one EMA was completed. The lapse risk estimator predicted 52.16% (181/347) of lapses based on EMAs completed 13-24 hours before a lapse. As the lapse drew closer in time, the cumulative percentage of detected lapses gradually increased. Given all EMAs completed within 24 hours of a lapse, the algorithm detected 68.93% (284/412) of lapses. Although group differences are not shown in Figure 2, the lapse risk algorithm consistently detected more lapses for participants in the QuitGuide group (Range: 66.67-81.31%) than for participants in the Smart-T group (Range: 37.57-55.56). Despite these differences, the lapse risk algorithm consistently detected heightened lapse risk more often than participants retrospectively perceived warning signs at all timepoints beyond one hour prior to lapse.

Figure 2. Cumulative Lapse Detection: Comparison of Lapse Risk Algorithm and Participant Perceived Warning Signs.

Note. Figure 2 shows the cumulative percentage of detected lapses (y-axis) across timepoints leading up to the lapse as detected by lapse risk algorithm (unfilled symbols) and participant perceived warning signs (filled symbols; replotted from Figure 1). Note that only EMAs and warning signs that occurred within 24 hours of the lapse are included. The x-axis shows lapse proximity in bins that correspond to the response options available when participants indicated how long ago they perceived warning signs. The fraction next to each data point shows the number of detected lapses (numerator) divided by the number of lapses preceded by at least one EMA at each timepoint (denominator for algorithm assessed risk) or the total number of lapses (denominator for participant perceived risk).

A GLMM indicated that the lapse risk algorithm was significantly more likely than participants to detect lapse risk first. Parameter estimates for the intercept-only and covariate-adjusted models are shown in Table 3. The intercept-only model indicates that the algorithm was 2.01 times as likely as participants to detect lapse risk before participants reported detecting warning signs, 95% CI [1.24-3.26]. After controlling for treatment group, study day, age, race, sex, and baseline cigarettes per day the algorithm was 3.34 times as likely as participants to detect elevated lapse risk first, 95% CI [1.50-7.42]. There were no significant fixed effects for any of the predictors added to the covariate-adjusted model.

Table 3.

Model Estimates for Predictors of the Algorithm Detecting Lapse Risk Before Participant Perceived Warning Signs

	Intercept-Only Model		Covariate-Adjusted Model
Predictors	OR (95% CI)	p	AOR (95% CI)	p
Intercept	2.01 (1.24–3.26)	.005	3.34 (1.50–7.42)	.003
Smart-T (ref: QuitGuide)			0.44 (0.17–1.12)	.085
EMA Day (cent)			0.99 (0.98–1.00)	.068
Age (cent)			1.01 (0.96–1.05)	.780
Non-White (ref: White)			1.29 (0.44–3.74)	.644
Male (ref: female)			0.36 (0.11–1.18)	.092
Cigarettes per day (cent)			1.05 (0.99–1.13)	.129

Coping Strategies

Figure 3 shows the percentage of lapses in which participants endorsed various coping skills that they believed could have helped them to avoid lapsing (note that participants could endorse multiple options). The most endorsed coping strategies were coping with urges/cravings related to smoking (72.83% of lapses) and coping with stress (60.31% of lapses). Participants also frequently endorsed coping skills related to negative affect—including coping with depression (26.01% of lapses), frustration/anger (36.61% of lapses), and restlessness (36.03% of lapses)—and interpersonal contexts—including coping with others smoking (24.47% of lapses) and coping with arguments (17.73% of lapses). Learning more about the risks of smoking (7.32% of lapses) and “other” coping strategies (9.63% of lapses) were endorsed the least frequently. Differences between groups were small, typically less than 10%.

Figure 3. Participant Endorsed Coping Skills That May Have Prevented the Lapse.

Note. Participants could endorse multiple coping skills after each lapse.

Discussion

The primary goal of this study was to understand the temporal characteristics of warning signs preceding a smoking lapse. In self-initiated EMAs shortly after lapse, participants reported that they perceived lapse warning signs in most cases (i.e., nearly 70% of lapses); however, participants typically only detected warning signs within the 2 hours preceding a lapse. Similarly, our lapse risk estimator (Businelle et al., 2016a) detected heightened lapse risk in nearly 70% of all lapses based on all EMAs completed within 24 hours; however, the detection rate only gradually decreased at distal timepoints, such that over half of lapses were detected based on EMAs completed 13-24 hours before the lapse. Overall, these findings suggest that our lapse risk algorithm (Businelle et al., 2016a) may identify elevated smoking lapse risk before participants become self-aware of elevated smoking lapse risk. Thus, this lapse risk estimator could be included as a useful tool in future interventions. With advanced notice of smoking lapse risk, there is a broader window in which app-based content could be provided to enable participants to anticipate and better manage lapse risk.

Although the QuitGuide group was associated with significantly higher odds of self-initiating smoking lapse assessments compared to the Smart-T group, there were no significant differences between groups in the odds of detecting participant perceived warning signs or in the odds of the algorithm detecting smoking lapse risk before participants perceived warning signs. It is possible that participants in the Smart-T group were less likely to self-initiate lapse assessments because they had fewer smoking lapses. That is, tailored messages delivered after each EMA may have increased the ability of participants in the Smart-T group to manage smoking lapse risk factors and consequently prevent lapses. However, the present findings suggest that when participants in the Smart-T group did lapse, they were not necessarily better at detecting those risk factors compared to participants in the QuitGuide group.

Relatively few studies have been conducted to identify the temporal contiguity of the onset of smoking lapse risk factors and smoking lapse. The findings of Shiffman et al. (1996) suggest there is a relatively narrow window of time for participants to resist lapse after an acute peak in urge. In contrast, participants in the present study perceived warning signs many hours in advance of some lapses. These differences may be in part due to the unrestricted definition of “warning signs” in the present study compared to the emphasis on “high urge” in the Shiffman et al. study. Further, the present findings suggest that there is considerable variability in the duration of lapse risk within and/or between risk factors. Indeed, risk factors such as stress (Selva Kumar et al., 2022) and alcohol consumption (Walters et al., 2024) have been shown to produce temporally extended periods of increased smoking lapse risk (see also, Businelle et al., 2016a; Hébert et al., 2021; Koslovsky et al., 2018). Future research should aim to characterize the temporal boundaries of different risk factors and examine the trajectory of lapse risk as ways to determine the urgency in which smartphone-based intervention content or participant coping strategies need to be deployed in response to specific risk factors.

Identifying strategies to manage lapse risk factors and prevent future smoking lapses continues to be of interest in smoking cessation research (Businelle et al., 2024; O’Connell et al., 1998; Shiffman, 1982). One goal of the present study was to identify coping skills that participants believed could have helped to prevent their smoking lapse. Participants most frequently reported that improving coping strategies related to managing urges/cravings and coping with stress would help them to prevent future smoking lapses. These participant-identified coping skills are generally consistent with findings in the literature that suggest heightened urges/cravings (Shiffman et al., 1996) and stress (Nakajima et al., 2020) are associated with increased smoking lapse during a quit attempt. Other frequently endorsed coping skills, related to negative affect and interpersonal contexts, may be worth considering as intervention targets in future studies (e.g., Businelle et al., 2016a; Shiffman, 1982).

Further, future research should aim to bolster and personalize coping skill-based components of smartphone-based interventions. In the present study, participants in both groups had access to on-demand tips related to common smoking lapse risk factors (including tips for coping with urge and stress) and participants in the Smart-T group received tailored messages after each EMA that addressed currently reported lapse risk factors. Despite the availability of on-demand tips and tailored messages, participants indicated that improving coping skills would have helped prevent lapses. It may be of interest in future research to further refine smartphone-based interventions by identifying strategies to increase participant engagement with in-app content, such as increasing the frequency of prompts to review on-demand quit tips, developing more personalized intervention content, or by including novel modalities for content delivery (e.g., in-app videos or interactive chatbots; Calle et al., 2024).

Strengths of the current study include the nationwide recruitment strategy, long duration (i.e., 13 weeks) of daily EMA collection, and use of intensive longitudinal data to identify smoking lapses and smoking lapse risk. In addition, the use of self-initiated assessments at the time of a lapse to obtain in-the-moment information on participant perceived warning signs and coping skills likely reduced recall bias (Bradburn et al., 1987) and strengthened implications of the results of this study.

Study limitations include the reliance on self-initiated EMAs and the lack of questions on which strategies participants actually used to try to prevent lapses. The reliance on self-initiated assessments could limit the generalizability of the present findings for several reasons. Importantly, less than half of the participants enrolled in the parent study self-initiated a lapse assessment. Given that participants were not compensated for completing self-initiated EMAs, this subsample may overrepresent participants that were highly motivated to seek help from their assigned intervention during their quit attempt. Additionally, participants may have only self-initiated EMAs at convenient times; thus, information surrounding lapses that occurred within some contexts may be missing. Another limitation is that participants were only asked about pre-lapse warning signs and coping skills that could prevent lapses after a lapse. Consequently, hindsight bias may have increased the likelihood of participants recalling pre-lapse warning signs. Future EMA studies may benefit from instructing participants to prospectively monitor and report moments of heightened lapse risk (e.g., event recording stressful moments; see, Shiffman & Waters, 2004) and by including questions about which coping strategies individuals use to successfully avoid smoking lapse during high-risk moments (see Shiffman et al., 1996). Finally, participants were not provided with a definition for what constituted a warning sign, thus, certain risk factors may have been detected by participants but not considered to be warning signs.

The present study adds to a growing body of research that suggests that EMA-informed, smartphone-based lapse risk estimators can predict smoking lapses with increasing precision and prior to participant self-perceptions of heightened lapse risk. Further, participant insights on lapse warning signs and coping skills highlight important future directions for improving the effectiveness of smoking cessation interventions. By detecting elevated lapse risk early, whether via algorithm or participant self-monitoring, there are more opportunities for intervention content or coping skills messaging to be deployed.

Highlights.

• Participants typically detected smoking-lapse warning signs within 2 hours of lapse

• A smoking lapse-risk algorithm quantified lapse risk based on six risk factors

• The algorithm detected heightened lapse risk earlier than participant detection

• Lapse-risk algorithms may improve smartphone-based smoking cessation interventions

• Improving coping skills related to urge/cravings and stress may help prevent lapses