Intervention

The intervention was comprised of 25 group sessions of decreasing frequency over 12 months, and included all features of standard behavioral weight loss including weight goals, regular weighing, calorie and physical activity goals/self-monitoring, and guidance in identifying and following a balanced deficit diet that would comport with calorie goals. Each 75-minute session consisted of a brief check-in from each group member, introduction or review of a weight control skill, and goal-setting. Treatment sessions were delivered by doctoral-level clinicians and a trainee who served as a co-leader. If a participant was absent from a particular group session, he/she completed a brief (20-minute) individual makeup session with the group co-leader to review the session material. Participants were asked to carefully track all food/beverage consumption and calculate caloric intake so as to ensure that they met their calorie goals. One participant from the parent study did not participate in the EMA protocol and was excluded from analyses.

Procedure

Participants were loaned an Android player (essentially a smartphone without phone capabilities; Samsung Galaxy Player 4.0) pre-loaded with a custom-designed EMA smartphone application (DrexelEMA). The EMA protocol occurred for 14 days within the first three weeks of treatment (baseline), 7 days at 6-months (mid-treatment), and 7 days at 12 months (end-of-treatment). Prior to each period of EMA data collection, participants were given oral and written instructions on how to use the EMA app and how to identify a dietary lapse (see Ecological Momentary Assessment, below).

Measures

Data descriptives

A total of 24,772 (13,402 baseline, 6,017 mid, 5,353 end-of-treatment) EMA recordings were provided by 186 participants representing 4,568 participant days. Descriptive information for survey type, response days, and compliance by assessment point and overall are provided in Table 1. As noted above, sample size was not equal across assessments (baseline n = 186, mid-treatment n = 162, end-of-treatment n = 143). Logistic regression revealed that baseline lapse frequency did not significantly predict missing data at mid- (χ² = .58, p = .45, df = 1) or end-of-treatment (χ² = .02, p = .89, df = 1), suggesting that participants who lapsed more frequently at baseline were not at significantly greater risk of attrition from the EMA protocol. Analyses were conducted with participants who had eligible data (i.e., compliance ≥ 40%) at any assessment point, and with completers only (i.e., individuals who completed EMA at all three assessment points with ≥ 40%, n = 138). A similar pattern of results was observed, so we present results using all eligible data.

Table 1

EMA survey type and compliance data by assessment point

	Assessment Point
	Baseline	Mid-Treatment	End-of-Treatment
Sample size	186	162	143
Total number of recordings	13,402	6,017	5,353
Prompted	11,541	5,065	4,736
User-Initiated	1,861	952	617
Recording days
Mean (SD)	13.3 (2.1)	6.9 (1.5)	6.9 (0.8)
Range	3 to 18	1 to 13	2 to 10
Compliance with prompted surveys
Mean (SD)	83.2% (11.5%)	82.7% (14.7%)	85.5% (13.2%)
Range	45% to 100%	42% to 100%	42% to 100%

Open in a separate window

Lapse frequency

Participants reported a total of 2,786 lapses (1,452 at baseline, 667 at mid-treatment, and 667 at end-of-treatment). A GEE revealed a significant quadratic effect of time on lapse frequency, b = .14, SE = .06, Wald χ² (1, N = 4,578) = 5.62, p = .02. The linear term was not significant, b = −.04, SE = .05, Wald χ² (1, N = 4,578) = .61, p = .43. The estimated mean number of lapses per day was 0.60 at baseline, 0.55 at mid-treatment, and 0.66 at end-of-treatment. When translated to the weekly level, mean number of lapses were 4.22 at baseline, 3.86 at mid-treatment, and 4.63 at end-of-treatment (see Figure 1). The number of participants who reported no lapses at each assessment point was as follows: 14 at baseline (7.52% of participants completing EMA), 29 at mid-treatment (17.90% of participants completing EMA); and 25 at end-of-treatment (17.48% of participants completing EMA). Thus, the actual number of participants contributing data for EMA analyses at each time point were 172 at baseline, 133 at mid-treatment, and 118 at end-of-treatment. The range in number of lapses per participant (adjusted to 7 day total) was 0 to 16 at baseline, 0 to 21 at mid-treatment, and 0 to 21 at end-of-treatment.

Open in a separate window

Figure 1

Estimated mean number of lapses per week per participant by assessment point.

Note: Values are based on predicted values from a GEE examining the effect of time on lapse frequency, when controlling for all other predictors in the model. Values are presented at the weekly level for interpretative purposes.

Lapse Type, Timing, and Location

Table 2 displays the percentage of each type of lapse, lapse timing, and lapse location reported at each assessment point. Across all assessments, eating an unintended type of food was the most commonly reported type of lapse. Lapses occurred most often at home. Evening was the most common time of lapse; early afternoon, late afternoon, and late evening were other common times of lapses. Of note, lapse type, timing, and location also were examined after first calculating person-level averages for each variable (so that individuals with more lapses did not skew the distribution); a nearly identical pattern was observed.

Table 2

Lapse characteristics by assessment point.

	Assessment Point
	Baseline (%)	Mid-treatment (%)	End-of-treatment (%)
Lapse Type
Unintended food	44.6%	42.2%	45.8%
Unplanned time	30.3%	27.6%	25.0%
Larger portion	25.2%	30.2%	29.2%
Lapse Location
Home	48.6%	46.0%	46.7%
Work/School	16.6%	14.6%	19.5%
Eating Out	19.8%	19.2%	16.9%
Other	16.6%	20.1%	16.8%
Time of Lapse
Early morning (6:00–8:59 AM)	2.0%	2.5%	3.7%
Late morning (9:00–11:59 AM)	10.5%	11.2%	11.8%
Early afternoon (12:00–2:59 PM)	18.9%	20.7%	20.5%
Late afternoon (3:00–5:59 PM)	18.3%	16.5%	19.3%
Evening (6:00–8:59 PM)	28.9%	31.8%	28.8%
Late evening (9:00–11:59 PM)	20.1%	14.8%	14.1%
Overnight (12:00 AM–5:59 AM)	1.3%	2.4%	1.6%

Open in a separate window

Note: Percentages reflect all available data from completed survey items. Percentages may not add up to 100% due to rounding.

Relationship of Lapses to Weight Change

Weight change during the EMA assessment period (M = −3.09%, SD = 1.50%) and between baseline and end-of-treatment (M = −12.04%, SD = 7.86%) were both positively predicted by number of lapses during the baseline EMA period (b = .08, SE = .03, p = .01, and b = .46, SE = .17, p = .009, respectively). A multiple linear regression model containing number of reported lapses per week at baseline and weight change during the baseline EMA period indicated that lapse frequency also predicted percent weight change during treatment (b = 0.34, SE = .17, p = .05) when controlling for weight change during the baseline EMA period (b = 1.49, SE = .37, p < .001).

Characterizing frequency, type, timing and location of lapses

Our findings revealed that participants in the weight loss program lapsed, on average, between 4 and 5 times per week. This frequency is somewhat higher than some previous reports (2.7 to 6.0 lapses/week; 8, 15), while lower than others (11.8 lapses/week; 9). One factor that may have contributed to the lapse rate in this study being greater than those observed in some previous studies is the difference in how dietary lapses were defined in the present study. Whereas several prior studies have defined lapses as “an incident where you felt you broke your diet (e.g., overeat, eat a forbidden food, etc.)” (8, 9, 17), lapses were defined as instances of “eating or drinking likely to cause weight gain, and/or put weight loss/maintenance at risk” in the present study, and participants were forced to categorize their lapse type (i.e., eating a larger portion than intended, eating at an unintended time, eating a food one was trying to avoid). Being required to identify a lapse type may have provided participants with a more elaborated and concrete understanding of what would constitute as a lapse in daily life, thus leading to better identification of and more frequent reports of lapses. Other possible explanations for the greater lapse rate are that the dietary guidelines in the current study were more clear-cut, making lapses more likely, obvious, and easier to report, and that the assessment of lapses in this study was more comprehensive in general and placed a greater emphasis on lapses (i.e., when training participants) than previous studies have, leading participants to be more aware of potential lapses. The one study that revealed a higher lapse rate (9) was conducted with undergraduate students who were self-dieting; differences in sample characteristics, approach to weight loss, and participants’ environment may help to explain why the lapse rate was lower in this study.

Lapse frequency was reasonably stable throughout the three assessment periods, but did show a hypothesized curvilinear relationship over time, such that lapse frequency appeared to decrease from baseline to mid-treatment and then increase from mid- to end-of-treatment. While no previous study has examined changes in lapse frequency over time, the observed pattern is highly consistent with the many findings of weight loss intervention participants demonstrating weight loss during active treatment and then weight regain (or at least slowed weight loss) as treatment tapers or stops (28, 32–34). Motivation is likely highest at the beginning of the program, in part because of the reinforcement of easier and more demonstrable weight loss (35). In addition, participants are benefiting at the outset of the program from skills training in shopping, food selection, cooking and normalizing eating to maximize satiety while minimizing calorie intake (36, 37). The increase in lapse frequency from mid- to end-of-treatment may be due to motivational degradation to practice key behaviors (35, 38). For instance, declines in motivation over time (likely resulting from plateaus or regaining weight) have been found to be associated with poor adherence to self-monitoring food intake (a key weight control behavior) (35). Such findings indicate that motivation is crucial for maintaining adherence to behavioral skills that ultimately impact outcomes, a finding echoed by other research (39). Further research should directly investigate the relationship between motivational states and lapse frequencies throughout behavioral weight loss programs.

It should be emphasized that reported lapses are also a reflection of the participants’ understanding of the dietary prescription and how stringently they were attempting to follow it. For example, it is conceivable that lapse reports at mid-treatment and end-of-treatment are more meaningful than those at baseline, because participants’ understanding of the dietary prescription they were attempting to follow deepened with exposure to treatment. In partial support of this hypothesis, lapse frequency at mid- and end-of-treatment were more strongly correlated (r = 0.73, p < .001) than were lapse frequency at baseline and mid-treatment (r = 0.52, p < .001) and at baseline and end-of-treatment (r = 0.56, ps < .001), suggesting participants’ standards for what qualified as a lapse may have been more similar later in treatment.

Results also revealed that frequency of lapses varied by type and location. For example, eating a food one had not intended to eat was by far the most frequent type of lapse perhaps because of the power of cues from desired foods. The home was the most common location of lapses, which is not surprising given the amount of time people spend at home and that it is a location where desirable food is likely to be readily available. On a related note, lapses occurred most often in the evenings, and were more likely to occur on the weekends than on weekdays. This result is consistent with other studies showing that weight control behaviors show different patterns on weekends versus weekdays, and makes sense given that lapsing is associated with being at home, having easy access to food and having unscheduled time (40). In addition, more lapses occurring later in the day is consistent with the limited resource model of self-control (41) which posits that individuals have a limited capacity for self-control, and that exerting self-control more often throughout the day can deplete one’s self-regulatory resources.

These results extend previous findings among individuals presenting for treatment that lapses occur most frequently at home and in the evening (8, 9, 17). As such, the current study’s findings add to the argument that modification of the home environment (to remove cues and easy access to high-calorie food), especially in the evening, represents a crucial intervention target (42). The present study showed that, while exposure to palatable foods decreased over time (indicating that interventions successfully reduce exposure to tempting foods), the momentary relationship between tempting food in the environment and dietary lapses strengthened. Perhaps some “tempting” foods (e.g., leftovers) could not necessarily be removed from the environment, suggesting that other interventions (e.g., portion control) may be necessary. Alternatively, stimulus control interventions may need to be further strengthened, for example, by involving partners or other family members and through home visits (43). Future research could investigate the utility of using EMA to identify individual-level triggers in order to personalize treatment targets (e.g., focusing on boredom and the workplace for one individual, but social occasions and evenings for another). In addition, current behavioral weight loss intervention components that aim to prevent dietary lapses (e.g., stimulus control, planning, reducing emotional eating) were not originally informed by research using ecologically-valid measurement methods, which suggests that this research could inform ways to enhance these intervention components.

Clinical significance of lapses and association with concurrent weight gain

The clinical significance of the reported lapses is difficult to judge given that we did not attempt to record how sizeable they were in calories. However, conservatively estimating that the average lapse was a modest 150 kcal produces an estimate of 600–750 kcal of out-of-plan consumption per week, on average. According to metabolic calculations, adding this amount of caloric intake to the diet of an individual in a weight control program would significantly impact weight trajectory (44). A more direct measure of the impact of lapses is that lapse frequency was associated with co-occurring weight change. In fact, the effect size indicated a 1.27 lb. difference in weight loss between a participant of mean weight for this sample at baseline (i.e., 222.0 lbs.) whose lapse frequency was one standard deviation below the mean (.80 lapses) and one standard deviation above the mean (7.37 lapses). This magnitude of difference over just a two-week period does appear to be clinically meaningful, especially given it could multiply across time and the fact that desired weight loss in behavioral treatments is 1–2 lbs. per week (7). Moreover, baseline lapse frequency predicted weight change at end-of-treatment, even controlling for co-occurring early weight loss, suggesting that early inadherence, above and beyond early weight loss success, is predictive of eventual success in the program. Put simply, early adherence appears to predict adherence in the program generally, which translates into overall weight loss by the end of the program.

Predictors of lapse

Potential triggers of lapsing were found to be associated with lapses in two different ways. First, participant’s average level of sadness, loneliness, boredom, anger/irritation, stress, hunger, deprivation, fatigue, and exposure to delicious foods was positively associated with likelihood of lapsing. In other words, people who were generally at higher levels of aversive internal states and who were more often around delicious foods during the EMA period lapsed more often. Interventions that provide skills to generally ease unpleasant emotional and physical states might then aid in weight control efforts, as would interventions designed to remove tempting food from the personal food environment. Second, an increase in an individual’s level of hunger and deprivation (relative to, and controlling for, his or her average level) and availability of delicious food at a given time point (controlling for average level) prospectively predicted lapse occurrence at the subsequent time point. In fact, the likelihood of a lapse increased 30–50% for each increase in level of a trigger variable. For example, someone who is two points above her average on the deprivation scale is 2.6 times as likely to lapse as when she is at her average level of deprivation. These particular results support the notion that momentary changes in the level of internal and external cues have a significant impact on eating decisions. Of note, the influence of these cues on a person’s eating decisions potentially takes place implicitly without his or her conscious knowledge (45–47)

The identification of specific cues that prospectively predict lapses from a formal weight control program has clinical implications. For example, intervention programs could emphasize the insidious nature of cues like hunger, feelings of deprivation, and tempting food, and introduce skills designed to be aware of and manage these states. Moreover, technology-driven, tailored, in-the-moment approaches, such as smartphone-based just-in-time adaptive interventions, could be developed that track cues known to prospectively predict lapses and to deliver specific coping skills designed to prevent a lapse from occurring (48). Of note, systems that help people recognize and interrupt patterns leading to dietary lapse might be equally useful for those in less formal weight loss programs as many of these individuals would face the same challenge of adhering to an eating plan and would be subject to the same forces pushing against adherence (49).

Of interest, momentary increases in physically aversive states, but not emotional states, significantly increased the likelihood of subsequent lapses. Perhaps negative reinforcement effects are stronger in the former case and thus play a larger role in driving off-plan eating. In any case, the pattern of findings are at odds with self-report studies showing that emotions drive overeating and that emotional eating is associated with overweight (50–52). One possibility is that self-reported emotional eating is a product of retrospectively accounting for behavior at odds with intentions. It must also be acknowledged that the triggers with the lowest mean levels across participants and the least variability were loneliness, boredom, and sadness. Therefore, it is possible that a more sensitive measure (perhaps with anchors to compare momentary states to normative emotional experiences) is required to detect more subtle distinctions in emotional states for the purpose of predicting lapse behavior.

The availability of delicious food was more predictive of lapses at certain assessment points compared to others. In particular, from mid- to post-treatment, lapse likelihood differed by average level of self-reported exposure to delicious foods such that the increased risk of lapsing from general level of exposure to delicious foods was attenuated. Interestingly, however, after controlling for individuals’ average level of exposure, the momentary risk of lapsing after exposure to delicious food increased over time. A post-hoc analysis revealed that the likelihood of reporting that delicious foods were available decreased slightly from baseline to end-of-treatment (42.4% to 39.1% of surveys; p = .05). Thus, perhaps participants became better at removing food cues from their personal environments (resulting in a weaker relationship between general food exposure level and lapse occurrence), meaning that when food was availability it was especially influential.

Limitations

The findings of the current study must be considered within the context of several limitations. First, our investigation is limited by the number of variables that could be included within the EMA protocol. Given the burden of completing EMA, we did not include certain variables that may be important in predicting dietary lapse, such as social or interpersonal interactions, TV watching, self-efficacy or motivation, exercising, lack of healthy food availability, and alcohol intake. In particular, social cues or interactions were not directly assessed despite their established influence on lapse frequency (9, 17, 41, 53). Second, the items used to assess physical states were constructed for the present study and were not subjected to a formal factor analysis. Although these items were modeled off of the PANAS, which has been previously used in EMA research, it is possible that the physical state questions functioned somewhat differently than the items assessing affective states. In addition, although EMA is more ecologically valid than retrospective self-reports, it is still subject to limitations of participant insight, awareness, and recall (15). Another limitation is the fact that reporting lapses is confounded with the participant’s understanding of what constitutes a lapse, stringency of the participant’s intended diet, and with the participant’s conscientiousness in reporting lapses. Further, the analyses implemented in the current study only examined whether a cue predicted lapse at the next survey, when in fact, a cue could increase likelihood of lapsing later than the next EMA time point. Moreover, we did not attempt to examine the interacting effect of cues (e.g., if boredom only increases the risk of lapsing when paired with readily available desirable food). Finally, the attrition from the EMA procedures was quite high. Although baseline lapse frequency did not predict missing data at mid- or end-of-treatment and a similar pattern of results was observed when running EMA analyses only with participants who provided EMA data at all three time points, it is possible that the pattern of results would differ if all individuals had completed all assessments.

This project was wholly funded by an R01 grant (R01DK095069) to the first author from the National Institute of Diabetes and Digestive and Kidney Diseases. We also want to thank the participants of this study as well at its research coordinator, Andrew Frohn.