Initial Weight Change and Long-Term Changes in Weight and Compensation during Supervised Exercise Training

Abstract

Introduction:

Our primary aim was to investigate the association between initial weight change and longer-term changes in weight and compensation (predicted weight loss - observed weight loss) during exercise. As secondary aims, we investigated if initial weight change was related to change in cardiometabolic risk markers and energy balance modulators.

Methods:

Two 6-month randomized controlled exercise trials conducted in individuals with overweight or obesity were analyzed (Study one: N=312; Study two: N=102). In both studies, participants in an exercise condition (4 kcals·kg⁻¹·week⁻¹ [KKW], 8 KKW, 12 KKW, or 20 KKW) were split into tertiles based on percent weight change from baseline to week 4. Tertiles 1 and 3 exhibited the least and most initial weight loss, respectively. Changes in endpoints were compared between tertiles.

Results:

At month 6, weight loss was lower in tertile 1 than tertile 3 (Study one: −3.6%, 95% CI −4.6 to −2.6; Study two: −1.8%, 95% CI −3.1 to −0.4; P≤0.034). Tertile 1 also showed greater compensation than tertile 3 in studies one (3.0 kg, 95% CI 2.2 to 3.9) and two (1.5 kg, 95% CI 0.3 to 2.6; P≤0.048). Changes in triglycerides and, in study one, high-density lipoprotein cholesterol were less favorable in tertile 1 versus tertile 3 (P≤0.043); however, changes in other cardiometabolic markers were similar (P≥0.209). In study two, tertile 1 increased energy intake and exhibited maladaptive changes in eating behaviors relative to tertile 3 (P<0.050). No between-tertile differences in cumulative exercise energy expenditure and physical activity were evident (P≥0.321).

Conclusions:

Less initial weight loss was associated with longer-term attenuated weight loss and greater compensation during aerobic exercise training. Individuals who display less initial weight loss during exercise may require early interventions to decrease compensation and facilitate weight loss.

INTRODUCTION

Excess body weight is associated with increased risk for various physical and psychological conditions (1), and over two-thirds of Americans are either overweight or obese (2). Exercise training is advocated in individuals with overweight or obesity because it reduces the risk for a plethora of chronic diseases (3) and can stimulate clinically significant reductions in weight (4). However, inter-individual variability in exercise-induced weight change exists. Within many exercise training studies, some individuals present clinically significant weight loss and others gain weight (5). While exercise training triggers weight-independent benefits (4), substandard weight loss or weight gain attenuates many improvements of exercise (6, 7). Thus, during exercise training, it is crucial to identify characteristics of individuals who experience attenuated weight loss or weight gain and determine the mechanisms underlying this response.

A reliable predictor of long-term weight loss within dietary interventions is initial weight change (e.g., weight change in first 4 weeks) (8). Individuals who exhibit less initial weight loss display poorer changes in weight and cardiometabolic endpoints at the end of studies (9, 10). This early indicator of long-term weight loss success allows interventionists to identify individuals who are likely to experience substandard weight-related outcomes and apply procedures to optimize endpoints (8). Such strategies look to decrease compensation, which is a discrepancy between weight loss achieved and expected (11), and can include more rigorous modifications of key dietary patterns, such as eating behaviors and attitudes (12).

No study has examined if initial weight change during exercise is related to longer-term changes in weight, compensation, and cardiometabolic risk markers. Such investigations are important, given the heterogeneity in exercise-induced weight loss and the unknown indicators of this variability (13). It is also important to examine modulators of energy balance to ascertain the mechanisms underlying differences in weight change and compensation, and to inform approaches that augment the benefits of exercise training.

Our primary aim was to investigate the association between initial weight change from baseline to week 4 and changes in weight and compensation after 6 months of exercise training. We hypothesized that individuals who lose less weight from baseline to week 4 would experience lower weight loss and greater compensation after 6 months of training. As secondary aims, we investigated the associations between initial weight change and 6-month change in cardiometabolic risk markers, components of energy expenditure and energy intake, eating behaviors and attitudes, and compensatory health beliefs.

METHODS

The current analysis utilizes data from two supervised 6-month exercise intervention studies: the Dose Response to Exercise in postmenopausal Women (DREW) study and the Examination of Mechanisms of Exercise-Induced Weight Compensation (E-MECHANIC) study. Trial design, randomization methods, trial dates, blinding procedures, and sample size calculations of both studies have been published (14–17). Previous interventions have found robust relationships between weight change from baseline to week 4 and prolonged weight change (18–20). Accordingly, we chose to examine the association between initial weight change at week 4 and weight change and compensation at the end of these studies. Compensation was defined as the difference between predicted weight loss and observed weight loss. Predicted weight loss was estimated in both studies utilizing a dynamic energy balance model, which accounts for metabolic adaptation and the body composition changes (fat mass and lean mass) during aerobic exercise training, and which overcomes limitations with conventional estimates that assume a 7700 kcal energy deficit leads to a 1 kg reduction in weight (see Appendix, Supplemental Digital Content, Supplementary Information) (17, 21, 22). This model has been validated on previous aerobic training studies (21).

To improve validity of our findings, participants were included if they: (1) were randomized to an exercise condition; (2) completed the trial; (3) performed weight measurements at week 4; and (4) achieved >75% compliance (number of exercise sessions/prescribed exercise sessions) and/or adherence (achieved exercise energy expenditure/prescribed energy expenditure) to the exercise regimen (17, 23, 24).

The DREW study

The DREW study (ClinicalTrials.gov: NCT00011193) was performed at the Cooper Institute and was approved by the institutes Institutional Review Board. All participants provided written informed consent before screening. The study included females with overweight or obesity (body mass index [BMI]: 25.0–43.0 kg·m⁻²) and elevated systolic blood pressure (120.0–159.9 mm Hg). Other exclusion criteria have been detailed (14).

Detailed descriptions of the exercise intervention have been published (14, 15). Participants were enrolled into either a no-exercise control group or one of three exercise groups for 6 months. The three exercise groups included a group that aimed to expend 4 kcals per kg body weight per week (KKW), one that aimed to expend 8 KKW, and another that aimed to expend 12 KKW. All participants expended 4 KKW during the first week; thereafter, participants enrolled to the 4 KKW group continued at this dose, whereas the 8 KKW and 12 KKW groups ramped up their exercise dose by 1 KKW every week until their prescribed dose was reached. Body weight measurements were collected before exercise sessions every week in the exercise training facility on an electronic scale (Siemens Medical Solutions, Malvern, PA). Exercise training was performed on semi-recumbent cycle ergometers and treadmills, with the intensity set at a heart rate equivalent to 50% of baseline peak V̇O₂. Using standard American College of Sports Medicine (ACSM) equations (25), the energy expenditure of exercise was calculated in real time based on the participant’s weight and either watts (cycle ergometer) or speed and gradient (treadmill). Exercise time was adjusted by dividing the stipulated daily caloric dose by energy expenditure rate. All sessions were monitored to ensure the prescribed exercise dose (energy expenditure) was closely met. The intervention was intended to take place over 24 weeks, but two additional weeks were allowed for participants who had not met their exercise doses (14, 15).

Outcomes

Outcome measures were assessed at baseline and month 6. Weight change and compensation at month 6 were primary endpoints, whereas others were secondary or exploratory. During clinical assessment visits, fasting body weight was measured on a calibrated electronic scale (Siemens Medical Solutions, Malvern, PA). Compensation was calculated as described previously, with predicted weight loss estimated using the dynamic energy balance model (17, 21).

Waist circumference was determined (26), and fitness tests were performed as documented to measure peak absolute and relative $\stackrel{.}{\text{V}}{\text{O}}_{\text{2}}$ (14). Further, triglycerides, cholesterol (total cholesterol, low-density lipoprotein cholesterol [LDL-C], high-density lipoprotein cholesterol [HDL-C]), glucose, and insulin concentrations were measured, and blood pressure was determined (14, 27).

The E-MECHANIC study

The E-MECHANIC study (ClinicalTrials.gov: NCT01264406) was conducted at Pennington Biomedical Research Center, with approval of the center’s Institutional Review Board. Males and females with overweight or obesity (BMI: 25.0–45.0 kg·m⁻²) were recruited, with all participants providing written informed consent before enrolment. Exclusion criteria have been reported (16).

The details of the intervention have been reported (16, 17). Participants were randomized to either a no-exercise control group, or two exercise groups: 8 KKW and 20 KKW (17). Participants assigned to the 8 KKW group performed their prescribed dose from the start; conversely, the 20 KKW group ramped up their exercise prescription from 8 KKW during week 1 to 14 KKW during week 2 and 20 KKW during week 3. A Tanita scale (Arlington Heights, IL) was utilized weekly before exercise sessions to measure body weight. All exercise was performed on a treadmill at a speed and gradient that kept participants within a heart range equivalent to 65% to 85% of baseline peak $\stackrel{.}{\text{V}}{\text{O}}_{\text{2}}$. Energy expenditure was calculated in real time based on treadmill speed, treadmill gradient, and participant weight using standard ACSM equations (28). Moreover, energy expenditure was measured using a metabolic cart at weeks 2, 4, 6, 8, 12, 16, and 20 to monitor changes in metabolic and/or biomechanical efficiency. The duration of exercise was adjusted to meet participant’s energy expenditure targets (16, 17). Participants aimed to achieve their total intervention energy expenditure within 24 weeks, though an additional three weeks were allowed if needed.

Outcomes

Outcomes were collected at baseline, month 6, and (for questionnaire data only) week 4. Weight change and compensation at month 6 were primary endpoints; all other endpoints were secondary. Body weight measurements at baseline and month 6 were the average of 3 fasting weights on a calibrated Tanita scale (Arlington Heights, IL) collected over a 14-day period (days 0, 7, and 14 of the 14-day period). Compensation was calculated using methods identical to DREW (see Appendix, Supplemental Digital Content, Supplementary Information) (17, 21).

Waist circumference was measured at baseline and month 6 in clinical assessment visits via a non-extensible tape measurer (Gulick II, Sammons Preston, Chicago, IL), and fat mass and lean mass were determined through dual-energy X-ray absorptiometry (DXA; iDXA, encore software version 13.60; GE Healthcare) prior to the exercise intervention and on completion of the trial. As described previously (16), fitness tests were performed at baseline and follow-up to measure peak and relative $\stackrel{.}{\text{V}}{\text{O}}_{\text{2}}$. The aforementioned cardiometabolic disease risk markers that were assessed in DREW were measured in E-MECHANIC (16, 17, 29).

Energy intake was estimated over a two-week period at baseline and month 6 utilizing doubly labelled water. This method, considered the gold-standard of free-living energy requirements (30), assesses total daily energy expenditure whilst accounting for body composition changes (31, 32). Daily steps were measured for two weeks using SenseWear armbands (Body Media, Pittsburgh, PA), with steps from exercise sessions excluded from the month 6 measurement period. After a 12-h fast, resting metabolic rate (RMR) was measured for 30 min with Max-II metabolic carts at baseline and month 6 (AEI Technologies, Pittsburgh, PA).

Several validated questionnaires were administered at baseline, week 4 and at month 6 to measure constructs of eating behaviors and attitudes, and physical activity. These included appetite ratings on visual analogue scales (VAS) (33) that were administered on two occasions: the laboratory after the consumption of a 190-kcal nutrition bar, and retrospectively during the previous week (34). The Activity Temperament Questionnaire (ATQ) (35), the Compensatory Health Belief Scale (CHBS) (36), the Eating Inventory (37), the Food Craving Inventory (FCI) (38), the Food Preference Questionnaire (FPQ) (39), and the Multifactorial Assessment of Eating Disorder Symptoms (MAEDS) (40) were also administered.

Statistical analysis

The present manuscript is a post hoc analysis of the DREW and E-MECHANIC studies; accordingly, the present analysis utilized the sample size obtained from both studies. In both studies, participants were divided into tertiles based on percent weight change from baseline to week 4, given residuals of changes were non-normally distributed and skewed. Participants in tertile 1 and tertile 3 had the least and most percent weight loss at week 4, respectively.

Similar analyses were performed on both DREW and E-MECHANIC data. Between-tertile differences in continuous and categorical measures were assessed via a one-way ANOVA and chi-squared test, respectively, for baseline and descriptive data. A one-way ANCOVA was used to compare change in primary and secondary endpoints from baseline, with age, exercise group, self-reported race, baseline values, and (for E-MECHANIC only) sex used as covariates. If between-tertile differences in cardiometabolic risk markers were evident, we also included percent weight change as a covariate to determine if variations were weight-dependent. As an exploratory analysis, for participants with all weekly weight data up to week 24, a two-way mixed (tertile-by-time) ANCOVA was performed utilizing the same covariates to examine weight change and compensation during the trial. ANCOVA analyses were used irrespective of normality (41). If data were aspherical, a Greenhouse-Geisser correction was applied for epsilon < 0.75, whereas the Huynh-Feldt correction was used for less severe asphericity (epsilon > 0.75). Where significance occurred, adjusted post hoc pairwise comparisons (Holm-Bonferroni) located differences. Analyses were performed utilizing SPSS version 25 (IBM Corp., Armonk, NY), and α was set at 0.05. Unless noted otherwise, baseline and descriptive data are reported as mean (± standard deviation [SD]), while outcome measures are presented as estimated marginal (EM) means (± 95% confidence intervals [CI]).

RESULTS

The DREW study

Participant characteristics

The DREW study enrolled 362 participants into the three exercise groups. In the present study, 312 participants were analyzed for the primary and secondary endpoints, with 50 participants removed for reasons related to attrition and abidance to the exercise regimen (Supplementary Figure 1, see Appendix, Supplemental Digital Content). The baseline and descriptive characteristics of the tertiles are shown in Table 1. Most variables were similar between tertiles (P ≥ 0.063), although there was a significant effect of tertile on BMI and glucose (P ≤ 0.049).

Table 1.

Demographic and baseline characteristics of participants in the Dose-Response to Exercise in Postmenopausal Women (DREW) study across tertiles of initial percent weight change.

Variable		All (N = 312)			Tertile 1 (N = 104) Least initial weight loss			Tertile 2 (N = 104)			Tertile 3 (N = 104) Most initial weight loss			P
		Mean	SD/%		Mean	SD/%		Mean	SD/%		Mean	SD/%
Ethnicity (N)														0.433
White		195	62.5		71	68.3		63	60.6		61	58.7
African American		96	30.8		28	26.9		33	31.7		35	33.7
Hispanic/other		21	6.7		5	4.8		8	7.7		8	7.7
Exercise group (N)														0.077
4 KKW		134	42.9		43	41.3		46	44.2		45	43.3
8 KKW		87	27.9		31	29.8		20	19.2		36	34.6
12 KKW		91	29.2		30	28.8		38	36.5		23	22.1
Age (years)		57.3	6.5		58.0	6.5		57.3	6.8		56.6	6.0		0.272
Weight (kg)		83.7	11.8		81.9	11.3		84.6	11.6		84.6	12.5		0.179
BMI (kg·m−2)		31.5	3.8		30.8	3.4		31.9	3.9		31.9	4.1		0.049
Waist circumference (cm)		100.3	11.7		99.5	11.0		101.8	11.7		99.7	12.5		0.299
Fitness variables
Peak absolute $\stackrel{.}{\text{V}}{\text{O}}_{\text{2}}$ (L·min−1)		1.29	0.24		1.25	0.24		1.31	0.24		1.30	0.25		0.175
Peak relative $\stackrel{.}{\text{V}}{\text{O}}_{\text{2}}$ (ml·kg−1·min−1)		15.5	2.9		15.4	2.6		15.6	2.9		15.5	3.0		0.798
Cardiometabolic disease risk markers
Triglycerides (mg·dL−1)		128.0	62.8		126.4	69.2		131.9	60.1		125.8	59.2		0.740
Total cholesterol (mg·dL−1)		201.1	29.6		198.5	28.2		201.8	31.9		203.0	28.8		0.536
LDL-C (mg·dL−1)		117.7	26.4		113.0	24.8		119.6	27.2		120.4	26.7		0.090
HDL-C (mg·dL−1)		58.0	14.6		60.6	14.6		55.9	12.5		57.4	16.1		0.063
Glucose (mg·dL−1)		94.5	8.6		94.5	8.5		96.0	7.5a		92.9	9.6		0.036
Insulin (pmol·L−1)		73.6	41.6		76.4	47.0		73.1	35.9		71.2	41.7		0.693
Systolic blood pressure (mm Hg)		139.0	13.3		140.8	13.8		138.1	13.1		138.1	13.0		0.236
Diastolic blood pressure (mm Hg)		80.8	8.7		81.5	9.1		79.6	8.3		81.3	8.6		0.209

Abbreviations: BMI, body mass index; HDL-C, high-density lipoprotein cholesterol; KKW, kilocalories·kilogram⁻¹·week⁻¹; LDL-C, low-density lipoprotein cholesterol; $\stackrel{.}{\text{V}}{\text{O}}_{\text{2}}$, oxygen uptake.

P is derived from ANOVA for continuous variables and contingency chi-square test for categorical variables.

^aSignificant difference in tertile 2 vs. tertile 3 (P = 0.03).

Values are mean and SD for continuous variables and number (n) and percentage for categorical variables.

Week 4 percent weight change differed between all tertiles per the design of this analysis (P < 0.001), with tertile 1 (least initial weight loss/initial weight gain), tertile 2 and tertile 3 (most initial weight loss) exhibiting 1.6% (± 1.0%), −0.2% (± 0.4%), and −2.9% (± 2.0%) mean (± SD) weight change, respectively, at week 4. Cumulative energy expended at exercise sessions was not different among tertiles at week 4 (tertile 1: 1547 ± 370 kcal; tertile 2: 1619 ± 325 kcal; tertile 3: 1569 ± 367 kcal; P = 0.321) and at the end of the trial (tertile 1: 15127 ± 7686 kcal; tertile 2: 14761 ± 6455 kcal; tertile 3: 13809 ± 6077 kcal; P = 0.351).

Change in outcome data at month 6

Weight and BMI change at follow-up were different between tertiles (P < 0.001; Table 2). Post hoc tests revealed tertile 1 (least initial weight loss/initial weight gain) exhibited less weight loss compared to tertile 2 (−1.0 kg, 95% CI −1.8 to −0.1; P = 0.022) and tertile 3 (most initial weight loss; −3.0 kg, 95% CI −3.8 to −2.2; P < 0.001), plus tertile 2 showed less weight loss compared to tertile 3 (−2.0 kg, 95% CI −2.8 to −1.2; P < 0.001; Table 2). Tertile 1 and tertile 2 also showed a smaller reduction in BMI compared to tertile 3 (P < 0.001), but no differences between tertiles for change in waist circumference (P = 0.270).

Table 2.

Predicted weight change, compensation, and change in outcome variables after 6 months of exercise in participants from the Dose-Response to Exercise in Postmenopausal Women (DREW) study.

Variable		Tertile 1 (N = 104) Least initial weight loss			Tertile 2 (N = 104)			Tertile 3 (N = 104) Most initial weight loss			P
		EM mean	95% CI		EM mean	95% CI		EM mean	95% CI
Weight (kg)		−0.2	−0.8 to 0.4		−1.2c	−1.8 to −0.6		−3.2ab	−3.8 to −2.6		<0.001
Weight (%)		−0.2	−0.9 to 0.5		−1.4c	−2.1 to −0.7		−3.8ab	−4.5 to −3.1		<0.001
BMI (kg·m−2)		−0.1	−0.4 to 0.1		−0.5	−0.7 to −0.2		−1.3ab	−1.5 to −1.0		<0.001
Waist circumference (cm)		−3.1	−4.4 to −1.8		−1.9	−3.2 to −0.6		−3.2	−4.6 to −1.9		0.270
Predicted weight change		−2.0	−2.1 to −1.9		−2.0	−2.2 to −1.9		−2.0	−2.1 to −1.9		0.856
Weight compensation (kg)		1.8	1.2 to 2.4		0.8c	0.2 to 1.4		−1.2ab	−1.8 to −0.6		<0.001
Fitness variables
Peak absolute $\stackrel{.}{\text{V}}{\text{O}}_{\text{2}}$ (L·min−1)		0.05	0.02 to 0.09		0.07	0.04 to 0.10		0.05	0.02 to 0.09		0.756
Peak relative $\stackrel{.}{\text{V}}{\text{O}}_{\text{2}}$ (ml·kg−1·min−1)		0.7	0.3 to 1.2		1.0	0.6 to 1.4		1.3	0.9 to 1.7		0.194
Cardiometabolic disease risk markers
Triglycerides (mg·dL−1)		3.6	−4.7 to 12.0		0.5	−7.8 to 8.8		−11.2a	−19.5 to −2.9		0.035
Total cholesterol (mg·dL−1)		−1.4	−5.7 to 2.8		3.8	−0.4 to 8.0		0.5	−3.7 to 4.8		0.229
LDL-C (mg·dL−1)		−0.4	−4.7 to 3.9		2.9	−1.4 to 7.1		1.6	−2.7 to 5.8		0.568
HDL-C (mg·dL−1)		−3.5	−4.9 to −2.1		−0.1c	−1.5 to 1.3		1.1a	−0.3 to 2.5		<0.001
Glucose (mg·dL−1)		−1.8	−3.1 to −0.5		−1.0	−2.3 to 0.3		−1.2	−2.5 to 0.1		0.690
Insulin (pmol·L−1)		−2.0	−7.9 to 3.9		−5.6	−11.3 to 0.2		−3.3	−9.1 to 2.4		0.690
Systolic blood pressure (mm Hg)		−0.6	−3.0 to 1.7		−1.5	−3.9 to 0.8		−0.5	−2.8 to 1.9		0.795
Diastolic blood pressure (mm Hg)		0.7	−0.7 to 2.0		−0.5	−1.9 to 0.9		0.7	−0.7 to 2.1		0.395

Abbreviations: BMI, body mass index; CI, confidence interval; EM, estimated marginal; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; $\stackrel{.}{\text{V}}{\text{O}}_{\text{2}}$, oxygen uptake.

P is derived from ANCOVA. When significant, post hoc comparisons among tertiles were adjusted with Holm-Bonferroni corrections.

^aSignificant difference between tertile 1 and tertile 3 (P < 0.05).

^bSignificant difference between tertile 2 and tertile 3 (P < 0.05).

^cSignificant difference between tertile 1 and tertile 2 (P < 0.05)

Values are estimated marginal means and 95% CI adjusted for age, ethnicity, group, and baseline.

Predicted weight change did not differ between tertiles (P = 0.856), yet one-way ANCOVA showed initial weight change tertile was linked to compensation (P < 0.001). Post hoc tests revealed that tertile 1 showed 3.0 kg (95% CI 2.2 to 3.9) more compensation than tertile 3 (P <0.001). Further, tertile 1 exhibited greater compensation compared to tertile 2 (P = 0.025), and tertile 3 displayed lower compensation than tertile 2 (P < 0.001; Table 2).

There was a significant effect of tertile on change in HDL-C (P < 0.001; Table 2), with tertile 1 having a decrease in HDL-C relative to tertile 2 (P = 0.002) and tertile 3 (P < 0.001). Moreover, change in triglycerides was different between tertiles (P = 0.035); specifically, tertile 1 showed an increase in triglycerides versus tertile 3 (P = 0.043; Table 2). The between-tertile variations remained significant for change in HDL-C (P < 0.001) but not change in triglycerides (P = 0.329) when controlled for percent weight change at month 6. Change in other cardiometabolic risk markers and fitness variables were similar between tertiles (all P ≥ 0.194; Table 2).

Weekly weight change and compensation

In participants with all weekly weights (N = 110), there was a main effect of tertile on weight change (P < 0.001), though no time or tertile-by-time effects were seen (P ≥ 0.175; Figure 1). Post hoc comparisons for main tertile effect showed that tertile 1 displayed less weight loss versus tertile 2 and tertile 3 (P ≤ 0.027), while weight loss was also lower in tertile 2 versus tertile 3 (P < 0.001). Although predicted weight was expected to decrease over time (main effect of time; P < 0.001), there were no effects of tertile and no tertile-by-time interaction for predicted weight change (P ≥ 0.083; Supplementary Figure 2, see Appendix, Supplemental Digital Content). There was a main effect of tertile on compensation (P < 0.001), with tertile 1 exhibiting greater compensation than tertiles 3 (P < 0.001) and tertile 2 showing higher compensation than tertile 3 (P < 0.001; Figure 1). No main effect of time and no tertile-by-time interaction was revealed, however (P ≥ 0.174). All weekly weight change and compensation data are shown in Supplementary Figure 3 (see Appendix, Supplemental Digital Content).

Figure 1.

(A) Weight change data for participants in the DREW study with all weekly weight data up to week 24 (N = 110 [tertile 1: N = 34; tertile 2: N = 41; tertile 3: N = 35]). (B) Compensation data for participants in the DREW study with all weekly weight data up to week 24 (N = 110 [tertile 1: N = 34; tertile 2: N = 41; tertile 3: N = 35]). Data are from weekly weight measurements performed before exercise sessions. Participants in tertile 1 and tertile 3 had the least and most percent weight loss at week 4, respectively. Black arrows represent point where tertiles were calculated. Values are estimated marginal means (95% CI) adjusted for age, ethnicity, group and baseline.

The E-MECHANIC study

Participant characteristics

In total, 133 participants were enrolled into exercise groups, although 102 were used in the analysis because of attrition, inadequate adherence/compliance, or no weight data at week 4 (Supplementary Figure 4, see Appendix, Supplemental Digital Content). There were no differences in most baseline characteristics among tertiles (P ≥ 0.053), though significant differences were seen for glucose (Table 3), CHBS total score, restraint, and FCI sweet score (all P ≤ 0.034; Supplementary Table 1, see Appendix, Supplemental Digital Content).

Table 3.

Demographic and baseline characteristics of participants in the Examination of Mechanisms of Exercise-Induced Weight Compensation (E-MECHANIC) study across tertiles of initial percent weight change.

Variable		All (N = 102)			Tertile 1 (N = 34) Least initial weight loss			Tertile 2 (N = 34)			Tertile 3 (N = 34) Most initial weight loss			P
		Mean	SD/%		Mean	SD/%		Mean	SD/%		Mean	SD/%
Ethnicity (N)														0.608
White		71	69.6		24	70.6		23	67.6		24	70.6
African American		29	28.4		10	29.4		11	32.4		8	23.5
Hispanic/other		2	2.0		0	0		0	0		2	5.9
Exercise group (N)														0.624
8 KKW		54	52.9		16	47.1		20	58.9		18	52.9
20 KKW		48	47.1		18	52.9		14	41.2		16	47.1
Sex (N)														0.371
Male		30	29.4		8	23.5		13	38.2		9	26.5
Female		72	70.6		26	76.5		21	61.8		25	73.5
Age (years)		48.8	11.9		50.7	11.8		47.3	12.3		48.6	11.6		0.506
Weight (kg)		87.1	15.4		85.7	15.8		87.1	15.1		88.5	15.7		0.753
BMI (kg·m−2)		31.1	4.5		30.8	4.9		30.9	4.4		31.6	4.3		0.732
Waist circumference (cm)		97.3	11.9		95.1	10.9		97.4	12.3		99.3	12.3		0.347
Fat mass (kg)		36.2	9.5		36.2	10.5		35.1	9.3		37.3	8.6		0.652
Lean mass (kg)		47.9	10.1		46.7	10.0		49.0	10.5		48.1	9.9		0.654
Fitness variables
Peak absolute $\stackrel{.}{\text{V}}{\text{O}}_{\text{2}}$ (L·min−1)		2.05	0.54		1.93	0.49		2.14	0.66		2.07	0.43		0.273
Peak relative $\stackrel{.}{\text{V}}{\text{O}}_{\text{2}}$ (ml·kg−1·min−1)		23.9	5.3		22.8	4.0		24.9	6.7		23.9	4.9		0.253
Cardiometabolic disease risk markers
Triglycerides (mg·dL−1)		109.5	51.8		108.3	62.4		114.6	47.7		110.1	44.4		0.695
Total cholesterol (mg·dL−1)		204.9	35.8		197.7	40.3		213.6	33.3		203.3	32.3		0.175
LDL-C (mg·dL−1)		123.4	27.3		117.9	30.6		128.2	26.7		124.2	23.9		0.295
HDL-C (mg·dL−1)		59.5	16.9		59.0	20.3		62.5	16.3		57.1	13.6		0.414
Glucose (mg·dL−1)		92.4	7.6		89.6	7.1		94.1	6.8a		93.3	8.2		0.034
Systolic blood pressure (mm Hg)		119.8	9.9		118.2	12.1		120.1	9.0		120.9	8.3		0.512
Diastolic blood pressure (mm Hg)		76.7	7.2		75.6	8.8		76.9	6.1		77.6	6.4		0.512
Energy intake (kcal·day−1)		2498	471		2389	479		2513	494		2591	430		0.207
Steps·day−1		6180	2254		5968	1939		6093	2179		6468	2612		0.645
RMR (kcal·day−1)		1519	284		1474	304		1502	276		1582	268		0.271

Abbreviations: BMI, body mass index; HDL-C, high-density lipoprotein cholesterol; KKW, kilocalories·kilogram⁻¹·week⁻¹; LDL-C, low-density lipoprotein cholesterol; RMR, resting metabolic rate; $\stackrel{.}{\text{V}}{\text{O}}_{\text{2}}$, oxygen uptake.

P is derived from ANOVA for continuous variables and contingency chi-square test for categorical variables.

^aSignificant difference in tertile 2 vs. tertile 1 (P ≤ 0.045).

Values are mean and SD for continuous variables and number (n) and percentage for categorical variables.

In accord with analysis design, mean (± SD) week 4 percent weight change was different between tertile 1 (least initial weight loss/most initial weight gain; 2.1% [± 0.9%]), tertile 2 (0.8% [± 0.3%]), and tertile 3 (most initial weight loss; −0.5% [± 0.7%]) at week 4 (P < 0.001). Cumulative energy expenditure at exercise sessions was not different among tertiles at week 4 (tertile 1: 3915 ± 1425 kcal; tertile 2: 3779 ± 1323 kcal; tertile 3: 3990 ± 1565 kcal; P = 0.829) and at the end of the trial (tertile 1: 28231 ± 11570 kcal; tertile 2: 26328 ± 12142 kcal; tertile 3: 27810 ± 13357 kcal; P = 0.802).

Change in outcome data at month 6

Initial weight change tertile influenced weight change at month 6 (P ≤ 0.042), with tertile 1 (least initial weight loss/most initial weight gain) presenting less weight loss compared to tertile 3 (most initial weight loss; −1.5 kg, 95% CI −2.7 to −0.3; P ≤ 0.042; Table 4). There was only a tendency for a difference in change in BMI between tertiles (P = 0.052), and change in waist circumference did not differ between tertiles (P = 0.227). Further, changes in fat mass and lean mass were not significantly different between tertiles (P ≥ 0.189).

Table 4.

Predicted weight change, compensation, and change in outcome variables after 6 months and 4 weeks (questionnaire data only) of exercise in participants from the Examination of Mechanisms of Exercise-Induced Weight Compensation (E-MECHANIC) study.

Variable		Tertile 1 (N = 34) Least initial weight loss			Tertile 2 (N = 34)			Tertile 3 (N = 34) Most initial weight loss			P
		EM mean	95% CI		EM mean	95% CI		EM mean	95% CI
Weight (kg)		−0.3	−1.2 to 0.5		−0.8	−1.7 to 0.0		−1.8a	−2.7 to −1.0		0.042
Weight (%)		−0.5	−1.4 to 0.5		−1.0	−1.9 to −0.0		−2.2a	−3.2 to −1.3		0.033
BMI (kg·m−2)		−0.2	−0.5 to 0.2		−0.3	−0.6 to −0.0		−0.7	−1.0 to −0.4		0.052
Waist circumference (cm)		−0.5	−1.6 to 0.6		−0.8	−1.9 to 0.3		−1.8	−2.9 to −0.7		0.227
Fat mass (kg)		−0.6	−1.3 to 0.1		−0.7	−1.4 to 0.4		−1.5	−2.2 to −0.7		0.189
Lean mass (kg)		−0.1	−0.5 to 0.3		−0.2	−0.6 to 0.1		−0.4	−0.8 to −0.1		0.512
Predicted weight change (kg)		−3.0	−3.1 to −2.9		−3.1	−3.2 to −3.0		−3.0	−3.2 to −2.9		0.641
Weight compensation (kg)		2.7	1.8 to 3.5		2.3	1.4 to 3.1		1.2a	0.4 to 2.0		0.043
Fitness variables
Peak absolute $\stackrel{.}{\text{V}}{\text{O}}_{\text{2}}$ (L·min−1)		0.18	0.11 to 0.26		0.17	0.10 to 0.24		0.17	0.09 to 0.24		0.933
Peak relative $\stackrel{.}{\text{V}}{\text{O}}_{\text{2}}$ (ml·kg−1·min−1)		1.5	0.2 to 2.8		2.0	0.7 to 3.3		2.0	0.7 to 3.3		0.831
Cardiometabolic risk markers
Triglycerides (mg·dL−1)		5.3	−5.8 to 16.3		−9.1	−20.1 to 1.9		−18.2a	−29.1 to −7.3		0.013
Total cholesterol (mg·dL−1)		−4.3	−10.5 to 2.0		−5.0	−11.3 to 1.2		0.3	−5.9 to 6.4		0.428
LDL-C (mg·dL−1)		−4.4	−9.8 to 1.0		−2.3	−7.6 to 3.1		2.2	−3.1 to 7.5		0.209
HDL-C (mg·dL−1)		−1.0	−3.2 to 1.1		−0.9	−3.1 to 1.3		1.8	−0.3 to 3.9		0.115
Glucose (mg·dL−1)		0.0	−2.0 to 2.0		−1.7	−3.7 to 0.3		−0.2	−2.1 to 1.8		0.424
Systolic blood pressure (mm Hg)		−2.5	−4.8 to −0.1		−4.7	−7.0 to −2.3		−4.7	−7.0 to −2.3		0.329
Diastolic blood pressure (mm Hg)		−1.4	−3.2 to 0.3		−1.9	−3.6 to −0.2		−1.1	−2.8 to 0.7		0.789
Energy intake (kcal·day−1)		321	155 to 487		76	−90 to 242		−92a	−259 to 75		0.003
Steps·day−1		−487	−1218 to 245		−329	−1024 to 366		−509	−1199 to 180		0.926
RMR (kcal·day−1)		17	−53 to 88		45	−28 to 117		72	−1 to 145		0.567
Questionnaires, Week 4
CHBS		−2.2	−4.0 to −0.4		1.0	−0.8 to 2.8		0.3	−1.4 to 2.1		0.039
FPQ, High fat		0.0	−0.5 to 0.5		−0.6	−1.1 to −0.1		−0.9a	−1.3 to −0.4		0.044
FPQ, High fat and high complex carbohydrates		0.2	−0.3 to 0.6		−0.4	−0.9 to 0.1		−0.9a	−1.3 to −0.4		0.018
FPQ, High sugars		0.0	−0.4 to 0.5		−0.5	−0.9 to −0.0		−0.8a	−1.2 to −0.4		0.040
FPQ, High protein		−0.0	−0.5 to 0.5		−0.7	−1.2 to −0.3		−0.9	−1.4 to −0.4		0.039
FPQ, Low fat		0.0	−0.4 to 0.5		−0.5	−1.0 to −0.1		−0.7	−1.1 to −0.3		0.049
FPQ, Low fat and high protein		−0.0	−0.6 to 0.5		−0.8	−1.3 to −0.3		−0.9	−1.4 to −0.4		0.043
FPQ, Low fat and high sugars		0.2	−0.2 to 0.6		−0.3	−0.7 to 0.1		−0.7a	−1.1 to −0.3		0.016
Questionnaires, Month 6
MAEDS, Avoidance of forbidden foods		−1.5	−3.4 to 0.4		0.1	−1.8 to 1.9		2.3a	0.4 to 4.2		0.024

Abbreviations: BMI, body mass index; CHBS, Compensatory Health Belief Scale; CI, confidence interval; EM, estimated marginal; FPQ, Food Preference Questionnaire; HDL-C, high-density lipoprotein cholesterol; KKW, kilocalories·kilogram⁻¹·week⁻¹; LDL-C, low-density lipoprotein cholesterol; MAEDS, Multifactorial Assessment of Eating Disorder Symptoms; RMR, resting metabolic rate; $\stackrel{.}{\text{V}}{\text{O}}_{\text{2}}$, oxygen uptake.

P is derived from ANCOVA. When significant, post hoc comparisons among tertiles were adjusted with Holm-Bonferroni corrections.

^aSignificant difference between tertile 1 and tertile 3 (P < 0.05).

Values are estimated marginal means (95% CI) adjusted for age, sex, ethnicity, group, and baseline.

There were no between-tertile differences in predicted weight change (P = 0.641). One-way ANCOVA showed a significant effect on compensation (P = 0.043). Post hoc comparisons demonstrated that tertile 1 had greater compensation relative to tertile 3 (1.5 kg, 95% CI 0.3 to 2.6; P = 0.048).

There was an effect of initial weight change on change in triglycerides at month 6 (P = 0.013). Individuals categorized in tertile 3 had a reduction in triglyceride levels compared to tertile 1 (P = 0.011; Table 4). The effect of initial weight change was reduced when percent weight change at month 6 was controlled, with only a tendency for a between-tertile difference in change in triglycerides seen (P = 0.050). Change in other cardiometabolic markers and fitness variables were similar among tertiles (all P ≥ 0.115; Table 4).

Change in energy intake differed between initial weight change tertiles (P = 0.031), with change in energy intake higher in tertile 1 compared to tertile 3 (P = 0.003). No between-tertile differences in change in steps·day⁻¹ and RMR were seen (all P ≥ 0.567; Table 4). Tertile 1 had a reduction in MAEDS avoidance of forbidden foods relative to tertile 3 at month 6 (P = 0.020; Table 4), yet no other between-tertile changes were seen (all P ≥ 0.085; Supplementary Table 2, see Appendix, Supplemental Digital Content).

Change in outcome data at week 4

There was an effect of tertile on change in CHBS score (P = 0.039; Table 4), although there was only a tendency for a difference after pairwise adjustments (P = 0.050). Tertile 3 showed a decrease in several FPQ-assessed food preferences compared to tertile 1, including high fat, high fat/high complex carbohydrate, high sugar, and low fat/high sugar foods (all P ≤ 0.041; Table 4). No other between-tertile differences in questionnaire change scores were seen (all P ≥ 0.081; Supplementary Table 3, see Appendix, Supplemental Digital Content).

Weekly weight change and compensation

For participants with all weekly weight data up to week 24 (N = 63), there was a main effect of tertile (P = 0.001) but no main effect of time and no tertile-by-time interaction (P ≥ 0.586; Figure 2). The main tertile effect indicated that tertile 1 showed less weight loss than tertile 3 (P = 0.001); further, tertile 2 had less weight loss compared to tertile 3 (P = 0.019). Two-way ANCOVA for predicted weight change showed weight was predicted to decrease (main effect of time; P < 0.001), yet no main effect of tertile (P = 0.961) and no tertile-by-time interaction (P = 0.840) was observed (Supplementary Figure 5, see Appendix, Supplemental Digital Content). There was a main effect of tertile on compensation (P = 0.001), with lower compensation in tertile 3 relative to tertiles 1 (P = 0.001) and 2 (P = 0.020; Figure 2); however, no time or tertile-by-time effects were shown (P ≥ 0.580). All weekly weight change and compensation data for participants are depicted in Supplementary Figure 6 (see Appendix, Supplemental Digital Content).

Figure 2.

(A) Weight change data for participants in the E-MECHANIC study with all weekly weight data up to week 24 (N = 63 [tertile 1: N = 17; tertile 2: N = 21; tertile 3: N = 25]). (B) Compensation data for participants in the E-MECHANIC study with all weekly weight data up to week 24 (N = 63 [tertile 1: N = 17; tertile 2: N = 21; tertile 3: N = 25]). Data are from weekly weight measurements performed before exercise sessions. Participants in tertile 1 and tertile 3 had the least and most percent weight loss at week 4, respectively. Black arrows represent point where tertiles were calculated. Values are estimated marginal means (95% CI) adjusted for age, ethnicity, group, baseline, and sex.

DISCUSSION

The present analysis demonstrated that during aerobic exercise training, individuals with overweight or obesity who showed less initial weight loss (or most initial weight gain) at week 4 displayed less weight loss, greater compensation, and poorer changes in blood lipids at the end of 6 months compared to those with most initial weight loss. Additionally, we showed that individuals with the least initial weight loss showed a greater rise in energy intake and maladaptive alterations in eating attitudes and behaviors relative to those with most initial weight loss, suggesting this drove differences in weight-related endpoints.

Exercise training is recommended for individuals with overweight or obesity due to the numerous health benefits it stimulates (3), but there is substantial unexplained heterogeneity in weight change, which can alter many improvements seen (5, 42). Our findings showed that individuals with less initial weight loss exhibited poorer long-term changes in weight in two exercise training studies. All individuals in tertile 1 showed weight gain at week 4 in both exercise studies. Generally speaking, this is at odds with dietary interventions which have characterized individuals with less initial weight loss as those who attain 0.5%−3.0% weight loss at 1–2 months (8). Although the degree of discrepancy could occur because—at odds with many dietary studies—weight loss was not a primary objective of exercise training in DREW and E-MECHANIC, the lower initial weight loss (and weight gain) shown in our study compared to these studies is unsurprising, as it is easier to provoke greater energy deficits through diet than exercise (17). It is less clear whether initial weight change affects long-term changes in body composition, because we observed non-significant between-tertile variations seen in fat mass and lean mass in E-MECHANIC participants. Still, our findings broadly support work demonstrating that weight change from baseline to week 4 is associated with weight loss after 12 weeks of exercise (43) and suggest, akin to dietary interventions (44), that initial weight change at week 4 can be monitored by interventionists to forecast weight loss during long-term exercise interventions. In light of previous trials (10, 19), it is tempting to speculate that the associations we observed occur for even longer periods after the onset of exercise training, but additional studies beyond 6 months are needed.

In dietary studies, individuals with less initial weight loss often display inadequate adherence, and this leads to lower energy deficits and in turn poorer changes in weight-associated endpoints (8, 19). Consequently, initial weight change can be used as an indicator of adherence to dietary restriction, and can pinpoint individuals who require early support to improve adherence, increase energy deficits and attain weight loss targets (8). We found that the amount of exercise performed is not implicated in the associations between initial weight change and weight change at month 6, as we saw no between-tertile differences in cumulative exercise energy expenditure. Participants who completed higher exercise doses in DREW (8KKW and 12 KKW) and E-MECHANIC (20 KKW) ramped up their exercise prescription during the initial phases of exercise training (14, 16), which would have lessened cumulative energy expenditure differences between exercise groups (doses) at week 4. It is interesting, however, that between-tertile differences in weight at month 6 also occurred independent of cumulative exercise energy expenditure. Our findings are consistent with the primary weight loss results from our two studies (5, 17) and potentially imply that at exercise doses typically recommended for health, inter-individual variations in weight occur and exercise energy expenditure has limited influence on initial and long-term weight change variability (21).

Our results indicate that the associations between initial weight change and long-term weight change are due to differences in compensation to exercise-induced energy deficits. Specifically, during exercise training, individuals who present less initial weight loss show greater weight compensation. Consistent with previous studies (17, 42), the variations in compensation in response to exercise were likely driven by differences in energy intake. Indeed, in E-MECHANIC, we saw a greater rise in free-living energy intake at month 6 in individuals who lost less weight (or gained most weight) initially compared to those who lost most weight, yet we saw no between-tertile differences in change in physical activity or RMR. We additionally saw that those with less initial weight loss presented a reduction in avoidance of forbidden foods at month 6 and a smaller decline in food preferences, including those foods high in fat and sugar, at week 4 compared to individuals with the most initial weight loss. Shifts in food preferences (45) and an elevation in avoidance of forbidden foods (46) affect food intake. It is thus possible, given the differences in week 4 food preferences, that changes in eating patterns started during the initial phases of exercise. This suggests that interventionists could implement early strategies that attenuate food preferences and increase avoidance of unhealthy foods in individuals who display substandard weight change promptly during exercise. Such strategies have been frequently utilized in dietary regimens (12) and could comprise early nutritional classes that assist individuals controlling portions sizes and food cravings, with a particular focus on unhealthy foods in food categories of the FPQ (cakes, doughnuts, and potato chips) (39). Similarly, as demonstrated by the findings of Unick and colleagues (47), early interventional support focussed on goal setting and meal planning may be effective (8). Nevertheless, studies with such interventions during exercise training are needed, particularly as changes in several other eating-related constructs (e.g., restraint and food cravings) were similar between tertiles.

Our findings show that individuals who present less initial weight loss experience poorer changes in blood lipids after exercise training. Intriguingly, the relationship between initial weight change and change in HDL-C at month 6 remained after weight change was statistically controlled. Consonant with some postulations (48), it is possible that those with less initial weight loss exhibited an increase in energy intake early in response to exercise, which resulted in decreases in HDL-C that were not mitigated by longer-term weight changes. However, between-tertile differences in triglyceride were attenuated when weight change was controlled, indicating the greater weight loss experienced by those in tertile 3 drove these findings. We also saw similar responses in other cardiometabolic disease risk markers, waist circumference, and fitness between tertiles. This implies exercise provokes metabolic and health improvements irrespective of weight loss, supporting previous work (15, 49).

There are two notable strengths of our study: it comprises a large sample of participants with overweight or obesity from two exercise training studies; and exercise sessions were supervised, with stringent monitoring of exercise doses. Our study has limitations, however. We indeed did not perform measurements of DXA at week 4, and although our primary endpoint measurements were performed in controlled conditions, our measurements of weight before exercise sessions were performed in testing facilities where less standardization procedures were enforced. Furthermore, most of our sample (93%) consisted of females, suggesting additional studies in males are required. We were also unable to obtain sophisticated measures of energy balance and body composition in the larger-powered DREW study. Nevertheless, despite the smaller sample size, we utilized gold-standard estimates of energy intake, physical activity, RMR, and weight-related constructs in E-MECHANIC at baseline and month 6.

In conclusion, less initial weight loss from baseline to week 4 was associated with diminished weight loss at 6 months in two supervised aerobic exercise interventions comprising individuals with overweight or obesity. Individuals who initially lost less weight or gained weight also showed greater compensation at month 6, and this was likely linked to an increase in energy intake and changes in eating behaviors and preferences conducive to poorer dietary patterns. Though exercise training should be universally advocated, individuals with overweight or obesity who show poor weight change initially during exercise training may benefit from early support to improve eating patterns, decrease compensation, and assist weight loss.