Does Increased Exercise or Physical Activity Alter Ad-Libitum Daily Energy Intake or Macronutrient Composition in Healthy Adults? A Systematic Review

Background The magnitude of the negative energy balance induced by exercise may be reduced due to compensatory increases in energy intake. Objective To address the question: Does increased exercise or physical activity alter ad-libitum daily energy intake or macronutrient composition in healthy adults? Data Sources PubMed and Embase were searched (January 1990–January 2013) for studies that presented data on energy and/or macronutrient intake by level of exercise, physical activity or change in response to exercise. Ninety-nine articles (103 studies) were included. Study Eligibility Criteria Primary source articles published in English in peer-reviewed journals. Articles that presented data on energy and/or macronutrient intake by level of exercise or physical activity or changes in energy or macronutrient intake in response to acute exercise or exercise training in healthy (non-athlete) adults (mean age 18–64 years). Study Appraisal and Synthesis Methods Articles were grouped by study design: cross-sectional, acute/short term, non-randomized, and randomized trials. Considerable heterogeneity existed within study groups for several important study parameters, therefore a meta-analysis was considered inappropriate. Results were synthesized and presented by study design. Results No effect of physical activity, exercise or exercise training on energy intake was shown in 59% of cross-sectional studies (n = 17), 69% of acute (n = 40), 50% of short-term (n = 10), 92% of non-randomized (n = 12) and 75% of randomized trials (n = 24). Ninety-four percent of acute, 57% of short-term, 100% of non-randomized and 74% of randomized trials found no effect of exercise on macronutrient intake. Forty-six percent of cross-sectional trials found lower fat intake with increased physical activity. Limitations The literature is limited by the lack of adequately powered trials of sufficient duration, which have prescribed and measured exercise energy expenditure, or employed adequate assessment methods for energy and macronutrient intake. Conclusions We found no consistent evidence that increased physical activity or exercise effects energy or macronutrient intake.


Introduction
Data from the 2009-2010 National Health and Nutrition Examination Survey (NHANES) suggest that 68 [2].
Medical expenditures associated with the treatment of obesity and obesity related conditions are estimated at greater than $147 billion annually [2]. Data from the NHANES (2003)(2004)(2005)(2006)(2007)(2008) indicated that among adults (18-54 years) approximately 75% of women and 54% of men expressed a desire to lose weight while 61% of women and 39% of men were actively pursuing weight control [3].
Exercise is recommended for weight management by several governmental agencies and professional organizations including the Association for the Study of Obesity [4], the Institute of Medicine [5], the U.S. Federal guidelines on physical activity [6], Healthy People 2020 [7] and the American College of Sports Medicine [8]. Compared with weight loss induced by energy restriction, weight loss achieved by exercise is composed predominantly of fat mass, while fat-free mass is preserved [9][10][11] and resting metabolic rate (RMR) is generally unchanged [12,13], or slightly increased [14,15], factors that may be associated with improved long term weight loss maintenance. However, several reports have demonstrated that the accumulated energy balance induced by an exercise intervention alone produces less of a negative energy balance than theoretically predicted for the imposed level of exercise-induced energy expenditure [16][17][18]. The energy balance induced by exercise training may be reduced due to compensatory changes in energy intake, non-exercise physical activity, or both [19][20][21]; thereby reducing the magnitude of observed weight loss. Although several narrative reviews regarding the impact of exercise on energy intake and appetite hormones have been conducted [22][23][24][25][26][27][28][29], we are aware of only one systematic review/meta-analysis on this topic. Schubert et al. [30] recently published a meta-analysis on the effect of acute exercise on subsequent energy intake that included only studies that assessed energy intake for #24 hours post-exercise in healthy (lean and/or obese), non-smoking individuals. To date, no systematic reviews on the effect of exercise and energy and macronutrient intake have been conducted that evaluated both the effects of acute exercise and exercise training and have included data from studies utilizing a variety of designs, e.g. cross-sectional, acutecrossover, non-randomized and randomized trials. Therefore, the aim of this systematic review was to identify and evaluate studies that have employed a variety of designs to assess the impact of both acute exercise and exercise training on energy and macronutrient intake. Results of this review will clarify our understanding of the association between exercise and energy intake and identify both exercise parameters including mode, frequency, intensity and duration and participant characteristics including age, gender, body weight, activity level that may impact this association. Such information will be useful for the design of weight management trials utilizing exercise, the potential identification of groups of participants for whom exercise may be most effective, and to identify areas for future investigation.

Methods
This systematic review was performed and reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines [31,32].

Objectives
The objective of this systematic review was to address the question: Does increased exercise or physical activity alter ad-libitum daily energy intake or macronutrient composition in healthy adults?

Eligibility Criteria
Primary source articles published in English in peer-reviewed journals were eligible for inclusion in this systematic review if data were presented on energy and/or macronutrient intake by level of exercise or physical activity or changes in energy or macronutrient intake in response to acute exercise or exercise training. Specific eligibility criteria included: Types of studies: Cross-sectional, acute/ short-term (exercise duration ranging from a single 30-min exercise bout to daily exercise over 14 days), and both nonrandomized and randomized trials. Types of participants: Healthy adults (age 18-65 years). Types of exercise interventions: Aerobic and resistance exercise. Types of outcome measures: No restrictions were placed on the assessment methods for the primary outcome (energy/macronutrient intake). Other criteria: There were no restrictions on the length of interventions or the types of comparisons. We included cross-sectional comparisons between participants differing by level of exercise or physical activity and longitudinal pre/post within group changes vs. non-exercise control or vs. a different level of exercise. Articles were excluded if they provided no data on energy or macronutrient intake by level of exercise or physical activity, manipulated or controlled energy intake, or were conducted in non-recreational athletes or individuals with chronic disease(s).

Information Sources
Studies were identified by searching electronic data bases, related article reference lists, and consulting with experts in the field. The search was applied to PubMed (1990-present) and adapted for Embase (1990-present). The last search was conducted on January 4, 2013. The search was developed as a collaborative effort of the research team in consultation with a Kansas University reference librarian and conducted by a co-author (SDH). No attempts were made to contact study investigators or sponsors to acquire any information missing from the published article.

Search Strategy
We used the following search terms in PubMed and Embase to identify potential articles with abstracts for review: exercise[ti (title), ab (abstract) ] or ''physical activity'' [ti,ab] [ti,ab]). Additional search terms were applied to eliminate case reports and studies involving participants with chronic disease, and to retrieve studies published in English and conducted in adults (age 18-65 years). Word truncation and the use of wildcards allowed for variations in spelling and word endings.

Study Selection
Retrieved abstracts were independently assessed for eligibility for inclusion in the review by 2 investigators and coded as ''yes'', ''no'' or ''maybe.'' All investigators who participated in eligibility assessments were trained regarding study inclusion/exclusion criteria and completed practice eligibility assessments on 50 test abstracts prior to actual coding. Eligibility assessments on the practice abstracts were reviewed by the primary author (JED) and any coding problems were discussed. Disagreements regarding eligibility for inclusion were resolved via development of consensus among all co-authors. Full text articles for abstracts coded as ''yes'' or ''maybe'' were retrieved and reviewed by 2 independent coauthors prior to inclusion in the review. An excel spread sheet was developed and used to track eligibility status. 4.14.5) data base [33]. A REDCap data extraction form was developed, pilot tested on a sample of 10 studies (at least 2 studies of each of the 4 study designs included in this review), and revised accordingly. Relevant data were extracted from each manuscript by one author and verified by a second author. Disagreements were resolved by group discussion. Data extracted from each article included basic study information (design, sample size, groups compared, exercise or physical activity groups/intervention(s), participant characteristics (age, gender, BMI, minority status), energy and macronutrient assessment method, and results.

Risk of Bias in Individual Studies
Risk of bias for randomized trials was independently evaluated by two authors using the Cochrane risk of bias tool [34]. Risk of bias was assessed in the following domains: selection bias, performance bias, detection bias, attrition bias, reporting bias, and other bias. A third reviewer resolved any discrepancies in bias coding. Studies were not excluded on the basis of risk of bias.

Synthesis of Results
Articles were grouped by study design: cross-sectional, acute/ short-term, non-randomized, and randomized trials. Considerable heterogeneity existed within study groups for several important study parameters. These parameters included: 1) participant characteristics (age, gender, BMI), 2) physical activity assessment methods (questionnaires, pedometers, accelerometers), 3) exercise prescriptions (mode, frequency, intensity, duration), 4) comparison groups (interventions: pre vs. post-exercise, exercise vs. nonexercise control, varying amounts), 5) intervention length, and 6) energy and macronutrient assessment methods (food frequency questionnaire, weighed and un-weighed food records, direct observation weigh and measure technique). A meta-analysis was therefore considered inappropriate. Results based on the extracted data were instead synthesized and presented grouped by study design.

Results
The initial database search plus hand searching identified 4,668 unique records of which 4,490 were excluded based on review of title and abstract. Full-text articles for the remaining 178 citations were reviewed of which 79 articles did not satisfy our inclusion criteria and were excluded. Thus 99 articles representing 101 studies were included in the review (Figure 1).

Cross-Sectional Studies
The 17 cross-sectional studies identified comprised ,17% of the total number of studies included in this review (Table 1).

Cross-Sectional Studies: Participant Characteristics
Age. The median (range) age across the 15 studies that reported age was 42 (18-60) years.

Cross-Sectional Studies: Results
Energy intake. Seven of 17 (41%) cross-sectional studies indicated significantly higher absolute (kcal/day) [35,[39][40][41]49] or relative (kcal/kg/day) energy intake [38,50] in active compared with less active groups. There were no apparent differences in either basic study design parameters including sample size, assessment methods for both energy intake and physical activity, the number and type of comparison groups, or participant characteristics such as age, gender or BMI between studies that did and did not report a significant association between physical activity and energy intake.

Acute Studies
The 40 acute studies comprised ,40% of the total studies identified for this review ( Table 2). All acute studies employed cross-over designs, which compared energy intake assessed over a time frame of 24 hours or less following an acute exercise bout.

Acute Studies: Study Characteristics
Sample size. The median (range) sample size across the 40 acute studies was 12 .

Acute Studies: Participant Characteristics
Age. The median (range) age across the 36 studies that reported age was 23 (20.1-50) years.
Minority status. No studies described the racial or ethnic composition of the study sample or reported post-exercise energy intake by race or ethnicity.
There was also a significant difference in the percentage of EI from fat by activity level with group 2 consuming a larger proportion of energy from fat than group 4. No significant between group differences in either PRO or CHO consumption.    There were nonsignificant increases in EI in both the M-EX (5%) and L-EX (8%) groups. Daily intake of carbohydrates, protein or fat averaged over the study period was not affected by exercise.      Macronutrient intake. Sixteen of the 40 acute studies (40%) reported data on macronutrient intake [54][55][56][57][58]60,68,72,[76][77][78][79]82,84]. Fifteen studies showed no effect, while one study indicated significantly higher fat and protein intake following exercise compared with non-exercise control [76].

Effect of Study Parameters on Energy Intake
Exercise mode. Three studies provided information relative to the effect of exercise mode on post-exercise energy intake. Balaguera-Cortes et al. [77] reported no effect of either aerobic (treadmill) or resistance exercise while King et al. [56] showed no effect of swimming on absolute post-exercise energy intake. These results are in contrast to those of Laan et al. [65] who showed an increase in absolute post-exercise energy intake following both aerobic (cycling) and resistance exercise; however, relative energy intake was lower following aerobic exercise compared to resistance exercise or control. Exercise intensity. Six acute studies reported the effect of exercise intensity on post-exercise energy intake. Four studies found no effect of exercise intensity on absolute energy intake following exercise [51,60,64,85]; however, Imbeault et al. [51] reported a lower relative energy intake following high intensity exercise (75% VO 2 max) compared with low intensity exercise (35% VO 2 max) or non-exercise control. One study [52] showed a significant increase in absolute energy intake for high (70% VO 2 peak) but not low intensity exercise (40% VO 2 peak); however relative energy intake was lower in both the high and low intensity exercise groups compared with non-exercise controls. One study reported a significant decrease in absolute energy intake following strenuous (40 min/90 W cycle ergometer) but not moderate exercise (40 min/30 W cycle ergometer) in non-obese but not in obese women [75].
Exercise duration. Two studies evaluated the role of exercise duration on post-exercise energy intake with divergent results. King et al. [60] reported no effect of exercise duration on post-exercise energy intake; however, Erdman et al. [64] reported that absolute energy intake was not significantly greater than control following cycle ergometer exercise bouts of 30 and 60 min, but was significantly greater than control following 120 minutes of exercise.
Exercise time of day. Two studies evaluated the effect of the time of day of aerobic exercise on post-exercise energy intake [55,63]. Both studies found no significant difference in absolute post-exercise energy intake between exercise performed in the morning (7 and 8:15 AM) compared to the same exercise performed in the evening (5 and 7:15 PM). However, O'Donoghue et al. [55] showed that relative energy intake at breakfast was lower after morning exercise compared with afternoon exercise or control while relative energy intake at dinner was lower post afternoon exercise compared with control.
Composition of test meals. Four studies evaluated the effect of the macronutrient composition of the test meal on postexercise energy intake. Three studies found no significant differences in absolute post-exercise energy intake compared to rest between low or high fat test meals [70,71,74]. King et al. [59] found no difference in absolute post-exercise energy intake when either high fat/low carbohydrate or low fat/high carbohydrate test meals were presented; however, relative energy intake was significantly lower in the low fat/high carbohydrate, but not the high fat/low carbohydrate condition compared with control.
Time between the end of exercise and the presentation of the test meal. In the one acute study that investigated the effect of time between exercise and presentation of the test meal on energy intake, Verger et al. [67] showed that absolute energy intake increased as the time post-exercise that the test meals were presented increased (immediate to 120 min).

Effect of Participant Characteristics on Energy Intake
Age. No studies evaluated the effect of age on post-exercise energy intake. Studies were generally conducted in young adults with a median age of 23 years.
Gender. Although 6 studies included both men and women [64][65][66][67][68] the data were presented separately in only one study. Verger et al. [67] showed significant increases in absolute EI following exercise (2 hours of non-stop submaximal aerobic athletic activities) in both men and women.
Weight status. Three studies provided data on the effect of weight status on post-exercise energy intake [73,75,86]. George et al. [73] found non-significant differences in absolute post-exercise energy intake between normal and overweight women. Kissileff et al. [75] reported significant decreases in post-exercise energy intake in non-obese, but not obese women, while Ueda et al. [86] found larger energy deficits (i.e. decreased energy intake) induced by exercise in obese compared with normal weight men.
Weight/Diet/Dietary restraint. Three studies evaluated the effect of combinations of weight, dieting status or level of eating restraint on post-exercise energy intake. Harris et al. [61] found no differences in post-exercise energy intake in a sample of men across 5 groups: 1) normal weight/low dietary restraint/nondieting; 2) normal weight/high dietary restraint/non-dieting; 3) overweight/low dietary restraint/non-dieting; 4) overweight/high dietary restraint/non-dieting; and 5) overweight/high dietary restraint/dieting. In a sample of normal weight young women, Lluch et al. [70] found increased absolute post-exercise energy intake in women classified as unrestrained eaters and decreased energy intake in restrained eaters. Relative energy intake compared with rest was greater in restrained compared with unrestrained eaters. Visonia et al. [62] demonstrated a significant interaction between dieting/eating restraint status and study condition (exercise vs. control) on 12-hour energy intake in sample of women. The mean difference in 12-hour energy intake between the exercise and control day was significantly higher for the dieting-high restraint group compared with the non-dieting high restraint group.
Activity level. Two studies evaluated the effect of activity level on post-exercise energy intake. Jokisch et al. [78] showed a significant decrease in post-exercise energy intake compared with control in inactive but not in active men. Larson-Meyer et al. [76] found non-significant differences between post-exercise energy intake and control in a normal weight sample of both habitual walkers ($3 days/wk for $60 min/day) and habitual runners ($ 32 km/wk); however, relative post-exercise energy intake was significantly lower in runners compared with controls, but not in walkers vs. controls.

Short-Term Studies
The 10 short-term studies comprised ,9% of the total studies identified for this review (Table 2). These studies employed crossover designs that compared energy intake assessed over a time frame of 2-14 days during which participants engaged in exercise with energy intake during an equivalent period of no imposed exercise.

Short-Term Studies: Study Characteristics
Sample size. The median (range) sample size for the 10 short-term studies was 9.5 (6-20) Trial length. The median (range) duration of short-term studies was 8 (2-14) days.

Short-Term Studies: Participant Characteristics
Age. The median (range) for age across all short-term studies was 28.3 (23-35) years.
Minority status. No studies describe the racial or ethnic composition of the study sample or reported an association between exercise level and energy intake by race or ethnicity.

Short-Term Studies: Results
Energy intake. Five of 10 short-term studies (50%) reported increased absolute energy intake (,200-335 kcal/day) over periods of 2 to14 days when exercise was imposed compared with a non-exercise control period [87,88,93,95,96]. Three studies that reported increased absolute energy intake showed relative energy intake at a level to maintain a negative energy balance during the exercise period [87,88,93]; however, Tremblay et al [95] showed that participants achieved a positive energy balance when presented with a high fat diet. Macronutrient intake. Seven of the 10 short-term studies (70%) reported macronutrient intake [87][88][89][90]92,94,96]. Four of 7 studies (57%) showed no effect of exercise on macronutrient intake [89,90,92,94]. Farah et al. [96] reported increased intake of carbohydrate and protein while Stubbs et al. [87] observed increased intake of carbohydrate and fat with exercise compared to control. Whybrow et al. [88] noted increased intake of carbohydrate, fat and protein with exercise vs. control in men but not in women.

Effect of Study Parameters on Energy Intake
Exercise mode. The one study that compared the effect of aerobic and resistance exercise reported no significant difference between exercise and control for short-term energy intake during either aerobic or resistance training [91].
Exercise intensity. No studies evaluated the effect of exercise intensity on short-term energy intake.
Composition of test meals. The one study that compared the effect of the composition of test meals (mixed, high fat, low fat) on short-term energy intake noted increased energy intake when high fat, but not low or mixed fat meals were presented [95].

Effect of Participant Characteristics on Energy Intake
Age. No short-term studies evaluated the effect of age on postexercise energy intake. Studies were generally conducted in young adults with a median age of 28.3 years.
Gender. Results from the two studies that evaluated gender differences in short-term energy intake with exercise reported that absolute energy intake increased in men but not in women [88,93]. Results from 2 separate studies that used identical exercise and energy intake protocols in samples of men [89] and women [87] found increased energy intake with exercise in women, but not in men.
Other participant characteristics. No studies evaluated the effect of weight, physical activity or level of dietary restraint on short-term changes in energy intake induced by exercise.

Non-Randomized Trials
The 12 non-randomized trials comprised ,12% of the total studies identified for this review (Table 3). Most trials (11/12) evaluated changes in energy intake in a single group (no control) assigned to complete a longitudinal exercise training program [18,21,[97][98][99][100][101][102][103][104][105] while one study observed differences in energy intake between women who participated in an 8 week exercise program at a commercial exercise facility with a group of nonexercise volunteer controls [106].

Non-Randomized Trials: Study Characteristics
Sample size/completion rate. The median (range) sample size across the 12 non-randomized trials was 31 . The median (range) rate of trial completion in the 5 trials that provided data on this parameter was 83% (68-87%) of participants who started the intervention [97][98][99][100]105].
Energy and macronutrient assessment. Five trials used non-weighed food records [97,98,100,101,105], 2 used weighed food records [21,102], 3 used test meals [18,103,104], 1 used 12 hour recall [106] and one used a combination of food records and 24-hour recalls [99]. Six studies assessed energy intake at baseline and end [18,21,98,101,104,105], 3 studies completed energy intake assessments at 4 time points [99,100,103], one study at 3 time points [106] while 2 studies collected daily estimates of energy intake over the course of the intervention [97,102].
Minority status. No non-randomized trials described the racial or ethnic composition of the study sample or reported an association between exercise level and energy intake by race or ethnicity.
Participant activity level. Eleven of 12 non-randomized trials described inclusion criteria for level of baseline physical
Macronutrient intake. None of the 8 non-randomized trials that evaluated the effect of exercise training on macronutrient intake reported significant change in the intake of carbohydrate, fat or protein [18,21,[98][99][100]102,105,106].

Effect of Study Parameters on Energy Intake
Exercise mode. No studies evaluated the effect of exercise mode on changes in energy intake in response to exercise training.
Level of exercise energy expenditure/duration. No studies evaluated the effect of exercise energy expenditure/ duration on energy intake in response to exercise training.
Exercise intensity. No studies evaluated the effect of exercise intensity on energy intake in response to exercise training.
Composition of test meals. Three non-randomized trials evaluated energy intake using test meals [18,103,104]; however, the effect of variations in the macronutrient composition of the test meals was not evaluated.

Effect of Participant Characteristics on Energy Intake
Age. No studies evaluated the effect of age on changes in energy intake in response to exercise training.
Gender. Three non-randomized trials provided data on gender differences [100,101,104]. Two trials reported no differences for change in energy intake between men and women in response to exercise training [100,104], while one trial found no effect in men or lean women, and a significant decrease in energy intake with exercise training in obese women [101].
Other participant characteristics. No non-randomized trials were identified that specifically evaluated the effect of weight status, level of physical activity or level of dietary restraint on the energy intake response to exercise training.

Randomized Trials: Study Characteristics
Sample size/completion rate. The median (range) sample size across the 24 randomized trials was 43.5 . The median (range) proportion of randomized participants who completed the intervention and provided data for energy intake for the 23 trials that provided data on this parameter was 74% (21-100%).
Trial length. The median (range) length of the 24 randomized trials was 21  wks.
Exercise mode. Eight of 24 randomized trials involved laboratory based aerobic exercise where participants used a variety of modalities including cycle ergometers, rowers, recumbent cycles, steppers and treadmills [16,110,113,114,117,118,123], 8 evaluated indoor/outdoor walk/jog [107,111,112,115,116,122,124,127], 5 trials employed primarily laboratory based treadmill walking/jogging [108,109,125,126,128], 4 trials involved resistance training only [108,119,120,123], 2 trials used a combination of resistance and aerobic training [118,123] and one trial each involved swimming [124] and laboratory cycle ergometer exercise [129].
Exercise prescription (frequency). The median (range) exercise frequency was 4 (2-7) days/wk in trials/groups random- Intermittent vs. continuous exercise. The one randomized trial that compared changes in energy intake in response to continuous (one-30 minute sessions/day) vs. intermittent exercise (2-15 minute sessions/day) reported no between or within group differences [128].
Composition of test meals. Rosenkilde et al. [121] found no differences in energy intake with exercise training when low or high carbohydrate test meals were offered.

Effect of Participant Characteristics on Energy Intake
Age. No randomized trials evaluated the effect of age on changes in energy intake in response to exercise training.
Gender. Two randomized trials provided data on gender differences for changes in energy intake in response to exercise training [109,120]. Donnelly et al. [109] reported no between group (exercise vs. control) for change in energy intake in response to 16 months of supervised exercise in either men or women. Similarly, Washburn et al. [120] found no between group differences (exercise vs. control) for change in energy intake in response to 6 months of supervised resistance training in either men or women.
Other participant characteristics. No randomized trials were identified that specifically evaluated the effect of weight status, level of physical activity or level of dietary restraint on the energy intake response to exercise training.
Risk of bias. The risk of bias for all randomized trials is presented in Table 5. The description of the procedures for random sequence generation were unclear in the majority of trials (16/24 -,67%). Six trials adequately described randomization procedures and were considered low risk of bias [109,110,113,121,123,124], while 2 trials were considered high risk for randomization bias based on failure to provide any description of the randomization process [119] or randomization based on level of occupational and lifestyle physical activity [127]. With the exception of the trial reported by Cox et al. [124], which adequately described procedures for allocation concealment (low bias) all other randomized trials (96%) provided no description of procedures for allocation concealment. Blinding participants and personnel is not feasible in an exercise trial. Blinding of personnel performing outcome assessments is feasible in exercise trials; however, this was described in only 2 trials [110,121]. Twenty one trials provided no information relative to blinding of outcome assessments, while one trial directly stated that outcome assessments were not blinded [113]. Based on an effectiveness study paradigm the risk of attrition bias is high in the majority of the 24 randomized trials included in this review. Fifty-four percent of trials reported completion rates of less than 80%; however, 22 of 24 of these studies were conducted as efficacy trials where data from participants who were non-adherence to the exercise intervention or outcome assessment protocols were not included in the analysis.

Summary of Evidence
In this paper we systematically reviewed 99 studies that employed a variety of study designs including cross-sectional, acute/short-term, non-randomized and randomized trials to address the question: Does increased exercise or physical activity alter ad-libitum daily energy intake or macronutrient composition in healthy adults? Our results can be summarized as follows.
Energy intake. It is commonly believed that individuals increase energy intake in response to increased physical activity or exercise training. However overall, we found no consistent, compelling evidence that any level of increased physical activity or exercise has any impact on energy intake. Forty-one percent of cross-sectional studies reported higher energy intake among active compared with inactive individuals. However, cross-sectional data precludes determination of cause and effect. Likewise, it is not possible to determine if the higher energy intake observed among inactive individuals meets or exceeds their level of daily energy expenditure or how between group differences in body weight may impact the results of cross-sectional studies. In agreement with the results of the recent meta-analysis by Schubert et al. [30] on acute exercise and subsequent energy intake, our results from both acute and short-term trials suggest that any observed increase in postexercise energy intake only partially compensates for the energy expended during exercise. Thus, in the short-term, exercise results in a negative energy balance. Results from both non-randomized and randomized trials are in agreement with the results from acute and short-term trials. Only 2 of 36 (,6%) non-randomized and randomized trials, ranging in duration from 3 to 72 weeks, report an increase in absolute energy intake in response to exercise training, thus implying that exercise does not result in a compensatory increase in energy intake. A limited number of studies across all study designs have evaluated the effect of exercise parameters and participant characteristics on energy intake. Our results suggest no effect of either exercise parameters including mode (aerobic, resistance), intensity, duration/energy expenditure or participant characteristics including age, gender, weight or physical activity level on energy intake. These results are in contrast to those of Schubert et al. [30] from their meta-analysis on the acute effect of exercise on absolute post-exercise energy intake where they noted individuals with low to moderate levels of physical activity were more likely to reduce energy intake in response to exercise compared to their more active counterparts.
Macronutrient intake. Data on macronutrient intake was reported in 67 of the 99 studies (68%) included in this review. Irrespective of study design we found no consistent evidence for an effect of exercise on macronutrient intake. Forty-four of 54 acute/ short-term, non-randomized and randomized trials (81%) reported no effect of exercise on macronutrient intake, while results of the 10 trials reporting an association were mixed. Thus, it appears individuals do not spontaneously alter the composition of their diets in response to physical activity or exercise.

Limitations in the Available Literature
There are several important limitations in the literature available for this systemic review. The most critical limitation in the available literature is the lack of studies that have been specifically designed and adequately powered to detect significant between or within group differences in energy intake in response to exercise training. With the exception of 2 acute studies [56,61] no studies included in this review were statistically powered to detect between or within group differences in energy or macronutrient intake. Only 5 of 12 non-randomized trials (,42%) [97,100,[102][103][104]] and 3 of 24 randomized trials (,13%) were conducted specifically to evaluate the effect of exercise training on energy and macronutrient intake [112,121,125]. The sample size in each of these 3 trials was ,20 participants/group. Inadequate statistical power may explain the disconnect between the results of our review, which found no effect of exercise training on energy intake and other trials that have evaluated the effect of exercise on body weight. For example, studies have reported high individual variability in weight loss in response to the same level of exercise energy expenditure [18] and no significant increase in weight loss in response to increased level of exercise energy expenditure [11,121]. Both of these observations suggest compensatory increases in energy intake in response to exercise training; however, changes in resting metabolic rate and non-exercise physical activity may also play a role. Because most exercise training studies did not prescribe exercise by level of energy expenditure, assess exercise energy expenditure, or employ adequate methods for the assessment of energy intake, it is not possible to determine the level of exercise induced energy imbalance actually achieved. Only 7 of the 36 non-randomized and randomized trials included in this review (,19%) prescribed exercise by level of energy expenditure [18,98,103,104,113,121,126] while only 3 of those trials included assessments of the actual level of exercise energy expenditure by indirect calorimetry [18,104,126]. Furthermore, the heterogeneity of exercise/physical activity prescriptions prohibits the identification of a specific level that may elicit compensatory changes in energy intake. Only 5 of 36 non-randomized and randomized trials employed more precise estimates of dietary intake such as weigh and measure test meals [18,103,104,121] or observed weigh and measure ad-libitum eating [109]. Two of the 3 trials that measured exercise energy expenditure used only 1 day assessments of energy intake using test meals at baseline and end study [18,104], while one trial employed 4 day food records at baseline, mid and end study [126]. Thus, reported estimates of energy balance should be cautiously interpreted. The available literature is also limited by a preponderance of data from acute/short-term studies as compared to non-randomized and randomized longitudinal trials. For example, approximately 50% of studies reviewed evaluated the acute or short-term (2-14 days) effect of exercise on energy and macronutrient intake. While acute/short-term studies have employed precise methods for the assessment of energy intake (weigh and measure test meals), the short time frame over which energy intake was assessed may be insufficient to demonstrate significant adaptations in energy intake that may be induced by changes in parameters such as aerobic fitness, body weight or hormonal status etc. associated with longer term exercise training. For example, in overweight men and women Kirk et al. [130] have shown that changes in aerobic fitness and weight takes 4 months and does not level off until 9 months in an exercise program with a typical progressive exercise protocol Approximately 60% of acute studies reviewed recruited participants who were relatively young (median age = 23 years), normal weight, and physically active or aerobically fit. Thus, results from acute trials do not generalize to the older, sedentary overweight population in need of weight management. The available literature is also limited by an insufficient number of studies that have evaluated the impact of exercise parameters (e.g. mode, frequency, intensity, duration, time of day, time course) or participant characteristics (e.g. age, gender, ethnicity, weight, activity level) on energy and/or macronutrient intake.

Limitations of this Review
Our conclusions should be cautiously interpreted as they are based on both data from sub-optimal study designs (e.g. crosssectional, acute/short-term, non-randomized trials) and from randomized trials with a high risk of one or more forms of bias. In addition, we did not contact authors to obtain missing data or for clarification of any information presented in the published reports; therefore missing information may reflect reporting bias as opposed to any limitations in the conduct of the study.

Conclusions
The present systematic review found limited evidence to suggest that acute exercise or exercise training has a significant effect on energy or macronutrient intake. However, as previously discussed the available literature on this topic suffers numerous methodological shortcomings. Therefore, we recommend additional randomized trials to specifically evaluate the impact of exercise training on energy and macronutrient intake that: 1) are powered specifically to detect clinically significant differences in energy and/or macronutrient intake; 2) utilize state-of-the-art techniques for the assessment of energy intake such as direct observation weigh and measure or picture-plate-waste and provide multiple measures across the duration of the study; 3) include assessments of exercise energy expenditure across the duration of the study; 4) evaluate and compare levels of exercise for weight management currently recommended by governmental agencies or professional organizations such as the International Association for the Study of Obesity, the Institute of Medicine, and the American College of Sports Medicine (i.e. 60-90 min/day, moderate intensity) to determine differential effects on energy and macronutrient intake; 5) include overweight and obese, sedentary middle-age or older adults; and 6) evaluate both the effect of exercise parameters (e.g. mode, frequency, intensity duration, time course) and participant characteristics (e.g. age, gender, body weight, activity level, ethnicity) on the association between exercise and energy and macronutrient intake.

Supporting Information
Checklist S1 PubMed complete search strategy.