![[Feedback]](/icons/banner/feedbackACT.gif){kind=icon}
![[For Subscribers]](/icons/banner/subscriberACT.gif){kind=icon}
![[Archive]](/icons/banner/archiveACT.gif){kind=icon}
![[Search]](/icons/banner/searchACT.gif){kind=icon}
![[Contents]](/icons/banner/tocACT.gif){kind=icon}
|
|
|
|
|
||
The Journal of Nutrition Vol. 128 No. 1 January 1998,
pp. 61-67
Neuropsychology and Behavioral Neuroscience Program, Department of Psychology, Georgia State University, Atlanta GA 30303-3083
 |
ABSTRACT |
|---|
Abstract
Introduction
Methods
Results
Discussion
References
|
|---|
A fundamental issue in understanding how energy balance is accomplished involves comprehending how changes in intake affect subsequent intake. This was investigated in free-living humans by reanalyzing the data previously collected from 733 adults who were paid to maintain a 7-d diary of everything they ate and when they ate it. Food energy intake during a day was found to only mildly affect intake on the subsequent day (mean r = 
0.07, P < 0.001), but was more strongly negatively related to intake occurring on the second (mean r = 0.18, P < 0.001) and third day (mean r = 0.10, P < 0.001) afterward. Each macronutrient was shown to have a maximal negative relationship with subsequent intake of that same macronutrient, with 2-d lag mean autocorrelations equal to 0.11, P < 0.001 for carbohydrate, equal to 0.18, P < 0.001 for fat, and equal to 0.13, P < 0.001 for protein. These effects on daily intake were found to result from separate negative feedback effects on meal size and frequency. The results suggest that intake affects subsequent intake by persistently setting a long-term bias that, integrated over time, produces a net shift in intake.
Modern society affords humans the opportunity to select foods from a wide array of attractive and nutritious alternatives with varying hedonic properties, nutrient compositions and energy contents. These foods are ingested in a variety of complex situations that contain potent sociocultural influences. The amounts ingested vary with the hour of the day (de Castro 1987 A fundamental issue in understanding how energy balance is accomplished involves comprehending how changes in intake affect subsequent intake. For regulation to occur, food energy intake must in some way restrain subsequent intake. That is, it must produce negative feedback. On a meal-to-meal basis, intake influences subsequent intake by affecting the amount of food remaining in the stomach at the beginning of the next meal. This has a negative influence on the amount of food energy ingested in that meal (de Castro et al. 1986 Accounting for the variance in daily intake has been difficult. For regulation to occur, food energy intake over a day must in some way restrain food energy intake on subsequent days. That is, it must produce negative feedback. Hence, the correlation between food energy intake on a day and that ingested on the next day, the autocorrelation, should be negative. Across-subjects daily intakes are positively correlated (Hankin et al. 1967 This study investigates how and when intake over a day affects subsequent intake by reanalyzing 7-d diet-diary reports of intake that have been collected over the last decade. Daily intake effects on subsequent day's intake were investigated with an autocorrelational analysis including univariate and linear structural modeling approaches. As in the earlier research, simple negative feedback was not detected. Rather, delayed negative feedback was reported, with the negative feedback from food energy intake delayed such that food energy intake affected subsequent food energy intake, not on the next day, but 2 and 3 d later.
The data used in this study were available in the data base accumulated during earlier research projects (de Castro 1987 Subjects.
Data were collected from 324 men and 409 women who were paid $30 to participate. They also received a detailed nutritional analysis based on their food intake for the 7-d reporting period. The subjects averaged 38.1 y (SD = 13.8), 69.1 kg (SD = 15.4) in weight, 24.4 kg/m2 (SD = 4.2) body mass index (BMI) and 1.68 m (SD = 0.10) in height. Informed consent was obtained from all subjects.
Procedure.
The subjects filled out a lengthy questionnaire, which requested information on subject demographics, lifestyle and eating habits. The subjects were given a small (8 × 18 cm) pocket-sized diary and were instructed to record, in as detailed a manner as possible, every item that they either ate or drank, the time they ate it, the amount they consumed and how the food was prepared. The subjects initially recorded this information for a day and were then contacted by the experimenter who reviewed the information, corrected any problems and answered any questions. The participants were then asked to record their intake for seven consecutive days. Data recording was not initiated on any particular day of the week; as a result, the first days' records were scattered over the week. After this recording period, the subjects were again contacted by the experimenter who reviewed the diaries, clarifying any ambiguities or missing data. After the completed diaries were submitted, two individuals who ate with the subject during the recording period were contacted and asked to verify the subject's reported intake. In some cases, difficulty was encountered in remembering exactly what the subject ate. However, in no case was the subject's diary report contradicted in substance or amount.
Data analysis.
An experienced registered dietician were assigned codes to the foods reported in the diaries from a computer file of over 3500 food items. The coder was unaware of the experimental hypotheses and did not interact directly with the subjects. Meals were then identified, and the nutrient compositions of the individual items composing the meal were summed. A number of definitions of a meal were employed to ensure that the results would be general and not definition specific. No attempt was made to separate meals and snacks. Meals were identified solely on the basis of the amount of food energy ingested and the time since the last meal. For a reported intake to be classified as an individual meal, it had to contain at least 209 kJ, or more stringently 418 or 837 kJ. It also had to be separated in time from the preceding and subsequent meals by at least 15 min; more stringent definitions of 45 and 90 min were also employed. Five different definitions of a meal were used combining these minimum criteria, 15 min/209 kJ, 45 min/209 kJ, 45 min/418 kJ, 45 min/837 kJ and 90 min/209 kJ.
There were quantitative differences among the results obtained for the five different meal definitions. Larger meal sizes and lower meal frequencies were apparent for the more stringent definitions. However, there were no significant differences among the correlations or LISREL results for the various meal definitions. Thus only the minimum 209 kJ, 45 min definition is presented as representative.
Daily energy intake.
The mean autocorrelations between the total amount of food energy ingested in a day and the amounts ingested on each of the four subsequent days are presented in the left panel of Figure 1. The autocorrelations significantly differed (P < 0.005). In particular, the autocorrelation between intake and the amount eaten 2 d later was significantly stronger than those for 1 d (P < 0.005) and 4 d later (P < 0.005). These data suggest that the amount ingested on a day has a small negative feedback on intake on the subsequent day, but a much larger effect 2 d later. This continues into the third, but vanished by the fourth day.
Daily macronutrient intake.
Figure 3 presents the mean autocorrelations between the amounts of macronutrients ingested in a day and the amounts ingested of each of the macronutrients on each of the four subsequent days. A three (macronutrient) × 4 (lag) ANOVA on these data for each of the macronutrients revealed significant main effects for macronutrient (P < 0.005), lag (P < 0.025) and for daily carbohydrate, fat and protein intakes. However, there were clear macronutrient-specific effects.
Daily meal size and frequency.
The relationship between daily average meal sizes and frequencies on subsequent days was analyzed with an elaborate simplex autoregressive LISREL model. In the model, average meal size was allowed to influence meal frequency on the same day, and both meal frequency and average meal size on the next day, and 2 and 3 d later. Similarly, meal frequency was allowed to influence average meal size on the same day, and both average meal size and meal frequency on the next day, and 2 and 3 d later. The results are depicted in Figure 5.
These findings are based upon diet-diary self-reports of intake maintained by normal humans, freely living in their natural environments. Even though the diet-diary technique has been demonstrated to be both reliable and valid (Adleson 1960 The author acknowledges the substantial contributions of Dixie K. Elmore, Sandor Goldstein, Sara Orozco, Sharon Pearcey, Margaret Pedersen and Marie Redd without whose assistance the work could not have been performed and Tim Bartness for helpful comments on the manuscript.
INTRODUCTION
Abstract
Introduction
Methods
Results
Discussion
References
), the day of the week (de Castro 1991b, Tarasuk and Beaton, 1991), the week of the month (de Castro and Pearcey 1994) and the month of the year (de Castro 1991; Tarasuk and Beaton 1991), and they can be markedly altered by an individual's psychological (de Castro and Elmore 1988) and/or physiological state (de Castro et al. 1986, de Castro 1993a and 1993b). Yet, somehow intake and expenditure are often balanced to produce stability in body weight. How this is accomplished has been studied intensively, yet remains a mystery.
). Intake of food energy also tends to reduce the level of subjective hunger at the beginning of a new meal. This in turn reduces the amount of food energy ingested in that meal (de Castro and Elmore 1988). However, these influences account for only ~6% of the variance in meal size; vastly more powerful stimuli affect intake (de Castro and de Castro 1989; de Castro and Brewer 1992), causing the daily amount ingested to vary greatly (Hankin et al. 1967, Hartman et al. 1990, Morgan et al. 1987, Tarasuk and Beaton 1991). In our studies, daily intake had a mean variance over 7 d of 465 kcal (1.95 MJ) (Pearcey and de Castro 1996).
, Hartman et al. 1990). That is, people who eat a lot on one day tend to be the same people who eat a lot on the next day. Within-subjects autocorrelations have been found to be small and predominantly positive (Morgan et al. 1987, Tarasuk and Beaton 1991). That is, when a person eats a lot on one day, that same person may or may not eat a lot on the next day. "The finding challenges the belief that some short-term homeostatic mechanism exists that causes high energy intakes to be followed by low ones" (Tarasuk and Beaton 1991). Hence, there is currently little understanding regarding how intake affects subsequent intake to produce energy balance.
SUBJECTS AND METHODS
Abstract
Introduction
Methods
Results
Discussion
References
, 1991a, 1991b, 1993a, 1993b, 1994, and 1995, de Castro et al. 1986, de Castro and Elmore 1988, de Castro and de Castro 1989, de Castro and Brewer 1992, de Castro and Pearcey 1994). Details regarding the methods are available in these publications. All protocols were approved by the Georgia State University Institutional Review Board.
). The mean correlation coefficients were then compared with 0 by using a t test. The mean autocorrelation coefficients were compared over the four lags with a one-way ANOVA. The mean autocorrelation coefficients calculated for the three macronutrients were compared with a 3 × 4 (macronutrient × lag) ANOVA. Individual comparisons were made with t tests.
). Simplex autoregressive models do not assume that the error terms are uncorrelated. The model allows the estimation of intraday path coefficients, which indicate the covariation of measurement over time and innovation (
), within-day variation, variance that is not accounted for by the earlier days' intakes. These can be estimated with a Linear Structural Relations (LISREL) computer program (Boomsma et al. 1989, Neale and Cardon 1992). This model was applied to investigate the factors controlling the day-to-day regulation of food energy intake. The simplex autoregressive models use data across subjects. Hence, to correct for individual differences, the proportion of the average daily intakes ingested on each day, rather than the absolute amounts, as employed in the analyses. The models were tested against observed data with LISREL-VII (Joreskog and Sorbom 1989). The model's fit is assessed with a likelihood ratio 
2 test. The significance of individual components of the model was assessed by dropping them from the model. The deleted parameter then was tested for significance with a difference 2 test (Neale and Cardon 1992). This process was continued until the removal of any remaining parameter produced a significant (P < 0.05) reduction in the model's fit to the data. All other statistics presented are significant at the 0.01 level.
RESULTS
Abstract
Introduction
Methods
Results
Discussion
References
View larger version (18K):
[in a new window]
Fig 1.
Autocorrelation coefficients between daily energy intake and intake on subsequent days (left panel) and daily macronutrient intake and energy intake on subsequent days (right panel). The first bar of each set of four represents the correlations calculated between the amounts ingested on a day and on the next day (lag 1), and 2 (lag 2), 3 (lag 3) and 4 (lag 4) d later. Values are means ± SEM, n = 733. 
indicates that the correlation coefficient is significantly (P < 0.05) different than zero as assessed with a t test.
, Neale and Cardon 1992) was applied to the prediction of the overall amount of food energy ingested on each of three subsequent days on the basis of the day's intake. The results are depicted in Figure 2. The LISREL analysis results parallel the autocorrelations. Only two of the six paths indicating the effect of a day's intake on the next day's intake were significant. This indicates, as seen with the 1-d lag autocorrelations, that the intake during a day has a small effect on intake during the subsequent day. On the other hand, all of the paths in the LISREL analysis from an intake during a day to intake occurring 2 or 3 d after were significant and negative. Hence, the LISREL analysis also supports the conclusion that intake during a day has its maximum negative effect on eating 2-3 d later.
View larger version (13K):
[in a new window]
Fig 2.
LISREL simplex autoregressive model applied to the prediction of daily energy intake. The numbers represent path coefficients produced by the analysis. Solid lines represent significant paths, whereas dashed lines represent nonsignificant paths. Removal of any solid-lined path presented in the figure produces a significant degradation of the model's fit as assessed with a 2 test (P < 0.05). represents variance that is not accounted for by the prior days' intakes.
View larger version (18K):
[in a new window]
Fig 3.
Autocorrelation coefficients between daily macronutrient intake and macronutrient intake on subsequent days for carbohydrate (top panel), fat (middle panel) and protein (bottom panel). The first bar of each set of four represents the correlations calculated between the amounts ingested on a day and on the next day (lag 1), and 2 (lag 2), 3 (lag 3) and 4 (lag 4) d later. Values are means ± SEM, n = 733. indicates that the correlation coefficient is significantly (P < 0.05) different than zero as assessed with a t test.
0.09 to 0.30. Hence, the LISREL analysis supports the conclusions from the autocorrelational analysis that the negative effect of ingestion of a macronutrient is greatest on the subsequent intake of that particular macronutrient 2 d later.
View larger version (38K):
[in a new window]
Fig 4.
LISREL simplex autoregressive model applied to the prediction of daily macronutrient intake. The numbers represent path coefficients produced by the analysis. Solid lines represent significant paths, whereas dotted lines represent nonsignificant paths. Removal of any solid lined path presented in the figure produces a significant degradation of the model's fit as assessed with a 2 test (P < 0.05). represents variance that is not accounted for by the prior days' intakes.
View larger version (42K):
[in a new window]
Fig 5.
LISREL simplex autoregressive model applied to the prediction of daily average meal sizes and frequencies. The numbers represent path coefficients produced by the analysis. Solid lines represent significant paths, whereas dotted lines represent nonsignificant paths. Removal of any solid lined path presented in the figure produces a significant degradation of the model's fit as assessed with a 2 test (P < 0.05). represents variance that is not accounted for by the prior days' intakes.
View larger version (35K):
[in a new window]
Fig 6.
LISREL simplex autoregressive model applied to the prediction of daily energy intake, average meal size and frequency. The numbers represent path coefficients produced by the analysis. Solid lines represent significant paths, whereas dotted lines represent nonsignificant paths. Removal of any solid lined path presented in the Figure produces a significant degradation of the model's fit as assessed with a 2 test (P < 0.05). represents variance that is not accounted for by the prior days' intakes.
DISCUSSION
Abstract
Introduction
Methods
Results
Discussion
References
, Gersovitz et al. 1978, Heady 1961, Krantzler et al. 1982, St. Jeor et al. 1983, see also Beaton 1994, de Castro 1994 for review), it is not without error because it has been shown to underestimate intake (Bandini et al. 1990, Goran and Poehlman 1992, Livingstone et al. 1990 and 1992). However, there is no reason to suspect any systematic relationship between recording errors and day-to-day intake. For the negative feedback effect, as reported here, to be an artifact, accurate reporting or underreporting on 1 d would have to be followed, not on the next day, but systematically 2 or 3 d later with underreporting. This would produce a decrease in the amount reported each day over the course of the week. However, there were no significant differences among the amounts recorded across recording days. That is, there was no decline in reported amounts of energy intake found over the 7-d recording period. Hence, underreporting would not appear to produce systematic error over days. Unsystematic error should obfuscate significant relationships, not produce them. The fact that significant relationships were discerned with a somewhat insensitive technique suggests that the effects reported may actually underestimate the significance of intake effects on subsequent intake.
, Hartman et al. 1990, Morgan et al. 1987, Tarasuk and Beaton 1991). On the other hand, the results of this study indicate that there is a 2- to 3-d delay before intake feeds back to affect subsequent intake. Edholm et al. (1995) monitored military cadets for a 2-wk period and found no correlation between intake and expenditure on the subsequent day, but did find a significant correlation between expenditure and intake 2 d later. Unfortunately, they did not calculate daily intake autocorrelations. However, in a subsequent study (Edholm et al. 1970) they reported, but did not note or discuss, intake autocorrelations in a table showing a significant negative autocorrelation after a 2-d lag and not for 0, 1 or 3 d. Hence, it would appear that both intake and expenditure produce negative feedback to subsequent intake, but not until two or more days later.
manipulated the fat and carbohydrate content of meals that subjects were required to eat. Although the subjects compensated for the change in food energy, there was no macronutrient-specific compensation. However, only compensation occurring during the day of the manipulation was investigated. The present findings suggest that compensation should occur two or more days later.
, Rolls 1995) and that fat intake may be the least subject to regulation (Cotton et al. 1996, DeGraaf et al. 1996). In contrast, the present results indicate that compensation occurs equally for the intake of each of the macronutrients. The difference may be due to a number of factors. First, the time frames of the studies differ. Most of the work investigating compensation for intake look only within a day or at most over one additional day. The present results suggest that a longer, 2- or 3-d, time frame is necessary to see compensation. Second, the studies that tend to find a lack of compensation for ingested nutrients typically employ a test meal format that appears to promote overeating. For example, the amount ingested in the test lunch in the study by DeGraaf et al. (1996) was 5 MJ or 40% of the total daily intake. Third, all of these studies manipulate the macronutrient composition of the diet while subjects are ingesting meals in unusual circumstances, meals whose composition may differ from their usual intake. On the other hand, this study investigated the usual, unmanipulated intakes of humans. Flatt (1995) suggested that metabolic processes adapt to the usual composition of the diet. If this is the case, then it would be expected that the system would be tuned to compensating for minor perturbations from usual levels. Finally, the amount of compensation seen in this study is very small. A comparable amount of compensation in other work is frequently viewed as insignificant. For example, in the study of DeGraaf et al. (1996), after food energy intake was reduced by 1.9 MJ by substituting a fat replacer, average daily intakes were reduced by 1.5 MJ. This suggests a compensation for 21% of the food energy. However, the authors focused their conclusions on the 79% of intake that was not compensated for.
0.19, which accounts for < 4% of the variance in daily intake. In the LISREL analyses, the largest amount of variance accounted for was 16.1% for any day's total or macronutrient intake, and 30.1 and 22% for meal frequency and size, respectively. Most effect sizes in the LISREL analyses were much smaller than these. Hence, prior intake accounts for only a small proportion of the variance in the daily or meal intake of nutrients.
ACKNOWLEDGMENTS
| |
FOOTNOTES |
|---|
Manuscript received 17 September 1996. Initial reviews completed 10 December 1996. Revision accepted 9 September 1997.
| |
LITERATURE CITED |
|---|
|
Abstract
Introduction
Methods
Results
Discussion
References
|
|---|
This article has been cited by other articles:
|
|
 |
|
  J. M. de Castro Varying Levels of Food Energy Self-Reporting Are Associated with Between-Group, but Not Within-Subject, Differences in Food Intake J. Nutr., May 1, 2006; 136(5): 1382 - 1388. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. M. de Castro Dietary Energy Density Is Associated with Increased Intake in Free-Living Humans J. Nutr., February 1, 2004; 134(2): 335 - 341. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. M. de Castro Age-Related Changes in the Social, Psychological, and Temporal Influences on Food Intake in Free-Living, Healthy, Adult Humans J Gerontol A Biol Sci Med Sci, June 1, 2002; 57(6): M368 - M377. [Abstract] [Full Text] [PDF]
|
|
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
