NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Ammerman A, Lindquist C, Hersey J, et al. The Efficacy of Interventions to Modify Dietary Behavior Related to Cancer Risk. Rockville (MD): Agency for Healthcare Research and Quality (US); 2001 Jun. (Evidence Reports/Technology Assessments, No. 25.)

  • This publication is provided for historical reference only and the information may be out of date.

This publication is provided for historical reference only and the information may be out of date.

Cover of The Efficacy of Interventions to Modify Dietary Behavior Related to Cancer Risk

The Efficacy of Interventions to Modify Dietary Behavior Related to Cancer Risk.

Show details

3Results

General Overview of Interventions Reviewed

Intervention Setting and Population Characteristics

As discussed in Chapter 2, our literature search and screening procedures resulted in the inclusion of a total of 104 eligible articles (reporting results from 92 studies) in the Evidence Tables. In our secondary analyses, the unit of analysis was the study (not the article); thus our analyses were based on the pool of 92 studies. We classified the interventions by setting; approximately 49 percent were conducted in health care settings,36-87 27 percent were conducted in community settings,98-122 14 percent were conducted in school settings,19-35 and 10 percent were conducted in worksite settings.88-97

The majority of interventions were conducted with adult populations (77 percent, or 71/92 studies); only 16 studies focused on children.19-30, 32-35, 42, 63-64, 99, 109 In five interventions, either the age of the study population was not reported or the study included both children and adults.31,100-102,112 The interventions were equally likely to target high-disease-risk populations (46 percent, or 42/92 studies) and general populations (47 percent, or 43/92 studies). Seven interventions (8 percent) either included both high-risk and healthy populations or did not report information about the risk status of the participants.53-64,82,88, 90,98-99,108

Intervention Characteristics and Study Quality

We abstracted detailed information about the content and duration of the interventions. Thirteen percent (12/92) of the interventions were classified as being of low intensity,29,39,60,62,71,78,84-85,92,105,114-115 57 percent (52/92 studies) as medium intensity,23,30-31,34,36,38,40-42,50-51,56-58,61,63-69,72-75,77,79-80,82,88,90-91,93-100,102-104,106-111,113,116-122 and 30 percent (28/92 studies) as being of high intensity.19-22,24-28,32-33,35,37,43-49,52-55,59,70,76,81,83,86-87,89,101,112 Nearly one-half of the interventions exclusively targeted dietary change (and did not include non-nutrition targets, such as exercise, smoking, and stress management). Although the delivery of the intervention was often difficult to determine (with multiple delivery agents being common), the results of our review indicated that 7 percent (6/92) involved self-administered interventions (such as computer modules).92,104,108,118,121-122 Another 27 percent (25/92 studies) were delivered by nonhealth professionals (such as classroom teachers, trained peers)25,19-28,31-35,48,73-74,81,90,95,99-101,105-106,109,113-115, 21 percent (19/92 studies) were delivered by health professionals (including doctors, nurses)29-30,41,49,51,56,58,60,65,71,77,78,82,84-85,88,116,119, and 36 percent (33/92 studies) were delivered by dietitians or nutritionists.36-39,42-47,52-55,57,59,61,63-64,66-70,72,76,79,80,83,86-87,89,93,98,102-103,110,112,120 In 10 interventions (11 percent), the delivery mode was not reported.40,50,62,75,91,94,96-97, 107,111,117

Of the 92 interventions, about one-half (43, or 47 percent) appeared to employ a theoretical framework in the design of the intervention, although, as mentioned previously, we were not able to determine the extent to which theory was involved in the design of the intervention. Because we were interested in the use of theory in recently published articles compared with the earlier generation of articles, we examined the proportion of articles published before and after 1995 that reported the use of theory. Slightly more studies reported the use of theory in or after 1995 (49 percent, or 25/51) than before 1995 (44 percent, or 18/41).

We also abstracted information on "special features" employed in the interventions we reviewed. About 25 percent of the studies (23/92) included a family component (in which either families were the primary target of the intervention or the family of the primary target was involved in some aspects of the intervention).23-28,32-36,39,42,45-47,60,63-64,76,89,96-103,107,112 Eighteen percent (17/92 studies) reported including a social support component in the intervention (e.g., support groups within the intervention).23,45-48,56,76,81,83,86-87,95-97,99-101,109 Twenty-one percent (19/92) of the interventions used small groups in the delivery of nutrition education (or other components of the intervention).30,40,42,44,48,52,54,65,76,81,83,89,95,99-101,107,112-113 Interactive activities involving food (e.g., taste tests, cooking demonstrations) were included in 26 percent -- 24/92 -- of the interventions.20,22-28,32-35,42,45-47,72,88-89,99-101,106-107,109,111-113,116,119 A common feature of the interventions we reviewed was the use of goal setting and self-monitoring (to track the participants' progress on the dietary goals set in the intervention); 33 percent (30/92) of the interventions included this component. 19-20,23,34,40,43-44,52,54-55,57,59-60,70,73-74,76,78,81,83,85,87,89,93,96-97,100-102,109,111,120 Only 10 of the interventions (11 percent) were designed to be ethnically or culturally specific (to the target population).76,87,95,98,101-102,107,112,116,119 Finally, only eight interventions (9 percent) included a tailoring component (which we defined using rather restrictive criteria),73-74,104,108,118-122 such as an interactive computer program that is tailored to the participant's responses.

As discussed in the previous chapter, several of the intervention components we explored are likely to be closely associated with one another (such as the use of theory and goal setting), resulting in a certain degree of confounding. We did not have a large enough pool of interventions to explore the independent effects of specific intervention components (by controlling for the other characteristics), but we did explore the extent to which intervention characteristics tended to overlap. For example, interventions that incorporated the use of theory were more likely also to employ the "special features" described above. Only 53 percent (26/49) of studies lacking an obvious theoretical framework included any special features, whereas 84 percent (36/43) of studies that did report the use of theory included special features. In addition, interventions that used any special feature at all were likely to use several such features. Of the 62 articles that reported any special features, about 42 percent (26) reported the use of only one special feature, and 58 percent (36) reported the use of two or more (with 21 percent employing three, eight percent employing four, and five percent employing five or more special features).

The interventions we reviewed varied in terms of quality (see the previous chapter for a description of our quality rating system). The quality scores assigned to the 92 studies ranged from 25 to 86 points (the theoretical range was 0 to 100). Equal numbers of studies were then classified as being of low,20-21,29,38-40,42,45-47,50,53,56-57,60,62,66-69,71,77,85-96,88,90,94,99,105-107,110,116-117 medium, 22,30-31,36-37,41,43,48-49,51-52,58-59,61,75,79-80,82-83,92,100,102-104,108,111-114,121 or high quality,19,23-28,32-35,44,54,55,63-65,70,72-74,76,78,81,84,87,89,91,93,95-98,101,109,115,118-120,122 based on the distribution of raw quality scores.

Dietary Outcomes

Of the dietary outcomes on which we focused in this review (dietary fat and fruit and vegetable intake), fat outcomes were the most commonly reported. Roughly 60 percent (55/92) of the interventions reported results for fat outcomes only, 13 percent (12/92) reported results for fruit and vegetable outcomes only, and 27 percent (25/92) reported results for both dietary fat and fruit and vegetable outcomes. The results for both sets of outcomes are presented in the following section.

The primary purpose of this report was to determine whether behavioral dietary interventions positively influence intake of dietary fat and fruits and vegetables. As described in detail in Chapter 2, we explored this issue using three main analysis strategies: (1) meta-analysis, (2) a standardized analysis of the magnitude of the intervention effect (median differences in effect sizes), and (3) a semiquantitative analysis (proportion of studies) summarizing the statistical significance of the intervention effect. In addition to determining the overall effect of interventions on various dietary outcomes, we considered the impact of interventions among population subgroups and the effect of specific intervention components on changing dietary behavior.

This chapter reports the results separately for fruits and vegetables and for dietary fats. For both sets of outcomes, the results are organized by these three analysis strategies. Within each subsection, we report the overall intervention effect and results by various intervention and population characteristics. We are not giving study-by-study reviews because of our decision to focus on these types of summative analyses.

Fruit and Vegetable Consumption

Meta-Analysis

After reviewing the articles for fruit and vegetable intake, we determined that the small number of articles eligible for a statistical meta-analysis and the diversity in study populations and intervention received did not support the application of this technique. In short, we reviewed all studies reporting results for daily servings of fruits and vegetables (the most commonly reported outcome and the fruit and vegetable outcome presumed to be most closely associated with cancer risk), using criteria developed by our meta-analysis expert that included both parametric and nonparametric methods.

Of the 14 studies including this outcome, we excluded five because of inadequate statistical reporting, which most commonly included lack of an indication of the variance of the estimate. Combining a mean difference requires estimates of standard deviations or standard errors. We reviewed the remaining nine studies for substantive comparability, and we decided that they were too dissimilar to allow a meaningful interpretation of a combined mean effect. For example, the majority of the eight interventions were conducted among special populations (e.g., low-income participants in the Special Supplemental Nutrition Program for Women, Infants, and Children [WIC] program, African-American churchgoers, Girl Scouts, middle-aged male workers), with high variability in population characteristics and the type of intervention conducted. Therefore, we concluded that the few statistically eligible studies were not substantively similar enough to be combined in a meta-analysis.

Standardized Analysis of the Magnitude of the Intervention Effect

The second analysis strategy we used to determine the effect of interventions on fruit and vegetable intake was the "difference-in-deltas" approach, which employs a standardized metric indicating the magnitude of change in fruit and vegetable intake between intervention and control groups across the various studies in this report. We calculated the differences in deltas for three outcomes: daily servings of fruits and vegetables, daily servings of fruits, and daily servings of vegetables. The results of this set of analyses are reported separately for fruits, vegetables, and fruits and vegetables combined because we wanted to examine the effects of interventions on fruits and vegetables separately. In addition, we tried to accommodate articles that reported results for fruits and vegetables either combined or separately. These outcomes are largely independent from one another (i.e., the summary results for fruits and vegetables combined are not necessarily the same as the results for fruits and vegetables as individual outcomes summed together), although there is some overlap in the articles reporting results for the various outcomes.

Because we focused on only three outcomes (based on specific measurement units), we excluded from this analysis studies reporting other fruit- and vegetable-related outcomes and studies that did not report statistical information necessary for the calculation of the differences in deltas. In the end, we used 17 of the 39 studies reporting results for fruit- and vegetable-related outcomes.

Overall Magnitude of Effects

The median differences between intervention and control groups in percentage change at the initial follow-up period (which varied in duration across the studies, but generally took place immediately after the completion of the intervention) for fruit and vegetable outcomes are presented in Table 17. To convey the variability of the differences in deltas across the included studies, the table also reports the range. Among the 17 studies we reviewed, interventions generally demonstrated the ability to moderately increase fruit and vegetable intake among intervention groups compared with controls. The median difference between intervention and control groups in the change in daily servings of fruits and vegetables at the first follow-up period was +16.6 percentage points (see the top row of Table 17), indicating that intervention groups increased their intake of fruits and vegetables about 17 percent more than did the control groups.

Table 17. Median differences between intervention and control groups in percentage change in fruit and vegetable intake.

Table

Table 17. Median differences between intervention and control groups in percentage change in fruit and vegetable intake.

We caution that any comparison of medians in effect sizes are suggestive only, because the number of studies was generally too small to support statistical tests of the differences in medians. Among the nine studies reporting results for fruit and vegetable consumption as individual outcomes, the magnitude of the intervention effect was higher for fruit intake (+16.9 percentage points) than for vegetable intake (+5.7 percentage points) (see the middle and bottom rows of Table 17).

Duration of Effects

Many studies reported the results for more than one point in time to determine long-term effects of the intervention. Thus, we considered the sustained impact of interventions on fruit and vegetable intake among the subset of articles reporting results for two follow-up periods. The median differences in percentage change in fruit and vegetable outcomes at the first and second follow-up periods are reported in Table 18. This table presents the results only for the subset of articles that reported results for both follow-up periods. Neither the first nor the second follow-up time point is standardized across the studies reviewed because of the large degree of variability in intervention durations and duration of follow-up measurement. However, in the majority of these studies, the first follow-up period took place immediately after the completion of the intervention.

Table 18. Median differences in percentage change in fruit and vegetable intake between intervention and control groups at Follow-Up 1 and 2.

Table

Table 18. Median differences in percentage change in fruit and vegetable intake between intervention and control groups at Follow-Up 1 and 2.

In looking at the longer-term differences (i.e., the second follow-up period) between intervention and control groups, we observed an inconsistent pattern. As indicated in Table 18, the magnitude of the intervention effect for total fruit and vegetable intake decreased from +16.8 percentage points at follow-up 1 to +6.7 percentage points at follow-up 2, indicating a reduction in the differences in fruit and vegetable intake between intervention and control groups over time. The pattern of decreasing magnitude effects from time 1 to time 2 was also evident for vegetable intake (as an individual outcome), with the median difference-in-deltas declining from 25.8 percent at follow-up 1 to 19.2 percent at follow-up 2. We observed, however, a slight increase in the difference-in-deltas for fruit intake from 24.1 percentage points at follow-up 1 to 27.4 percentage points at follow-up 2, suggesting that the difference between intervention and control groups grew larger over time for fruit intake.

Effects by Population Characteristics

The next step of the difference-in-deltas analysis involved examining the magnitude of the intervention effect across various population and intervention characteristics. We required a minimum of five studies in each cell for any given comparison, which eliminated any comparisons for fruits and vegetables as individual outcomes. In addition, the small number of articles eligible for the difference-in-deltas analysis for fruit and vegetable intake (combined) that focused on children or high-risk populations also prevented the determination of any population differences in change in fruit and vegetable intake.

Effects by Intervention Characteristics

We were able to explore the impact of three intervention characteristics on the difference in change in fruit and vegetable intake -- namely, social support components, interactive activities involving food, and goal-setting components (Table 19). Other potential comparisons (including individual tailoring, cultural or ethnic specificity, intervention intensity, or delivery mode) were not possible because the studies did not deal with these factors or did not report relevant data.

Table 19. Median differences in percentage change in fruit and vegetable outcomes by demographic and intervention characteristics.

Table

Table 19. Median differences in percentage change in fruit and vegetable outcomes by demographic and intervention characteristics.

None of the three intervention strategies appeared to be associated with a greater magnitude of change in intervention groups compared with control groups for daily servings of fruit and vegetable intake. Interventions that incorporated social support components, food-related activities, or goal-setting components appeared to be no more successful at increasing fruit and vegetable intake than were interventions that did not employ such strategies. Unfortunately, the small number of articles eligible for the difference-in-deltas analyses and the limited variability in other intervention characteristics precluded further comparisons.

Analysis of the Significance of the Intervention Effect

Our final approach to determining the efficacy of dietary interventions at increasing fruit or vegetable intake was the most inclusive in that it accommodated the diversity of outcomes and statistical reporting evident in the articles we reviewed. Indeed, although this strategy is less standardized than is the difference-in-deltas approach, and although it does not allow for determination of the magnitude of the intervention effect, it has one primary advantage -- it permitted us to include 36 of the 39 articles reporting fruit- and/or vegetable-related outcomes. In addition, it enabled us to explore the influence of several population and intervention characteristics at promoting change in fruit and vegetable intake that we were not able to consider using the difference-in-deltas approach (because of the small number of articles eligible for that analysis). As described in Chapter 2, we categorized the various outcomes measures reported in the studies into three sets of outcomes (fruit and vegetable intake, fruit intake, and vegetable intake), regardless of the specific units of measurement, and simply summarized whether studies reported a significant intervention effect for each of these outcomes.

Overall Magnitude of Effects

Table 20 presents the results of this analysis strategy. Similar to the findings based on the difference-in-deltas approach, the majority of the 36 studies reported significant intervention effects for all sets of fruit and vegetable outcomes. Among the articles reporting results for fruit and vegetable intake combined (n = 22), 17 reported a statistically significant intervention effect. Similar to the pattern observed with the difference-in-deltas approach, studies reporting results for fruit and vegetable intake as individual outcomes were more likely to report a significant intervention effect for fruit intake than for vegetable intake (14/22 studies compared with 10/23 studies, respectively).

Table 20. Differences in the proportion of studies reporting significant intervention effects for fruit and vegetable outcomes by demographic and intervention characteristics.

Table

Table 20. Differences in the proportion of studies reporting significant intervention effects for fruit and vegetable outcomes by demographic and intervention characteristics.

Effects by Population Characteristics

Using the summary of significant findings approach, we were able to consider the role of age and risk status as mediators in the efficacy of behavioral interventions. As shown in Table 20, approximately equal proportions of studies focusing on children and adults reported a significant intervention effect for fruit and vegetable intake (combined). However, when fruit and vegetable consumption were reported as separate outcomes in these studies, interventions conducted with children appeared slightly more successful at increasing fruit intake and less successful at increasing vegetable intake compared with interventions conducted with adults.

Although we did not have enough articles to compare combined fruit and vegetable intake by differences in the risk status of the study population, a clear pattern was evident for the separate fruit and vegetable outcomes. Interventions conducted on high-risk populations were nearly twice as likely to report a significant intervention effect as were interventions conducted on general populations for both fruit (6/6 studies compared with 8/14 studies) and vegetable (5/7 studies for high-risk populations compared with 5/14 general-risk studies) intake.

Effects by Intervention Characteristics

In exploring whether significant changes in fruit and vegetable intake would be associated with various intervention characteristics, we detected several interesting findings. First, studies employing a theoretical framework were consistently more likely to report a significant intervention effect than studies that did not. Keeping in mind that each dietary category consisted of 22 to 23 studies, 20 percent to 30 percent more studies using a theoretical framework reported significant findings than did studies not reporting the use of theory. This pattern was evident for the combined and separate fruit and vegetable outcomes. Among the subset of articles published in 1995 or later, which ranged from 16 to 20 per dietary category, the differences in the proportions reporting significant effects between articles that did or did not use theory were still evident for combined fruit and vegetable intake and for fruit intake as a separate outcome. For fruit and vegetable intake (combined) and fruit intake (individually), nearly twice as many studies employing theory reported significant effects as did studies not using a theoretical framework. Interestingly, we observed the opposite pattern with vegetable intake (i.e., more than twice as many articles not using a theoretical framework reported significant findings as did the articles that did use theory).

A dose-response relationship between the quality of the study and the likelihood of significant findings being reported was evident for both fruit and vegetable intake (as separate outcomes); increasingly greater proportions of studies classified as being of "high" quality reported a significant intervention effect. Among the high-quality interventions, 6/9 to 8/9 studies reported significant findings for the two outcomes; for low-quality interventions, the range was only 0/5 to 2/5 studies.

The presence of non-nutrition components in the interventions was not consistently associated with the likelihood of reporting a significant intervention effect. Roughly similar proportions of studies reported effects for combined fruit and vegetable intake (4/5 studies with non-nutrition components compared with 13/17 studies without such components). By contrast, slightly more studies including non-nutrition components reported significant findings for fruit intake (5/7 studies with non-nutrition components versus 9/15 without), and the reverse pattern was evident for vegetable intake (3/8 studies versus 7/15).

We were able to explore the effectiveness of several specific intervention strategies (or "special features") aimed at increasing fruit and vegetable intake. Specifically we considered the impact of including a family component in the intervention, employing social support, incorporating interactive activities involving food, and goal setting.

Our results indicated that family components were not consistently associated with more favorable fruit and vegetable intake, although this feature was positively associated with fruit intake (7/9 studies using a family component reported significant changes in fruit intake compared with 7/13 studies not using this component). Including a social support component in the intervention did appear to be positively associated with more favorable results, particularly when fruit and vegetable intakes were measured as separate outcomes. The results for interactive activities involving food were not as consistent. Although a greater proportion of studies including food activities reported significant effects for fruit and vegetable intake (combined) and fruit intake, this pattern was not evident for vegetable intake (for this outcome, interventions that included food-related activities were less successful at increasing vegetable intake). Finally, goal setting was consistently associated with more favorable intervention effects for all three outcomes. The differences between studies that did and did not include goal setting as a part of the intervention were particularly pronounced when fruit and vegetable intake were measured separately. Thirteen of 16 articles that included goal setting reported significant changes in fruit and/or vegetable intake; for vegetable intake alone, this proportion is particularly notable because it is nearly twice as high (4/5 studies, or 80 percent) as the proportion (10/23 studies, or 43 percent) found in the overall pool of articles reporting results for change in vegetable intake.

Dietary Fat Consumption

This section considers the efficacy of behavioral dietary interventions for reducing intake of dietary fats. Our analyses included the same set of strategies reported in the previous section: (1) a meta-analysis, (2) median differences in effect sizes, and (3) a summary of the proportion of studies reporting significant intervention effects. Within each subsection, we report the overall intervention effect and results by various intervention and population characteristics.

Meta-Analysis

Our meta-analysis involved 28 eligible articles in five major groupings: (1) school-based interventions with healthy children, (2) worksite and community interventions with healthy adults, (3) health care setting interventions with healthy adults, (4) health care setting interventions with adults at risk of cancer or type 2 diabetes, and (5) health care setting interventions with adults diagnosed with cancer, type 2 diabetes, or cardiovascular disease. Table 21 presents the mean differences in change in total fat intake (as a percentage of energy intake) between intervention and control groups for all articles and groupings included in the meta-analysis.

Table 21. Mean differences in change in total fat (percentage of energy) between intervention and control groups.

Table

Table 21. Mean differences in change in total fat (percentage of energy) between intervention and control groups.

Because of differences in populations, interventions, and lengths of follow-up, we could not combine all studies within the major groupings into five overall estimates of the effects of dietary intervention. Within each group, however, we were able to combine different sets of studies that had similar populations, interventions, and follow-up times (as shown in the italicized entries in Table 21). The studies typically indicated small to moderate changes in the intervention group that were significantly larger than the changes in the comparison group. As discussed in the sections that follow, the magnitude of effects varied by setting and population and by the length and intensity of the intervention.

Group 1. School-Based Interventions with Healthy Children

Three studies investigated school-based interventions on healthy elementary and middle-school children.19,24,25,35 Each study included classroom education plus other nutrition- and health-related activities, such as cafeteria modification, smoking prevention, and exercise programs. The timing and extent of the classroom nutrition education differed among the studies, as did the types of non-nutrition components.

We started by computing and plotting a difference-in-difference estimate of percentage fat intake (i.e., the difference in percentage points between the change over time in the intervention group and the change over time in the comparison group) for each study. The estimates for the school-based interventions are reported in Table 21 and plotted in Figure 2. The mean estimates were very close in magnitude, ranging from 1.8 to 2.9 percentage points. We then computed and plotted a pooled estimate. The mean of the pooled estimate indicated an improvement in fat intake by 2.2 percentage points (95 percent confidence interval [CI], 1.5 to 2.9 percentage points). In other words, on average, children in school-based interventions showed a 2.2 percentage point greater change (e.g., greater decrease) in consumption of dietary fat than did children in the comparison group.

Figure 2. Plotted Estimates of the Mean Differences in Change in Total Fat Intake (Percentage of Energy) Between Intervention and Control Groups for Group 1 (School-Based Interventions with Healthy Children).

Figure

Figure 2. Plotted Estimates of the Mean Differences in Change in Total Fat Intake (Percentage of Energy) Between Intervention and Control Groups for Group 1 (School-Based Interventions with Healthy Children).

Group 2. Worksite and Community Interventions with Healthy Adults

Four studies examined worksite or community interventions conducted with healthy adults.93,96-97,107,112 Two of these studies included low-income mothers and their children aged 7 to 12 years, but our analysis included only the mothers' data.107,112 The intervention components of all four studies included nutrition education or group meetings, although curriculum, duration, and added components all varied. The two studies of low-income mothers reported short-term outcomes only (i.e., the difference in the percentage fat intake immediately after the conclusion of the 12-week intervention). The third study was a one-time brief education session for which the investigators reported change in fat intake at 16 to 20 weeks following the intervention.93 The final study had fewer nutrition sessions than did the studies for mothers, and the sessions were spread out over the first year, with newsletters sent out quarterly in the second year.96-97

We combined the results for the two studies of low-income mothers (which had short-term follow-up) and compared the pooled result with the results of the other two studies shown separately. The mean estimates are reported in Table 21 and plotted in Figure 3. The pooled difference in fat intake at 12 weeks was statistically significant, with a mean of 6.4 percentage points (95 percent CI, 3.5 to 9.3 percentage points). The effect for the two studies with low-income mothers, in which the intervention consisted of 60- to 90-minute classes once a week for 12 weeks, was significantly greater than the effects for two worksite studies (discussed in the following paragraph) with less intensive interventions.

Figure 3. Plotted Estimates of the Mean Differences in Change in Total Fat Intake (Percentage of Energy) Between Intervention and Control Groups for Group 2 (Worksite and Community Interventions with Healthy Adults).

Figure

Figure 3. Plotted Estimates of the Mean Differences in Change in Total Fat Intake (Percentage of Energy) Between Intervention and Control Groups for Group 2 (Worksite and Community Interventions with Healthy Adults).

The study of a worksite intervention program that had five nutrition sessions over 1 year, reinforced by a monthly newsletter, found a small (1.0 percentage point), albeit significant, effect at the 1-year follow-up, but this effect was no longer statistically significant by the end of the second year.96-97 A study of a brief worksite program (consisting of a 20-minute education session and a mail reminder after 6 weeks) did not find statistically significant differences in effects.93

Group 3. Health Care Setting Interventions with Healthy Adults

Four studies were conducted with healthy adults drawn from patient lists in different health care settings.62,70,78,87 The interventions ranged in intensity from a self-help booklet with a brief physician message to 2 years of dietary counseling reinforced by ongoing group sessions. Two studies included only females,70,87 and one included only males.62 The studies also had different follow-up times (ranging from 3 months to 1.5 years).

We combined the estimates of the change in percentage of fat intake at 6 months of follow-up for the two studies that included only women. In both cases, the intervention was group counseling sessions conducted over at least 6 months. As indicated in Table 21 and Figure 4, the combined estimate has a mean effect of 10.8 percentage points (95 percent CI, 10.1 to 11.5 percentage points). One of these studies continued the sessions for another 6 months.87 For this study, the difference in the changes between the intervention group and the comparison group was statistically significant and roughly of the same magnitude after a full year as after the first 6 months.

Figure 4. Plotted Estimates of the Mean Differences in Change in Total Fat Intake (Percentage of Energy) Between Intervention and Control Groups for Group 3 (Health Care Setting Interventions with Healthy Adults).

Figure

Figure 4. Plotted Estimates of the Mean Differences in Change in Total Fat Intake (Percentage of Energy) Between Intervention and Control Groups for Group 3 (Health Care Setting Interventions with Healthy Adults).

The intervention targeted to men that provided only nutrition advice appeared to have no significant impact at 6 months. However, when the intervention was coupled with an exercise program, the investigators observed a statistically significant change of 4.5 percentage points (95 percent CI, 1.5 to 7.5 percentage points).

The nutrition booklet provided to adults by physicians with a brief nutrition message had a small but statistically significant effect -- a mean effect of 1.3 percentage points (95 percent CI, 0.7 to 1.7 percentage points).

Group 4. Health Care Setting Interventions with High-Risk Adults

Eight studies investigated the effect of nutrition education on fat intake among high-risk adult patients from various health care settings.38,44,61,72,79,80,83,111 Again the populations, interventions, and follow-up times differed markedly from study to study. Upon further investigation, we dropped one study because it was conducted with adults at high risk for type 2 diabetes (participants in the remaining studies were at high risk for cancer) and because the intervention was less intensive than those in the other studies.80

We combined the effect size (i.e., difference-in-change) estimates for (1) four studies that investigated the effect of at least 6 months of dietary counseling reported at the end of 6 months,38,44,83,111 (2) five studies that investigated the effect of at least 12 months of counseling reported at the end of 12 months,38,44,61,83,111 and (3) two studies that investigated the effect of 2 years of counseling at the end of 2 years. 61,79

As indicated in Figure 5, our analyses suggested that the major benefit of nutritional counseling was evident within 6 months. The studies of interventions in medical settings that employed nutritional counseling with high-risk adults found mean effects after 6 months of 15.3 percentage points (95 percent CI, 13.8 to 16.8 percentage points). Continuing counseling for longer than 6 months helped to maintain these effects but did not appear to result in further improvements. After 12 months of nutritional counseling, the mean effect size was 13.9 percentage points; after 24 months of counseling, it was 13.0 percentage points. (The differences between these results and the 6-month results were not statistically significant.) Another study also suggested that continuing intervention can help to maintain behavioral change.44 Estimates from a study providing 12 months of counseling indicated that without continuous counseling, the effect had started to wear off by the end of the second year of follow-up (decreasing from 13.0 percentage points to 9.2 percentage points). Nonetheless, despite a decrease over time, the effects of a similar intervention in another study were still evident after 7 to 8 years of follow-up.72 After 7 years, the impact remained statistically significant, with a mean effect size of 3.6 percentage points (95 percent CI, 0.7 to 6.5 percentage points).

Figure 5. Plotted Estimates of the Mean Differences in Change in Total Fat Intake (Percentage of Energy) Between Intervention and Control Groups for Group 4 (Health Care Setting Interventions with High-Risk Adults).

Figure

Figure 5. Plotted Estimates of the Mean Differences in Change in Total Fat Intake (Percentage of Energy) Between Intervention and Control Groups for Group 4 (Health Care Setting Interventions with High-Risk Adults).

Group 5. Health Care Setting Interventions with Adults Diagnosed with Specific Diseases

In this grouping, we included seven studies conducted with patients with type 2 diabetes,40,43,52,57,76,73-74 two studies conducted with cancer patients,55,81 and one study conducted with patients with cardiovascular disease.48 Because the interventions for the three diseases were not sufficiently comparable, we did not combine the effect size estimates across diseases. Two studies of computer-assisted interventions for cancer patients were dropped from this group after further consideration because their interventions were very different from the others and from one another.73-74,104

We combined the results for (1) three studies among type 2 diabetes patients receiving less than 3 months of group sessions with various other activities,40,43,76 (2) three studies among type 2 diabetes patients receiving similar interventions and reporting results at 6-months follow-up,43,52,76 and (3) two studies conducted with cancer patients receiving less than 6 months of group sessions with other activities at 6-months follow-up.55,81 The mean difference-in-differences for the final grouping are reported in Table 21 and plotted in Figure 6.

Figure 6. Plotted Estimates of the Mean Differences in Change in Total Fat Intake (Percentage of Energy) Between Intervention and Control Groups for Group 5 (Health Care Setting Interventions with Disease-Diagnosed Adults).

Figure

Figure 6. Plotted Estimates of the Mean Differences in Change in Total Fat Intake (Percentage of Energy) Between Intervention and Control Groups for Group 5 (Health Care Setting Interventions with Disease-Diagnosed Adults).

The results for the type 2 diabetes patient interventions were all quite low. Only the 6-month follow-up estimate was statistically significant (mean effect of 2.6 percentage points; 95 percent CI, 0.07 to 5.1 percentage points).

In contrast, after 6 months, the studies with cancer patients had an effect of 9.7 percentage points (95 percent CI, 7.8 to 11.6 percentage points). The CVD patients who received intensive nutrition education (including a week-long retreat and biweekly group sessions for a year) had a significant mean effect of 24.1 percentage points (95 percent CI, 17.1 to 31.1 percentage points).48

Summary of Meta-Analysis Findings

The results of this meta-analysis lend support to the proposition that these interventions can promote decreases in total fat intake. In addition, our findings suggest that behavioral interventions appear to be more successful with higher-risk populations -- although we observed that interventions with higher-risk populations are often more intense than interventions with lower-risk populations.

The greatest effects were observed among populations that had been diagnosed with potentially life-threatening diseases such as cancer and CVD. We observed the largest effect size (a difference in improvement after 12 months of 24.1 percentage points) among patients diagnosed with CVD. However, this intervention was also particularly intense, including a week-long retreat and two group meetings per week over 1 year, suggesting that the intensity of the intervention may have been more influential than the risk status of the population at promoting dietary change.

We also observed mean effects at 6 months of 15.3 percentage points among women at high risk for cancer who were participating in counseling and monthly sessions. Similarly, among women diagnosed with cancer who were involved in counseling and ongoing support sessions, we observed mean effects at 1 year of 9.7 percentage points.

The biggest departure from the pattern of larger effects being observed among high-risk populations was that individuals diagnosed with type 2 diabetes had a mean effect of only 2.6 percentage points. Part of the reason for this relatively lower effect is likely the fact that this set of interventions was not as intense or long lasting as the intervention conducted with cancer and CVD patients. Beyond this, we also speculate that type 2 diabetes may be perceived by patients more as a chronic condition than as a life-threatening illness, and that this may be associated with the lower efficacy of interventions with this population.

Although interventions conducted outside health care settings appeared to have modest effects, in many cases the interventions themselves were less intense. However, such interventions may have the potential to reach large numbers of people, and preventive interventions may benefit from working with general populations who can be reached through worksite, community, and school settings.

Finally, our results seem to indicate that interventions that have multiple components (e.g., counseling plus materials rather than materials alone) are more effective; furthermore, continuing reinforcement of interventions may be associated with maintenance of changes in behavior. For instance, even though the effects of interventions were often apparent after 6 months, interventions that extended for a longer period of time (through the use of ongoing group sessions) were largely able to maintain those changes. This suggests that research into the various components of interventions and the maintenance of behavior change may be fruitful areas for continuing research.

Standardized Analysis of the Magnitude of the Intervention Effect

Our next analysis strategy for determining the efficacy of dietary interventions on fat intake involved the difference-in-deltas approach, which we applied to the three most commonly reported fat outcomes: total fat (percentage of energy intake), total fat (grams), and saturated fat (percentage of energy intake). This analysis included 34 of the 80 articles reporting results for dietary fat outcomes.

Similar to the analyses for fruits and vegetables, we calculated the median differences in change in fat intake between the intervention and control groups. In addition to indicating the magnitude of the intervention effect for total fat and saturated fat outcomes that had not been included in the meta-analysis just described, this approach enabled us to include more studies reporting total fat than were eligible for the meta-analysis (because the metric we calculated does not require an indication of the variance of the estimate).

This section reports the overall magnitude of the intervention effect at the first follow-up period (Table 22) as well as additional follow-up periods among the subset of eligible studies reporting results for more than one follow-up period (Table 23). Finally, we present the results of various population and intervention characteristics considered as "predictors" of the efficacy of interventions at decreasing intake of total and saturated fat (Table 24). We applied the same minimum cell size requirement to the difference-in-deltas analyses for dietary fat outcomes that were applied to the fruit and vegetable outcomes. Thus, a minimum of five studies had to be included in each cell in order for a particular statistic to be included in the tables. This requirement prevented our carrying out several analyses for the outcome of total fat.

Table 22. Median differences in percentage change in dietary fat intake between intervention and control groups.

Table

Table 22. Median differences in percentage change in dietary fat intake between intervention and control groups.

Table 23. Median differences in percentage change in dietary fat intake between intervention and control groups at Follow-Up 1 and 2.

Table

Table 23. Median differences in percentage change in dietary fat intake between intervention and control groups at Follow-Up 1 and 2.

Table 24. Median differences in percentage change in dietary fat outcomes by demographic and intervention characteristics.

Table

Table 24. Median differences in percentage change in dietary fat outcomes by demographic and intervention characteristics.

Overall Magnitude of Effects

As shown in Table 22, the interventions were consistently successful at reducing intake of total fat (expressed as both percentage of energy intake and grams per day) and saturated fat. The magnitude of the intervention effect was generally similar to that observed for the fruit and vegetable outcomes. The median difference in the percentage change between the intervention and control groups was similar for total fat (-15.7) and saturated fat (-14.5), suggesting that studies were equally successful at reducing intake of total fat and saturated fat (as a percentage of energy intake). The magnitude of the intervention effect was notably higher for absolute intake of fat (-38.0) than for the proportion of total energy intake from fat (-15.7).

Duration of Effects

In considering the longer-term differences in fat intake between treatment groups (Table 23), we found the intervention effect to be relatively stable from follow-up 1 to follow-up 2 for both total fat and saturated fat. An insufficient number of studies reporting results for total fat intake (in grams) at more than one follow-up period prevented us from examining longer-term effects for this outcome.

Effects by Population Characteristics

The difference-in-deltas for the outcomes of total fat and saturated fat are presented by population and intervention characteristics in Table 24. We were not able to explore any population (or intervention) characteristics for absolute fat intake because of an insufficient number of studies. As indicated in Table 24, risk status was the only population characteristic for which the number of articles was large enough to conduct a comparison. Similar to the pattern observed for fruit and vegetable intake, studies conducted with populations at risk of (or diagnosed with) disease observed greater changes in dietary intake than studies conducted among general populations. The magnitude of the reductions in both total fat and saturated fat were notably higher among interventions conducted among high-risk populations than general populations (by 12 percentage points for total fat and 15 percentage points for saturated fat).

Effects by Intervention Characteristics

We considered the influence of several intervention characteristics on the magnitude of change in total fat and saturated fat intake. These characteristics included theoretical basis, study quality, the use of non-nutrition components in the intervention, family components, social support, small group components, food-related activities, and goal setting. However, as indicated in Table 24, the minimum cell size requirement precluded several comparisons for saturated fat intake.

Unlike the pattern among fruit and vegetable outcomes, the use of a theoretical framework in designing or implementing the intervention was not associated with the magnitude of the intervention effect using the difference-in-deltas approach. In fact, studies employing a theoretical framework reported substantially lower decreases in total fat and saturated fat intake than studies not using theory (by 8 to 10 percentage points).

Interestingly, the quality score assigned to the studies also did not appear to be associated with the magnitude of the intervention effect using this analysis strategy. As indicated in Table 24, studies for which we assigned the highest quality rating scores actually reported the lowest magnitude of change in total fat intake (a decrease of 9.2 percentage points).

Contrary to our expectations, studies employing intervention components other than dietary education (e.g., physical activity components, smoking components) had slightly higher decreases in both total and saturated fat than did studies that focused solely on nutrition (our rationale was that among studies with non-nutrition components, the nutrition message would be diluted by the ancillary components). However, the differences in this grouping were quite small.

Of the special intervention features, only social support and goal setting appeared effective at reducing intake of total fat. Goal setting was not associated with a greater decrease in saturated fat, however. Interventions that incorporated social support components had substantially larger decreases in total fat (-26.7 percentage points) than did interventions not incorporating social support (-10.4 percentage points). An insufficient number of articles prevented us from exploring the impact on saturated fat. The use of family components and interactive activities involving food were actually associated with smaller decreases in fat intake than the decreases observed among studies that did not incorporate these components into the intervention.

Corroboration of Changes in Dietary Fat Outcomes Using Biochemical Indicators

One advantage of measuring biochemical outcomes among participants in behavioral interventions is the ability to determine whether physiological changes accompany self-reported behavioral changes in dietary fat intake. As part of our review, we abstracted information for two biochemical outcomes associated with intake of dietary fat: total blood cholesterol (TC) and low-density lipoprotein (LDL) cholesterol. However, because an insufficient number of articles reported results for LDL cholesterol, this report presents only the results for our secondary analyses of the relationship between change in dietary fat intake and TC. For studies reporting results for both fat intake and TC, we conducted a separate series of analyses to determine the relationship between change in dietary fat intake and change in biochemical indicators over the intervention period. This series of analyses used the difference-in-deltas approach -- that is, we calculated the difference in change in TC between intervention and control groups.

Table 25 presents the median difference-in-deltas for biochemical outcomes and fat outcomes for the two eligible comparisons (i.e., TC and total fat as a percentage of energy intake and TC and saturated fat as a percentage of energy intake). As indicated in the table, the differences in the changes in biochemical outcomes between intervention and control groups were consistently of a much smaller magnitude than changes in dietary outcomes over the same time period.

Table 25. Median differences in percentage change in dietary fat outcomes and total blood cholesterol.

Table

Table 25. Median differences in percentage change in dietary fat outcomes and total blood cholesterol.

We also explored the linear bivariate relationship between difference-in-deltas for biochemical outcomes and the difference-in-deltas for fat outcomes using Pearson correlations. Because of the small number of studies reporting results for both biochemical and dietary fat-related outcomes for which difference-in-deltas could be calculated, we decided to focus primarily on two relationships: change in TC compared with change in total fat and change in TC compared with change in saturated fat. While the goal of disease risk reduction is typically to lower LDL (but not high-density lipoprotein cholesterol [HDL]) cholesterol levels through reductions in saturated fat intake, an insufficient number of studies prevented us from examining other comparisons; for example, only three studies reported both total fat and either TC or LDL cholesterol,48,76,80 only one study reported both saturated fat and LDL cholesterol,76 and only two studies reported both total fat (grams) and LDL cholesterol.48,63-64

The results of the correlation analyses suggest that whereas change in total fat intake does appear to be associated with decreases in TC, changes in saturated fat are not. The correlation coefficient for the relationship between the difference-in-deltas for total fat and TC was 0.763 (p = 0.004). In contrast, the correlation between saturated fat and TC was -0.066 (p = 0.877).

In short, the differences between intervention and control groups in reducing intake of total fat observed among the studies we reviewed are substantiated by a significant correlation with concomitant reductions in TC. The difference-in-deltas for saturated fat were not associated with similar reductions in TC. The small number of eligible studies may have influenced our ability to detect a statistically significant correlation between saturated fat intake and TC. In addition, the significant correlation between TC and total fat intake likely reflects the fact that total fat is a marker of saturated fat intake, because intake of total fat independent of the relative content of different types of fatty acids is not thought to be associated with blood cholesterol. Finally, the significant correlation between TC and total fat observed in our analyses may reflect the fact that energy balance influences plasma cholesterol, and because we were not able to adjust for the influence of weight change in our secondary analyses, the association may simply reflect an energy deficit (regardless of the actual composition of the diet).123

Analysis of the Significance of the Intervention Effect

The final set of analyses summarize the likelihood that investigators reported a significant intervention effect for dietary fat outcomes. To accommodate the tremendous variability in the measurement of dietary fat and to include as many articles (and outcomes) as possible in our analyses, we grouped dietary fat variables into five major sets of outcomes (see Chapter 2): (1) intake of total fat, (2) intake of saturated fat, (3) general fat intake, (4) intake of individual high-fat foods or specific high-fat eating behaviors or practices, and (5) intake of individual low fat foods or specific low-fat eating behaviors or practices. For each grouping, we calculated the percentage of studies reporting a significant intervention effect and then compared differences among various population and intervention characteristics. This analysis strategy included approximately 74 of the 80 articles reporting results for dietary fat outcomes.

Table 26 reports the proportion of articles reporting a significant intervention effect for four of the five sets of fat outcomes (we did not present the results for intake of individual low-fat foods or specific low-fat eating behaviors because we were not able to conduct any analyses by population or intervention characteristics, due to the small number of articles reporting results for this outcome). As indicated in Table 26, we explored the influence of several population and intervention characteristics on the likelihood that a significant intervention effect would be reported. In order to prevent unstable estimates, we used the same minimum cell size requirement employed in the previous analyses. Therefore, in order for us to compare the proportion of articles within a particular grouping characteristic, a minimum of five studies had to be included in each cell of the grouping characteristic.

Table 26. Differences in the proportion of studies reporting a significant intervention effect for change in dietary fat outcomes by demographic and intervention characteristics.

Table

Table 26. Differences in the proportion of studies reporting a significant intervention effect for change in dietary fat outcomes by demographic and intervention characteristics.

Overall Magnitude of Effects

Of the studies we reviewed, the vast majority reported a significant intervention effect for changes in dietary fat intake. Depending on the outcome reported in the articles we reviewed, the percentage of studies reporting significant changes in fat intake ranged from 76 percent, or 13/17 studies (for general fat intake outcomes), to 88 percent, or 14/16 studies (for outcomes measuring the intake of specific high-fat foods or eating practices). Although the data are not shown in Table 26, only 64 percent (7/11) of studies reporting intake of low-fat foods or low-fat eating practices reported a significant intervention effect. Similar to the results for the difference-in-deltas approach, we did not observe any differences in the efficacy of interventions at reducing total fat intake compared with saturated fat intake.

Effects by Population Characteristics

Very few studies explored changes in fat intake among children. We were only able to compare age differences in changes in total fat and saturated fat intake. While a greater proportion of interventions conducted with children reported significant reductions in total fat intake (around 17 percent more studies), slightly fewer reported reductions in saturated fat intake (28 percent fewer studies). However, the extremely small number of studies reporting results for children requires that caution be used in interpreting these results.

As indicated in Table 26, the influence of risk status on fat outcomes was inconsistent. While studies conducted with higher-risk populations were more likely to report significant decreases in saturated fat intake than studies conducted with general-risk populations (94 percent, or 16/17 studies, compared with 70 percent, or 7/10 studies -- similar to the pattern observed using the difference-in-deltas approach), such studies were not more likely to report reductions in total fat intake. Higher disease risk was associated with more favorable results for outcomes related to intake of high-fat foods (or eating behaviors), with all eight eligible studies reporting a significant effect for this set of outcomes (compared with only 3/5 studies conducted in general-risk populations).

Effects by Intervention Characteristics

Insufficient variation among the studies we reviewed prevented us from comparing the influence of intervention settings, delivery mode, or intensity on any of the fat outcomes explored in this tier of analyses. However, we were able to determine the effects of several other intervention characteristics. Among the total pool of articles we reviewed, the use of a theoretical framework in designing or implementing the intervention was not consistently associated with the significance of findings. However, studies that used a theoretical framework were clearly more likely to report significant reductions in total fat intake (20/20 studies) than studies not employing theory (22/29 studies, or 76 percent). This same pattern was evident when examining the subset of articles published in 1995 or later (all 12 studies using theory compared with 11/14 studies that did not). We observed the opposite pattern, however, for saturated fat and outcomes measuring intake of high-fat foods (or eating practices); for these outcomes, interventions reporting a theoretical framework were slightly less likely to report significant findings. For example, 100 percent of studies with some theoretical underpinnings produced a significant intervention effect for total fat, compared with 76 percent of those not reporting a theoretical framework.

The quality score assigned to the studies we reviewed was associated with the likelihood of reporting significant findings for intake of total fat in a dose-response fashion (9/14 low-quality studies reported a significant intervention effect, 12/14 medium-quality studies, and 100 percent -- 21/21 -- of high-quality studies). We did not observe this pattern for intake of saturated fat (although high-quality studies were more likely than low-quality studies to report significant decreases in saturated fat) or for general fat intake scores (for which low-quality studies were the most likely to report significant findings). An insufficient number of articles and limited variability in quality scores precluded this comparison for the high-fat and low-fat intake-related outcomes.

Equal proportions of articles that focused exclusively on dietary change or included non-nutrition components reported significant intervention effects for all sets of fat outcomes. While a higher percentage of "nutrition only" interventions reported significant reductions in total fat intake than did studies that targeted non-nutrition behaviors, this difference was slight (i.e., only eight percent more articles reported significant findings for total fat intake).

We had a large enough number of studies (and variation within the studies) to consider the effects of several specific strategies employed in the studies. Several of these strategies appeared particularly effective at producing changes in dietary behavior. Interventions that included family components were substantially more likely to report significant reductions in total fat intake (13/13 studies compared with 29/36 -- 81 percent) and slightly more likely to report significant reductions in saturated fat intake (9/10 studies compared with 17/20). Similar to the findings based on the difference-in-deltas approach, interventions that incorporated social support and used small groups to provide nutrition education produced greater changes in intake of total fat and saturated fat. The use of small groups was also associated, though to a lesser extent, with reductions in saturated fat intake. We observed a positive association between reductions in total fat intake and the inclusion of interactive activities involving food (such as taste tests and cooking classes). This pattern was not evident for saturated fat intake (with equal proportions of articles reporting significant effects for saturated fat).

Goal setting was consistently associated with a greater likelihood of reporting significant intervention effects among the studies eligible for this analysis. Though the differences for total and saturated fat intake were small, interventions that incorporated goal setting (or self-monitoring) to promote dietary change were more likely to report significant reductions in these fat outcomes. The differences in the proportion of studies reporting significant reductions in fat using general measures of fat intake were substantial (all six studies that employed goal setting reported significant effects, compared with 7/11 studies that did not include this strategy). Finally, the impact of culturally or ethnically specific interventions was consistently associated with reductions in dietary fat, although we were only able to make this comparison for two sets of outcomes (total and saturated fat intake). All five studies that were designed to be culturally or ethnically specific (to the study population) reported significant decreases in total or saturated fat intake.

Views

  • PubReader
  • Print View
  • Cite this Page

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...