BMI: body mass index; CI: confidence interval; HDL: high density lipoprotein; LDL: low density lipoprotein; MD: mean difference; RCT: randomized controlled trial.

RCTs included in the meta-analysis reported baseline total fat intakes of 29–43% of total energy intake and reductions in body weight were observed across all baseline values, with most RCTs having baseline intakes of 30% or greater. In the published systematic review, all measures of body fatness were grouped together for GRADE assessment and assigned a collective rating of high. For this guideline the measures of body fatness were assessed individually, and all outcomes except body fat percentage were assigned a rating of high (body fat percentage was assigned a rating of moderate).

Number of comparisons provided in parentheses.

Most of the studies included in the meta-analysis (18 out of 26) appeared to use appropriate methods of random sequence generation and almost half had good allocation concealment. None of the studies were blinded; however, blinding in dietary trials is generally very difficult with the exception of a small number of trials that provide foods in which macronutrients have been covertly modified. It was not clear in most studies if blinding of outcome assessment had been achieved. Incomplete outcome reporting was variable across studies, with many having a high risk of attrition bias as defined in this analysis (studies that lost more than 5% of participants per year were considered to be at high risk of attrition bias), though selective reporting was assessed as probably not having a significant impact on risk of bias. Most included studies had systematic differences in care (i.e. intervention arms had more time or attention devoted to them than the control arm); however, sensitivity analysis in which studies with systematic differences in care were excluded did not significantly change the effect observed on body weight, suggesting that systematic differences in care were not an important source of bias in these studies. Most studies did not suggest bias resulting from variable compliance, and sensitivity analysis in which studies with suspected problems with compliance were excluded did not significantly change the effect observed on body weight, suggesting that issues with compliance were not an important source of bias in these studies. No other significant biases noted.

I² = 75%, indicating a significant level of heterogeneity. The heterogeneity was partly explained by the degree of reduction of fat intake, by the BMI of participants, and by the level of fat intake in control arms, which together explained 16% of between-study variance (in meta-regression). The only inconsistency (where heterogeneity arose) was in the size of this effect. Otherwise, the direction of effects in these RCTs was remarkably consistent – 30 of the 33 comparisons showed a decrease in body weight. The reduction in weight in those on reduced fat diets was seen in very different populations and from 6 months to several years and was resistant to all sensitivity analyses, including the exclusion of: studies that gave additional support to the low-fat arms, studies that delivered additional dietary interventions (on top of the change in dietary fats), studies with suspected compliance issues, the largest study (1), and when using fixed – rather than random-effects meta-analysis. Effects on body weight are supported by similar effects on BMI, waist circumference and percentage of body fat. Therefore, although formal analysis indicated a significant level of heterogeneity that was only partially explained, additional evidence indicated that the effect was robust and consistent across a large number of studies and was therefore not downgraded.

All included RCTs directly compared (and randomized participants) to lower versus usual fat intake. Participants were directly relevant because they came from all parts of the world, included men and women, and people who were healthy, with risk factors or with long-term conditions at baseline. The studies all addressed measures of body fatness or potential undesirable effects directly and did not use proxy measures.

Large numbers of participants (a minimum of approximately 2400, and for most outcomes many more) were included in RCTs of at least 6 months duration. The 95% CI does not cross a threshold of irrelevant benefit or important harm.

The funnel plot did not suggest publication bias. The consistent reduction in body weight observed, despite the fact that none of the studies included intended to alter weight in either arm, suggests that publication bias for this outcome is unlikely.

Most of the studies included in the meta-analysis (12 of 14) appeared to use appropriate methods of random sequence generation although most did not have clear reporting on allocation concealment. None of the studies were blinded; however, blinding in dietary trials is generally very difficult with the exception of a small number of trials that provide foods in which macronutrients have been covertly modified. It was not clear in most studies whether blinding of outcome assessment had been achieved. Incomplete outcome reporting was variable across studies, with many having a high risk of attrition bias as defined in this analysis (studies that lost more than 5% of participants per year were considered to be at high risk of attrition bias), though selective reporting was assessed as probably not having a significant impact on risk of bias. Most included studies had systematic differences in care (i.e. intervention arms had more time or attention devoted to them than the control arm); however, sensitivity analysis conducted for body weight (which has many studies in common with those reporting BMI) in which studies with systematic differences in care were excluded did not significantly change the effect observed on body weight, suggesting that systematic differences in care were not an important source of bias in these studies. Most studies did not suggest bias resulting from variable compliance, and sensitivity analysis conducted for body weight (which has many studies in common with those reporting BMI) in which studies with suspected problems with compliance were excluded did not significantly change the effect observed on body weight, suggesting that issues with compliance were not an important source of bias in these studies. No other significant biases were noted.

I² = 60%, indicating a moderate level of heterogeneity. The only inconsistency (where heterogeneity arose) was in the size of this effect. Otherwise, the direction of effects in these RCTs was remarkably consistent – 15 of the 17 comparisons showed a decrease in body weight. The reduction in BMI in those on reduced fat diets was seen in different populations and from 6 months to several years. Effects on BMI are supported by similar effects on body weight, waist circumference and percentage of body fat. Therefore, although formal analysis indicated a moderate level of heterogeneity, additional evidence indicated that the effect was robust and consistent across a large number of studies and was therefore not downgraded.

Both studies included in the meta-analysis appeared to use appropriate methods of random sequence generation and no significant issues with incomplete outcome data, or selective reporting, but neither was free from systematic differences in care. One study did not report sufficient information to make an assessment on allocation concealment, blinding of outcome assessment or compliance. No other significant biases were noted.

I² = 21%, indicating a low level of heterogeneity.

All three studies included in the meta-analysis appeared to use appropriate methods of random sequence generation, had good allocation concealment and no significant issues with blinding of outcome assessment, incomplete outcome data, selective reporting or compliance. One study was not free from systematic differences in care. No other significant biases were noted.

I² = 0%, indicating no heterogeneity.

Of the two studies reporting percentage body fat, one (1) carried 98.8% of the weight in meta-analysis and therefore contributed virtually all the data. This study was conducted in postmenopausal women from different ethnic backgrounds living in the United States of America. Although the effect observed for percentage body fat in this population was consistent with the effects observed for body weight, BMI and waist circumference in larger, more heterogeneous populations, and there was no evidence to indicate that the physiological response to a change in total fat intake would be significantly different between this group and the general adult population, this outcome has been downgraded for indirectness as a conservative measure.

Although a small, clinically insignificant effect was observed and the 95% CI crosses the null, it does not cross a threshold of important benefit or harm. Taken together, this was considered as precise evidence of no effect and therefore not downgraded.

Most of the studies included in these meta-analyses appeared to use appropriate methods of random sequence generation but many did not provide sufficient information to assess allocation concealment. None of the studies were blinded; however, blinding in dietary trials is generally very difficult with the exception of a small number of trials that provide foods in which macronutrients have been covertly modified. It was not clear in most studies whether blinding of outcome assessment had been achieved. Incomplete outcome reporting was variable across studies, with many having a high risk of attrition bias as defined in this analysis (studies that lost more than 5% of participants per year were considered to be at high risk of attrition bias), although selective reporting was assessed as probably not having a significant impact on risk of bias. Most included studies had systematic differences in care (i.e. intervention arms had more time or attention devoted to them than the control arm); however, sensitivity analysis conducted for body weight (which has many studies in common with those reporting on blood lipids) in which studies with systematic differences in care were excluded did not significantly change the effect observed on body weight, suggesting that systematic differences in care were not an important source of bias in these studies. Most studies did not suggest bias resulting from variable compliance, and sensitivity analysis conducted for body weight (which has many studies in common with those reporting on blood lipids) in which studies with suspected problems with compliance were excluded, did not significantly change the effect observed on body weight, suggesting that issues with compliance were not an important source of bias in these studies. No other significant biases were noted.

I² = 72%, indicating a significant level of heterogeneity. The only inconsistency (where heterogeneity arose) was in the size of this effect. Otherwise, the direction of effects in these RCTs was remarkably consistent – 22 of the 27 comparisons showed a decrease in total cholesterol. The reduction in total cholesterol in those on reduced fat diets was seen in different populations and from 6 months to several years. Therefore, although formal analysis indicated a moderate level of heterogeneity, additional evidence indicated that the effect was robust and consistent across a large number of studies and was therefore not downgraded.

The funnel plots were difficult to interpret but did not suggest publication bias.

I² = 57%, indicating a moderate level of heterogeneity. The only inconsistency (where heterogeneity arose) was in the size of this effect. Otherwise, the direction of effects in these RCTs was remarkably consistent – 19 of the 22 comparisons showed a decrease in LDL cholesterol. The reduction in LDL cholesterol in those on reduced fat diets was seen in different populations and from 6 months to several years. Therefore, while formal analysis indicated a moderate level of heterogeneity, additional evidence indicates that the effect is robust and consistent across a large number of studies and was therefore not downgraded.

I² = 23% indicating a low level of heterogeneity.

Most of the studies included in the meta-analysis (4 of 5) appeared to use appropriate methods of random sequence generation and three of them had good allocation concealment. None of the studies were blinded; however, blinding in dietary trials is generally very difficult with the exception of a small number of trials that provide foods in which macronutrients have been covertly modified. Blinding of outcome assessment was achieved in one study and the others did not provide sufficient information to assess. About half the studies did not report issues with incomplete outcome data, selective reporting or systematic differences in care. All but one reported good compliance. No other significant biases were noted.

I² = 23%, indicating a low to moderate level of heterogeneity.

I² = 57%, indicating a moderate level of heterogeneity. There was little evidence of overall effect with about half the studies reporting a clinically insignificant decrease and half a clinically insignificant increase. All but one of the 95% CIs overlapped. Therefore, although formal analysis indicated a moderate level of heterogeneity, additional evidence indicated that the heterogeneity is not important and was therefore not downgraded.

Most of the studies included in the meta-analysis (9 of 10) appeared to use appropriate methods of random sequence generation and four of them had good allocation concealment. None of the studies were blinded; however, blinding in dietary trials is generally very difficult with the exception of a small number of trials that provide foods in which macronutrients have been covertly modified. Less than half the studies reported no issues with blinding of outcome assessment, selective reporting or systematic differences in care; most did not provide sufficient information to assess. More than half the studies did not report issues with incomplete outcome data. All but two reported good compliance. No other significant biases were noted.

I² = 9%, indicating a low level of heterogeneity.

The funnel plots were difficult to interpret, but suggested that studies with smaller reductions or small rises in blood pressure may be missing. If such studies were included, then the effect would move closer to zero. The 95% CI does not cross a threshold of irrelevant benefit or important harm. Not downgraded.

I² = 7%, indicating a low level of heterogeneity.

This single, large and well-conducted study appeared to use appropriate methods of random sequence generation, had good allocation concealment and no issues with blinding of outcome assessment, incomplete data, selective reporting or compliance. This study did have systematic differences in care. No other significant biases noted.

Single trial only, no inconsistency but no evidence of consistency. Downgraded once.

This study (1) was conducted in postmenopausal women from different ethnic backgrounds living in the USA. Although the effect observed for percentage body fat in this population was consistent with the effects observed for body weight, BMI and waist circumference in larger, more heterogeneous populations, and there was no evidence to indicate that the physiological response to a change in total fat intake would be significantly different between this group and the general adult population, this outcome was downgraded for indirectness as a conservative measure.

This study included more than 40 000 participants with several years of follow-up. The 95% CI does not cross a threshold of irrelevant benefit or important harm.

BMI, body mass index; CI: confidence interval; MD: mean difference; RCT: randomized controlled trial.

Baseline intakes of total fat were 30% of total energy intake or higher in the RCTs conducted in children, and the goal of the intervention was to reach intakes of less than 30% of total energy intake.

Only single trials were undertaken for each outcome so it was not possible to assess inconsistency.

There were too few studies to formally assess publication bias.

This study was downgraded once for serious risk of bias because it was unclear whether allocation concealment was achieved.

The study was conducted in the population of interest and assessed the effect of reduced total fat intake on the priority health outcome decided on before initiating review.

The 95% CI did not cross a threshold of irrelevant benefit or important harm.

Additional evidence on the relationship between reduced total fat intake and measures of body fatness in children comes from additional follow-up data of effects on BMI in this study at 1 year (MD −0.30 kg/m>²; 95% CI −0.75, 0.15; 620 participants; low certainty in the evidence) and approximately 7 years (MD −0.10 kg/m>²; 95% CI −0.75, 0.55; 576 participants; low certainty in the evidence). The last measurements in this study were made on a final visit of the children, which was at approximately 7 years after initiation of the study for many participants. The variability in the actual time of follow-up as well as attenuation of the intervention after 3 years of follow-up decreased confidence in the data for the final follow-up period. The data for the final visit is consistent with the data at both 1 year and 3 years of follow-up, in that no significant effect was observed at any timepoint.

This was a well-conducted RCT with methods in place to minimize risk of selection, performance, detection, attrition and reporting bias.

Downgraded once for serious indirectness: participants were children with raised blood lipids, and therefore results may not be directly generalizable to all children.

The 95% CI crosses a threshold of potentially relevant benefit or important harm.

Additional evidence on the relationship between reduced total fat intake and measures of body fatness in children comes from additional follow-up data of effects on body weight in this study at 1 year (MD −0.50 kg/m2; 95% CI −1.78, 0.78; 620 participants; low certainty in the evidence)

Weight and weight-for-age z-scores may not be sufficiently indicative of changes in body fatness because these measures do not factor in changes and differences in height, which are known to vary greatly across children throughout development. Already downgraded for indirectness.

Unclear risk of bias across all domains.