Intermittent energy restriction (IER), also known as intermittent fasting, is one of the most popular weight-loss diets, with celebrity endorsements and widespread public interest, even in the Republic of Korea. Unlike energy restriction diets, IER focuses on meal schedules rather than energy intake or meal composition. Although IER can include any type of meal schedule that alternates between non-fasting and fasting or energy-restriction periods, the usual IER strategy includes periodic fasting (or energy-restriction intake), such as 5 days non-fasting and 2 days fasting (5:2), 4 days non-fasting and 3 days fasting (4:3), alternate day fasting (ADF), daily time-restricted eating (TRE) with an 8-hour eating window and 16-hour fast (8 hr TRE) or 12-hour eating and fasting periods (12 hr TRE). IER has become popular because it allows people who want to lose extra body weight to experience periods of free eating.

Previous systematic reviews and meta-analyses of IER^1-6 concluded that it induced modest weight loss and metabolic improvements in people with obesity, but there is a lack of long-term, largescale trials on this technique. Three studies^2,3,6 reported that IER produced weight loss and metabolic improvement comparable to that with continuous energy restriction (CER). However, most obesity specialists have not yet adopted IER. In fact, guidelines on obesity management published in the past 5 years do not even mention IER.^7-9

In 2022, the Korean Society for the Study of Obesity, the Korean Diabetes Association, and the Korean Society of Hypertension issued a consensus statement on the effects of carbohydrate-restricted diets and intermittent fasting on obesity, type 2 diabetes mellitus, and hypertension management.^10,11 The consensus statement does not make a clear recommendation for using IER in adults with overweight or obesity due to the lack of long-term studies and the heterogeneity of results from previous studies; moreover, the statement recommends avoiding IER in patients with type 2 diabetes because of a lack of evidence on its benefits, harms, and associated risk of hypoglycemia. The consensus statement and the results of previous meta-analyses of IER differ because the consensus project searched for only level-I evidence from randomized controlled trials (RCTs) that directly compared IER and CER with study durations of longer than 8 weeks. Therefore, those researchers included only eight RCTs and refrained from making clear recommendations because of the lack of evidence.

Another issue with previous reviews and meta-analyses on this subject is the inclusion of the maintenance period in the study duration. Considering the nature of diet therapy, the treatment strength of IER could weaken in the maintenance period. Therefore, there is a need to gather more high-level evidence on IER versus CER, including the latest clinical trials, which considers only the active weight-loss period, to determine the clinical usefulness of IER in obesity treatment. In this meta-analysis, we analyze high-level evidence from RCTs.

This meta-analysis followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines.¹²

Data search and selection process

A systematic literature search was conducted to answer the research question: “Can IER be a good alternative diet tool to CER for alleviating obesity?” The inclusion criteria for this meta-analysis were as follows: (1) study design of RCT; (2) active weight loss period of ≥12 weeks; (3) study population of human adults with overweight or obesity, with/without diabetes; (4) intervention comprising IER, ADF, or TRE; (5) CER group as control; (6) outcomes of anthropometric measurements, body composition, or blood metabolic markers available; (7) report type of original journal articles written in English; and (8) publication date between January 1, 2011 and December 31, 2021. The electronic databases of Embase, PubMed, PubMed Handheld PICO, and Cochrane Central Register of Controlled Trials were initially searched on April 30, 2022, and reassessed on July 26, 2022. The detailed search items used in the electronic databases are listed in Supplementary Table 1.

We found a total of 200 records, which we stored in EndNote 20 and then exported as a tab-delimited list. Using that list, the first author removed duplicated records and records only with trial identification numbers from clinicaltrials.gov using Microsoft Excel (Microsoft, Redmond, WA, USA). Two researchers (KKK and JHK) screened the remaining 121 records independently and retrieved their reports. After the separate evaluations, disagreements were resolved by consensus of all three authors. One study¹³ from a previous report was added to the remaining 25 records.^14-38 Among the selected 26 articles, one trial with patients with type 1 diabetes mellitus²³ reported most of the targeted outcomes as medians and ranges without variances; therefore, we excluded it from our calculation of integrated mean differences (MDs) because of the unfeasibility of calculating the standard deviation (SD) from the given data. In the end, we included 16 RCTs from 25 articles comprising 1,438 participants in this meta-analysis (Fig. 1).

Data collection

Two researchers (JHK and EMK) extracted data from divided sets of the articles, and the third researcher (KKK) extracted data from the full set of articles independently. Only data for active weightloss interventions were extracted, and the outcome data for the maintenance period were not included. The following details were collected, including the relative changes in their levels after the intervention (%): body weight (kg), relative change in body weight (%), body mass index (BMI; kg/m²), waist circumference (WC; cm), body fat mass (kg), fat free mass (kg), body fat percentage (%), systolic blood pressure (SBP; mmHg), diastolic blood pressure (DBP; mmHg), blood levels of fasting glucose (mg/dL, mmol/L), glycated hemoglobin (HbA1c; %, mmol/mol), fasting insulin (μU/mL, pmol/L); and lipids: total cholesterol, triglycerides, highdensity lipoprotein cholesterol (HDL-C), and low-density lipoprotein cholesterol (LDL-C; mg/dL, mmol/L). The primary outcomes of the studies included in this meta-analysis were body weight, WC, and body fat percentage. As a secondary outcome for this study, we gathered blood pressures and the blood levels of fasting glucose, HbA1c, and lipids (triglycerides, HDL-C, and LDL-C).

Other variables sought from the source articles were the first author’s last name, year of publication, type of IER diet intervention (5:2, 4:3, 1:3:3, ADF, 8-hour TRE, or 12-hour TRE), number of participants at the beginning and at each time point of assessment (number for intention-to-treat analysis was preferred), characteristics of the participants, duration of the run-in period, and period of active weight loss or maintenance. Regarding the type of IER, 1:3:3 refers to 1 day of free eating with 3 days of a low-energy diet and 3 days of a very low-energy diet. As a subgroup analysis, 4:3 and 1:3:3 were regarded as ADF, and the IER methods were classified into 5:2, ADF, and TRE groups.

Bias and certainty assessment

We assessed the risk of bias for the included trials with RoB 2, a revised tool for assessing the risk of bias in RCTs.³⁹ The RCTs were independently assessed by two researchers (KKK and JHK), and any disagreement between them was resolved by consensus. The GRADE (Grading of Recommendations, Assessment, Development and Evaluations) approach was used to assess the certainty of the evidence.⁴⁰

Effect size calculation and conversion

To measure the amount of variation, the interquartile ranges, 95% confidence intervals (CIs), and standard errors were converted to SDs according to the method in the Cochrane Handbook for Systematic Reviews of Interventions.⁴¹ Because most of the selected studies reported the measurements at each planned visit and not the change-from-baseline data, the mean and SD of the changefrom- baseline data were calculated to determine the MD. The SDs of the mean change-from-baseline data were missing in most studies and were thus imputed using a correlation coefficient (Corr), according to the method in the Cochrane Handbook⁴¹ as follows:

SDchange=SDbaseline2+SDfinal2−(2×Corr×SDbaseline×SDfinal)

The Corr was calculated using the data in Headland’s trial²⁸ according to the method in the Cochrane Handbook⁴¹ as follows:

Corr=SDbaseline2+SDfinal2−SDchange22×SDbaseline×SDfinal

Because the calculated Corr values ranged from 0.9024 to 0.9854, it was set at 0.9.

The relative change in body weight (%) was converted to body weight for each time point by multiplying the body weight change (%) with the baseline body weight or 100 kg if the baseline body weight was unavailable. The relative change in cholesterol (%) was converted to a cholesterol level for each time point by multiplying the cholesterol change (%) with the baseline cholesterol level. We converted mmol/L to mg/dL, mol/mol to %, and pmol/L to μU/mL.

Visual display of results of individual studies

Forest plots are used to display the results of the included studies. Standardized mean differences (SMDs; Hedges’ g) were calculated, and pooled effect sizes (weighted mean differences [WMDs]) were calculated using the inverse variance method with the common fixed-effect model. Unstandardized MDs and their WMDs were also calculated with the same model to determine the certainty of the evidence (Supplementary Table 2). The heterogeneity of a meta-analysis can be measured using the Tau² and chi-square statistics. We estimated the level of heterogeneity with the I² statistic, which indicates the percentage ratio of the variability in effect estimates caused by heterogeneity rather than chance. Subgroup analyses were conducted to explore the possible causes of heterogeneity among the study results. R version 4.2.1 (R Foundation for Statistical Computing, Vienna, Austria) and meta package version 5.5-0 were used for all meta-analyses in this study.

To visualize the pattern of change according to period, measurement time points were divided into three groups: 8 to 12 weeks, 16 to 24 weeks, and 52 weeks; forest plots were built using subgroups of the measurement time points. However, because some studies used repeated measurements at different time points and the results of those studies were duplicated in one analysis, the total pooled effect of the intervention was not visualized in forest plots according to the measurement time points.

Sensitivity analysis

Two methods were used to perform sensitivity analyses. First, to evaluate the effect of including studies with a high risk of bias, the whole meta-analysis was repeated after excluding studies with high risk of bias in at least one domain. Second, to evaluate the influence of an outlier on overall effect sizes with high heterogeneity, a sensitivity analysis using the leave-one-out method was executed when the I² for heterogeneity was greater than 75%.

Characteristics of the included studies

The characteristics of each individual study are summarized in Table 1. Regarding the study populations, one study²¹ included only men, and four studies^14,16,33,34 included only women. One study²⁰ included only people with type 2 diabetes, and two studies^13,37 included only people with metabolic syndrome. Most studies included people whose BMI was lower than 30 kg/m², but four studies^21,22,25,33 included only people whose BMI was 30 kg/m² or higher. The BMI range in the inclusion criteria was 24–45 kg/m², and the age range was 18–70 years.

As IER protocols, eight studies used 5:2; six studies used ADF; and two studies used TRE. All studies included in this meta-analysis used CER as the comparator to investigate the efficacy of IER. Most studies tried to match total energy intake per week between the intervention groups. The range of duration of the active interventions was 12 weeks to 52 weeks. Although the article by Headland et al.²⁸ did not clearly describe the durations of the active weight loss and weight maintenance periods, the visit schedule and study flow chart suggest that the initial 8-week period was the intensive weight loss period. However, the participants’ body weight decreased steadily in all groups until at least 28 weeks; therefore, we considered the duration of their study to be 52 weeks.

Risk of bias

The results of the risk of bias assessment for the included trials are shown in Supplementary Fig. 1. A high risk of bias existed in the randomization process (domain 1) and missing outcome data (domain 3). Overall, 25% of the included studies had a high risk of bias. The certainty of evidence scores of the study variables are shown in Supplementary Table 2, along with their unstandardized MDs.

Attrition

Of the 1,438 participants in 16 RCTs randomized to the CER and IER groups, 1,074 participants finished the active weight-loss intervention, for an attrition rate of 25.3%. The attrition rates were 26.6% and 24.1% for the IER and CER groups, respectively.

Changes in body weight, WC, and body composition

The changes in body weight (WMD, –0.06; 95% CI, –0.18 to 0.05), WC (WMD, –0.12; 95% CI, –0.28 to 0.05), and body fat percentage (WMD, –0.14; 95% CI, –0.34 to 0.05) did not differ significantly between the two intervention groups (Fig. 2A, C, and E); significant differences in those values were not found even in the subgroup analyses for types of IER (Fig. 2A, C, and E) and measurement time points (Fig. 2B, D, and F). There were no significant WMDs in changes in BMI, body fat mass (kg), or fat free mass (kg) between the groups (Supplementary Figs. 2-4).

Upon examining individual studies, we found that in the 16 studies with data on body weight, the ranges of mean weight loss were 0.6–13.9 kg and 0.5–12.5 kg in the IER and CER groups, respectively. Except for one study by Razavi et al.³⁷ that reported significantly superior reductions in weight and WC in the IER group, compared with the CER group, all the studies reported non-significant between-group differences. Regarding body composition, the study by Bowen et al.¹⁹ reported a significantly larger reduction in body fat mass in the CER group, but all the other studies that measured body composition reported non-significant between-group differences in changes to body fat percentage and body fat mass.

Changes in blood glucose and HbA1c levels

A meta-analysis of 10 studies that measured blood glucose levels showed that the change in blood glucose levels was in favor of CER and not IER (WMD, 0.16; 95% CI, 0.02–0.30), and the same result was found in the subgroup analysis of measurement time points for 8−12 weeks (WMD, 0.20; 95% CI, 0.04–0.36) (Fig. 3A and B). The superior effect of CER to IER on glucose levels was not observed at measurement time points of 16−24 weeks and 52 weeks (Fig. 3B).

However, there was significant heterogeneity in the effect sizes of changes in blood glucose levels (chi²=58.25, P<0.01, I²=85%) (Fig. 3A). Schübel et al.²⁴ and Bowen et al.¹⁹ reported significantly higher reductions in blood glucose level in the CER group (–6.7± 3.4 mg/dL, –3.6±5.3 mg/dL) than the IER group (–2.8±3.3 mg/dL, –1.8±3.2 mg/dL), and Gray et al.³⁴ reported a significant increase in the blood glucose level in the IER group (5.4±4.0 mg/dL) and no changes in the CER group (0.0±4.0 mg/dL). On the contrary, Headland et al.²⁸ reported a significantly greater reduction in blood glucose in the IER group (–3.6±4.8 mg/dL) than the CER group (0.0±5.3 mg/dL). Considering the characteristics of these four studies with significant SMDs, we note that the studies by Schübel et al.²⁴, Headland et al.,²⁸ and Gray et al.³⁴ used the 5:2 IER method, and the study by Bowen et al.¹⁹ used the 1:3:3 method. The study by Gray et al.³⁴ included women with a history of gestational diabetes mellitus, and their blood glucose levels decreased at 3 months in both groups and increased at 12 months, compared with the baseline glucose levels.

Only five studies measured HbA1c levels, and the meta-analysis of those studies showed no significant between-group differences in pooled effect sizes (Fig. 3C and D). Heterogeneity in the effect size was lower for HbA1c (chi²=8.55, P=0.07, I²=53%) than for blood glucose. Among the four studies with significant SMDs in blood glucose levels, only the study by Gray et al.³⁴ measured HbA1c levels; although it reported a significant increase in the blood glucose level in the IER group, the HbA1c level decreased in both groups, with a higher decrease in the CER group (IER=–0.1% vs. CER=–0.3% and IER=–0.1% vs. CER=–0.2%, at 3 and 12 months, respectively).

Six studies measured fasting blood insulin levels, and a metaanalysis of those studies showed similar levels of decrease between the groups; however, the study by Schübel et al.²⁴ showed a higher reduction in the CER group, and the study by Gray et al.³⁴ showed a higher reduction in blood insulin levels in the IER group (Supplementary Fig. 5).

Changes in SBP and DBP values

SBP was significantly lower in the CER group than the IER group after the intervention (WMD, 0.21; 95% CI, 0.05–0.36) (Fig. 4A). However, DBP did not differ significantly between the groups, except for the results of the subgroup analysis for TRE, which was in favor of CER (WMD, 0.47; 95% CI, 0.10–0.83) (Fig. 4C). According to the measurement time points, no significant differences were found in SBP and DBP changes between the groups (Fig. 4B and D). In the sensitivity analysis of studies without an overall high risk of bias, no significant difference was found in the SBP changes between the groups (WMD, 0.15; 95% CI, –0.01 to 0.31; not shown in figures).

Changes in lipid profile

The overall changes in total cholesterol, triglyceride, HDL-C, and LDL-C levels did not differ significantly between the groups (Supplementary Fig. 6A, Fig. 5A, C, and E). In the subgroup analysis for ADF, CER showed significantly favorable changes in triglyceride (WMD, 0.34; 95% CI, 0.06–0.61) and HDL-C (WMD, –0.34; 95% CI, –0.61 to –0.08) levels (Fig. 5A and C). However, the subgroup analysis comparing ADF with CER showed significant heterogeneity in the effect sizes for changes in triglyceride (chi²=5.12, P=0.02, I²=80%) and HDL-C (chi²=9.75, P<0.01, I²=79%) levels (Fig. 5A and C) because of the study by Bowen et al.¹⁹ that used the 1:3:3 IER method. In addition, in the subgroup analysis of measurement time points, CER showed significantly favorable effects on triglyceride levels at 16–24 weeks and 52 weeks (WMD, 0.20; 95% CI, 0.01–0.40 and WMD, 0.44; 95% CI, 0.04–0.85), respectively (Fig. 5B). The 5:2 method induced more favorable changes in LDL-C and total cholesterol levels than CER (Fig. 5E, Supplementary Fig. 6A).

Sensitivity analysis

Among the included studies, four had a high risk of bias in at least one domain (Supplementary Fig. 1B). Excluding those four trials did not produce changes in the significance of any of the analyses except SBP. Without those four trials, the overall differences in SBP changes became insignificant between the two diet interventions (WMD, 0.15; 95% CI, –0.01 to 0.31). In the sensitivity analysis for blood glucose level, the leave-one-out method was used because the overall I² heterogeneity was high. In the absence of outliers, the overall differences in glucose level changes became insignificant between the two diet interventions (WMD, 0.04; 95% CI, –0.11 to 0.19).

The results of this meta-analysis show that IER can induce considerable weight loss and reduce metabolic complications in people with overweight or obesity, with efficacy comparable to that of CER. However, IER was not superior to CER in weight reduction or metabolic improvement. The findings of this meta-analysis were robust in our sensitivity analyses and consistent with previous reviews and meta-analyses.^1,2,4-6,42

Obesity is a complex chronic disease and a major burden to public health. Obesity and its related complications are becoming increasingly common not only in adults but also in children in many countries, including the Republic of Korea. It is time to put an emphasis on treatment, along with prevention. Diet therapy is a basic treatment for obesity, and a balanced energy-restricted diet is generally recommended by obesity specialists. Adherence to CER, however, is difficult due to its restrictive nature. IER is suggested as an alternative to CER because it includes regular near-normal or free eating periods in the midst of calorie restriction and moves the focus from the amount of energy intake to the timing of that intake.

Achieving comparable weight reduction and metabolic improvement with less stressful diet methods appears to be an attractive strategy. However, two clinical implications should be noted. First, interestingly, the number of people who continued IER until the end of active treatment was not higher than the number that continued CER. Although IER looks easier to maintain than CER in theory, it was apparently not easier in practice. At least during the active weight loss period, CER could be sustained well. Second, although no outcome variable showed significant between-group differences in the sensitivity analysis, in the primary analysis, the changes in blood glucose, triglyceride, HDL-C levels, and blood pressure favored CER. On the other hand, the changes in total cholesterol and LDL-C levels favored IER. Therefore, it is unclear whether IER or CER better improves metabolic parameters.

The usefulness of IER for people with diabetes is still uncertain. Among the two relevant trials included in this meta-analysis, one trial that included 10 patients with type 1 diabetes²³ was a pilot study whose results could not be summated to determine the total effect size because it reported its endpoints as medians and ranges; thus, only one trial that included people with type 2 diabetes^17,20,27 was included here. Both trials^17,20,23,27 used the 5:2 IER method. Concerns about complications from fasting and very low energy diet periods in diabetes patients include hypoglycemia and the loss of fat free mass. Both of the trials we evaluated reported that IER was safe and unlikely to cause hypoglycemia with appropriate medication changes and that it induced a mild loss of fat free mass. Although IER is known to improve beta cell function and glycemic control,⁴³ the study of type 2 diabetes patients²⁰ did not report the level of blood insulin or an index for insulin sensitivity, maybe because of the participants’ use of insulin and its secretagogues. At this time, the evidence for IER in diabetes is scant.

In this meta-analysis, we tried to select articles that were relatively homogenous and used non-contentious methods. All our included studies were RCTs with active weight loss periods of ≥12 weeks and at least one group using any type of IER and another group using CER. All trials reported their results in their publications and included adults aged 18 years or older. Even with those very strict criteria, our analysis included double the number of RCTs considered in the latest consensus statement project; therefore, the level of evidence for the clinical usefulness of IER is clearer in this metaanalysis than in previous works.

This meta-analysis has two limitations. First, we imputed the SDs of group differences by calculating the Corr from the body weight change data. Researchers use different methods to calculate their effect sizes and CIs. When calculating the effect size in a meta-analysis, one hurdle is determining the variance in between-group differences because many clinical trials report outcomes at each time point or differences in the outcomes from the baseline only for the individual intervention groups without specifying between-group differences. The method used to impute the SD of the MD can thus affect the results of meta-analyses. We imputed the SDs of the MDs by calculating the Corr from data in Headland’s trial,²⁸ which reported all SDs of baseline measurements, measurements at different time points, and changes in body weight. However, it is unclear whether the calculated Corr could be applied to other variables. Second, we found statistically significant differences between the standardized and unstandardized MDs in triglyceride and LDL-C levels. Given that the SMDs in triglyceride and LDL-C levels showed no statistical significance and their unstandardized MDs showed marginal statistical significance, the possibility of imprecision cannot be excluded.

In conclusion, IER can be used as an alternative to CER because it induces comparable weight loss and metabolic improvement. However, the effect of IER was not superior to that of CER, and its attrition rate was not lower than that of CER.

We would like to thank Johng Ho Kim for editing of figures.

Supplementary materials can be found online at https://doi.org/10.7570/jomes22050.