Michael L. Dansinger

Meta-analysis: The Effect of Dietary Counseling for Weight Loss

Key Summary Points This systematic review suggests that, on average, dietary counseling interventions for weight loss have resulted in a net loss of approximately 2 BMI units (6%) at 12 months compared with usual care. This weight difference narrows considerably over the subsequent 4 years. Long-term trials that evaluate the time course of weight changes are needed to confirm these conclusions. Factors that may alter the effectiveness of dietary counseling interventions require additional investigation. Future studies must improve their analyses, reporting of study design and findings, and follow-up of study participants. Obesity-related medical problems are among the most serious health problems facing U.S. adults. Approximately 65% of U.S. adults are overweight (body mass index [BMI] >25 kg/m2), and approximately half of overweight adults are obese (BMI >30 kg/m2) (1). Coronary heart disease is twice as common in obese people as in normal-weight people, and obesity substantially exacerbates all metabolic cardiac risk factors (2, 3). Obesity is associated with decreased longevity (47) and quality of life (8). Dietary and lifestyle modification efforts are the primary methods for treating and preventing obesity. Several important systematic reviews show that dietary-based lifestyle modification efforts can statistically significantly improve body weight and decrease related medical problems (917). The average weight change due to dietary counseling compared with usual care is unclear, particularly over the long term. We systematically reviewed and quantitatively synthesized published data on the net effect of dietary-based counseling compared with usual care over time. We also evaluated various sources of heterogeneity on the effectiveness of weight loss strategies. Methods Data Sources and Searches The starting point for the literature search was an extensive systematic review published in 1998, on which current clinical guidelines regarding identification, evaluation, and treatment of overweight and obesity in adults are based (9). For that report, a 24-member panel of experts methodically and critically examined all relevant scientific literature published between January 1980 and September 1997. After identifying all relevant studies from that report, we conducted a systematic literature search in all languages in MEDLINE and in the Cochrane Central Register of Controlled Trials databases from January 1997 through July 2006 to identify the effect on BMI of dietary-based weight-loss counseling interventions. We found additional studies from reference lists of other systematic reviews and included studies. Terms used in the searches of recent studies were weight loss, body mass index, obesity, diet, behavior therapy, and lifestyle. The search was limited to randomized, controlled trials. Study Selection We included randomized trials that reported original data on the effect of dietary counseling (advice to change dietary patterns) on body weight or BMI compared with the effect of control interventions (usual care or a minimal intervention, such as general verbal or written advice given at baseline, which was designed to approximate usual care). We excluded studies in which patients were younger than 18 years, the goal of the intervention was not weight loss, exercise was the only intervention, mean baseline BMI was less than 25 kg/m2, or SEs could not be determined. In keeping with the eligibility criteria of the 1998 evidence report (9), we also excluded studies with interventions that lasted fewer than 12 weeks and those that did not report effects at a minimum of 16 weeks. Figure 1 shows the number of studies that did not meet the eligibility criteria and the reasons for their exclusion. Figure 1. Study flow diagram. Cochrane = Cochrane Central Register of Controlled Trials; NHLBI/NIDDK = National Heart, Lung, and Blood Institute/National Institute of Diabetes and Digestive and Kidney Diseases. *Not randomized or quasi-randomized; no usual care control; weight loss not goal of intervention; exercise-only intervention; intervention duration less than 12 weeks or follow-up duration less than 16 weeks; baseline body mass index less than 25 kg/m2; no weight change data; insufficient data to calculate standard error; duplicate publication of study; abstract or letter. Data Extraction and Quality Assessment One of 3 reviewers used standardized forms to extract all studies that met the eligibility criteria. A second reviewer reviewed all extracted data. When necessary, disagreements were resolved by consensus of 2 or more authors. For all included studies, we extracted or estimated the net change in BMI and the SE of the net change from the reported data. Net change was defined as the change from baseline in the treatment group minus the change from baseline in the control group. We did not analyze change from baseline in the treatment groups alone (without subtracting changes in the control groups). When necessary, we calculated the change in BMI by using the ratio of baseline BMI to kilograms as the conversion factor. For the 9 studies that did not report such data, we assumed a ratio of 2.7 (equal to a height of 1.64 m), which corresponded well to ratios from other studies. Within studies, we preferentially chose data from intention-to-treat analyses; however, we retained data from all studies regardless of whether their analyses omitted missing data, used last observations carried forward, or replaced missing data with baseline data. For each study that reported change in BMI at multiple time points, we calculated the slopes of the net change in BMI across the different time points. When necessary, the SE of the net change was estimated from the SEs of the changes in BMI in the intervention and control groups, assuming that these values were independent of each other. Similarly, when necessary, the SEs of the intervention and control group changes in BMI were calculated from the SEs of the baseline and final values by using the following equation: where SE1, SE2 and SE12 are the SEs for baseline values, final values, and change in values, respectively, and is the correlation between SE1 and SE2 (18). We arbitrarily chose to be 0.50, the midpoint value. In secondary analyses, we used correlations of 0.25 and 0.75 (18). For the meta-analysis, we calculated the SE of the net BMI slopes by using the same methods. We also extracted data that described the counseling and usual care interventions, participants, study design, and adverse events. Active and maintenance phases of weight loss were defined according to the designations provided by the authors of each study. All studies were assessed for methodological quality according to the design, conduct, and reporting of the clinical study. We used a 3-level classification of study quality (19). Studies rated as good mostly adhered to the commonly held concepts of good quality, including clear description of the sample, setting, intervention, and comparison groups; blinding of outcome assessors; appropriate statistical and analytic methods and reporting of these methods, including randomization technique and intention-to-treat analysis; no obvious reporting errors; participant withdrawal less than 20%; clear reporting of those who withdrew; and no obvious bias. Fair-quality studies had some deficiencies (but none that were likely to cause major bias) or may have had missing information that made it difficult to assess limitations and potential problems. Poor-quality studies had serious errors in design, analysis, or reporting or may have had a large amount of missing information, discrepancies in reporting, or high rates of withdrawal. Data Synthesis and Analysis To summarize the net changes in BMI over time, we used different methods to analyze the data. All analyses were performed separately for data from the active and the maintenance phases of the behavioral weight loss programs. We graphed the net change in BMI for each study against time from the start of the intervention. Separate random-effects model meta-analyses assign a weight to each study on the basis of individual study variance and between-study heterogeneity (20) at each time point for which data were available from at least 3 cohorts of patients. For studies that reported data at multiple time points, not including baseline, we calculated slopes for each available period (for example, if data were reported at 3, 6, and 12 months, we calculated slopes for 3 to 6, 3 to 12, and 6 to 12 months). For periods where we had data from at least 3 cohorts of patients, we meta-analyzed the slopes by using a random-effects model. The calculated SEs of the slopes were used only for weighting the studies in the meta-analysis, not for estimating the statistical significance of the slopes. These SEs capture the relative variances of the net weight changes from baseline at the multiple time points but do not accurately estimate the SEs of the slopes themselves. For each period, we compared the effect of diabetes (inclusion vs. exclusion of patients with diabetes) and intervention (diet and exercise vs. diet alone) by using 2-sample t tests. In addition, all data were analyzed in a random-effects model meta-regressiona meta-analytic technique of multivariable linear regression across studiesaccording to the method of Morris (21) as described by Berkey and colleagues (22). This model is similar to the DerSimonian and Laird (20) random-effects model meta-analysis. We regressed net change BMI against time in months. We also conducted analyses with study-level variables that were potentially associated with the magnitude of the treatment effect, based on known associations within individual studies or on what we considered to be clinically or methodologically relevant from previous studies. These variables included intervention type (diet vs. diet and exercise) (23), frequency of support meetings (prorated for the first

Dietary Fructose and Glucose Differentially Affect Lipid and Glucose Homeostasis

Absorbed glucose and fructose differ in that glucose largely escapes first-pass removal by the liver, whereas fructose does not, resulting in different metabolic effects of these 2 monosaccharides. In short-term controlled feeding studies, dietary fructose significantly increases postprandial triglyceride (TG) levels and has little effect on serum glucose concentrations, whereas dietary glucose has the opposite effects. When dietary glucose and fructose have been directly compared at approximately 20-25% of energy over a 4- to 6-wk period, dietary fructose caused significant increases in fasting TG and LDL cholesterol concentrations, whereas dietary glucose did not, but dietary glucose did increase serum glucose and insulin concentrations in the postprandial state whereas dietary fructose did not. When fructose at 30-60 g ( approximately 4-12% of energy) was added to the diet in the free-living state, there were no significant effects on lipid or glucose biomarkers. Sucrose and high-fructose corn syrup (HFCS) contain approximately equal amounts of fructose and glucose and no metabolic differences between them have been noted. Controlled feeding studies at more physiologic dietary intakes of fructose and glucose need to be conducted. In our view, to decrease the current high prevalence of obesity, dyslipidemia, insulin resistance, and diabetes, the focus should be on restricting the intake of excess energy, sucrose, HFCS, and animal and trans fats and increasing exercise and the intake of vegetables, vegetable oils, fish, fruit, whole grains, and fiber.

Effects of eicosapentaenoic acid and docosahexaenoic acid on cardiovascular disease risk factors: a randomized clinical trial

BACKGROUND Eicosapentaenoic acid (EPA) and docosahexaenoic acid (DHA), the primary omega-3 fatty acids in fish oil, have been shown to reduce cardiovascular disease (CVD) risk. OBJECTIVE This study aimed to examine the independent effects of EPA and DHA on lipid and apolipoprotein levels, as well as on inflammatory biomarkers of CVD risk, using doses often used in the general population. DESIGN A blinded, randomized 6-week trial was performed in 121 healthy, normolipidemic subjects who received olive oil placebo 6g/d, EPA 600mg/d, EPA 1800mg/d, or DHA 600mg/d. The EPA was derived from genetically modified yeast. RESULTS The subjects tolerated the supplements well with no safety issues; and the expected treatment-specific increases in plasma EPA and DHA levels were observed. Compared to placebo, the DHA group had significant decreases in postprandial triglyceride (TG) concentrations (-20%, -52.2mg/dL, P=0.03), significant increases in fasting and postprandial low-density lipoprotein cholesterol (LDL-C) (+18.4%, 17.1mg/dL, P=0.001), with no significant changes in inflammatory biomarkers. No significant effects were observed in the EPA 600mg/d group. The high-dose EPA group had significant decreases in lipoprotein-associated phospholipase A2 concentrations (Lp-PLA2) (-14.1%, -21.4ng/mL, P=0.003). CONCLUSIONS The beneficial effects of EPA 1800mg/d on CVD risk reduction may relate in part to the lowering of Lp-PLA2 without adversely affecting LDL-C. In contrast, DHA decreased postprandial TG, but raised LDL-C. Our observations indicate that these dietary fatty acids have divergent effects on cardiovascular risk markers.