Promoting Weight Loss in Type II Diabetes SHARON A. BROWN, PHD, RN SANDRA UPCHURCH, PUD, RN, CDE ROBERTA ANDING, MS, RD, CDE
MARY WINTER, MSN, RN GILBERT RAMIREZ, DRPH
OBJECTIVE — To examine strategies—behavioral therapies, exercise, diet, anorectic drugs, surgery, or a combination of strategies—used for promoting weight loss in people with type 11 diabetes. RESEARCH DESIGN A N D METHODS— Meta-analysis was used to synthesize research of promoting weight loss in this population. Literature search strategies involved reviewing bibliographies, conducting computer searches and surveys of relevant master's degree programs, and contacting representatives of the Centers for Disease Control. The final sample consisted ol 89 studies involving 1,800 subjects. Data were extracted on 80 variables characterizing the sample of studies/subjects and on 23 outcome variables, including weight, metabolic control, lipids, and other physiological parameters. RESULTS — Diet alone had the largest statistically significant impact on weight loss (—201b) and metabolic control (—2.7% in glycosylated hemoglobin). All diets significantly improved lasting blood sugar. Behavioral programs alone had a statistically significant impact on weight loss (—6.4 lb) and metabolic control (—1.5%) but effects were less than for diet alone. Data from tin- few exercise studies indicated that weighted average effects for exercise on weight loss (—3.4 11)1 and metabolic control (—0.8%) were less than diet alone. Behavioral therapy plus diet plus exercise was associated with statistically significant effect size estimates for weight loss (—8.5 lb) and metabolic control (—1.6%). Diet alone achieved belter results. Effects of weight promotion strategies, in general, were smaller in experimental studies and for individuals over age 55. CONCLUSIONS— Dietary strategies are most effective for promoting short-term weight loss in type II diabetes. A number of gaps exist in the extant literature—descriptions of subjects, interventions, or longitudinal outcomes beyond 12 months after intervention.
O
f the people with type II diabetes, 80-90% are obese (1). Promoting weight loss in people with type II diabetes is one of the most important and dillicu.lt goals to accomplish in patient management. Achievement and maintenance of ideal body weight improves metabolic control and, in some cases, eliminates the need for insulin or other types of serum glucose-lowering medications. Weight reduction enhances insulin sensitivity; improves glucose control; and reduces hypertension, hypertriglyceridemia, and hypercholesterolemia (2,3,52). However, Wing etal. (100) demonstrated that type 11 diabetic subjects experienced more difficulties losing weight than did
their overweight nondiabetic spouses. And, as in nondiabetic people, any significant weight loss was usually followed by weight regain within 1-2 years. Failure to achieve and maintain weight loss may not be indicative of noncompliance but may be more reflective of the difficulties of losing weight when coupled with the altered metabolism of diabetes. Physiological mechanisms that mediate relationships between obesity, diabetes, and metabolic control of diabetes remain elusive. Obesity is a complex entity, even more so when associated with diabetes. Hereditary, metabolic, environmental, and behavioral factors appear to influence expression of obesity and ability
1 loin the School of Nursing (S.A.B.), the University of Texas at Austin, Austin; the School of Nursing (S.U., R.A., M.W.), the University of Texas-Houston Health Science Center, Houston; and the San Antonio Coi-hv.uu' Center (G.R.), Audie I.. Murphy Memorial Veterans Hospital, San Antonio, Texas. Address correspondence and reprint requests to Sharon A. Brown, PhD, RN, the University of Texas at Aiwin, 1700 Red River, Austin, TX 78701. E-mail:
[email protected]. Received for publication 5 June 1995 and accepted in revised form 25 January 1996. ADA, American Dietetic Association; PSMF, protein-sparing modified fast; VLCD, very-low-calorie diet.
D i u u n s CARE, VOLUME 19, NUMBER 6, JUNE 1996
to achieve and maintain ideal body weight. For example, upper body fat (central or android obesity) poses a greater risk for developing diabetes than tat stored in hips and thighs (lower body or gynecoid obesity) (4,5). The phenomenon of repeated cycles of weight loss and regain has been studied in animal models, providing early evidence that each weight loss attempt is slower and regain is faster (6). Weight loss appears to reduce ?A-\\ energy expenditure, but it is not yet dear whether this is responsible for the regain of weight (7,52). IIWeight reduction lowers metabolism, can metabolism be adjusted, through exercise for example, to avoid weight regain? Are behavioral therapies commonly used to treat obesity in type II diabetes (patient education programs, counseling, behavior modification) effective in promoting and maintain ing weight loss? While many studies oi strategies to promote weight loss in type 11 diabetes have been conducted, no critical evaluations or syntheses of this literature have been reported. Wing and Jeffrey (8), however, conducted a synthesis of literature on outpatient treatments for obesity in nondiabetic individuals (145 studies, 1966-1977). Weight loss-promotion methods of behavior therapy, exercise, diet, and anorectic medications were compared. Anorectic medications produced the greatest weight loss; however, behavioral therapy was found to produce the best maintenance of weight losses. Although the study of Wing and Jeffrey was conducted prior to recent refinements in meta-analytic procedures, important data were generated on the most effective strategies for promoting weight loss in otherwise healthy obese individuals. The study reported here addresses the same intervention strategies as in the study by Wing and Jeffrey. Strategies included diet, behavioral therapies (patient education, counseling, and behavior modification), exercise, anorectic drugs, surgery, or a combination of any of these strategies. In addition, studies of surgical strategies for the treatment of obesity in people with type II diabetes were located early in the literature search process; consequently, this intervention was added to the systematic search process and was included in
613
Weight loss in type II diabetes
the analyses. Other variables of interest included the subjects' ages and study variables of research design, research quality, and date of research report. Recently refined meta-analytic procedures were applied to a literature base that included both published and unpublished data on weight loss in type II diabetes. Few decisions can be made on the basis of one study, and frequently, multiple studies produce conflicting results. Consequently, the need to synthesize data and to explain conflicting results warrant application of the meta-analytic approach to the important area of weight loss in diabetes. Understanding effects of weight losspromotion strategies is essential for improving diabetes care and for preventing the long-term debilitating consequences of the disease.
RESEARCH DESIGN AND METHODS— This study used metaanalytic procedures of Glass et al. (9) and Hedges and Olkin (10). Quality of studies was analyzed as well; Cooper (11) and Glass et al. (9) recommended that conclusions about the effect of quality of a study on outcomes should be drawn a posteriori, that is, research design aspects should be coded so one can determine if outcomes are related to how studies were conducted. Locating the sample Literature search methods. A rigorous literature search was undertaken to locate published and unpublished data related to the promotion of weight loss in obese individuals with type II diabetes. Bibliographies of each review article and research study relating to interventions used to promote weight loss were examined. A computer was used to search Medline (1966-1994), Combined Health Information Data Base (1978-1994), Psychological Abstracts (1967-1994), ERIC (1966-1994), and Dissertation Abstracts (1961-1994) under weight loss limited to diabetes, relating this to behavioral strategies (patient education, counseling, and behavior modification), diet, exercise, anorectic drugs, and surgery. To identify other unpublished studies, master's degree programs in nursing, public health, and dietetics/nutrition were surveyed for theses potentially relevant to this study. Contact was made with representatives of the Centers for Disease Control, which supports diabetes research centers across 614
the U.S., to locate other researchers investigating weight loss. Inclusion criteria. Inclusion criteria developed for selection of the sample were that each included study must have involved the following: 1) a sample of obese adults with type II diabetes; 2) a behavioral, dietary, exercise, anorectic drug, surgical, or combined strategy to promote weight loss; 3) a measure of weight loss as an outcome of the intervention; 4) the same setting for both treatment and control groups; 5) an ex post facto, one-group pretest-posttest, quasiexperimental, or experimental design; and 6) data that permitted calculation of effect sizes. The literature search located 912 documents on strategies to promote weight loss in diabetes; 395 were eliminated either because the document was not a research report or subjects were not clearly identified as overweight/obese individuals with type II diabetes. An additional 404 did not meet criteria of relevance and acceptability for inclusion, and 24 were duplicative, i.e., the same data were reported in both reports. The investigator and two consultants, both of whom were certified diabetes educators, independently applied inclusion criteria. Agreement of three raters independently applying inclusion criteria to a random subset of 10 studies was 92.5%. Disagreements in applying the criteria were reviewed and discussed among the coders until consensus was reached. The final sample consisted of 89 studies (18-105) that involved a total of 1,800 subjects. (Note that Wing et al. [98] reported the results of two separate intervention studies that were relevant to this meta-analysis.) Coding procedures Once studies were selected for analysis, information was extracted and features of each study were coded on forms developed for this purpose. A detailed code book defined the coding process for each of the 80 variables that characterized the subjects/studies and for 23 outcomes. Studies were coded for descriptive data (such as source of the study and the date of publication), methodology (such as sample size and type of group assignment), research quality, and substantive features (characteristics of subjects, such as mean age or ethnicity, or of interventions, such as the use of group versus individualized approaches). Six quality criteria were applied: 1) type of study
design; 2) description of sample selection (e.g., random versus convenience); 3) specification of illness or condition with regard to the use of replicable diagnostic criteria; 4) completeness of the intervention description; 5) clarity of definitions of the outcome constructs; and 6) directness and longitudinal nature of outcome measures. (Procedures for applying quality criteria were adapted from similar previously reported criteria developed by S.A.B. [12]). Outcome variables included indicators of the following: 1) weight loss (e.g., body weight, BMI, and percentage ideal body weight); 2) metabolic control (e.g., glycosylated hemoglobin levels); 3) lipid metabolism (e.g., triglycerides); and 4) other physiological parameters (e.g., blood pressure). Selection of coding variables was determined by a number of factors: 1) suggestions by published experts on meta-analysis (9,10); 2) previous meta-analysis experience of the investigators; and 3) a preliminary review of the sample of studies by S.A.B. and two consultants before code sheet development. There is disagreement regarding the best index of interrater reliability for estimating coding decisions in research syntheses. Because percent agreement has been the most widely used index in previous research syntheses (13), intercoder agreement percentage rates were used in this study. Intercoder agreement (94.5%) of the code sheet was determined by four coders coding five randomly selected studies. The coding instrument was modified twice to improve reliability. However, all studies ultimately were coded independently by at least two coders to ensure the accuracy of the coding process. Any disagreements in coding were discussed until differences were resolved and 100% agreement was achieved. Finally, several studies included multiple experimental and/or comparison groups within the same study. To maintain independence of the data for analysis, decision criteria (Table 1) were identified so that the most relevant data would be extracted. These decision criteria were adapted from criteria used in other reported meta-analyses (14,15). The criteria are designed to maximize the number of experimental studies that are included in each outcome analysis.
Data analyses To ensure accuracy, effect sizes were computed by two individuals using two difDIABETES CARE, VOLUME 19,
NUMBER 6, JUNI:
1996
Brown and Associates
Table 1—Decision criteria 1. The experimental group was the method being tested; the control group was always the placebo group or the group that received routine care. 2. If a study compared two types of the same strategy (e.g., an ADA diet versus a VLCD), the more intensive intervention was the experimental group, in this case, the VLCD. Also, outcome data were coded for each group as one-group pretest-posttest designed study so that comparisons within strategy groups could be made, e.g., to determine which types of diets were most effective in producing weight loss. 3. If a study compared two types of the same strategy (e.g., behavior modification and contracting) and no intensive intervention was identifiable, outcome data were coded for each group as a one-group pretest-posttest designed study. 4. If a study compared two different types of strategies (e.g., diet versus exercise), each strategy was treated as a one-group pretest-posttest designed study. ferent types of computer software; one person used SPSS and one person used DSTAT (16), a statistical computer program specifically designed for metaanalysis. Effect sizes obtained from the two statistical programs were compared and determined to be identical (17). Homogeneity analyses, when appropriate, were conducted using DSTAT. Significance testing was based on the hypothesis that the obtained weighted-effect size was significantly different from zero. For all analyses, statistical significance was set at a 2% on pretest data, effect sizes were calculated from pretest-posttest change scores for each group. When this procedure was used, the pooled posttest standard deviation was used to calculate the relevant weighted-effect size estimate (L.V. Hedges, personal communication). Partitioning of effect sizes. Effect sizes were partitioned according to the major outcome variables of this study, i.e., weight loss, glycosylated hemoglobin levels, etc. Next, effect sizes were calculated for each outcome under each specific strategy. Homogeneity analyses were performed, one on each outcome by each intervention type. Using this procedure, outlier studies, in most instances one to two studies, were removed one at a time until homogeneity was achieved, and then weighted-effect size estimates were calculated for each intervention. Further analyses were conducted on groups within each major intervention variable. For example, weighted mean effect sizes were determined for groups within the diet strategy (American Dietetic Association [ADA] diet versus the very-low-calorie diet [VLCD]). Additional similar analyses were planned for other major strategies, such as behavioral therapy and exercise. However, partitioning of the effect sizes for these interventions resulted in too few studies per cell, which precluded further testing. To address the effects of various subject/study variables by weight loss strategy, weighted-effect size estimates were pooled for those variables for which primary authors provided sufficient data. Five such variables were believed to warrant such analyses: subjects' ages, research design of the study, publication date, quality of the primary study, and
length of time after the intervention that measurements were made. The mean age of subjects, as reported by authors of primary studies, was divided into patients 1 2 months. Weighted effect size estimates were calculated for each time period for each outcome variable grouping, both for the overall effect (heterogeneity assumed) and for the effect of each individual strategy (homogeneity analyses conducted), when possible. Again, because breakdown of variables resulted in too few subject/studies per cell for some outcomes, only those variables with more than three studies contributing data are reported. (Analyses originally planned on other subject/study variables were precluded by lack of primary data [i.e., degree of obesity, gender, ethnicity, socioeconomic status, type of diabetes treatment, and compliance measures]). Also, only a few unpublished studies were located, precluding comparisons of outcomes reported in published versus unpublished research reports. (A more detailed description of studies included in these analyses, the code book, and the list of 80 variables and 23 outcomes can be obtained from S.A.B. upon request.) RESULTS Characteristics of the sample of studies/subjects Of the 89 studies selected for the sample, the majority (92%) was published since 1980 (range, 1965-1994), reflecting an 615
Weight loss in type II diabetes
tion or reported a 2- to 3-lb greater weight loss in the control group. All interven-60 tions produced a statistically significant reduction in mean body weight, except -50 for anorectic drugs and behavioral therapies plus exercise. Behavioral therapies Mean Body " Glycosylated alone and exercise alone produced the Weight Loss Hemoglobin (lbs.) smallest changes in mean body weight, (%) — 6.4 and —3.4 lb respectively, although standard 7 the changes were statistically significant. deviation I -20 value H For the major metabolic control variable, allied glycosylated hemoglobin, the largest decreases were calculated for diet alone and for the combination strategy of behavioral therapies plus diet plus exercise. Diet Behavior Exercise Anorectic Surgery Behavior Behavior Behavior Other important physiological + + + Only Only Only Drugs variables were measured in these studies Diet Exercise Diet as indicators of improved health and, Exercise therefore, were incorporated into this metaFigure 1—Major outcomes of intciycntion strategics. analysis. Table 2 shows the weighted average effect size estimates resulting from hoincreased interest in weight loss promo- years of age (range, 29-71) and weighed mogeneity testing for each of these tion in type II diabetes in recent years. The 211 lb (range, 148-314) or 147% of their remaining outcomes. The data, however, majority of the studies were published ideal body weight (range, 122-204%). were not as consistently reported in the journal articles (90%); a nurse was the The overall sample was fairly evenly dis- primary studies as were data on body first author in 6.7% (n = 6) of them and a tributed between women (51%) and men weight and glycosylated hemoglobin levdietitian was the first author in only one (49%). Out of a possible score of 21 qual- els. In fact, for some strategies, results are study. Of the interventions, 55% were ity points, the average total quality score based on only three or four studies and conducted in outpatient settings or on was 10, ranging from 5 to 17. Primary should be interpreted with caution. Howmetabolic/research units. Approximately reasons that studies received low quality ever, all results are presented for com20% of the studies involved incomplete scores were as follows: 1) lack of random- pleteness and to demonstrate those areas descriptions of the interventions that were ization to treatment groups; 2) lack of de- needing further investigation. These findstudied. Of the studies, 40% involved di- scription of the intervention, particularly ings are presented in effect size estimates etary interventions; 20% involved behav- of the behavioral strategies; and 3) lack of to compare the relative responsiveness of ioral interventions; 10% were exercise in- direct longitudinal measures of outcome. the various outcome indicators with the terventions; and the remaining 30% were Overall attrition rates were able to be de- weight loss-promoting interventions. In distributed among drug, surgery, or com- termined from 88 studies; the mean at- Table 2, data indicate the standard deviabination strategies. Dietary interventions trition rate was 7.8%. Only 4 of the 89 tions and the number of studies that conprimarily involved investigations of ADA studies reported that a double-blind tributed to the individual effect size estimates. reduced-calorie, very-low-calorie, or pro- procedure was used. tein-sparing modified fast diets. Both With regard to BMI, all single group and individual dietary instructions Major outcomes of intervention strategies (diet, behavioral therapies, and were used equally in these studies. Stud- strategies exercise) produced a sizable and signifiies investigating behavioral strategies in- Figure 1 shows the weighted average cant effect, as did the combination varivolved information giving (41%), behav- postintervention changes in the major able of behavior plus diet plus exercise. ior modification (24%), or other types of outcome variables of this investigation, For fasting blood sugar levels, as well as similar interventions. On average, behav- body weight and glycosylated hemoglo- for the 2-h postprandial blood sugar, ioral interventions involved a series of 10 bin levels, for each strategy when compar- again dietary strategies alone produced sessions (range, 1-25), each session meet- ing individuals in the experimental con- the largest effect sizes. The only exception ing for ~78 min (range, 22 min to 2 h); dition with those in the controlled was the effect size for surgical approaches 60% of the studies involved group ap- condition. used in the treatment of obesity. Serum proaches. Exercise interventions primarExcept for surgery, the largest cholesterol (total as well as HDL and LDL) ily involved aerobic approaches, such as change in body weight was attained was measured primarily as a result of dibicycling, jogging, walking, calisthenics, through dietary strategies alone: ~20 lb etary interventions. Only the total cholesetc. On average, subjects exercised four reduction in weight in the treatment terol level was associated with a statistitimes per week for 50 min each time. Few groups compared with the control cally significant effect, and that result was studies provided any information relative groups. Of the 89 studies included in an outcome of diet alone strategies. Trito exercise intensity, such as distance (n = these analyses, 9 studies either reported glyceride levels were associated with sta4) or calories expended (n = 2). no weight differences between experi- tistically significant effects as a result of On average, subjects were 52 mental and control groups after interven- any intervention that included a dietary -70
40
616
DIABETES CARE, VOLUME 1 9 , NUMBER 6, JUNE. 1 9 9 6
Brown and Associates
Table 2—Effect
sizes for major intervention
Outcomes
Behavior only
Diet only
0.71 ±0.53 (7)* BMI 0.63 ± 0.26 (9)* Serum insulin 0.70 ± 0.42 (16)* lasting blood sugar 1.76 ± 0.59 (27)* 2-h postprandial blood 1.35 ± 0.31 sugar (7)* Cholesterol 0.62* ± 0.43 (15) HDL 0.17 ±0.37 (9) I.OI. 0.12 ±0.30 (4) Triglycerides 0.56 ± 0.40 (16)* Systolic blood pressure 0.79 ± 0.16 (4)* Diastolic blood pressure 0.72 ± 0.30 Ideal body weight
strategies
Exercise only
Anorectic drugs
Surgery
Behavior plus exercise
Behavior plus diet plus exercise
—
—
0.39 ± 0.27
—
—
0.95 ± O.bl
—
—
0.48 ± 0.33 (5V:;
—
—
±0.33
—
0.11 ± 0 . 3 0
0.12 ±0.46 (3)
—
0.00 ± 0.21
Behavior plus diet
0.18 ±0.31 _ _ _ (7) 0.57 ± 0.08 0.46 ± 0.41 — — (4) (3) — 0.27 ±0.19 — — (6) 0.31 ± 0.29 0.34 ± 0.23 0.44 ± 0.66 3.01 ± 0.89 (11)* (6)* (3)* (3)* — — — — 0.09 ± 0.16 0.06 ± 0.36 (4) (6) — 0.02±0.57 (5) _ _ 0.12 ± 0.05 0.20 ± 0.49 (3) (5) 0.56 ± 0.08 -0.12 ± 0.09 (3)* (3) — 0.00 ± 0.50
(4r
—
—
_
0.25
(4) —
— _
_
_
—
0.41 ± 0.41
—
_
—
0.31 ± 0.10
—
0.72 :t 0.7(-» (3)* .
(3)* —
— _
_
_
_
_
(3)
Data are means ± SD (number of studies contributing data) and are weighted-effect size estimates. Effect sizes reflect the weighted-effect sine estimate on a homogeneous set of studies. There were not enough studies of diet plus exercise to allow analyses of this combination strategy. °P
