Interpreting weight losses from lifestyle modification trials: using categorical data

Abstract

Although studies in obese subjects using weight loss medications typically report mean and categorical weight loss, results from diet and exercise intervention trials typically only report mean weight change from baseline along with a level of significance. These data alone do not give clinicians or administrators the data needed to determine the probability that an individual will achieve clinically relevant weight loss. Thus, it is difficult to decide which patients, employees or health plan enrollee would benefit from the type and level of support used in a clinical trial. Our goal was to assess what fraction of subjects enrolled in lifestyle modification interventions achieved clinically significant weight loss. Thus, we requested categorical weight loss data from several investigators who had published results from studies involving either a high- or low-intensity lifestyle modification intervention arm. These categorical data indicate that a substantial fraction of subjects in each lifestyle modification intervention achieved clinically meaningful weight loss, even when the average weight loss is modest.

Overweight and obesity are significant causes of preventable deaths, morbidity, adverse health conditions, as well as use of health-care resources and cost.¹ Evidence reviews,² meta-analyses^3-8 and cost-effectiveness analyses^9,10 of weight management interventions are available to provide guidance to clinicians, employers, health-care managers, and benefit designers creating programs and providing supports to help people achieve and maintain a healthy weight.

The Food and Drug Administration draft guidance for studies of weight loss medications encourages the reporting of not only mean weight loss data, but also the reporting of categorical weight loss efficacy benchmarks.¹¹ However, most often results from diet and or exercise intervention trials with a weight loss outcome report mean change from baseline with a confidence interval and P-value. These data alone do not give clinicians or administrators the data needed to determine the probability that an individual will achieve a clinically relevant weight loss. Thus, it is difficult to decide which patients, employees, or health plan enrollees would benefit from the type and level of support used in a clinical trial.

To gain insight into the potential value of using categorical weight loss data in assessing various lifestyle modification interventions, we requested categorical weight loss data from several investigators who had published results from studies conducted in the United States with one or more lifestyle modification intervention arm. We did not examine data from either pharmacotherapy or surgical trials. We selected studies that reported 12-month weight loss outcome data. We chose 12 months rather than 6 months, 18 months or 24 months because many studies have shown an erosion of impact after 6 months; and we wanted a time point where we could get a reasonable number of studies to include. Although we did not conduct a systematic review, we requested data from all studies with one or more lifestyle modification intervention arms that were included in recent evidence reviews and meta-analyses.^2-10

Some studies included categorical data in their published results. Among studies that did not include categorical results, 65% of investigators from whom we requested additional data provided us with this information. Investigators were contacted multiple times each.

Categories were segmented as follows: study participants who, at 12 months had maintained a loss of ≽10, 5–9.9 or 0.1–4.9% of their baseline weight. Several reports have concluded that weight loss in the 5–9% range contributes to important health benefits and this is the efficacy bench-mark used by the Food and Drug Administration in evaluating weight loss medications.^3,10,12 Compelling evidence from bariatric surgery studies and other evidence reviews indicate that weight loss ≽10% of baseline weight produces greater health benefits, with resolution of diabetes in some patients.^2,13 Although weight loss of <5% has not been considered clinically significant, a recent analysis of Diabetes Prevention Program lifestyle intervention data showed that weight loss was the dominant determinant of the reduced risk of diabetes. For every kilogram lost, there was a 16% reduction in risk. This effect began with losses as low as 2% of body weight.¹⁴

Using intention-to-treat reported data, results in Table 1 are segmented by two types of studies, split at the median hours of treatment received by participants. ‘Intense lifestyle modification interventions’ were defined as those in which participants were asked to attend sessions where the intervention was delivered for an average of 37 h (range 13–52 h); whereas ‘less intense lifestyle modification interventions’ were those where sessions required participants to attend a average of 5 h (range 1–9 h). Among participants in intense interventions, 28% of individuals lost ≽10% of baseline weight, 26% lost 5–9.9% and 38% lost 0.1–4.9%. In less intense interventions, 13% of participants lost ≽10% or more, 16% lost 5–9.9% and 27% lost 0.1–4.9% of body weight. Although data were not available from all studies, an average of 26.3% or participants in intense interventions and 43.6% in low/less intense interventions lost no weight or gained weight.

Table 1.

The percentage of participants who fit each weight change range—baseline to 12 months (sorted by hours of intervention)

	Number of completers in group	Mean weight change at 12 months (%)	≽ 10% loss (%)	5–9.9% loss (%)	0.01–4.9% loss (%)	No loss/ weight gain (%)	Estimated total hours of intervention during year 1	Data source
Less intense interventions (x̄ of 5 h)
Self-help	159	−1.5	11.0	9.0	33.0	47.0	0.7	Heshka. JAMA 2003; 289: 1792–1798.
eDiets	88	−2.8		29.0			2.0	Gold BC. Obesity 2007; 15: 155–164.
Type 2 diabetes	141	−0.9	2.1	19.9	31.4	46.6	2.8	Christian JG. Arch Intern Med 2008; 158: 141–146.
Atkins	21	−2.1	7.5	15.0	12.5	65.0	4.0	Dansinger ML. JAMA 2005; 293: 43–53.
Ornish	20	−3.2	10.0	10.0	20.0	60.0	4.0	Dansinger ML. JAMA 2005; 293: 43–53.
Weight watchers	26	−3.1	10.0	15.0	27.5	47.5	4.0	Dansinger ML. JAMA 2005; 293: 43–53.
Zone	26	−3.2	15.0	12.5	22.5	50.0	4.0	Dansinger ML. JAMA 2005; 293: 43–53.
Diabetes support  and education	2463	−0.7	3.5	6.0			7.7	Look AHEAD. Diabetes Care 2007; 30: 1374–1383.
Atkins	68	−5.5	32.4	17.6	22.1	27.9	9.0	Gardner CD. JAMA 2007; 297: 969–977.
Zone	61	−1.9	11.4	21.3	32.8	34.5	9.0	Gardner CD. JAMA 2007; 297: 969–977.
LEARN	61	−3.1	25.0	13.3	40.0	21.7	9.0	Gardner CD. JAMA 2007; 297: 969–977.
Ornish	59	−2.6	13.6	20.3	30.5	35.6	9.0	Gardner CD. JAMA 2007; 297: 969–977.
Mean		−2.5	12.9	15.7	27.2	43.6	5.4
Intense interventions (x̄ of 37h)
Combined energy  restricted groups	147	−6.3	22.5	22.5			12.7	Rolls BJ. Obesity Res 2005; 13: 1052–1060.
RF + RF and FV	71	−5.8	23.7	21.6			21.3	Ello-Martin JA. Am J Clin Nutr 2007; 85: 1465–1477.
Weight loss group	547	−3.3	13.8	19.0	37.6	29.7	26.0	Stevens VJ. Ann Intern Med 2001; 134: 1–11.
Time-calorie  replacement	147	−6.1	36.6	31.9			35.5	Fitzwater SL. J Am Diet Assoc 1991; 91: 421–426.
Vtrim	62	−5.5		42.0			38.0	Gold, BC. Obesity 2007; 15: 155–164.
Intense lifestyle  intervention	2496	−8.6	38.0	30.0			46.7	Look AHEAD. Diabetes Care 2007; 30: 1374–1383.
Weight watchers	148	−4.8	21.0	17.0	39.0	23.0	52.0	Heshka. JAMA 2003; 289: 1792–1798.
Low-calorie diet	41	−9.7	22.0				52.0	Wing RR. Am J Med 1994; 97: 354–362.
Very low-calorie diet	38	−13.4	47.0				52.0	Wing RR. Am J Med 1994; 97: 354–362.
Mean		−7.1	28.1	26.3	38.3	26.3	37.3

Abbreviations: RF, reduced fat; RF and FV, reduced fat and increased intake of water-rich foods.

By analyzing outcomes using categorical criteria, weight loss impacts can be explored further by a specific type of intervention (for example, physician supported, utilization of meal replacements) or by other factors such as intensity of treatment, percentage of individuals who completed the intervention or baseline subject characteristics (for example, weight loss readiness, gender, ethnicity, socioeconomic or health insurance status). Although intention-to-treat analyses are most useful for payors, clinicians may find greater value in seeing categorical outcome data for subjects who complete the intervention. Associations of success in weight loss (measured by categorical outcomes) could be correlated with these socio-demographic, clinical and treatment factors.

The limitations of the current analysis include the limited number of studies examined, the heterogeneity of the interventions and the lack of data for some cells. However, these categorical data indicate that a substantial fraction of subjects in each lifestyle modification intervention achieve clinically meaningful weight loss, even when the average weight loss is modest. As would be expected, more intensive interventions resulted in a higher percentage of subjects achieving 5 or 10% weight losses. In a staged intervention strategy,¹⁵ many individuals experience initial success with less intense lifestyle interventions. As has been seen in studies of weight loss medications,¹⁶ some individuals will respond to a given intervention, others will not. However, the type of intervention and probability of a clinically important response can be used for planning purposes. Thus, we propose that investigators reporting the results of lifestyle weight loss interventions, in addition to reporting mean weight change, should always report categorical weight loss outcomes.