This article is more than 5 years old. Information may no longer be current.
Look AHEAD: Overall neutral treatment benefit masks better, worse outcomes in subgroups
Click Here to Manage Email Alerts
Among patients with well-controlled or poorly controlled type 2 diabetes and a good self-perception of their health, an intensive lifestyle intervention was associated with reduced risk for a cardiovascular event. Patients with well-controlled diabetes and a negative self-perception of health, however, saw an increase in cardiovascular risk with intensive weight loss, according to a post hoc analysis of the Look AHEAD trial.
In a new analysis of data from the study, conducted by data science researchers at the Arnhold Institute for Global Health at the Icahn School of Medicine at Mount Sinai, researchers found that the cohort’s overall neutral treatment findings, which led to the study being halted in 2012, masked treatment benefits among several subgroups.
“As with many trials that have reported negative or neutral average treatment effects, statistical commentators have been concerned that the average study result could mask important heterogeneous treatment effects, or systematically different outcomes among different types of
study subjects,” Aaron Baum, PhD, lead economist at the Arnhold Institute for Global Health and assistant professor at the Icahn School of Medicine at Mount Sinai, and colleagues wrote. “Traditional subgroup analyses will typically fail to identify such [heterogeneous treatment effects], because they are underpowered and are susceptible to estimation bias and multiple testing errors. Additionally, subgroup analyses generally only consider one factor at a time, rather than combinations of factors that are typically thought to generate [heterogeneous treatment effects].”
Baum and colleagues applied causal forest analysis, developed through machine learning, to fit Cox proportional hazard models to the Look AHEAD data, using a study sample that was included in the National Institute of Diabetes and Digestive and Kidney Diseases registry (n = 2,450; mean age, 59 years; 59% women; average baseline BMI, 36 kg/m²). Eligible patients, recruited between August 2001 and April 2004, were aged 45 to 75 years with type 2 diabetes and overweight or obesity, and were randomly assigned to intensive lifestyle intervention (n = 1,231) or a control group (n = 1,219; brief diet and exercise education sessions and social support). Primary outcome was defined as the Look AHEAD trial’s primary composite CV outcome, which was first occurrence of death from CV causes, nonfatal myocardial infarction, nonfatal stroke or hospitalization for angina, as well as three composite secondary CV outcomes. Researchers studied 84 baseline predictors from four categories (sociodemographic variables, medical history, laboratory values and behavior measures) to estimate heterogeneous treatment effects.
During a mean 8.5 years of follow-up, 199 patients in the intervention group (16.2%) and 186 patients from the control group (15.2%) experienced a primary outcome event.
According to researchers, the causal forest model revealed that two covariates — baseline HbA1c and general health self-reported on the 36-item Short Form Health Survey — were of primary importance in distinguishing individuals with high vs. low benefit from the intensive weight-loss intervention; self-reported mental health status was also a predictor (P < .0001).
The model then stratified trial participants by six subgroups; researchers analyzed two subgroups with covariates defined as primary: group 1, including patients with baseline HbA1c of 6.8% or less and baseline SF-36 general health score of 48 or less; and group 2, including patients with baseline HbA1c of 6.8% or less and baseline SF-36 general health score of 48 or more.
Using the testing data set, researchers found among that patients not in subgroup 1 — patients with well-controlled or poorly controlled diabetes and a SF-36 score of at least 48 — the number needed to treat to prevent one primary outcome event was 28.9 for 9.6 years, for an absolute risk reduction of 3.46% (95% CI, 0.21-6.73). In contrast, those patients in subgroup 1 experienced an absolute risk increase of 7.41% for a primary outcome event (95% CI, 0.6-14.22). Patients in subgroup 1 also reported fewer minutes of exercise in the first 6 months and last 6 months of the intervention year compared with those not in subgroup 1, the researchers noted.
"Our analysis demonstrates that recent advances in machine learning for causal inference can increase the quantity of clinically relevant findings generated from large randomized trials," Baum said in a press release. "As researchers and data scientists, we are always concerned that an overall study result could mask important disparities in benefit or harm among different types of patients, which is exactly what this study revealed. Being able to identify individuals that could benefit from an intervention is fundamental to patient care."
In commentary accompanying the study, Edward W. Gregg, PhD, and Rena Wing, PhD, of the division of diabetes translation at the CDC, said the study of heterogeneity in intervention response is important to guide personalized care.
“Variation in health behaviors remains an overwhelming determinant of chronic disease risk,” Gregg and Wing wrote. “As we enter an era characterized by increasing life spans but persistent high prevalence of chronic disease and multimorbidity, more diverse menus of lifestyle interventions will be necessary to improve population health.” – by Regina Schaffer
Disclosures: The authors report no relevant financial disclosures. Gregg and Wing report no relevant financial disclosures.
Click Here to Manage Email Alerts