But some patients in LOOK Ahead fared worse with intervention
by Kristen Monaco, Contributing Writer, MedPage Today July 17, 2017
Action Points
• Among a majority of patients with type 2 diabetes, intensive weight loss intervention reduced the risk of adverse cardiovascular outcomes.
• Note that 14% of the study participants reported significantly worse outcomes following such intensive lifestyle intervention.
Among a majority of patients with type 2 diabetes, intensive weight loss intervention reduced the risk of adverse cardiovascular outcomes, a post hoc analysis of the Look AHEAD (Action for Health in Diabetes) trial found.
In the study, led by Aaron Baum, PhD, lead economist at the Arnhold Institute for Global Health in New York City, and colleagues, 86% of the overall population (n=2,101 of 2,451) reduced the risk of composite cardiovascular outcomes via intensive lifestyle modification — 167 (16%) of 1,046 primary outcome events for intervention versus 205 (19%) of 1,055 for control; absolute risk reduction 3.46%, 95% CI 0.21-6.73%, P=0.038. This group included those with any HbA1c levels and self-reported good general health.
The findings were published online in The Lancet Diabetes & Endocrinology.
However, 14% of the study participants (n=350 of 2,451) reported significantly worse outcomes following such intensive lifestyle intervention (27 [16%] of 171 primary outcomes events for intervention versus 15 [8%] of 179 primary outcomes events for control; absolute risk increase 7.41%, 0.60-14.22, P=0.003). These patients were identified as having well-controlled diabetes at baseline (A1c <6.8%) and poor self-reported general health.
“We were surprised to find that weight loss interventions may not be beneficial for all patients,” Baum, who is also affiliated with the Icahn School of Medicine at Mount Sinai, told MedPage Today. “The intervention seems to lower the risk of cardiovascular events and mortality for the majority of patients, but it may have had a negative impact on a small subgroup patients, thus rendering the overall average effect neutral.”
The original analysis of the 2013 Look AHEAD trial, published in the New England Journal of Medicine, was stopped early after a median follow-up of nearly a decade due to the lack of significant findings between the primary outcomes of long-term cardiovascular disease morbidity and mortality with intensive weight loss intervention among people with type 2 diabetes.
The multicenter trial included 5,145 overweight or obese individuals with type 2 diabetes, who were randomized to undergo intensive lifestyle intervention, marked by diet and exercise to attain at least a 7% total weight loss, or a control group, which included diabetes education and support.
In the current analysis, Baum’s group used new machine learning techniques to re-address the original data by implementing a causal forest analysis, designed to identify heterogeneous treatment effects.
Baum said that after attending a talk on new machine learning techniques aimed at causal inference problems, Baum said, “it struck me that the newer [machine learning] methods would be well suited for increasing the quantity of knowledge we can learn from the clinical trial data we already have,” and that this new type of model was particularly well suited to analyze health disparities.
“From a technical point of view, the method was attractive, because it learns from half of the data — in this case, looking for combinations of characteristics of the patients who benefited the most from the weight loss intervention, and uses the rest of the data to test those hypotheses and look for complex interactions in the data, thus avoiding multiple hypothesis testing or p-hacking,” he added.
A total of 4,901 participants from the original trial were included and randomized into the training set (n=2,450) to identify the factors via the causal forest model, or to a testing set (n=2,451) used to validate the findings. Baum’s group employed the same primary outcome definition of composite cardiovascular (CV)-outcome: first occurrence of death from CV-causes, non-fatal myocardial infarction, non-fatal stroke, or hospitalization for angina.
The causal forest analysis divided the testing set into subgroups based upon baseline HbA1c measures (A1c<6.8%: well-controlled; ≥6.8%: moderately or poorly controlled), and general health self-reported with the Short Form-36 survey (score≥48: good health; <48: poor). Self-reported mental health was also assessed using the SF-36 with a mental component summary.
In an accompanying editorial, Edward W. Gregg, PhD, of the Centers for Disease Control and Prevention, and Rena Wing, PhD, of Alpert Medical School of Brown University in Providence, R.I., called the study’s method a “novel approach to assessing such heterogeneity,” noting, however, that the findings are difficult to interpret.
Gregg and Wing suggested that good self-reported health may be indicative of improved lifestyle intervention adherence, thus rendering improved outcomes, although “the lack of any significant difference between these subgroups in compliance or metabolic risk factors in the supplemental analyses leaves this explanation unsatisfying.” Instead, the editorial posed the idea that it may be “simply a chance finding” in regard to the perplexing 14% of those in the study who had worse CV-outcomes with intervention.
“Computer science methods for causal inference will continue to improve,” Baum added. “We are working on applying them to targeting hypertension management goals, balancing risk and reward for cancer screenings, and understanding the impacts of insurance expansions.”