There may be subgroups of patients with type 2 diabetes who experience marked health improvements in response to intensive weight-loss interventions, while others may experience substantially worse outcomes, the results of a reanalysis of data from a randomized controlled trial indicate.
The original Action for Health in Diabetes (Look AHEAD) trial of a weight-loss intervention in individuals with type 2 diabetes was stopped early in 2012 by the National Institutes of Health after finding no difference in a composite of cardiovascular events between the groups. Results were first fully reported and published in 2013.
However, the current researchers used machine learning to conduct an analysis of data on almost 5000 participants from the trial, finding that approximately 85% of patients with a particular combination of baseline plasma glucose levels and self-reported general health had a clinically meaningful response to weight loss.
In contrast, the remaining 15% of patients with well-controlled or mild diabetes at baseline but a negative perception of their health status experienced a significant increase in cardiovascular events during follow-up.
Study author Ronald Tamler, MD, medical director at the Mount Sinai Clinical Diabetes Institute, New York, commented in a press release: “This analysis restores my faith in basic common sense.
“For the vast majority of people with diabetes, a healthy lifestyle with weight loss carries significant benefits; however, it’s not for everyone. Thanks to this work, clinicians can infer which patients will benefit the most from such a lifestyle intervention.”
Study coauthor Aaron Baum, PhD, from Arnhold Institute for Global Health, Icahn School of Medicine at Mount Sinai, New York, added: “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 who could benefit from an intervention is fundamental to patient care.”
However, the clinical meaning of the findings, which were published online in the Lancet Diabetes & Endocrinology on July 12, has been questioned by some experts, who describe the results as “puzzling” in an accompanying comment.
Digging Deeper: Reanalyzing the Analysis
Look AHEAD was a randomized, controlled, open-label trial that involved 5145 patients at 16 sites in the United States aged 45 to 75 years with a history of type 2 diabetes and overweight and obesity, who were enrolled between 2001 and 2004.
Of those, 2570 were assigned to an intensive weight-loss intervention involving healthy eating and increased physical activity, and 2575 were assigned to a diabetes support and education control group.
Hypothesizing that the overall neutral treatment outcome masked heterogeneous treatment effects from the intensive weight-loss intervention, the researchers decided to dig deeper and conduct a post hoc analysis on 4901 trial participants included in the National Institute of Diabetes and Digestive and Kidney Diseases Repository.
They performed causal forest modeling on a randomly selected training set of 2450 participants, constructing causal decision trees using 84 baseline predictors. These were then tested in the remaining 2451 participants.
Over a mean follow-up of 8.5 years, 199 (16.2%) of 1231 patients in the training set from the intervention group experienced the primary outcome, defined as a composite of death from cardiovascular causes, nonfatal myocardial infarction, nonfatal stroke, or hospitalization for angina. The primary outcome also occurred in 186 (15.2%) of 1219 control patients in the training set.
The new analysis revealed that baseline HbA1c levels and self-reported general health on the Short Form Health Survey (SF-36) general health score were able to distinguish participants with a high vs low benefit from the weight-loss intervention.
Specifically, Cox models revealed that, among 2015 (83.7%) participants in the training set and 2101 (85.7%) in the testing set with an HbA1c ≥6.8% or HbA1c <6.8% and an SF-36 general health score of ≥48, there was an absolute risk reduction of the primary outcome of 3.46% (P = .038), at a number needed to treat of 28.9 to prevent one event over 9.6 years.
In contrast, among 400 (16.3%) of 2450 participants in the training set and 350 (14.3%) participants in the testing set with an HbA1c <6.8% and an SF-36 general health score <48, there was an absolute risk increase in the primary outcome of 7.41% (P = .033).
These latter subgroups reported significantly fewer minutes of exercise during both the first 6 months and last 6 months of the intervention than those in the first subgroup, at -495 minutes and -924 minutes, respectively (P < .0009 for both).
Among participants with an HbA1c ≥6.8% or HbA1c <6.8% and an SF-36 general health score of ≥48, there was also evidence of greater improvements in HbA1c, self-reported mental health, and blood pressure related to the intervention compared with other patients.
Data-Driven Methodology Can Reveal Hidden Benefits, or Can it?
Discussing the study, the authors write that their data-driven methodology “can reveal otherwise-undiscovered and clinically meaningful relationships between interventions, outcomes, and subgroups and can complement expert-based preregistered subgroup hypotheses.”
While noting that there is a need for further, prospective examination of their findings, they add: “Identifying robust subgroup treatment effects can increase the quantity of clinically relevant findings generated by clinical trials and enable clinicians to better individualize patient care.”
But in an accompanying comment, Edward W Gregg, PhD, from the Centers for Disease Control and Prevention, Atlanta, Georgia, and Rena R Wing, PhD, from the Alpert Medical School of Brown University, Providence, Rhode Island, say the findings are “difficult to interpret.”
While they acknowledge that “speculatively, good baseline health status might prime participants for better adherence to the lifestyle intervention, and a more favorable improvement in health and lower baseline HbA1c might be a marker of less severe diabetes or better adherence to treatment regimens — the combination might therefore potentiate a better health outcome.”
They write, however, that this explanation is “unsatisfying” due to the lack of significant difference in compliance or metabolic risk factors between the subgroups.
They continue: “The finding of increased relative risk of the primary outcome among participants with low HbA1c and poor health status is intuitively puzzling, and the very low absolute hazard ratio in the trial control group in this ostensibly unhealthy subgroup raises the question of whether this observation is simply a chance finding.”
Nevertheless, Drs Gregg and Dr Wing note, “Despite the questions left unanswered by this analysis, novel and rigorous study of the heterogeneity in intervention response is important to guide personalized care, risk stratification, and translation of prevention studies to clinical and community settings” and will becoming increasingly relevant as lifespans increase.
The authors and editorialists report no relevant financial relationships.
Weight-Loss Benefit Only in T2D Subgroups