In this long-term follow-up of participants in the Look AHEAD study, we observed three main findings related to the implications of achieving diabetes remission. First, although 11% of intervention participants achieved remission at year 1 of follow-up, the percentage of participants with remission had decreased to 4% by the 8th year of the study. Second, despite the relatively short-lived durations of most episodes of remission, we found that any achievement of remission was associated with 33% and 40% lower rates of CKD and CVD, respectively, compared with participants who did not achieve remission, and risk reduction was even greater (55% and 49%, respectively) among those who had evidence of at least 4 years of remission. Third, participants with a short duration of diabetes, low starting HbA1c and a large magnitude of weight loss were most likely to experience remission.

Remission from type 2 diabetes may be associated with lower rates of CKD, CVD and other long-term health outcomes through several pathways. First, the sustained reduction in HbA1c and insulin resistance may benefit the vascular endothelium, microcirculation and reduce atherosclerosis progression [21]. Second, remission benefits may occur through the diverse physiological effects of extensive weight loss, including reductions in hyperglycaemia, BP, insulin resistance, inflammation and hepatic fat levels [2, 22]. In these analyses, participants with any remission had a slightly greater net weight loss than those without remission (2%), but those with extended remission had an 8% greater weight loss. Previous post hoc cohort analyses of the Look AHEAD study found that participants who met the weight loss goal of 10% at 1 year had a 20% lower incidence of the primary CVD outcome despite the overall null finding of the Look AHEAD intervention [17, 23]. Third, the behavioural changes associated with greater weight loss and intervention adherence, including better dietary quality, increased physical activity and higher attained physical fitness, may drive further health benefits. Finally, the achievement of remission may be a marker for other unmeasured differences or advantages arising from care and risk factor management. The observational study design limits inferences about which of these mechanisms of action account for the long-term benefit.

Compared with our study, the DiRECT study and the DIADEM-I study (Diabetes Intervention Accentuating Diet and Enhancing Metabolism) found considerably greater remission and generally similar weight loss over the first 2 years [4, 5, 24]. The DiRECT study found remission rates of 46% and 36% among intervention participants at 1 and 2 years, respectively (compared with 4% and 3% of control participants), accompanying 9.5% and 5.3% net weight losses, respectively, after 1 and 2 years [5, 25]. The DIADEM-I study reported a net 1-year remission rate of 61% (compared with 12% of control participants) and a 6.1% net weight loss [24]. By comparison, the Look AHEAD study reported 12% and 10% remission rates years 1 and 2, respectively, and a 1-year net weight loss of 8%. However, the DiRECT and DIADEM-I studies differed from the Look AHEAD study in other ways. Both studies employed total diet replacement, with targets of approximately 800–850 kcal/day, and importantly, employed an active protocol-driven approach to medication removal. After 6 months, the DiRECT study re-introduced food [5] and provided advice to increase physical activity during the maintenance period [26]. Several other studies have observed similarly profound effects on HbA1c with dietary replacement and intensive weight loss over a shorter time period than the DiRECT and Look AHEAD Studies [27,28,29,30]. In contrast, the Look AHEAD study aimed for diets of 1200–800 kcal to achieve long-term weight loss, and passively relied upon primary care providers to remove medications as appropriate [19]. The Look AHEAD study was also distinct in terms of its relatively strong emphasis on physical activity, the longer duration of diabetes at enrolment of participants (mean 6 years), and the inclusion of people with prior coronary heart disease. We are aware of only one study examining long-term outcomes following remission in non-surgical settings: a 7-year follow-up of a cohort of adults with diabetes in Southern England that found a 20–40% lower incidence of CVD and microvascular disease outcomes among participants who experienced diabetes remission, but did not collect information on intervention participation [16].

Collectively, the results of these trials suggest that remission in people with recently diagnosed type 2 diabetes is achievable through lifestyle intervention resulting in substantial weight loss, but that its effectiveness may depend upon a strong response to interventions in selected subgroups of participants. Previous analyses of Look AHEAD data showed that the percentage of participants achieving remission was twice as great in participants with a 1-year weight loss >6.5% and <2 years diabetes duration, and about 50% greater among participants with HbA1c <48 mmol/mol (6.5%) or in the top tertile of fitness change [4]. In the DiRECT study, duration of diabetes did not predict remission probability within the trial, but the average duration was only 3 years and all participants had a duration of fewer than 6 years [5, 8, 31]. Further, in post hoc analyses, weight loss and programme attendance were the strongest predictors of remission [31].

Observational studies have shown that rates of diabetes remission in the community in the absence of intensive interventions are particularly low [32,33,34]. In the UK National Diabetes Audit, which did not assess intentional weight loss, remission was only 1% in the overall population with diabetes and 3% in those with a recent diagnosis, and 8% among the subset with a large weight loss [34]. Achievement of and duration of remission were associated with several differences at baseline, including shorter duration of diabetes, higher levels of education, and lower starting HbA1c, fasting glucose levels and systolic BP. Our analyses adjusted for these factors and thus probably did not confound the association of remission with long-term outcomes. However, these differences may indicate that people who are earlier in the natural history of diabetes are more likely to achieve and benefit from attempts at remission. Our observation of higher LDL-cholesterol levels in participants achieving remission is surprising, but may be a reflection of the higher statin use among participants in the DSE group, as observed in the primary trial.

There are several limitations to these analyses. First, we lacked the necessary data to directly replicate the newly recommended clinical definition of remission, and instead relied upon an epidemiological definition as a proxy [1]. Second, we lacked data to pinpoint the onset of diabetes remission with precision, and instead used the number of visits with remission as a proxy for time without remission, thus underestimating the time in remission for some and overestimating it for others, and increasing the error variance in the estimates of risk reduction. Third, our analyses excluded approximately 10% of the overall sample due to the participants having bariatric surgery, inadequate data or already meeting the remission definition at baseline. Fourth, because the study did not have a randomised design, it is also possible that participants who achieved remission sought more intensive care and risk factor management, that drove the reduction in outcomes. Thus, there may be additional unmeasured differences in the characteristics of participants who achieved remission that also correlate with better health outcomes. Fifth, we lacked the power to stratify our core analyses by intervention status. Remission also occurred in DSE participants, albeit less frequently (7%, compared with vs 22% in ILI participants), and accounted for about one third of those with remission. Finally, we lacked the necessary data to determine which factors explained the association of remission with reductions in the incidence of CKD and CVD, including whether sex played a role or whether findings varied by sex. We speculate that remission in the DSE group was driven by similar lifestyle and risk factor changes to the factors driving remission in the intervention group, but our study was not adequately powered to examine predictors of remission within the DSE group, as only 11 CVD events and 20 CKD events occurred in DSE participants with diabetes remission. Finally, although the magnitude of weight loss appears to be a key driver in the reduction in CKD and CVD risk, whether targeting remission to influence subsequent CKD and CVD risk is more effective than targeting optimal management of the risk factors themselves cannot be tested by these analyses.

Despite the promising efficacy and outcomes of lifestyle-based remission in this and other studies, the viability of focusing on remission with lifestyle interventions as a major priority for clinical and public health efforts remains undetermined. On the one hand, these findings may drive a paradigm shift whereby selected subsets of the population are actively encouraged to strive beyond prevention of diabetes and its complications, to regression in risk status, in order to optimise long-term health outcomes [14, 15]. On the other hand, the long-term sustainability of such intensive interventions is unclear, and their incremental benefit above and beyond what may be achieved by targeting optimal risk factor management and more modest weight loss has not been tested in experimental settings. These questions underscore the need for continued follow-up in remission studies, as well as rigorous evaluation of real-world programmes of remission as they develop in the future.