Characteristics of cohort 1

Of the 11,254 individuals in cohort 1 (ESM Fig. 1b), 3970 had a first evaluation within 2 years after diagnosis, and 3902 had available data to determine exposure (at least two body weight measurements within 5 years after diagnosis). We retained 1934 participants with available data for the primary outcome (change in HbA1c), who were divided into those who lost ≥10% of their initial body weight (n=308; 15.9%) and those who did not (n=1626). Mean age was 62.7 years and 42.7% of participants were female, mean baseline BMI was 29.5 kg/m2 and mean HbA1c was 64 mmol/mol (8.0%). The mean known diabetes duration was 0.8 years. No participants underwent bariatric surgery. The two groups significantly differed for sex, baseline BMI, HbA1c, degree of comorbidities and use of metformin, sulfonylurea and insulin (Table 1). The median (IQR) duration of observation was 10.8 (7.3–15) years, with a maximum of 24.8 years. The main characteristics of individuals who were excluded because of missing data are shown in ESM Table 1.

Table 1 Baseline characteristics of the study cohorts Weight and glycaemic trajectory

After adjusting for baseline weight in the model, the weight curves diverged early by design and remained separated for up to 20 years (Fig. 1a). The difference was greater at 5 years (~8 kg) and the total adjusted mean difference was 5.9 kg (95% CI 5.1, 6.6). In logistic regression analysis, ≥10% weight loss was associated with female sex, higher BMI, not being on sulfonylurea or insulin, and comorbidity status (ESM Table 2a).

Fig. 1

Main effects of weight loss. The figure compares selected endpoints between participants who lost ≥10% body weight and those who lost <10%. For cohort 1 (a–c) and cohort 2 (d–f), the panels show the change over time in body weight (a, d) and HbA1c (b, e) analysed with the MMRM. (c) and (f) show the HRs of insulin initiation analysed with the Cox model. Adjusted effects are reported for each outcome along with the respective 95% CIs

From initial values of 64–66 mmol/mol (8.0–8.2%), HbA1c declined rapidly in the first year, reaching 51 mmol/mol (6.8%) and 55 mmol/mol (7.2%) in the ≥10% and <10% weight loss groups, respectively (Fig. 1b). The two curves remained separated for at least 10 years and the adjusted mean difference was 2.1 mmol/mol (95% CI 0.3, 3.9 [0.19%; 95% CI 0.03%, 0.36%]). The difference was greater during the first 5 years (3.4 mmol/mol [0.31%]; ESM Table 3).

The burden of therapy (number of GLM classes) increased steadily in the two groups (ESM Fig. 2a), but the ≥10% weight loss group had a marginally lower burden of therapy by 0.2 drug classes (p=0.002). Among participants who were insulin-free at baseline, 21.4% initiated insulin and there was no significant difference in the adjusted rate of insulin initiation between groups (HR 1.22; 95% CI 0.92, 1.61; p=0.165; ESM Fig. 2c).

Remission was achieved by 145 individuals, 20.2% in the group with ≥10% weight loss and 5.5% in the group with <10% weight loss (adjusted HR 4.2; 95% CI 3.0, 5.9; Fig. 1c). The median (IQR) duration of remission was 1.8 (0.9–3.4) years and was similar in the two groups.

Cardiometabolic parameters and complications

Systolic BP declined early and remained significantly lower for up to 20 years in the ≥10% weight loss group, with a difference that was more evident during the first 5 years (ESM Fig. 3a). Triglyceride levels declined early in the ≥10% weight loss group and the mean adjusted difference vs the group with <10% weight loss in the first 5 years was −0.13 mmol/l (−11.2 mg/dl; p<0.001; ESM Fig. 3c), but the difference was lost on longer observation. HDL-cholesterol remained significantly higher by 0.04 mmol/l (1.4 mg/dl) in the ≥10% weight loss group for up to 20 years (ESM Fig. 3d). LDL-cholesterol declined in both groups, but was higher (by 0.18 mmol/l; 6.8 mg/dl) in the ≥10% weight loss group, as was total cholesterol (ESM Table 3).

eGFR (median of 11 values per participant) increased early in the ≥10% weight loss group and remained higher by 2.4 ml/min per 1.73 m2 (ESM Fig. 4a). The change in UACR was highly variable and did not differ between groups (ESM Fig. 4c), but only a median of four values per participant was available. The rate of new-onset CKD, examined in those who were free from CKD at baseline, did not differ between groups (adjusted HR 0.93; 95% CI 0.79, 1.24; p=0.928).

The incidence of new micro- or macroangiopathy was evaluated in participants who were free from these complications at baseline (n=1561 and n=1810, respectively). During observation, 55.1% and 55.5% developed new micro- or microangiopathy, respectively. There was no difference between groups in the rates of incident micro- or macroangiopathy (adjusted HR 1.05 and 1.01, respectively).

Replication in cohort 2

Of 145,198 individuals with type 2 diabetes in cohort 2, 50,080 had available data within 2 years after diagnosis of type 2 diabetes, 29,536 had body weight data to assess exposure and 13,277 had outcome data (ESM Fig. 1b). Participants had a mean age of 61.6 years and 58.9% were male; mean baseline BMI was 30.1 kg/m2 and mean HbA1c was 57 mmol/mol (7.4%). The median (IQR) duration of diabetes at the beginning of observation was 0 (0–2) years. Weight loss of ≥10% was achieved by 1167 participants (8.8%). The two groups differed significantly for sex, diabetes duration, baseline BMI, HbA1c, lipids, prevalence of macroangiopathy, comorbidity index and therapy with DPP-4 inhibitors, statins and anti-platelet agents (Table 1). The median (IQR) observation period was 4.0 (2.0–6.3) years.

The mean adjusted difference in body weight between the groups was 9.0 kg (95% CI 8.4, 9.6; Fig. 1d). Variables significantly associated with weight loss were female sex, shorter diabetes duration, higher BMI, HbA1c and comorbidity index (ESM Table 2b).

HbA1c declined more in the ≥10% weight loss group and remained lower by 2.8 mmol/mol (0.26%) than in the <10% weight loss group, but the curves converged after 6 years (Fig. 1e). The rates of remission were 13.2% in the group with weight loss ≥10% and 4.1% in the group with <10% weight loss (adjusted HR 2.62; 95% CI 2.18, 3.15; Fig. 1f). The median (IQR) duration of remission was 1.3 (0.6–2.1) years and was similar in the two groups.

The number of GLM classes used by the participants increased similarly in the two groups (ESM Fig. 2b) and 4.5% in the two groups combined initiated insulin. The ≥10% weight loss group had a significantly greater risk of initiating insulin (HR 1.76; 95% CI 1.39, 2.25; p<0.001; ESM Fig. 2d).

In the ≥10% weight loss group, there were significantly greater improvements in systolic BP (ESM Fig. 3b), serum triglycerides (ESM Fig. 3d) and HDL-cholesterol (ESM Fig. 3f), while LDL-cholesterol declined similarly (ESM Table 3).

No significant differences were observed in the trends over time in eGFR and UACR (ESM Fig. 4b, d).

During observation, 33.0% developed a new microangiopathy (yearly incidence 8.2%) and 23.9% developed a new macroangiopathy (yearly incidence 6.0%) among participants who were free from these complications at baseline. The incidence rates of new micro- and macroangiopathy were similar between the two groups (Fig. 3).

Features and consequences of remission

In cohort 1, remission was associated with shorter diabetes duration, lower prevalence of macrovascular disease and lower use of metformin and sulfonylurea. In cohort 2, remission was associated with shorter disease duration, lower BMI, higher BP and HbA1c, lower rates of macroangiopathy and lower use of metformin, sulfonylureas, DPP-4 inhibitors, insulin and statins (ESM Table 4). HbA1c and body weight trends in those with and without remission are shown in Fig. 2a, b (cohort 1) and Fig. 2d, e (cohort 2).

Fig. 2

Main effects of diabetes remission. The figure compares selected endpoints between participants with diabetes remission and those without. For cohort 1 (a–c) and cohort 2 (d–f), the panels show the change over time in body weight (a, d) and HbA1c (b, e) analysed with the MMRM. (c) and (f) show the HRs of insulin initiation analysed with the Cox model. Adjusted effects are reported for each outcome along with the respective 95% CIs

In both cohorts, remission led to a marked reduction in insulin initiation (cohort 1: adjusted HR 0.33; 95% CI 0.18, 0.60; Fig. 2c; cohort 2: adjusted HR 0.09; 95% CI 0.03, 0.27; Fig. 2f).

We evaluated the association between remission and occurrence of new complications in cohort 2 (Fig. 3). In the unadjusted analysis, individuals who achieved remission had significantly lower rates of microangiopathy (HR 0.77; 95% CI 0.65, 0.90; p=0.001) and macroangiopathy (HR 0.84; 95% CI 0.73, 0.96; p=0.011). When adjusted for clinical variables significantly associated with remission, the protection from new microangiopathy was still evident (HR 0.84; 95% CI 0.73, 0.97; p=0.019), whereas significance was lost for macroangiopathy (HR 0.90; 95% CI 0.80, 1.02; p=0.100).

Fig. 3

Rates of complications. The forest plot reports crude rates of new microangiopathy and new macroangiopathy in the groups indicated. HRs with 95% CIs and p values are also shown. PYE, patient-years of exposure

Sensitivity analyses

We evaluated mean weight loss by remission status in the two groups. ESM Fig. 5 shows that diabetes remission in the group with <10% weight loss was associated with small weight differences. Within the ≥10% weight loss group, remission was associated with a faster weight loss in the first 2 years, but the difference was lost in the long run.

Finally, for both cohorts, we performed a three-group comparison based on weight loss: <5% (group 0); 5% to <10% (group 1); and ≥10% (group 2; ESM Tables 5, 6). Compared with group 0, group 1 (28.2% in cohort 1 and 17.4% in cohort 2) demonstrated a significant but marginal improvement in HbA1c (mean adjusted difference −1.3 mmol/mol [−0.12%] in cohort 1 and −1.4 mmol/mol [−0.13%] in cohort 2; ESM Table 7), which was inferior to that observed in group 2 and which was no longer present after 8–10 years (ESM Fig. 6b, e). Remission was strikingly more frequent in group 2 than in group 1, especially in cohort 1 (HR vs group 0: 5.2 and 1.7, respectively; ESM Fig. 6c). BP and lipids improved marginally in group 1 and significantly more in group 2 than in group 1 (ESM Table 7).