Summary Background Our objective was to quantify and predict diabetesriskreduction during the DiabetesPreventionProgramOutcomes Study (DPPOS) in participants who returned to normalglucoseregulation at least once during the DiabetesPreventionProgram (DPP) compared with those who consistently met criteria for prediabetes. Methods DPPOS is an ongoing observational study of participants from the DPP randomised trial. For this analysis, diabetes cumulative incidence in DPPOS was calculated for participants with normalglucoseregulation or prediabetes status during DPP with and without stratification by previous randomised treatment group. Cox proportional hazards modelling and generalised linear mixed models were used to quantify the effect of previous (DPP) glycaemic status on risk of later (DPPOS) diabetes and normalglucoseregulation status, respectively, per SD in change. Included in this analysis were 1990 participants of DPPOS who had been randomly assigned to treatment groups during DPP (736 intensive lifestyle intervention, 647 metformin, 607 placebo). These studies are registered at ClinicalTrials.gov, NCT00004992 (DPP) and NCT00038727 (DPPOS). Findings Diabetesrisk during DPPOS was 56% lower for participants who had returned to normalglucoseregulation versus those who consistently had prediabetes (hazard ratio [HR] 0·44, 95% CI 0·37–0·55, p<0·0001) and was unaffected by previous group assignment (interaction test for normalglucoseregulation and lifestyle intervention, p=0·1722; normalglucoseregulation and metformin, p=0·3304). Many, but not all, of the variables that increased diabetesrisk were inversely associated with the chance of a participant reaching normalglucoseregulation status in DPPOS. Specifically, previous achievement of normalglucoseregulation (odds ratio [OR] 3·18, 95% CI 2·71–3·72, p<0·0001), increased β-cell function (OR 1·28; 95% CI 1·18–1·39, p<0·0001), and insulin sensitivity (OR 1·16, 95% CI 1·08–1·25, p<0·0001) were associated with normalglucoseregulation in DPPOS, whereas the opposite was true for prediction of diabetes, with increased β-cell function (HR 0·80, 95% CI 0·71–0·89, p<0·0001) and insulin sensitivity (HR 0·83, 95% CI 0·74–0·94, p=0·0001) having a protective effect. Among participants who did not return to normalglucoseregulation in DPP, those assigned to the intensive lifestyle intervention had a higher diabetesrisk (HR 1·31, 95% CI 1·03–1·68, p=0·0304) and lower chance of normalglucoseregulation (OR 0·59, 95% CI 0·42–0·82, p=0·0014) than did the placebo group in DPPOS. Interpretation We conclude that prediabetes is a high-risk state for diabetes, especially in patients who remain with prediabetes despite intensive lifestyle intervention. Reversion to normalglucoseregulation, even if transient, is associated with a significantly reduced risk of future diabetes independent of previous treatment group.