Background: Hypoglycaemia unawareness and severe hypoglycaemia can increase fear of hypoglycaemia and the risk of subsequent hypoglycaemic events. We aimed to assess the safety and efficacy of insulin pump therapy with integrated continuous glucose monitoring (CGM) and a suspend-before-low feature (Medtronic MiniMed 640G with SmartGuard) in hypoglycaemia-prone adults with type 1 diabetes. Methods: SMILE was an open-label randomised controlled trial done in people aged 24–75 years with type 1 diabetes for 10 years or longer, HbA1c values of 5·8–10·0% (40–86 mmol/mol), and at high risk of hypoglycaemia (recent severe hypoglycaemia or hypoglycaemia unawareness defined by a Clarke or Gold score ≥4). Participants were enrolled from 16 centres (eg, clinics, hospitals, or university medical centres) in Canada, France, Italy, the Netherlands, and the UK. After baseline run-in phase (2 weeks), participants were randomly assigned to the MiniMed 640G pump (continuous subcutaneous insulin infusion) with self-monitoring of blood glucose (control group) or to the MiniMed 640G system with the suspend-before-low feature enabled (intervention group), for 6 months. The study statistician analysing the data was masked to group assignment until final database lock; because of the nature of the intervention, participants and treating clinicians could not be masked to group assignment. The primary outcome was the mean number of sensor hypoglycaemic events, defined as 55 mg/dL (3·1 mmol/L) or lower, and was analysed on an intention-to-treat basis in all randomly assigned participants. This trial is registered with ClinicalTrials.gov, number NCT02733991, and is completed. Findings: Between Dec 7, 2016, and March 27, 2018, 153 participants with a mean age 48·2 [12·4] years were randomly assigned: 77 to the control group (mean age 47·4 [12·5] years) and 76 to the intervention group (mean age 49·0 [12·2] years). After 6 months, the intervention group had significantly fewer hypoglycaemic events per participant per week (1·1 [SD 1·2] vs 4·1 [3·4] mean events, model-based treatment effect −2·9 [95% CI −3·5 to −2·3]; p<0·0001) and fewer severe hypoglycaemic events (instances requiring third-party assistance with carbohydrate or glucagon administration, or other resuscitative actions) overall (three vs 18; p=0·0036). The most common adverse events were hypoglycaemia (observed in ten [13%] of 77 participants in the control group vs four [5%] of 76 in the intervention group) and hyperglycaemia (observed in seven [9%] of 77 vs seven [9%] of 76). No serious adverse device effects or episodes of diabetic ketoacidosis were reported. Interpretation: Insulin pump therapy with integrated CGM and a suspend-before-low feature reduced the frequency of sensor hypoglycaemic and severe hypoglycaemic events in hypoglycaemia-prone adults compared with use of continuous subcutaneous insulin infusion without real-time CGM. These results suggest that this technology could be beneficial in this high-risk population. Funding: Medtronic International Trading Sàrl and Medtronic Canada.