Main

Closed-loop automated insulin delivery systems (also commonly referred to as artificial pancreases) are progressive technologies designed to treat type 1 diabetes by automating insulin delivery from a portable insulin pump with a wearable glucose sensor and a mathematical dosing algorithm1. Commercially available systems are termed hybrid systems because they automate the insulin delivery between meals but require the user to initiate prandial insulin2,3,4,5,6. Although these systems improve glucose control compared to SAP therapy, hyperglycemia above 10 mmol l−1 is excessive for many patients, on average 6–8 h per day2,3,4,5,6.

Sodium/glucose cotransporter 2 (SGLT2) inhibitors inhibit glucose reabsorption in the kidney, which allows more glucose to be excreted in the urine and thus lowers blood glucose levels in an insulin-independent manner7,8. In type 1 diabetes, SGLT2 inhibitor therapies have been shown to improve HbA1c, weight and blood pressure without increasing hypoglycemia9,10. Moreover, post hoc analysis of randomized trials suggests that SGLT2 inhibition provides renal protection to individuals with type 1 diabetes and albuminuria11,12. However, due to its mechanism of action, SGLT2 inhibitors can raise ketone concentrations and may potentiate diabetic ketoacidosis13.

Empagliflozin is a highly selective compound from the SGLT2 inhibitor class that has been tested in phase 3 clinical trials in individuals with type 1 diabetes on either multiple daily insulin injections or conventional insulin pump therapy14. Similar to trials of dapagliflozin15, sotagliflozin12 and ipragliflozin16, empagliflozin has been tested at the high doses used in the management of type 2 diabetes (10 mg and 25 mg per day in the case of empagliflozin, associated with a placebo-corrected HbA1c reduction of 0.54%) (ref. 14). However, a type 1 diabetes-specific dose of 2.5 mg, associated with similar magnitude of glucosuria as observed in individuals with type 2 diabetes on the higher doses17, was observed to have clinically significant, yet attenuated, HbA1c reduction (0.28%) compared to higher doses.

In this study, we assessed the glycemic efficacy and safety of adding 25 mg d−1 empagliflozin to a closed-loop automated insulin delivery system in adults with type 1 diabetes. In a randomized, placebo-controlled, crossover multicenter trial, we compared, in a 2 × 2 factorial design, 4 weeks of each of closed-loop therapy plus empagliflozin, closed-loop therapy plus placebo, SAP therapy plus empagliflozin and SAP therapy plus placebo. The primary outcome was the time spent in the glucose target range (3.9–10.0 mmol l−1) calculated over the entire 4-week period.

Results

Twenty-eight participants were enrolled in the study between 2 August 2019 and 6 January 2021 (CONSORT diagram in Fig. 1). One participant withdrew before randomization due to circumstances surrounding the COVID-19 pandemic. Two (7%) participants withdrew from the study due to symptomatic ketosis (without acidosis), both within the first few days of their first interventions, which were closed-loop therapy plus empagliflozin and closed-loop therapy plus placebo. The data of one participant were excluded from the analysis due to a pharmacy error that prevented accurate unblinding. Participant characteristics are shown in Table 1: 56% were women, the mean age was 38 ± 15 years, weight was 83 ± 19 kg, body mass index (BMI) was 29.5 ± 6.6 kg m−2, HbA1c was 7.7 ± 0.9% (61 ± 10 mmol mol−1), average duration of diabetes was 24 ± 15 years and average total daily insulin was 0.63 ± 0.21 U kg−1.

Fig. 1
figure 1

CONSORT (Consolidated Standards of Reporting Trials) Diagram Detailing Flow of Trial Participants.

Full size image
Table 1 Baseline characteristics of the 27 randomized participants with type 1 diabetes
Full size table

Figure 2 shows the comparison glucose profiles during each of the interventions. Glucose sensor readings were available 89 (78–98), 90 (77–94), 88 (82–92) and 88% (78–91) of the time during closed-loop therapy plus empagliflozin, closed-loop therapy plus placebo, SAP therapy plus empagliflozin and SAP therapy plus placebo, respectively. The system was operational in closed-loop mode 86 (77–91) and 89% (73–94) of the time during the closed-loop therapy plus empagliflozin and closed-loop therapy plus placebo interventions, respectively (P = 0.91).

Fig. 2: The median (IQR) profiles of glucose levels during the interventions from midnight to midnight.
figure 2

n = 24 independent participants in each of the four settings. Median values according to time of day are represented by the bold blue and red lines. Interquartile ranges by time of day are represented by the shaded light blue and light red areas. The dashed lines indicate the upper threshold levels for the target ranges analyzed as outcomes, specifically 3.9 to 10.0 mmol l−1 and 3.9 to 7.8 mmol l−1.

Full size image

The primary outcome of time spent in the target range over the entire 4-week period (3.9–10.0 mmol l−1) was 75.5 ± 8.8, 68.2 ± 9.1, 69.3 ± 10.7 and 57.9 ± 13.2% with closed-loop therapy plus empagliflozin, closed-loop therapy plus placebo, SAP therapy plus empagliflozin and SAP therapy plus placebo, respectively (Table 2). Empagliflozin improved time in range with SAP therapy compared to placebo by 11.4% (7.7–15%, P < 0.0001) (corresponding to 2.7 h (1.8–3.6 h)). Empagliflozin also improved time in range with closed-loop therapy compared with placebo (P < 0.0001) but by a lower magnitude of 7.2% (3.5–10.9) (1.7 h (0.8–2.6 h)). No treatment-by-period interaction was found and no difference was observed due to the sequence or order of the interventions (P > 0.87 for the regression model coefficients). Daytime (6:00–24:00) and nighttime (24:00–6:00) outcomes followed similar patterns (Supplementary Material; Tables 1–3 for daytime and Tables 4–6 for nighttime). HbA1c, blood pressure and estimated glomerular filtration rate (eGFR) outcomes measured at baseline and at the end of drug intervention are reported in Supplementary Table 8.

Table 2 Comparison of glucose, insulin and safety outcomes between (1) closed-loop therapy plus empagliflozin versus closed-loop therapy plus placebo and (2) SAP therapy plus empagliflozin versus SAP plus placebo
Full size table

To evaluate whether the use of empagliflozin precludes the need for closed-loop therapy, we compared closed-loop therapy plus empagliflozin versus SAP therapy plus empagliflozin (Table 3). On the background of empagliflozin, closed-loop therapy improved time in range compared to SAP therapy by 6.1% (2.5–9.8%; 1.5 h (0.6–2.4 h)), which was significant (P < 0.0001) but lower than the improvement on the background of placebo of 10.3% (6.6–14%, P < 0.0001) (2.5 h (1.6–3.4h)). Combined closed-loop therapy and empagliflozin compared to SAP therapy plus placebo was associated with the largest improved time in range of 17.5% (13.8–21.2%) (4.2 h (3.3–5.1 h)).

Table 3 Comparison of glucose, insulin and safety outcomes between (1) closed-loop therapy plus empagliflozin versus SAP plus empagliflozin and (2) closed-loop therapy plus empagliflozin versus SAP plus placebo
Full size table

Hypoglycemia outcomes

No severe hypoglycemia was observed in any intervention period. Time spent in hypoglycemia <3.0 mmol l−1 was rare in all interventions (<0.7%). Time spent in hypoglycemia <3.9 mmol l−1 was 2.5 (2.1–3.8), 2.1 (1.5–2.8), 4.6 (2.4–5.5) and 2.8% (1.3–4.4) with closed-loop therapy plus empagliflozin, closed-loop therapy plus placebo, SAP therapy plus empagliflozin and SAP therapy plus placebo, respectively (Table 2). The difference in time spent <3.9 mmol l−1 between empagliflozin and placebo was 0.3 (−0.7 to 1.3, P = 0.53) (4.3 min (−10.1 to 18.7)) and 0.9% (−0.1 to 1.9, P = 0.0368) (12.1 min (−1.4 to 27.4)) with closed-loop and SAP therapy, respectively.

Consensus glycemic targets

Extended Data Fig. 1 shows the proportion of participants achieving consensus guidelines of time spent in the target range (>70%), time spent in hypoglycemia <3.9 mmol l−1 (<4.0%) and time spent in hypoglycemia <3.0 mmol l−1 (<3.0 mmol l−1)18. The proportion of participants achieving simultaneously all three targets increased approximately twofold between arms and was 8, 17, 30 and 63% in SAP therapy plus placebo, SAP therapy plus empagliflozin, closed-loop therapy plus placebo and closed-loop therapy plus empagliflozin, respectively.

Daily insulin dose requirements

Empagliflozin reduced average daily insulin dose requirements compared to placebo with closed-loop therapy (P = 0.0003; Table 2) and SAP therapy (P = 0.0003; Table 2) by an identical value of −5.3 U d−1 (−8.4 to −2.1), corresponding to a relative reduction of 9%. The average daily insulin dose was 54.2 ± 28.0 U d−1 with closed-loop therapy plus empagliflozin, 59.8 ± 31.2 U d−1 with closed-loop therapy plus placebo, 54.2 ± 28.0 U d−1 with SAP therapy plus empagliflozin and 59.2 ± 29.8 U d−1 with SAP therapy plus placebo (Table 2). With SAP therapy, the reduction in insulin dose due to empagliflozin was primarily driven by reductions in insulin boluses (−4.5 U d−1 (−7.1 to −1.8), P = 0.0003) rather than insulin basal (−0.8 U d−1 (−2.6 to 1.1), P = 0.33) (Extended Data Fig. 2). On placebo, due to its continuous corrective actions, closed-loop therapy had a higher basal delivery compared to SAP therapy (4.2 U d−1 (2.3–6), P < 0.0001), which led to lower insulin boluses (−3.8 U d−1 (−6.5 to −1.1), P = 0.0017), partially due to the resulting lower daytime glucose levels (−0.6 mmol l−1 (−1.0 to −0.3), P = 0.0004), which impacts correction boluses. However, on empagliflozin, the increase in total daily basal insulin was not observed with closed-loop therapy compared to SAP therapy (0.0 U d−1 (−1.8 to 1.9), P = 0.9809), likely due to reduced need for continuous insulin corrective actions due to the glucose-lowering effect of empagliflozin. The average number of daily meals and snacks announced to the bolus calculator to deliver prandial boluses was similar between interventions. Specifically, these were 3.9 ± 1.0 and 3.7 ± 1.0 meals and snacks during closed-loop therapy plus empagliflozin and closed-loop therapy plus placebo, and 4.0 ± 1.1 and 4.3 ± 1.1 during SAP therapy plus empagliflozin and SAP therapy plus placebo, respectively. The average meal and snack size announced was 42 g carbohydrate in all 4 arms.

Adverse events, ketone levels and ketoacidosis

There were no severe hypoglycemia or diabetic ketoacidosis events in any interventional period. Two (7%) participants withdrew within the first few days on closed-loop therapy plus placebo intervention and closed-loop therapy plus empagliflozin intervention, respectively, due to symptomatic ketosis (1.8 and 3.7 mmol l−1, respectively) without acidosis. These were 1 male and 1 female participant with a baseline daily insulin dose of 26 U (0.34 U kg−1) and 27 U (0.56 U kg−1), BMI of 24.8 kg m−2 and 18.7 kg m−2 and baseline HbA1c of 5.1% (32 mmol mol−1) and 7.5% (58 mmol mol−1), respectively. Among participants who completed the study, five (20%) experienced ketosis (>1.5 mmol l−1) on closed-loop plus empagliflozin compared to one participant (5%) in each of the other interventions (Table 2).

Empagliflozin increased mean fasting ketone level with SAP therapy from 0.11 ± 0.08 mmol l−1 to 0.17 ± 0.08 mmol l−1 (P = 0.0100) and with closed-loop therapy from 0.12 ± 0.06 mmol l−1 to 0.27 ± 0.15 mmol l−1 (P < 0.0001). Compared to the baseline mean fasting ketone level with SAP therapy plus placebo, the increase with closed-loop therapy plus empagliflozin (0.15 mmol l−1 (0.1–0.2), P < 0.0001) was double the additive increase with closed-loop therapy plus placebo (0.01 mmol l−1 (−0.04 to 0.06)) and SAP therapy plus empagliflozin (0.06 mmol l−1 (0.01–0.1)), suggesting a synergic impact of empagliflozin and closed-loop therapy on ketone levels. When the data from the closed-loop and SAP therapy interventional arms were combined, the mean fasting ketone level was 0.24 ± 0.14 mmol l−1 in women compared to 0.18 ± 0.10 mmol l−1 in men (P = 0.19).

In the post hoc analysis, BMI was inversely correlated to mean ketone level, particularly in the empagliflozin arms (Extended Data Fig. 3). When assessing mean ketone level in the empagliflozin arms (closed-loop and SAP therapy interventional arms combined) with regard to the BMI cutoff in the approved indication of SGLT2 inhibitors in several countries (27 kg m−2) (ref. 19), the mean fasting level was lower in participants with a BMI > 27 kg m−2 (0.15 ± 0.05 mmol l−1) than participants with a BMI < 27 kg m−2 (0.27 ± 0.16 mmol l−1, P = 0.0057). Moreover, the increase in mean ketone level with empagliflozin (closed-loop and SAP therapy interventional arms combined) relative to placebo was lower in participants with a BMI > 27 kg m−2 (difference 0.07 mmol l−1 (0.05–0.10), P < 0.0001) than participants with a BMI < 27 kg m−2 (difference 0.13 mmol l−1 (0.06–0.20), P < 0.0001).

Two female participants in the closed-loop therapy plus empagliflozin arm reported an episode of genital mycotic infection, each mild in severity. In the SAP therapy plus empagliflozin arm, one participant reported an ear infection (mild severity) and one participant had an upper respiratory tract infection (mild severity). In the closed-loop therapy plus placebo arm, one participant reported flu-like symptoms. Two days after the end of the closed-loop therapy plus placebo arm, one participant was hospitalized with chest pain following a traumatic stress, diagnosed as Takotsubo cardiomyopathy unrelated to the study interventions. In the SAP therapy plus placebo arm, one participant reported a common cold.

Discussion

In this study, we assessed, using a factorial crossover design, the individual and combined efficacy of the SGLT2 inhibitor empagliflozin (compared to placebo) and closed-loop automated insulin delivery (compared to SAP) on glycemic control in type 1 diabetes. The efficacy of empagliflozin compared to placebo on the background of closed-loop therapy led to an increase in time spent in the glucose target range of 7.2% compared to closed-loop therapy alone. As per international consensus guidelines18, an increase of more than 5% is considered clinically meaningful. Despite this substantive contribution to glycemic control, the increase in time in range when empagliflozin was added to closed-loop therapy was lower than that when empagliflozin was added to SAP therapy (7.2 versus 11.4%). This is likely because, as opposed to SAP therapy, both empagliflozin and closed-loop therapy work in a continuous glucose-responsive manner that might have led to a partial overlap in their action. However, we emphasize that the combination of closed-loop therapy and empagliflozin was associated with 17.5% greater time in range compared to SAP therapy and placebo.

In our study, no episode of diabetic ketoacidosis was observed with empagliflozin in either SAP or closed-loop therapy interventions, probably due to daily ketone monitoring, the mitigation and education strategies put into place in the trial (Extended Data Fig. 4 and 5), the relatively small sample size and the short duration of the interventions. Nevertheless, the mean fasting ketone level was increased when empagliflozin was added to SAP therapy; closed-loop therapy amplified this empagliflozin-induced increase by twofold. Moreover, more participants experienced ketosis (>1.5 mmol l−1) with the closed-loop therapy plus empagliflozin intervention compared to the other interventions. Taken together, these results indicate that efforts must be taken to prevent ketoacidosis when adding a 25-mg empagliflozin dose to closed-loop therapy20, including enhanced patient education, minimization of insulin reductions, clinical follow-up and ketone monitoring. Technologies for continuous ketone monitoring could provide people with diabetes constant surveillance with an alarm in the context of ketosis as an additional measure to prompt self-management interventions to prevent progression to diabetic ketoacidosis18,20,21. Although of potential benefit to all people with type 1 diabetes for ketoacidosis prevention, uptake and use could first be targeted toward those with the greatest magnitude of risk (such as those using adjunct-to-insulin SGLT2 inhibitors), in combination with clinical risk mitigation strategies22,23 to reduce the risk of diabetic ketoacidosis24,25.

Additionally, aspects of patient selection could further mitigate the risk of diabetic ketoacidosis. Post hoc analysis of randomized trials showed that a higher BMI shifts the benefits versus risk profile of SGLT2 inhibitors with a reduced risk of diabetes ketoacidosis15, probably due to lower rates of lipolysis in this population compared to lean individuals and the requirement of higher doses of insulin compared to those of lower body mass26,27. In the current study, mean ketone levels with empagliflozin in participants with a BMI ≥ 27 kg m−2 was significantly lower than in those with a BMI < 27 kg m−2. Moreover, combined empagliflozin and closed-loop therapy remained effective in improving time in range when analysis was restricted to participants with high BMI (Supplementary Table 7). Dedicated long-term studies in this subpopulation are warranted.

Even though we tested one specific closed-loop system with the McGill automated insulin dosing algorithm, the results of our study likely generalize to other available and commercialized closed-loop systems. Acknowledging that comparison between systems across studies is complex due to differences in baseline patient characteristics and study designs that can influence study outcomes28, our closed-loop system plus placebo in this study performed similarly to current commercial systems tested in other studies. The time in range with our closed-loop system plus placebo was 68.2%, which approximated the targets observed with other systems in randomized trials (67–71%; refs. 2,3,4,5,6). Moreover, when compared to SAP therapy, our closed-loop system plus placebo improved time in range by 10.3% (Table 4), which is within what is observed with other systems (9–12%; refs. 2,3,4,5,6). However, with the introduction of new rapid-acting insulins and glucagon or pramlintide in dual-hormone systems, closed-loop systems will continue to improve. It is unknown if adding SGLT2 inhibitors will continue to provide a 7.2% improvement in time in range. Irrespective of the glycemic benefits, the potential cardiorenal protective benefits with SGLT2 inhibitors could justify using this class of drug in type 1 diabetes29, especially since mortality from any cause or from cardiovascular causes is still high compared to the general population even in those who achieve HbA1c below 7.0% (ref. 30).

Table 4 Glucose, insulin and safety outcomes for each of the four interventions
Full size table

Despite the use of a randomized 2 × 2 factorial design, the inclusion of two sites and the outpatient setting of the interventions, this study has some limitations. As is customary with device technologies, the closed-loop therapy intervention could not be blinded, while the SGLT2 inhibitor intervention was blinded to participants and investigators. We did not test low-dose options for SGLT2 inhibitors in this study. Longer and substantively larger studies are needed to quantify the diabetic ketoacidosis risk with closed-loop therapy plus empagliflozin intervention. Two earlier studies assessed SGLT2 inhibitors and closed-loop therapy in type 1 diabetes but were small and pilot in nature and in highly supervised settings. Biester et al. compared 10 mg dapagliflozin twice daily to placebo under fully closed-loop conditions in a 24-h inpatient study31. Haidar et al. assessed empagliflozin 25 mg once daily under 3 prandial insulin strategies with closed-loop therapy in 9–14-h supervised interventions32. Both studies demonstrated improved control with the addition of SGLT2 inhibitors. This study builds on these proof-of-concept smaller studies and was conducted in an outpatient free-living setting over four weeks without remote monitoring; thus, it allowed more robust assessment of efficacy and safety outcomes.

Our study showed that combining high-dose empagliflozin with closed-loop therapy brings synergic benefits to glucose control during the day and night in patients with type 1 diabetes, although it amplifies the increase in ketone levels and ketosis. Clinical use of empagliflozin as an adjunct therapy to closed-loop systems could be appropriate in individuals with a low risk of diabetic ketoacidosis and implementing strategies for early ketoacidosis detection.

Methods

The study was approved by the local ethics committees (Mount Sinai Hospital Research Ethics Board, approval no. 18-0165-A, and McGill University Health Center’s Research Ethics Board, approval no. 2019-4946) and Health Canada.

Participants

From 2 August 2019 to 6 January 2021, participants were enrolled at Mount Sinai Hospital, Toronto and the Research Institute of McGill University Health Centre, Montreal, Canada. Participants were adults aged 18 years or older, diagnosed with type 1 diabetes for more than 1 year, had used an insulin pump for more than 3 months and had an HbA1c ≤ 10%. Exclusion criteria included reduced eGFR (< 60 ml min 1.73 ml−2), use of non-insulin antihyperglycemic drugs (for example, glucagon-like peptide-1 analogs or metformin), use of loop diuretics, pregnancy, diabetic ketoacidosis in the last 3 months, recent history of genital or urinary infection and use of hydroxyurea medications. Participants provided written informed consent.

Study design

We conducted a randomized, crossover, placebo-controlled, 2 × 2 factorial trial in 28 adults with type 1 diabetes. Each participant entered an open-loop 14-d therapy optimization period, was then randomized to empagliflozin 25 mg d−1 or placebo and (within each empagliflozin and placebo intervention) allocated to closed-loop and SAP therapy in random order. Each intervention lasted four weeks; on average, the total study duration was five months per participant. The interventions were separated by 11 d (8–13) when switching drugs (empagliflozin and placebo) and 1 d (0–2) when switching insulin delivery method (closed-loop and SAP therapy) to eliminate the effects of empagliflozin and insulin delivery method between interventions. (The half-lives of empagliflozin and insulin are around 12 h (ref. 33) and 50 min34, respectively.) The trial was registered at ClinicalTrials.gov (identifier NCT03979352) and the study protocol is provided in the supplementary materials.

Randomization, concealment and masking

We used blocked randomization with equal ratios of size eight (separate blocks for each site) to generate the allocation sequences. Sequences were generated by the Mount Sinai Hospital pharmacy and were placed in opaque envelopes, which were distributed to both sites and opened by the study coordinators after the therapy optimization period. The order of drug interventions (empagliflozin and placebo) was randomized and so was the order of the insulin delivery method (closed-loop and SAP therapy) within each drug intervention. The empagliflozin and placebo doses were placed in identical capsules. Participants and investigators were blinded to the allocation of the drug but were not blinded to the allocation of the insulin delivery method.

Study procedures

After study entry, participants had an optimization period of 14 d with a glucose sensor (Dexcom G5 or G6; Dexcom) and using their own pump. During this optimization, a member of the team reviewed the participant’s pump and sensor data and adjusted their pump settings if substantial hyperglycemia or hypoglycemia were observed. Dexcom G5 was discontinued in Canada during the study; thus, the last three participants used Dexcom G6.

On the SAP therapy arms, participants used the Dexcom glucose sensor and their own pump for four weeks. On the closed-loop therapy arms, participants used the iPancreas system35 (Oregon Health & Science University) with our dosing algorithm (the McGill dosing algorithm, version Lucie-3.9.0) (refs. 36,37) for four weeks. The system consists of a Dexcom glucose sensor, a noncommercial t:slim TAP3 insulin pump (Tandem Diabetes Care) and a smartphone. After the optimization period, participants were trained on how to use the iPancreas system. The iPancreas system was the same in both closed-loop therapy interventions (placebo and empagliflozin).

The closed-loop system was initialized using participants’ total daily insulin dose, carbohydrate ratios and programmed basal rates. A new basal rate was calculated every 10 min based on a model predictive control dosing algorithm36,37, which used the sensor data as an input. The computed basal rate was communicated wirelessly to the pump. The system glucose target was set at 6.0 for the basal rate changes. Participants were made aware of the exercise feature in the system, which would raise the glucose target by 3 mmol l−1.

Participants were instructed to manually enter the carbohydrate content of the meals and snacks into the iPancreas system, which calculated the prandial boluses. Prandial boluses were calculated using the carbohydrate-to-insulin ratios, premeal glucose levels and insulin on board. Participants could also manually deliver correction boluses through the system at any time. The system had a glucose target of 6.0 mmol l−1 for the correction boluses. The system does not administer automatic boluses outside mealtimes. The system switches to open-loop mode (delivering the participant’s usual basal rates) if communication between the phone and pump or sensor is lost for more than 20 and 30 min, respectively.

If a participant’s HbA1c was less than 8% before the start of a closed-loop or SAP therapy intervention, then the programmed basal rates, carbohydrate ratios and insulin sensitivity factors were altered, based on the investigators’ judgment, to deliver 10% less insulin to account for the glycemic lowering effect of empagliflozin. Since the participants and investigators were masked to the drug intervention, these initial adjustments were made regardless of drug assignment (empagliflozin or placebo). Participants were subsequently contacted on the first 3 d of closed-loop and SAP therapy interventions, as well as once weekly afterward to discuss any unexpected events or technical problems. Downloads of glucose and insulin data on days 3 (±2) and 7 (±2) were also reviewed by the research team during both closed-loop and SAP therapy interventions and further adjustments to insulin therapy parameters were made if necessary. Thus, even if the systems were the same in both closed-loop therapy interventions (placebo and empagliflozin), the adjustable parameters (for example, carbohydrate ratios) were generally different between the two closed-loop therapy interventions.

Study teams were on call throughout the interventions to provide technical support. For both closed-loop and SAP therapy interventions, sensor alarm thresholds were determined by the participants. Participants were asked to treat hypoglycemia as per their usual practice. Participants were instructed to measure their blood ketone levels using a capillary ketone meter every morning during the interventions, as well as in case of any symptoms of diabetic ketoacidosis, irrespective of glucose levels, and were instructed to contact the research team if their blood ketone levels were >0.6 mmol l−1 at any time. Participants were given a brochure for the management of ketone levels while on empagliflozin (Supplementary Information and Extended Data Figs. 4 and 5).

Study outcomes

The primary comparisons were aimed at assessing the benefits of empagliflozin as an adjunct therapy to both closed-loop and SAP therapy. Therefore, the two primary comparisons were the percentage of times that the sensor glucose readings were in the target range (3.9–10.0 mmol l−1) (refs. 15,16) between (1) closed-loop therapy plus empagliflozin versus closed-loop therapy plus placebo and (2) SAP therapy plus empagliflozin versus SAP therapy plus placebo. If the primary comparisons were achieved, then we aimed to assess (1) whether the use of empagliflozin with SAP therapy could overcome the need for a closed-loop system and (2) the benefits of combining both closed-loop therapy and empagliflozin. Therefore, the two conditional comparisons were the percentage of times that the sensor glucose readings were in the target range between closed-loop therapy plus empagliflozin versus SAP therapy plus (1) empagliflozin and (2) placebo. Secondary outcomes included the time spent below and above the target range, glucose variability and insulin delivery15,16. Analysis was by intention to treat. Outcomes were calculated for the entire four-week study period. For the closed-loop therapy arms, outcomes included all available data including when the system was in open-loop mode; 5-min glucose sensor data were used in the calculations. If glucose sensor values were missing for less than 2 h, then missing values were interpolated.

Statistical analysis

We aimed to recruit 28 participants (14 per site), a reduction from the initial planned recruitment owing to funding decisions before and during the pandemic. Assuming an s.d. in time spent in the target range of 13 and 15% for the empagliflozin arms and a correlation between arms of 0.50, 28 participants would give 73% and 60% power to detect 8% and 7% differences in time in range, respectively, at a significance level of 2.5%. After the filing of the protocol, preliminary data became available that indicated that we had overestimated the s.d. in time in range during exposure to empagliflozin and we felt confident that power was substantially higher.

Analyses were performed on an intention-to-treat basis. The effect of the interventions was estimated using a linear mixed-effects model: specifically, least squares mean estimates of treatment effect were determined and two-sided tests of the resulting t-statistic were applied. To examine for carryover effect, a model was fitted with the treatment-by-period interaction term. The model accounted for repeated observations as well as the period and randomization sequence. Residual values were examined for normality; if skewed, data were transformed using the square root function or the Wilcoxon signed-rank test was used. The two primary comparisons shared an overall Alpha level of 0.05 (that is, P < 0.025 was regarded as significant). Since testing the 2 conditional comparisons was conditional on the rejection of the 2 primary hypotheses (that is, prespecified ordering of the null hypotheses, with 2 sequential testing), we did not further adjust for multiple comparisons and a 5% significance level was shared between the 2 conditional comparisons (that is, P < 0.025 was regarded as significant). The results of the secondary analysis are regarded as hypothesis-generating and exploratory rather than conclusive; therefore, no formal multiplicity adjustments were made. Results are reported as the median (interquartile range (IQR)) or mean (s.d.). SAS v.9.4 (SAS Institute) was used for the statistical analysis.

Data sharing

The raw data (that is, insulin delivery, glucose levels and individual participant data) and informed consent form will be shared by the corresponding author, for academic purposes, subject to a material transfer agreement and approval of the Mount Sinai Hospital Research Ethics Board and McGill University Health Center’s Research Ethics Board. All data shared will be deidentified.

Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.