AUTONOMY: The First Randomized Trial Comparing Two Patient-Driven Approaches to Initiate and Titrate Prandial Insulin Lispro in Type 2 Diabetes

The management of patients with type 2 diabetes generally requires stepwise intensification of therapy beginning with lifestyle changes and oral antidiabetic drugs (OADs) progressing to noninsulin injectable antidiabetic agents and, given the progressive deterioration in β-cell function, to the addition of exogenous insulin (1–4). The results of the UK Prospective Diabetes Study (UKPDS) support the need for treatment intensification with exogenous insulin in combination with OADs in a significant percentage of patients to achieve and maintain metabolic control (5).

The Treat-to-Target trial investigated the efficacy and safety of adding basal insulin glargine (GLA) or NPH insulin in patients on OADs with poorly controlled type 2 diabetes, in North America, and it established a clinical standard for basal insulin treatment trials: ∼60% of patients achieved the HbA_1c target of <7.0% (53.0 mmol/mol) recommended by the American Diabetes Association using both GLA and NPH insulins; however, GLA resulted in lower rates of mild nocturnal hypoglycemia (3,6,7). The prevention of hypoglycemic episodes in the management of type 2 diabetes is critical because hypoglycemia may limit the adoption of further insulin intensification and has been shown to increase the risk for cardiovascular disease and other adverse events, particularly in older adults (8–10). Additionally, hypoglycemia is a primary source of fear that negatively impacts patients’ adherence to treatment regimens and quality of life (10,11). In contrast to the Treat-to-Target trial, only 32–43% of patients in a European trial which investigated the efficacy of initiating once-a-day GLA versus NPH insulin at bedtime, in combination with glimepiride, reached goal HbA_1c of <7.5% (58.5 mmol/mol), and the mean end point HbA_1c in all groups was higher than in Treat-to-Target (6,12).

The Treating to Target in Type 2 Diabetes (4-T) Trial investigated the efficacy of adding a basal, prandial, or biphasic insulin regimen to OAD therapy (13). The 3-year results from this study demonstrated that a greater proportion of patients on basal and prandial interventions reached an HbA_1c of ≤6.5% (≤47.5 mmol/mol) than those treated with biphasic premixed insulin (13). Given limited data and myriad treatment approaches, there is currently no global clinical consensus for the approach to treatment intensification with insulin therapy. A meta-analysis and a systemic review of randomized controlled trials suggests that the most effective use of insulin is achieved using a basal-bolus regimen (14,15). Basal-bolus therapy allows for more effective control of postprandial glucose excursions than basal insulin alone and provides greater flexibility for mealtime insulin timing and titration than premixed biphasic insulin therapies.

There is a need, particularly in the generalist setting, for evidence to support the implementation of simple approaches to prandial insulin therapy that empower the patient and promote individualized treatment. No randomized, controlled studies in patients with type 2 diabetes have investigated treatment escalation with prandial (bolus) insulin using patient-driven treatment intensification. The AUTONOMY trial was designed to compare the efficacy and safety of two patient-based self-titration algorithms for initiation and titration of prandial insulin lispro therapy in patients with type 2 diabetes who could not achieve adequate glycemic control on basal insulin plus OADs. The study provides the first comparison of two self-titration insulin algorithms for the escalation of prandial insulin therapy in a large, multicountry, randomized, controlled trial.

AUTONOMY was a 14-country/1-territory (Argentina, Austria, Brazil, Canada, Croatia, Denmark, France, Lithuania, Mexico, Poland, Romania, Russian Federation, South Africa, and U.S./Puerto Rico), multicenter, randomized, open-label, parallel trial in subjects with type 2 diabetes who had inadequate glycemic control on basal insulin plus OADs. The trial was designed as two independent studies (study A: N = 528; study B: N = 578) using a single protocol to corroborate substantial evidence of reproducibility. The data for each study were analyzed separately and independently. For maintenance of integrity, each investigator was assigned to one of the two studies according to an allocation plan specified before initiation of the trial. Stratification variables included baseline HbA_1c (≤8.0% and >8.0% [≤63.9 and >63.9 mmol/mol]), country, and sulfonylurea/meglitinide use. Approximately 44% of trial sites were in primary care (nonspecialist) settings. The trial enrolled subjects with type 2 diabetes (16), 18–85 years of age, a BMI <45 kg/m², and HbA_1c >7.0% (53.0 mmol/mol) and ≤12.0% (107.7 mmol/mol); treated with at least 20 units/day insulin GLA, NPH, lispro protamine suspension (NPL), or detemir; and who had been using metformin, meglitinide, sulfonylurea, pioglitazone, sitagliptin, or a combination of these for ≥3 months. The exclusion criteria included prior rapid- or short-acting insulin therapy, excessive insulin resistance (>2 units/kg), morbid obesity (BMI ≥45 kg/m²), pregnancy or planned pregnancy, cancer, recent history of severe hypoglycemia, and moderate-to-severe cardiovascular/renal/hepatic/hematologic disease. Patients were excluded from the study if they were taking the following medications: GLP-1 receptor agonists, α-glucosidase inhibitors, dipeptidyl peptidase-4 inhibitors (except sitagliptin), and rosiglitazone within 3 months or glucocorticoids within 2 weeks of screening.

All subjects provided informed consent, and the trial was conducted in compliance with the International Conference on Harmonization Guidelines for Good Clinical Practice and the Declaration of Helsinki (17).

Subjects treated with GLA (Lantus; Sanofi) at entry who had HbA_1c >7.0% (53.0 mmol/mol) and fasting blood glucose (FBG) ≤120 mg/dL did not require a lead-in period and were randomized to one of the two treatment arms. Those who required conversion to, or optimization of, GLA underwent a 6-week lead-in period during which the dose was adjusted by investigators every 3–7 days based on the treat-to-target algorithm (6). As GLA is widely used in clinical practice, with prior studies supporting comparable efficacy and safety to other basal insulins, the protocol was designed for use of this single basal insulin. After randomization, bedtime doses of GLA were only adjusted based upon clinical judgment of the investigator. Subjects were randomized 1:1 at baseline (randomization) to begin insulin lispro (Humalog; Eli Lilly and Company) therapy with either the every day (Q1D) or every 3 days (Q3D) self-titration algorithm (Supplementary Data). Assignment to treatment groups was designated by a computer-generated random sequence using an interactive voice-response system. Subjects continued the use of OADs at prestudy dose, and those on sulfonylurea or meglitinide discontinued that drug at randomization and increased GLA dose by 10% of their total daily total dose (TDD). A 24-week intervention duration after optimization and randomization was selected to allow sufficient time for prandial insulin therapy intensification and to stabilize glycemic control as measured by HbA_1c. Primary and secondary outcome measures were mainly recorded at baseline and 7, 12, and 24 weeks.

Safety was monitored throughout the study, and the occurrence and nature of all adverse events were recorded. Hypoglycemia was considered an adverse event, and a severe hypoglycemic event was recorded as a serious adverse event. Hypoglycemia was defined as anytime the subject experienced a sign or symptom associated with hypoglycemia or a blood glucose reading ≤70 mg/dL even if it was not associated with signs or symptoms.

Subjects were assigned to either the Q1D algorithm or Q3D algorithm, which were developed based on the pharmacokinetic/pharmacodynamics properties of lispro, where the first dose of insulin lispro was administered before the subject’s first meal of the day (prebreakfast). If the patient did not eat breakfast (he/she consumed only water, black coffee with no sugar or cream, or noncaloric drink) the individual began with the prelunch dose. The lispro dose started at 10% of the total daily GLA dose. If necessary, the investigators sequentially added bolus lispro injections at subsequent meals (prelunch followed by predinner) for a maximum of three mealtime injections per day. Subjects followed the algorithms using especially created logbooks. The algorithms were designed to titrate independent of the subjects’ food intake or carbohydrate counting to simplify the dosing of mealtime insulin.

The Q1D algorithm was self-titrated every day based on premeal glucose readings from the previous day; for example, when adjusting the prebreakfast dose, subjects used their prelunch readings from the previous day. The premeal target blood glucose was 85–114 mg/dL. If this target was not achieved, the subject increased the dose by 1 unit/day until the target was reached. If the subject had a blood glucose reading of 56–84 mg/dL, the dose was decreased by 1 unit, and if the subject had a reading of <56 mg/dL, the dose was decreased by 2 units.

The Q3D algorithm was self-titrated every 3 days based on the median blood glucose readings from the 3 days before: to adjust the prebreakfast dose, the subject used the median prelunch blood glucose reading from the past 3 days. If the median reading was 85–114 mg/dL, there was no change in insulin lispro dose; if the median was 115–144 mg/dL, the subject increased the dose by 2 units; if the median was ≥145 mg/dL, the dose was increased by 4 units; if the median was 56–84 mg/dL, the dose was decreased by 2 units; and if the median was <56 mg/dL, the dose was decreased by 4 units.

The primary efficacy measure, the HbA_1c change from baseline to the end of the study (week 24 after randomization), was compared between Q3D and Q1D algorithms. Secondary outcome measures included incidence and annualized rate of self-reported total, severe, and nocturnal hypoglycemia. Additional secondary outcome measures included proportion of subjects achieving the target HbA_1c of ≤7.0% (53.0 mmol/mol), change in FBG, 7-point self-monitored blood glucose (SMBG) profile, weight change from baseline, dose of basal (GLA), prandial (lispro) insulin at the end of the study, and change in 1,5-anhydroglucitol (1,5-AG), a marker of hyperglycemia, particularly in the postprandial state and is useful in assessing glycemic control (18). In addition, change in HbA_1c, hypoglycemia (incidence and rate), FBG, and proportion of subjects achieving target were compared between the two algorithms in the subjects ≥65 years of age.

The sample size calculation was based on the primary outcome: change in HbA_1c from baseline to week 24. It was estimated that 640 completers would provide ∼98% probability of reaching a conclusive outcome using the classification method, assuming an SD of 1.1% (12.0 mmol/mol), no treatment difference, and a noninferiority margin of 0.4% (4.7 mmol/mol) (19). The early dropout rate was monitored, and to be conservative, the enrollment was continued to enroll 1,096 subjects (548 in each study) to reach the target number of completers.

Because there was no prior projection, preference, or historical evidence on which self-titration algorithm performs better, a classification method was applied to the analysis of the primary efficacy measure (19).

All safety outcomes were assessed in the entire randomized population (all subjects who entered the study completed the GLA optimization lead-in period [if applicable] and were randomized to one of the two treatment arms). All efficacy analyses were based upon the full analysis set (subjects in the all-randomized population who took at least one dose of insulin lispro). A sensitivity analysis was conducted for the primary efficacy measure based upon the all-completer population.

All efficacy and safety analyses were conducted at an α-level of 0.05. All CIs were computed as two-tailed using a 95% significance level. Continuous efficacy and safety variables measured repeatedly were evaluated using a mixed-model, repeated-measure (MMRM) approach using the restricted maximum likelihood method, including the following independent variables: fixed effects for treatment algorithm, all stratification variables, visit, treatment-by-visit interaction, and baseline outcome variable as the covariate (20). Treatment–by–age-group (≥65 years and <65 years) interaction for the change in HbA_1c was tested using another MMRM model with additional items including subgroup and subgroup-by-treatment algorithm interaction. For categorical measures, including adverse events and hypoglycemia incidence, Fisher exact test or Pearson χ² test was used. The hypoglycemia incidence was also analyzed with a logistic model with terms for treatment algorithm and all stratification variables as sensitivity analysis. The rate of total, nocturnal, and severe hypoglycemia per year during the treatment phase was analyzed using last observation carried forward (LOCF) applying a negative binomial model with terms for treatment algorithm, HbA_1c stratum, sulfonylurea/meglitinide use, and the logarithm of the exposure time (in days) as an offset variable and compound symmetry as variance-covariance structure (21,22). A Wilcoxon rank-sum test was conducted as a sensitivity analysis. The percentages of subjects achieving HbA_1c targets at the end of the study (LOCF) were analyzed using a logistic regression model with terms for treatment algorithm and strata. All data are expressed as least square mean (LSM) ± SE unless otherwise stated, and a P value of <0.05 was considered significant.

The patient disposition was based on all randomized subjects. After screening and the lead-in period, 1,112 subjects were allocated separately into study A (531 subjects) and study B (581 subjects) (Supplementary Data). Percentages of discontinued subjects and reasons for withdrawal were similar comparing treatment algorithms in both studies (study A: Q1D 16.8%, Q3D 20.2%; study B: Q1D 15.6%, Q3D 17.5%). The percentages of subjects on specific OAD regimens were equivalent: ∼89% of subjects were taking biguanides, and approximately ∼43% were taking two or more OADs. There were no statistical differences between the treatment algorithms in either study A or B regarding baseline demographics (Table 1).

At weeks 7, 12, and 24, there were significant decreases in HbA_1c from baseline for both Q1D and Q3D algorithms in studies A and B (Fig. 1A and B). The 95% CIs for the LSM difference from both studies were within the interval (–0.4% to 0.4% [–4.4 mmol/mol to 4.4 mmol/mol]) and contain 0% (0 mmol/mol] (i.e., Q3D was noninferior to Q1D, Q1D was noninferior to Q3D, and neither was superior to the other), indicating that Q1D and Q3D were clinically equivalent (Fig. 1A and B). The all-completer population concluded the same outcome. No statistically significant two-way interaction (treatment by age-group) was evident for the change in HbA_1c (study A: P = 0.656; study B: P = 0.364). There was no difference in treatment effect among those taking sulfonylureas or meglitinides, prior to randomization, and those not taking these medications.

The overall percentage of subjects reaching the goal HbA_1c of ≤7.0% (53.0 mmol/mol) at the end of the study (LOCF) was not statistically significantly different between the Q3D (study A = 42.5%; study B = 42.4%) and the Q1D adjustment (study A = 49.8%; study B = 49.3%) for study A (odds ratio 0.75; 95% CI 0.52–1.09; P = 0.128) and study B (odds ratio 0.77; 95% CI 0.53–1.11; P = 0.162). Similarly, there was no statistical difference in the percentage of subjects ≥65 years of age reaching target in study A between Q3D (58.0%) and Q1D (58.5%; odds ratio 1.17; 95% CI 0.52–2.67; P = 0.701). The percentage of subjects ≥65 years of age reaching target in study B was significantly lower for those randomized to Q3D algorithm (46.2%) than to the Q1D algorithm (67.9%; odds ratio 0.32; 95% CI 0.13–0.80; P = 0.015); however, it is notable that four subjects started with an HbA_1c ≤7.0% (53.0 mmol/mol) in the Q1D group compared with none in the Q3D group.

There was a significant decrease from baseline to week 24 in the 7-point SMBG profile at all time points except for morning premeal values in both studies A and B for both algorithms (Fig. 2). There was no statistical difference in the change from baseline in 7-point SMBG between Q3D and Q1D in study A. In study B, there was a significantly greater decrease from baseline to week 24 in blood glucose concentrations in subjects using Q1D than in those using Q3D at midday premeal (LSM Q3D–Q1D 95% CI 0.1–12.3; P = 0.045), bedtime (LSM Q3D–Q1D 95% CI 1.7–20.6; P = 0.020), and 0300 h (LSM Q3D–Q1D 95% CI 0.5–17.5; P = 0.037) (Fig. 2). The change in FBG at week 24 was not significant in study A: For Q1D, the LSM change from baseline was 1.4 ± 4.0 mg/dL, and for Q3D, it was 6.6 ± 4.1 mg/dL, with no difference between algorithms (P = 0.238). There was a significant difference in the change from baseline to week 24 in FBG between Q3D (8.0 ± 3.7 mg/dL) and Q1D (−6.5 ± 3.8 mg/dL) in study B (P = 0.002). The change in FBG from baseline in subjects ≥65 years of age was not statistically different (P = 0.242) between Q3D (18.1 ± 8.2 mg/dL) and Q1D (7.5 ± 7.9 mg/dL) in study A. The change in FBG from baseline in subjects ≥65 years of age was not statistically different (P = 0.082) between Q3D (17.2 ± 8.2 mg/dL) and Q1D (−3.1 ± 9.1 mg/dL) in study B.

There was a significant increase at week 24 from baseline in 1,5-AG levels (µg/mL) in both Q1D and Q3D in study A, as well as for study B. In addition, there was no statistical difference in the change from baseline in 1,5-AG levels in both study A and B between Q3D and Q1D (Fig. 1C and D).

There was no difference in baseline body weight in either study A or study B between Q3D and Q1D treatment algorithms (Table 1). In both studies, subjects gained weight from baseline regardless of titration algorithm. Subjects in study A using the Q3D algorithm gained more weight from baseline than subjects using the Q1D algorithm (3.0 ± 0.3 vs. 2.2 ± 0.3 kg; P = 0.014), while there was no difference in weight gain between Q3D (2.0 ± 0.2 kg) and Q1D approaches (2.5 ± 0.2 kg) in study B (P = 0.108).

In study A and study B, GLA doses at week 24 were not statistically different between Q3D and Q1D algorithms, and the GLA doses were stable throughout the 24-week treatment. There was no significant difference between treatment algorithms in insulin lispro dose at week 24 of either study. The percentages of basal and bolus doses for the TDD in study A were as follows: Q1D basal –58.2%, bolus –41.8%; Q3D basal –53.8%, bolus –46.2%. The percentages of basal and bolus doses for the TDD in study B were as follows: Q1D basal –57.4%, bolus –42.6%; Q3D basal –57.2%, bolus –42.8%. Approximately 61% of subjects required two or fewer injections (Table 1).

In the overall subject population and those ≥65 years of age, the incidences and annualized rates of overall, nocturnal, and severe hypoglycemic episodes during the treatment phase (LOCF) were not statistically different between treatment algorithms Q3D and Q1D in either study (Table 2). There was no difference in the rate of hypoglycemia in those subjects taking sulfonylureas or meglitinides prior to randomization compared with those not taking these medications.

The incidence of serious adverse events in study A was similar between Q1D (n = 18 [6.7%]) and Q3D (n = 12 [4.6%]). The incidences of serious adverse events in study B were similar in Q1D (n = 21 [7.3%]) and Q3D (n = 25 [8.6%]) (Supplementary Data).

AUTONOMY uniquely demonstrates the comparable effectiveness of two unique self-titration algorithms when mealtime insulin lispro is added to appropriately optimized GLA. AUTONOMY addressed the need for approaches to escalate prandial insulin treatment in patients with type 2 diabetes in a real-world setting. Both patient-driven algorithms (Q1D and Q3D) demonstrated statistically significant and clinically equivalent reductions in HbA_1c, significant increases in 1,5-AG, and improved 7-point SMBG profiles in studies A and B. By implementing either algorithm, ∼50% of subjects, who had previously failed to reach goal HbA_1c of ≤7.0% (53.0 mmol/mol) with basal insulin optimization plus OADs, achieved the American Diabetes Association goals for glycemic control with less glucose variability. Moreover, the sequential addition of prandial insulin lispro injections resulted in ∼61% of subjects only requiring two or fewer doses rather than a full basal-bolus regimen, which simplifies treatment and could enhance therapy compliance.

The improved metabolic control with the initiation and escalation of lispro, regardless of titration algorithm, was accomplished with low incidences and rates of nocturnal and severe hypoglycemia in both the overall study population and the elderly subgroup (≥65 years of age). The efficacy and safety of treatment intensification in the elderly is critical because of the aging population and higher prevalence of type 2 diabetes in this group (23). These findings are consistent with those of the A₁chieve study and show that a basal-bolus therapy can be initiated in the elderly without increased risk of hypoglycemia (24). Moreover, in a pooled analysis, Lee et al. (25) described that adding insulin GLA in an elderly patient population had low rates of hypoglycemia with decreases in HbA_1c similar to those in younger patients; however, AUTONOMY expanded on these results to demonstrate that therapy intensification and self-titration, starting with a single prandial dose of lispro, can safely occur in the elderly.

In support of the current study, results from the Self-Titration With Apidra to Reach Target (START) and FullSTEP trials demonstrated similar decreases in HbA_1c when a basal-bolus regimen was initiated in patients inadequately controlled on basal insulin plus OADs (26,27). The START Study showed that similar glycemic control can be achieved by patients using a breakfast preprandial insulin titration approach compared with a physician-managed strategy (26). The FullSTEP study demonstrated that a stepwise insulin approach resulted in greater patient treatment satisfaction with fewer hypoglycemic events than a full basal-bolus regimen (27). However, the study initiated the first prandial insulin dose at the largest meal, the prandial dose was adjusted by the study investigators, and both AUTONOMY and START began the prandial therapy at the first meal of the day. Nevertheless, AUTONOMY used two self-titration algorithms, initiated the prandial dose based on a percentage of the total basal dose, and added other mealtime doses as necessary.

The majority of diabetes management is performed in a generalist setting, in which substantial clinical inertia—the failure to intensify treatment—exists (28–30). Two retrospective studies in the U.K. determined that patients with suboptimal glycemic control remained poorly controlled for >7 years before insulin treatment initiation, with a mean HbA_1c of ∼9.0–10.0% (74.9–85.8 mmol/mol) (31,32). In AUTONOMY, the average duration of diabetes at entry was 12 years, further supporting that this inertia prevents early glycemic control and timely treatment intensification with exogenous insulin. A 10-year follow-up of the UKPDS showed that a legacy effect exists from early intensive glycemic control, reducing the long-term risk for cardiovascular complications associated with type 2 diabetes (33). The complexity of therapy with multiple medications, the fear of hypoglycemia, and weight gain are major barriers to treatment intensification, especially with insulin (34–37). AUTONOMY demonstrated that prandial insulin can be initiated in an adult population, including the elderly, to lower HbA_1c and limit mealtime glucose excursions safely, with either patient-driven algorithm, in the endocrinology and generalist setting. Using Q1D and Q3D algorithms simplified insulin therapy by not requiring patient training on carbohydrate counting or insulin correction factor and reduced the number of OADs in those treated with sulfonylurea or meglitinide. In 2009, Oyer et al. (38) reported on the self-titration of twice-daily biphasic insulin in insulin-naïve patients with type 2 diabetes. While AUTONOMY further supports the concept of self-titration, it does this, in those already optimized on basal insulin, with a basal-bolus algorithm traditionally considered more complex. Subjects gained 2–3 kg of weight, regardless of treatment algorithm, with the initiation of prandial insulin; however, a previous study determined that treatment satisfaction increased and regimen-related distress decreased with the addition of rapid-acting insulin analogs to basal insulin despite any weight gain as a side effect (37). This simple patient-centric approach has the potential to empower patients and to limit barriers to achieve glycemic goals while improving treatment satisfaction.

A limitation of this study was the exclusion of subjects with BMIs ≥45 kg/m², which, with the growing health care burden associated with obesity, could be an important study population. Future research needs to address whether the safety and efficacy proven with applying either of the self-titration algorithms is applicable to Asian populations because no Asian countries were included; however, a large and multinational sample population was studied. Although numerical differences were observed, which seem to benefit the use of Q1D versus Q3D, these were not statistically significant and only should be considered hypothesis generating. Furthermore, the AUTONOMY algorithms were based on pharmacokinetic/pharmacodynamics modeling of GLA and lispro insulins; there may not be substantial differences with the use of other short-acting prandial insulin analogs. As in other insulin trials, the risk of hypoglycemia increased when subjects were initiated on a prandial insulin regimen (39,40). Although this trial investigated the use of basal-bolus as an approach to controlling postprandial glucose excursions, other options, such as the combination of GLP-1 receptor agonists with insulin, may be considered (41).

In summary, the AUTONOMY trial provides novel data and the basis for the initiation and escalation of lispro therapy using two simple, self-titration regimens in patients with type 2 diabetes who failed to achieve adequate glycemic control on appropriately titrated basal insulin plus OADs. The trial demonstrated that a basal-bolus regimen can effectively and safely be initiated in the endocrinology and generalist settings by empowering patients to self-titrate their bolus insulin in order to achieve glycemic goals with less glucose variability and low rates of nocturnal and severe hypoglycemia.

Acknowledgments. The authors thank Deborah Wimberley (Eli Lilly and Company) for the management of the trial and William Huster (Eli Lilly and Company), Yongming Qu (Eli Lilly and Company), Rong Qi (Eli Lilly and Company), Chunxue Shi (Ventiv Health), and Cheng Shao (Ventiv Health) for statistical support. Additionally, the authors thank Jeff Bonner (Eli Lilly and Company) for support and assistance in writing and preparing the manuscript.

Duality of Interest. S.V.E. serves on an advisory board for Eli Lilly and is on a Lilly speakers’ board. S.V.E. has advised for Tandem, Merck, Boehringer Ingelheim, Bristol-Myers Squibb, Dexcom, Novo Nordisk, Sanofi, and Abbott. The research was supported by Eli Lilly and Company. R.L., J.J., and L.C.G. are full-time employees of Eli Lilly and Company and are also minor stock owners as part of an employee offering program. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. S.V.E. analyzed and interpreted data, reviewed and edited the manuscript, contributed to the discussion, and confirmed final approval. R.L., J.J., and L.C.G. contributed to the design of the study, analyzed and interpreted data, reviewed and edited the manuscript, contributed to the discussion, and confirmed final approval. L.C.G. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the 49th Annual Meeting of the European Association for the Study of Diabetes, Barcelona, Spain, 23–27 September 2013, and at the 73rd Scientific Sessions of the American Diabetes Association, Chicago, IL, 21–25 June 2013.