ability to provide morning near-normoglycemia and as a means to predict initial insulin requirements in NIDDM. Twenty-seven pa- tients with poorly controlled ...
0021-972X/97/$03.00/0 Journal of Clinical Endocrinology and Metabolism Copyright © 1997 by The Endocrine Society
Vol. 82, No. 8 Printed in U.S.A.
An Overnight Insulin Infusion Algorithm Provides Morning Normoglycemia and Can Be Used to Predict Insulin Requirements in Noninsulin-Dependent Diabetes Mellitus* CATHERINE S. MAO, MARY ELIZABETH RIEGELHUTH, DEBORAH VAN GUNDY, COLLEEN CORTEZ, SUNNY MELENDEZ, AND ELI IPP Departments of Medicine and Pediatrics, Harbor-University of California-Los Angeles Medical Center, Torrance, California 90509 ABSTRACT Initial insulin requirements in noninsulin-dependent diabetes mellitus (NIDDM) are difficult to estimate because of individual variability in insulin sensitivity and secretion. We evaluated a simple, nurse-managed algorithm for overnight delivery of insulin, for its ability to provide morning near-normoglycemia and as a means to predict initial insulin requirements in NIDDM. Twenty-seven patients with poorly controlled NIDDM were studied on 30 occasions. A 12-h iv insulin infusion was begun at 2000 h, and bedside blood glucose concentrations were measured at hourly intervals. The rate of insulin infusion was adjusted according to blood glucose levels. We estimated the preprandial insulin dose requirement for the following day in 16 patients based on overnight insulin requirements to maintain normoglycemia. Preprandial insulin doses were adjusted for pre-
vailing blood glucose concentrations. At 2000 h, the mean (6SEM) blood glucose concentration was 265.7 6 10.8; at 0300 h, it was 122.8 6 3.4; and at 0700 h, it was 123.8 6 5.1 mg/dL. On the next day, mean blood glucose levels (before and 2 h after a meal) were: breakfast, 102.5 6 5.9 and 177.3 6 19.2; lunch, 138.9 6 15.5 and 136.3 6 11.4; dinner, 105.7 6 7.2 and 178.1 6 15.7 mg/dL. There was no significant difference between mean calculated and administered total insulin dosage the next day (84.2 6 7.0 vs. 78.2 6 8.2 U). Thus, a weight-based algorithm for iv insulin infusion induced near-normoglycemia in NIDDM and successfully predicted the insulin dose requirement. We conclude that initiating insulin therapy in NIDDM patients can be achieved rapidly and efficiently based on a nurse-managed overnight insulin infusion. (J Clin Endocrinol Metab 82: 2466 –2470, 1997)
cess, in which low doses of insulin are used initially, and increasing quantities are introduced as required to improve glucose control. When there is a need for more rapid induction of tight control, as during hospital admission for surgery, there are few reported methods to induce normoglycemia effectively and safely in patients with NIDDM. In insulin-dependent diabetes mellitus, methods have been developed to induce normoglycemia and also estimate initial insulin requirements (3–5), but to our knowledge no rapid and accurate methods have yet been established for patients with NIDDM. A previous study (6) that included some patients with NIDDM, induced and maintained normoglycemia, but did not attempt to predict subsequent sc insulin requirements. This study was designed to develop a feedback-controlled insulin delivery system for patients with NIDDM to induce normoglycemia and also to determine whether the overnight response to insulin can help estimate the insulin requirement for the following day. The insulin infusion algorithm used in this protocol was adapted from the algorithm used by White et al. (3), designed for lean patients with insulin-dependent diabetes.
ONINSULIN-DEPENDENT diabetes mellitus (NIDDM) is a disease process most commonly treated with diet and oral antidiabetic agents (1). When this approach fails, the next step to achieve improved glucose control is insulin administration. When insulin therapy is indicated, blood glucose in these patients is usually out of control, and attempts to improve glucose control are based on incremental adjustments of insulin in response to blood glucose measurements. The estimation of exact dose requirements in NIDDM is difficult. This is due to the fact that insulin replacement doses are based on multiple factors, particularly sensitivity to insulin action and the patient’s endogenous insulin-secreting capacity. In any individual patient, these factors are not easy to estimate clinically. Although the weight of the patient is an important factor in determining insulin resistance (2) and is often used as a basis for estimating insulin requirements, clinical experience indicates that precise insulin requirements cannot be predicted based upon weight alone. Initiation of insulin therapy is, therefore, an empirical proReceived February 25, 1997. Revision received April 25, 1997. Accepted April 30, 1997. Address all correspondence and requests for reprints to: Eli Ipp, M.D., Harbor-University of California-Los Angeles Medical Center, 1000 West Carson Street, Torrance, California 90509-2910. * This work was supported in part by grants from the NIH to the General Clinical Research Center at Harbor-University of California-Los Angeles Medical Center (MO1-RR00425), the American Diabetes Association, and Pfizer-Roerig.
Subjects and Methods Patients Twenty-seven subjects with poorly controlled NIDDM were studied. As wide a spectrum of patients as possible was entered into the study. Their age range was 29 – 64 yr, and the duration of their diabetes was 1–19 yr. Seventeen were women, and 10 were men. The mean body mass
INSULIN INFUSION ALGORITHM FOR NIDDM index was 30.9 6 0.9 kg/m2 (range, 22.8 – 44.8 kg/m2). The mean fasting blood glucose level was 265.7 6 10.8 mg/dL. Twenty-one patients were taking oral hypoglycemic agents, three patients were taking insulin, and three were new-onset, untreated patients. The three patients taking insulin had been treated with oral hypoglycemic agents for many years and were recently switched to insulin due to poor glucose control. All studies were performed at the Clinical Research Center at Harbor-University of California-Los Angeles Medical Center (Torrance, CA). All patients gave their informed consent to participate in these studies, and they were approved by the institutional review board. Patients who were in a state of acute metabolic distress, had an active infection, or had a recent history of symptomatic coronary artery disease or renal or hepatic insufficiency were excluded from the study.
Procedures Overnight insulin infusion. Thirty studies were performed in the 27 subjects. The studies were begun at 2000 h with the initiation of a 12-h iv insulin infusion using a nurse-managed algorithm. Blood glucose was measured at 60-min intervals by glucometer (Ames Division, Miles Laboratories, Elkhart, IN). The iv insulin infusion rate was adjusted by the nursing staff according to an algorithm based on the blood glucose level (Table 1). The goal of the algorithm was to induce and maintain near-normal blood glucose concentrations by the following morning (i.e. 90 –130 mg/dL). A euglycemic period was defined as the period that began when blood glucose concentrations first fell below 130 mg/dL and ended with the end of the infusion. Both insulin infusion rate and frequency of blood sampling were varied in response to the blood glucose concentration. Blood glucose sampling was performed at hourly intervals until the blood glucose level fell below 130 mg/dL. After this, blood sampling continued at 60-min intervals as long as glucose measurements remained between 90 –130 mg/dL. If the glucose concentration was less than 90 mg/dL or greater than 130 mg/dL, sampling frequency was switched to 30-min intervals. The increased frequency of sampling permitted a more rapid response to glucose concentrations that were outside of the targeted euglycemic range. In three subjects, the overnight insulin infusion was repeated because the algorithm (Table 1) failed to reduce blood glucose into the target range. The repeat infusion used an adjusted algorithm that provided increased insulin at each level of glucose measurement. Preprandial insulin treatment. The overnight insulin infusion algorithm was tested on 11 subjects. Once it was clear that this protocol was successfully achieving its goal, we attempted to predict the insulin dose required to control blood glucose on the following day. This was performed in another 16 patients. Calculations were based upon the basal insulin requirements necessary to maintain euglycemia (,130 mg/dL) during the overnight insulin infusion, i.e. the euglycemic insulin requirement (EIR). This was determined from the following parameters: TABLE 1. Algorithm for insulin infusion: weight: (place 50 U regular insulin in 1 L normal saline). Glucose (mg/dL)
.250 201–250 171–200 141–170 121–140 101–120 81–100 61– 80 ,60
Insulin infusion rate (mL/kg z h)
iv infusion rate mL/h
70 60 45 30 25 18 15 10 0
A nurse-managed iv insulin infusion algorithm based on the blood glucose level was designed to induce and maintain near-normal blood glucose concentrations by the following morning. (Blood glucose sampling is performed at hourly intervals until the blood glucose level falls below 130 mg/dL. After this, blood sampling continues at 60-min intervals unless glucose concentrations are less than 90 mg/dL or more than 130 mg/dL; then sampling occurs at 30-min intervals until glucose is back in the desired range.)
1) the duration of the euglycemic period (hours), i.e. measured from the time that blood glucose first dropped below 130 mg/dL; 2) the sum of the hourly rates of insulin infusion during the euglycemic period as defined in the first parameter (the hourly rates of insulin infusion are obtained from the chart shown in Table 1; when their sum is divided by the duration of the euglycemic period, this is the EIR, and is expressed as units per h); and 3) the basal insulin requirement (BIR), expressed as units per 12 h, was then estimated using the data obtained in the second parameter and multiplied by 12. Insulin requirements for the following day were then calculated. The estimated BIR provided the total amount of insulin required if the patient had been maintained at euglycemia from 2 h after dinner (2000 h) until breakfast (0800 h). The 12-h BIR was then used to calculate the total insulin due for the following day; BIR was assumed to constitute 20% of total 24-h insulin needs (7). Insulin need for the next 12 h thus would constitute 80% of the total 24-h insulin requirement and was split into three preprandial doses of regular insulin. Preprandial insulin doses for breakfast, lunch, and dinner were calculated by splitting the total daytime dose into the following ratio: 40:30:30. Insulin was injected 30 min before each meal. The prebreakfast dose of insulin was administered 30 min before the infusion was stopped. Preprandial insulin doses obtained by the above calculation were designated the predicted dose and were the primary guide for dose calculation. The predicted dose was adjusted only if the premeal blood glucose level did not reach a target range of 80 –120 mg/dL.
Analytical procedures Blood glucose measurements at the bedside were performed by trained staff using a glucometer device. The accuracy of these measurements was verified initially by comparing the glucometer measurements with plasma samples obtained at 3-h intervals in early studies (mean plasma glucose, 155.4 6 10.6 mg/dL; corresponding mean blood glucose concentration, 156.3 6 9.6 mg/dL; n 5 55 paired glucose samples; correlation: y 5 1.09x 2 0.80; r 5 0.82; P , 0.0001). Plasma glucose was measured using the hexokinase method with an Abbott autoanalyzer (Abbott, North Chicago, IL) (8). Statistical analysis was performed using Student’s t test for paired samples. Data are presented as the mean 6 sem.
Results Effect of overnight insulin infusion
Figure 1 demonstrates the effect of the overnight insulin infusion on blood glucose concentrations during all 30 infusions. The mean blood glucose at 2000 h was 265.7 6 10.8 mg/dL. The mean blood glucose concentration fell to 129.6 6 5.1 mg/dL by the fifth hour and remained at a plateau within a range of 120.0 6 3.5 to 129.3 6 4.4 mg/dL throughout the night. When the infusion was discontinued, the mean blood glucose concentration was 123.8 6 5.1 mg/dL. Figure 1 also displays the mean insulin infusion rates administered during this protocol. The infusion was started at 2000 h at a mean rate of 63.2 6 2.7 mU/kgzh. The mean infusion rate decreased along with glucose concentrations, reaching a plateau after 5 h. The mean infusion rate at the fifth hour was 23.7 6 1.7 mU/kgzh and remained between 20.7 6 1.4 and 23.9 6 1.5 mU/kgzh until the following morning. At the time the infusion was discontinued, the mean insulin infusion rate was 22.6 6 2.4 mU/kgzh. Efficacy of the overnight insulin infusion
Mean blood glucose concentrations reached near-normoglycemic levels (130 mg/dL or less by glucometer) by 0100 h with the insulin infusion algorithm (Fig. 1) and were maintained at near-normoglycemic levels between 0100 – 0800 h. The range of blood glucose concentrations observed in in-
MAO ET AL.
JCE & M • 1997 Vol 82 • No 8
FIG. 1. The effect of the overnight insulin infusion on blood glucose concentrations (n 5 30). The mean (6SEM) blood glucose is shown in the lower panel, and the mean insulin infusion rates administered during this protocol are shown in the top panel.
dividual subjects during this time was between 69 –190 mg/ dL. Seventy-nine percent of the blood glucose readings during the euglycemic period were within the range of 90 –130 mg/dL. Of the 21% of measurements that remained outside this range, 8% were below the target range (,90 mg/dL), and 13% were above it (.130 mg/dL). The total number of infusions was 30. In 27 of those infusions, glucose concentrations reached the target euglycemic range of below 130 mg/dL. On 3 occasions, the infusion failed to bring the blood glucose into the target range. In these patients, we repeated the infusion the following evening using an adjusted algorithm to provide an increased insulin dosage at each level of glucose measurement. This successfully reduced plasma glucose levels to below 130 mg/dL (data not included). The 3 patients who failed the initial insulin infusion did not differ from other subjects with respect to body mass index, sex, age, duration of diabetes, or initial fasting blood glucose level. Effect of predicted sc insulin doses on blood glucose concentrations
Sixteen patients received sc insulin injections the following day in an attempt to test the predicted insulin doses using the results of the infusion algorithm. Figure 2A demonstrates the effect of administered preprandial insulin doses on blood glucose concentrations. Mean blood glucose concentrations before breakfast, lunch, and dinner were 102.5 6 5.9, 138.9 6 15.5, and 105.7 6 7.2 mg/dL, respectively. Mean 2 h postprandial glucose concentrations for breakfast, lunch, and dinner were 177.3 6 19.2, 136.3 6 11.4, and 178.1 6 15.7 mg/dL, respectively. Our goal was not to produce extremely
FIG. 2. The effect of preprandial insulin doses on blood glucose concentrations in 16 patients who received sc insulin on the day after the overnight insulin infusion. In the lower panel, the mean blood glucose concentrations before and 2 h after breakfast, lunch, and dinner are shown. In the top panel, the mean administered preprandial insulin doses for breakfast, lunch, and dinner are compared to the mean predicted preprandial insulin doses.
tight control. As a result, of 96 glucometer measurements taken before and after meals, only 1 was less than 60 mg/dL. Comparison between administered and predicted insulin dose
The mean total sc insulin dose administered was 78.2 6 8.2 U compared with an estimated mean total requirement of 84.2 6 7.0 U. The administered insulin represented an average requirement of 0.9 6 0.07 U/kg. In Fig. 2B, the mean predicted preprandial insulin doses for breakfast, lunch, and dinner are compared to the mean administered preprandial insulin doses. The mean predicted preprandial insulin doses, calculated from data derived from the insulin infusion, for breakfast, lunch, and dinner were 33.7 6 2.8, 25.6 6 2.2, and 24.9 6 2.1 U, respectively. Administered doses of insulin were dependent upon the clinical situation at the time of treatment and were not necessarily identical with the predicted dose. Actual doses of insulin administered were based upon the predicted dose, but were modified depending upon the prevailing preprandial blood glucose measurements and also the response to regular insulin during the previous meal. Mean administered preprandial insulin doses for breakfast, lunch, and dinner were 33.6 6 2.9, 24.4 6 3.4, and 20.2 6 2.6 U, respectively. Statistical analysis revealed no significant
INSULIN INFUSION ALGORITHM FOR NIDDM
difference between administered and predicted insulin doses for breakfast (P 5 0.85) or lunch (P 5 0.57). However, there was a significant difference between administered and predicted insulin doses for dinner (P 5 0.004). The last finding was the result of a consistent reduction of predinner insulin doses from the predicted dose due to the appearance of relatively low postprandial glucose responses to insulin at lunch. Figure 3 displays the correlation between administered and predicted preprandial insulin doses for each individual dose on the day after the overnight insulin infusion (n 5 48 doses; y 5 1.1x 2 5.4; r 5 0.88; P , 0.001). Safety concerns
The possibility of hypoglycemia was of great concern, but only 34 of 438 measurements (7.8% during the entire protocol, including the insulin infusion and next day sc insulin injection) were below 90 mg/dL, and only 2 were below 60 mg/dL (,0.5%). One of the hypoglycemic episodes occurred during insulin infusion, and the infusion was discontinued for 30 – 60 min. Therapy (orange juice) was instituted in both instances. Discussion
We have developed a clinically useful methodology to reliably and safely initiate insulin therapy in patients with NIDDM. The algorithm used in the present study rapidly established normoglycemia in 90% of patients with type II diabetes studied in this project and maintained euglycemia without causing significant hypoglycemia. In addition, the insulin dosage required to maintain overnight euglycemia was used successfully to predict sc regular insulin requirements on the following day. This simple combined approach has not been previously described. Fairly rapid control of hyperglycemia is often required in patients with diabetes admitted to the hospital for surgery, other invasive procedures, or deteriorated blood glucose control. In these cases, a trial and error method or sliding scale is often used to obtain improved glucose control. There is no standardized approach to the problem. Previous algo-
FIG. 3. Correlation of the administered and the predicted preprandial insulin doses for the day after the overnight insulin infusion (n 5 48 doses; y 5 1.1x 2 5.4; r 5 0.88; P , 0.001).
rithm-based studies have involved mainly patients with type I diabetes (3, 9, 10). In the study by Mokan and Gerich (6), which also used a weight-based algorithm, only a small number of subjects with type II diabetes (9 of 29 subjects) were included. In contrast to that algorithm, ours used a lower rate of insulin infusion during the initial hyperglycemic phase. This probably accounts for the slower attainment of euglycemia (5 h vs. their 2– 4 h) and a lower incidence of hypoglycemia in our study. This algorithm appears to be useful in a relatively wide range of NIDDM, typical of many clinical situations. Because this is a weight-based algorithm, it was equally effective in both obese and lean type II diabetes mellitus patients. However, the initial overnight infusion was not successful in reducing blood glucose into the near-normoglycemia range in 3 of 30 subjects. This was probably due to greater insulin resistance in these 3 subjects, who had a body mass index that ranged between 28.6 –38 kg/m2. When we repeated the infusion on a subsequent night using an adjusted algorithm that provided higher doses, blood glucose reached nearnormoglycemia. Although the adjusted (increased dose) insulin algorithm is effective, we chose not to use it routinely in all patients because of an anticipated increased risk of hypoglycemia in more insulin-sensitive patients. Rather, a high dose insulin algorithm can be reserved for those patients that have shown themselves to have greater insulin resistance and, therefore, increased insulin requirements. We consider a 10% failure rate (to achieve near-normoglycemia) using this algorithm to be an excellent result, consistent with our attempt to achieve near-normoglycemia and minimize excessive hypoglycemia. During the induction of normoglycemia, blood glucose was sampled at 60-min intervals. Thirty-minute sampling is too frequent for routine clinical use, but was used initially until the protocol was shown to be safe. We adopted a 60-min sampling frequency for induction of normoglycemia when analysis of the data indicated that this frequency would be adequate. However, after correction of hyperglycemia (,130 mg/dL), we recommend the use of 30-min sampling when blood glucose strays out of range, below 90 mg/dL or above 130 mg/dL. This allows for a more rapid response to deviations from the desired range. We had only one episode of hypoglycemia during the establishment of euglycemia with this algorithm. In contrast, algorithms with higher insulin infusion rates and more frequent blood sampling (6) have induced hypoglycemia that necessitated iv glucose injection. The algorithm used in this study resulted in only two measurements less than 60 mg/dL, one during induction of normoglycemia and one the following day. Both of these episodes occurred in the same subject, and neither of the episodes required iv glucose injection. This algorithm has the added advantage of providing a basis for estimating insulin requirements for sc injection. The protocol was designed to calculate three preprandial regular insulin doses as an initial preparatory step, which would allow typical doses of medium and short acting insulin to be prescribed thereafter, but not tested here. The results indicate that overall daily insulin requirements were accurately predicted, although a small (but statistically significant) decrease in the insulin dose before dinner resulted in mean
JCE & M • 1997 Vol 82 • No 8
MAO ET AL.
postprandial glucose concentrations that were higher after dinner than after lunch. It is likely that if the insulin doses originally predicted were actually given, glucose values after dinner would have been even better controlled. We, therefore, conclude that estimation of insulin dosage for sc delivery is reasonably accurate and is also safe. However, there is an important caveat to the estimation of insulin requirements: the algorithm should not be used if the majority of glucose values during euglycemia are greater than 130 mg/ dL. The calculated EIR under these circumstances may overestimate insulin requirements for the following day. The mean overall glucose concentration for the next day was 137.1 6 6.1 mg/dL (n 5 16). This is reasonably good control [if maintained, it would be equivalent to a hemoglobin A1c level of 6.7% (11), in keeping with the recommendations of the American Diabetes Association (1)]. Thus, based on the glucose concentrations achieved the next day, the EIR appears to be a valid basis for calculation of subsequent insulin needs. The EIR reflects the insulin infusion rates necessary to maintain basal glucose concentrations at near-normal concentrations. These are influenced by both the prevailing insulin sensitivity and residual insulin secretion. The EIR, therefore, represents a composite of complex physiological information, which may explain its accuracy in predicting insulin requirements compared with that of other commonly used clinical parameters of insulin need, such as weight or body mass index. The poor predictive value of weight to estimate specific insulin requirements can be derived from our data. Had we used a simple per kg basis for initiating insulin requirements in these patients, using the mean sc dose administered (0.9 U/kg), many subjects would have been given inappropriate insulin doses. Two of 16 subject would have received an excess of 20 U or more, and 5 would have been undertreated by 20 U or more. It should be noted that the insulin doses used in this study were demonstrated to be effective for only 1 day. As a subject remains euglycemic over a longer period of time, insulin needs may also change due to a reduction in glucose toxicity (12). After the first day, the overnight insulin infusion can be repeated, or a dose of NPH insulin can be given to correspond with the overnight basal requirement (BIR) as calculated from the first night. Continued glucose monitoring is essential to further fine-tune the insulin dosage to the patient’s needs.
In summary, the availability of a safe and effective bedside algorithm to rapidly improve blood glucose is important in selected clinical situations faced by patients with NIDDM. This study demonstrates the feasibility of an algorithmdriven iv insulin infusion to normalize blood glucose in NIDDM that may be performed by nursing staff using a simple glucose meter and an infusion pump. This protocol can be used reliably and safely to induce near-normoglycemia overnight and can also be used to estimate sc insulin requirements. This methodology for estimating initial insulin needs to achieve good diabetes control appears to be safe and effective. Acknowledgments The authors are indebted to the nurses, dietary staff, and core laboratory technicians of the Clinical Study Center at Harbor-University of California-Los Angeles Medical Center for their excellent assistance in the performance of these studies.
References 1. ADA Consensus Statement. 1995 The pharmacological treatment of hyperglycemia in NIDDM. Diabetes Care. 18:1510 –1518. 2. Olefsky JM, Kolterman OG. 1981 Mechanism of insulin resistance in obesity and noninsulin-dependent (type II) diabetes. Am J Med. 70:151–168. 3. White NH, Skor D, Santiago JV. 1982 Practical closed-loop insulin delivery. A system for the maintenance of overnight euglycemia and the calculation of basal insulin requirements in insulin-dependent diabetes. Ann Intern Med. 97:210 –213. 4. Lambert AE, Buysschaert M, Lambotte L. 1979 Use of an artificial pancreas as a tool to determine subcutaneous insulin doses in juvenile diabetes. Diabetes Care. 2:256 –264. 5. Irsigler K, Kritz H, Kaspar L, Brandle J, Franetzki M. 1979 Use of glucosecontrolled insulin infusion system for improvement of subcutaneous insulin regimen. Horm Metab Res. 8(Suppl):134 –140. 6. Mokan M, Gerich JE. 1992 A simple insulin infusion algorithm for establishing and maintaining overnight near-normoglycemia in type I and type II diabetes. J Clin Endocrinol Metab. 74:943–945. 7. Schiffrin A, Belmonte M. 1981 Combined continuous subcutaneous insulin infusion and multiple subcutaneous injections in type I diabetic patients. Diabetes Care. 4:595– 600. 8. Genter PM, Ipp E. 1994 Accuracy of plasma glucose measurements in the hypoglycemic range. Diabetes Care. 17:595–598. 9. Clarke W, Haymond M, Santiago J. 1980 Overnight basal insulin requirements in fasting insulin-dependent diabetics. Diabetes. 29:78 – 80. 10. De Feo P, Perriello G, Ventura M, et al. 1986 Studies on overnight insulin requirements and metabolic clearance rate of insulin in normal and diabetic man: relevance to the pathogenesis of the dawn phenomenon. Diabetologia. 29:475– 480. 11. Nathan DM, Singer DE, Hurxthal K, Goodson JD. 1984 The clinical information value of the glycosylated hemoglobin assay. N Engl J Med. 310:341–346. 12. Leahy JL, Bonner-Weir S, Weir GC. 1992 Beta-cell dysfunction induced by chronic hyperglycemia. Current ideas on mechanism of impaired glucoseinduced insulin secretion. Diabetes Care. 15:442– 455.