Improving the Estimation of Mealtime Insulin Dose in Adults With Type 1 Diabetes: The Normal Insulin Demand for Dose Adjustment (NIDDA) study

We recruited adults with type 1 diabetes on insulin pump therapy from the patient population attending the Sydney insulin pump clinic. Eligibility criteria included the following: aged between 18 and 70 years inclusive; type 1 diabetes diagnosed for ≥1 year; use of insulin pump therapy, including proficiency with use of a bolus dose calculator for ≥2 months; and HbA_1c ≤9.0% and reliably performing self-monitoring of blood glucose at least four times daily. Exclusion criteria included eating disorders and individuals using medication that may influence blood glucose. The protocol was approved by the Human Research Ethics Committee of the University of Sydney, and subjects gave informed consent.

The participant recruitment started 1 month before the commencement of testing. Patients were instructed by the credentialed diabetes educator (CDE) to perform routine self-monitoring of blood glucose during the run-in period to verify individual basal rates and insulin to carbohydrate ratio (ICR). One week before testing, the diabetes team at the Sydney Insulin Pump Clinic (CDE and clinical dietitian) conducted small group meetings (n = 2∼3 patients) to discuss detailed requirements of the study, assess carbohydrate counting skills, and review bolus-wizard settings on the basis of self-monitoring of blood glucose data and sophisticated online diabetes management software (Medtronic Diabetes CareLink Personal Therapy Management Software). The purpose of the meeting was to optimize fasting blood glucose levels and fine-tune ICR and basal rates. The patients were instructed to make every effort to eliminate confounding variables that may otherwise affect blood glucose variation, including avoiding unusual exercise and alcohol intake during the 4-day study period.

The study was carried out under supervision in a dedicated research room. During the test week, participants were required to attend the testing room on 4 half-days (Monday to Thursday). On Monday afternoon, the real-time continuous glucose monitoring system (CGMS) device (MiniMed Paradigm, Medtronic, Northridge, CA) was inserted and calibrated for each attended subject. Subjects were instructed to enter four finger-prick blood glucose levels per day into the monitor at a time when blood glucose levels were stable for calibration (i.e., time before meal or before bed). Standardized dinner meals with precise amounts and same type of carbohydrate were distributed to standardize the impact of the evening meals on morning glycemia. Measuring cups and a kitchen scale were supplied to facilitate adjustment of meal size and amount of carbohydrate. On Tuesday, Wednesday, and Thursday mornings, subjects attended the clinic in the fasted state and consumed the test breakfasts with pre-assigned insulin algorithms (Supplementary Fig. 1). The order of three test meal–bolus algorithm combinations was randomized using a random digit table.

Each participant consumed two different breakfast meals (meals A and B), devised to have equal energy and fiber content, glycemic index (GI), and calculated insulin demand on the basis of the weighted FII of the component foods, but meal A was designed to have double the carbohydrate content of meal B (Table 1). Meal A was consumed only once, with insulin dosing according to the subject’s usual ICR, as verified during the run-in week. Meal B was consumed on two occasions (B1 and B2): once using carbohydrate counting as the basis for the insulin dose (about half the dose of insulin used in meal A) and once using the FII algorithm (the same dose as meal A, since both meals have the same insulin demand [FII = 60]). In the context of the current study design, insulin dosing for meal A was identical whether based on carbohydrate counting or the FII algorithm (Supplementary Fig. 2). The test meals were prepared on the testing day before subjects arrived, and the component foods were weighed using a Tanita kitchen scale (model KD-160; Tanita Corporation, Arlington Heights, IL) (2 kg max/1 g resolution). All the component foods were prepared according to the same procedure used for testing the original FII values (11).

On each test day, participants arrived at the testing room 0800 h after a minimum 10-h overnight fast. Fasting blood glucose level was recorded. Rapid-acting insulin was administered immediately before meal consumption. Subjects consumed the test meal within 20 min and remained sedentary throughout the session. No additional food or drink (except water) was consumed during the 3-h postprandial period unless required to treat symptomatic hypoglycemia. Blood glucose levels were monitored using the real-time CGMS and finger-prick blood testing with a blood glucose monitor (HemoCue; HemoCue Limited, Angelhom, Sweden) at 0.5-h intervals over the 3-h test period. At the conclusion of the test (Thursday at noon), CGMS data were downloaded for analysis. The CDE and clinical dietitian monitored the patients throughout the test sessions, and hypoglycemia was treated and recorded if it occurred. On any given morning, if the preprandial blood glucose level was >13 or <4 mmol/L, the subject was treated and asked to return for an additional test day the following week.

A sample size of 31 patients provided 80% power to detect a difference in glucose incremental area under the curve (iAUC) of ∼220 mmol × min/L between carbohydrate counting and the FII algorithm at the 5% significance level, assuming a within-person SD of differences in iAUC of ∼400 mmol × min/L. The blood glucose profile during the 3-h postprandial period was analyzed using the SPSS statistical package (version 19.0; SPSS, Chicago, IL). A one-way repeated-measures ANOVA was used to analyze the following parameters between the comparable test conditions: 1) time within normal blood glucose level range (4.0∼10.0 mmol/L), 2) the iAUC as described previously (12), 3) peak blood glucose level excursion, 4) fasting blood glucose level, 5) time to peak blood glucose level, and 6) time to return to fasting blood glucose level. For significant variables, post hoc test comparisons of means were carried out using least significant differences without any adjustment for multiple comparisons. Differences in coefficients were considered statistically significant if P was < 0.05 and highly significant if P was < 0.01 (two-tailed). Cochran Q test was used to assess the differences in numbers of hypoglycemia episodes (defined as blood glucose level ≤3.9 mmol/L) between carbohydrate counting and the FII algorithm. Data following treated hypoglycemic episodes were excluded from analysis. Results are presented as means ± SD unless otherwise stated.

A total of 31 adults with type 1 diabetes on insulin pump therapy participated in the study. One participant failed to finish the three test sessions because of food intolerance. One participant had an inappropriate starting blood glucose level on the first test day and was not available to repeat this test session. One participant missed CGMS-derived blood glucose level data in one test session because of equipment failure. The results of 28 participants (11 men, 17 women) were available for analysis.

The mean age of participants was 37.8 ± 14.4 (range 21–62) years. The subjects had been diagnosed with type 1 diabetes for 19.6 ± 11.4 years and had been on insulin pump therapy for 2.7 ± 2.3 years. The mean HbA_1c was 7.8 ± 0.9%. The preprandial blood glucose level for each of the three test conditions was not significantly different (P = 0.41) (Table 2).

Time within the normal blood glucose level range (4.0–10.0 mmol/L) in the 3-h postprandial period after consuming meals is shown in Table 2. Compared with carbohydrate counting, the FII algorithm significantly improved the time within the optimal blood glucose level range by 31% (88 ± 69 vs. 128 ± 57 min, P = 0.001, n = 28). There was no significant difference in the percent time within normal range between meal A with carbohydrate counting and meal B using the FII algorithm (108 ± 46 vs. 128 ± 57 min, P = 0.20, n = 28) (Fig. 1).

Using the FII algorithm resulted in a 52% decrease in iAUC compared with carbohydrate counting after the same meal B (197 ± 220 vs. 409 ± 373 mmol × min/L, P = 0.013, n = 28) (Table 2). iAUC was also calculated using the data derived from the HemoCue blood glucose meters. The FII algorithm elicited a 58% reduction in iAUC (P < 0.001, n = 28) (Fig. 1).

The FII algorithm also produced a 1.7 mmol/L lower (41% decrease) peak blood glucose excursion compared with the carbohydrate-counting algorithm (2.4 ± 1.9 vs. 4.1 ± 3.1 mmol/L, P = 0.01, n = 28). The mean time to peak blood glucose excursion for the FII algorithm occurred at 59 min, which was significantly earlier than that for carbohydrate counting (at 97 min) (P = 0.002). In contrast, as predicted, there was no difference in time to peak blood glucose level between the FII algorithm after the consumption of meal B and carbohydrate counting with meal A (P = 0.86), although the mean peak blood glucose level excursion for meal B using the FII algorithm was significantly lower (2.4 ± 1.9 vs. 4.0 ± 2.8 mmol/L, P = 0.012, n = 28) (Table 2).

Mean time taken to return to fasting blood glucose level was significantly longer for carbohydrate counting in meal B relative to the FII algorithm (158 ± 41 vs. 118 ± 55 min, P < 0.001, n = 28). The majority of participants (61%) using carbohydrate counting for meal B failed to return to fasting blood glucose level at the 3-h time point compared with only 21% of participants using the FII algorithm. There was no significant difference in time to reach fasting blood glucose level between meal B with the FII algorithm and meal A with carbohydrate counting (P = 0.99) (Table 2).

No severe hypoglycemia episodes were observed. For the total 84 test occasions, 17 mild hypoglycemic episodes that required treatment were recorded, and 4% of data time was excluded from analysis because of hypoglycemia treatment during the 3-h postprandial period. Hypoglycemia was reported in all three test conditions, and most (12 of 17) occurred at the tail end of the monitoring period (150∼180 min). Ten episodes occurred after meal A using carbohydrate counting, six episodes followed meal B with the FII algorithm, and one episode followed meal B with carbohydrate counting. The differences between rates under carbohydrate counting and the FII algorithm were not significant in either case (FII algorithm/meal B vs. carbohydrate counting/meal A: P = 0.57; FII algorithm/meal B vs. carbohydrate counting/meal B: P = 0.31).

The current study demonstrates that a novel FII algorithm derived from insulin responses in healthy individuals improved acute postprandial glycemia compared with the use of carbohydrate counting in subjects with well-controlled type 1 diabetes using intensive insulin pump therapy. The FII algorithm improved time spent in the postprandial normoglycemic range, produced a significantly lower iAUC and a smaller peak blood glucose excursion, and reduced the time to reestablish the fasting blood glucose level. Taken together, the findings provide the first stage of clinical evidence that use of a FII based on relative insulin demand in healthy subjects may be an effective tool for estimating mealtime insulin dose in type 1 diabetes.

Matching insulin to the carbohydrate amount of a meal is a proven strategy in achieving glycemic control (13). Nonetheless, numerous studies have shown that other dietary factors, including GI, protein or certain insulinotropic amino acids, fat, and fiber content, also affect the degree of postprandial insulinemia in both healthy and type 2 diabetic subjects (7,14–17). Moreover, these factors significantly affect mealtime insulin required in type 1 diabetes (18,19). Although carbohydrate counting is universally promoted in routine practice, many individuals with type 1 diabetes continue to report elevated postprandial blood glucose levels and remain at high levels for 2 or 3 h, particularly after the ingestion of meals with higher protein and/or fat content, such as pizza (19,20) and other fast foods (9). Epidemiological evidence has shown that elevated postprandial glucose increases the risk of cardiovascular disease, atherosclerosis, and mortality (21,22). It is therefore clinically important that in the context of everyday life, other dietary factors that influence insulin secretion should be considered.

Theoretically, use of carbohydrate counting may better suit meals with high carbohydrate content and less protein and fat. Meal A was designed to contain 75 g carbohydrate with only 20% of energy derived from protein and fat. The bolus-wizard settings and ICR of all participants were verified during the run-in period before testing. Blood glucose data after meal A was therefore the “reference” meal in the current study. Meal B was designed specifically to have approximately half the carbohydrate content of meal A but the same insulin demand based on knowledge of the FII values of the component foods. The FII algorithm was compared with carbohydrate counting after the consumption of meal B. As predicted, the mean blood glucose profile obtained using the FII algorithm closely mirrored the glucose profile of meal A. In contrast, the carbohydrate-counting algorithm produced very different glucose profiles (Fig. 1). This evidence supports our hypothesis that meals with the same insulin demand (assessed by FII = 60) should be administered the same dose of insulin. Nevertheless, meal A showed a significantly higher peak blood glucose excursion than meal B with the novel FII algorithm, which was not hypothesized. The result may reflect differences in the action of exogenous versus endogenous insulin on meals with a high glycemic load (glycemic load of meal A was twofold higher than that of meal B). Injected insulin must be absorbed and transported to the site of action and may therefore not act as promptly as endogenous insulin does in a healthy subject consuming a high–glycemic load meal. Several studies in type 1 diabetes have demonstrated that a high–glycemic load meal produces much higher blood glucose responses than a low–glycemic load meal with a standardized insulin-dosing algorithm (5,14,23).

Notably, the FII algorithm did not significantly increase the risk of occurrence of hypoglycemic episodes, although the total number of episodes appeared to be higher than in previous studies of this nature (14,23,24). The real-time CGMS device incorporated a low blood glucose level alarm system so that hypoglycemia was more likely to be identified than during routine management. The instant blood glucose level profiles shown on the CGMS screen made it difficult to blind the patients and study personnel. This situation is different from previous studies in which blood glucose profiles were only available at the end, and hypoglycemia was not treated unless the individual felt symptomatic (14,23). It is also possible that the standard bolus delivering 100% rapid-acting insulin immediately before the meal might not be suitable for low-GI meals such as meals A and B (both GI = ∼52). In a previous study, hypoglycemic episodes occurred two times more frequently with standard bolus compared with dual-wave bolus when low-GI meals were consumed (23).

Several limitations must be acknowledged. First, patients were not blinded to the real-time CGMS device. The instant readings of hyper- or hypoglycemia shown on the CGMS screen could induce symptomatic episodes, compromising the patient’s compliance with the study protocol. The FII algorithm was applied for a short period of time to a single meal, and the effects of daylong or chronic feeding on glycemic control remain to be explored. Dose-response relationships (i.e., insulin demand–to–insulin ratio) also need to be established before the FII algorithm can be put into routine clinical practice in the management of type 1 diabetes. Further clinical studies should clarify these important issues.

Currently, our database includes the FII of >120 common foods covering the most important energy sources in western diets (11). Other researchers therefore have the tools to undertake further clinical studies in this area to determine the practicality in using an insulin demand–to–insulin ratio to calculate mealtime insulin dose instead of a carbohydrate-to-insulin ratio. It can be argued that there is no additional burden integrating the FII algorithm into routine practice because FII counting replaces carbohydrate counting. In food tables, a new number (the insulin demand per serving) would replace the grams of carbohydrate per serving of a particular food. Of course, a major hurdle is the fact that a food’s FII does not currently appear on food labels, while carbohydrate does. We acknowledge that implementation of these findings represents advanced insulin therapy, which may best be incorporated when basic nutrition knowledge is established.

In summary, the NIDDA study supports further studies of a novel insulin algorithm on the basis of physiological insulin demand evoked by foods in healthy subjects as a tool for estimating mealtime insulin dose, resulting in improvement of postprandial glycemia. Given the fact that optimal postprandial control is classified as one of the most challenging aspects of type 1 diabetes management (25), continuing efforts for optimizing advanced insulin therapy to achieve physiological postprandial blood glucose profiles are warranted.

Support for this study was provided by the University of Sydney.

No potential conflicts of interest relevant to this article were reported.

J.B. conceived the study, interpreted data, conducted the study, prepared the meals, collected data, analyzed data, and wrote the manuscript. H.R.G. conceived the study and interpreted data. R.G. and D.M. conducted the study, prepared the meals, and collected data. G.H. monitored and supervised the whole test period as an endocrinologist. P.P. analyzed data. S.C. conceived the study. J.C.B.-M. conceived the study, interpreted data, and wrote the manuscript. All authors contributed to discussion of the manuscript.

Parts of this study were presented in abstract form and was awarded the Best Oral Presentation at the Australian Diabetes Society and Australia Diabetes Educators Association joint annual scientific meeting, Sydney, Australia, 1–3 September 2010.

This study has been presented at the 1st Asia Pacific Congress on Controversies to Consensus in Diabetes, Obesity and Hypertension (CODHy), Shanghai, China, 2–5 June 2011 and will be presented at 47th European Association for the Study of Diabetes (EASD) Annual Meeting, Portugal, Lisbon, 12–16 September 2011.