Precision and accuracy of hyperglycemic clamps in a multicenter study

Abstract

Application of glucose clamp methodologies in multicenter studies brings challenges for standardization. The Restoring Insulin Secretion (RISE) Consortium implemented a hyperglycemic clamp protocol across seven centers using a combination of technical and management approaches to achieve standardization. Two-stage hyperglycemic clamps with glucose targets of 200 mg/dL and >450 mg/dL were performed utilizing a centralized spreadsheet-based algorithm that guided dextrose infusion rates using bedside plasma glucose measurements. Clamp operators received initial and repeated training with ongoing feedback based on surveillance of clamp performance. The precision and accuracy of the achieved stage-specific glucose targets were evaluated, including differences by study center. We also evaluated robustness of the method to baseline physiologic differences and on-study treatment effects. The RISE approach produced high overall precision (3%–9% variance in achieved plasma glucose from target at various times across the procedure) and accuracy (SD < 10% overall). Statistically significant but numerically small differences in achieved target glucose concentrations were observed across study centers, within the magnitude of the observed technical variability. Variation of the achieved target glucose over time in placebo-treated individuals was low (<3% variation), and the method was robust to differences in baseline physiology (youth vs. adult, IGT vs. diabetes status) and differences in physiology induced by study treatments. The RISE approach to standardization of the hyperglycemic clamp methodology across multiple study centers produced technically excellent standardization of achieved glucose concentrations. This approach provides a reliable method for implementing glucose clamp methodology across multiple study centers.

NEW & NOTEWORTHY The Restoring Insulin Secretion (RISE) study centers undertook hyperglycemic clamps using a simplified methodology and a decision guidance algorithm implemented in an easy-to-use spreadsheet. This approach, combined with active management including ongoing central data surveillance and routine feedback to study centers, produced technically excellent standardization of achieved glucose concentrations on repeat studies within and across study centers.

INTRODUCTION

The glucose clamp technique imposes a controlled exposure to intravenous insulin and/or glucose with frequent monitoring and adjustments of infusion to “clamp” the glucose at a particular target concentration. It was developed in the 1950s and 1960s in laboratories at the National Institutes of Health and elsewhere (1–4), and a publication from the NIH in the 1970s has since provided the reference approach for this methodology (1). Work incorporating the non-glucose secretagogue l-arginine has added to this methodological background (5, 6). Glucose clamp methods provide a highly repeatable and controlled method for measuring metabolic physiology and have been widely adopted in clinical studies focused on assessing insulin sensitivity and/or β-cell function. These studies have generally implemented clamp methodology at a single institution, but, in a few cases, groups have performed measurements at multiple sites. These have primarily consisted of studies performing hyperinsulinemic euglycemic clamps (measuring insulin sensitivity), including the Relationship between Insulin Sensitivity and Cardiovascular Disease (RISC) study (19 centers) (7); an eight-center study in a subset of the T1D exchange study centers (8); a six-center study evaluating the metabolic effects of troglitazone (9); the European Network on Functional Genomics of Type 2 Diabetes (EUGENE2) Consortium at five centers (10); and the Botnia Study of the epidemiology of glucose metabolism at four centers (11, 12). Multicenter application of hyperglycemic clamps (measuring β-cell function) has been undertaken in a study evaluating effects of acarbose in screen-detected type 2 diabetes (13), and a study comparing liraglutide against insulin glargine with the goal of β-cell preservation (three centers), which also performed hyperinsulinemic euglycemic clamps (14, 15).

Standardization of glucose clamp methodology and approach across multiple study centers is a challenge because sequential decisions made in the performance of the procedure can influence the outcome measurement (1, 16). Approaches to methodological standardization in the above studies have ranged from the distribution of an instructional video to centralized training with or without distribution of infusion rate guidance calculators [personal communication from study investigators (7, 8, 12–15, 17, 18)]. Ongoing data quality surveillance, reeducation, and decision support similarly varied in intensity across these studies. Little information is available about the technical results of such multicenter standardization efforts.

The Restoring Insulin Secretion (RISE) Consortium undertook a multicenter study with repeated measurements of β-cell function and insulin sensitivity using a hyperglycemic clamp method in youth and adults with impaired glucose tolerance or recently diagnosed type 2 diabetes (19). RISE was performed at seven centers across the United States, including four centers performing measurements in pediatric populations and four performing measurements in adult populations (one center studied both) (19). At the stage of study design, the RISE investigators set out to standardize the application of the hyperglycemic clamp methodology across these centers, to ensure a uniform methodological approach, and to minimize between-center variance in the performance of the study measurements. Here, we describe the methods used for the RISE hyperglycemic clamp, including the steps taken to achieve standardization across multiple study sites.

METHODS

The Restoring Insulin Secretion (RISE) Consortium undertook three randomized studies in parallel exploring approaches to the preservation of β-cell function. Descriptions of the study designs and primary outcomes have been published (19–22). In brief, an adult medication study evaluated four medical therapies (metformin alone, liraglutide with metformin, insulin glargine followed by metformin, or placebo), a pediatric medication study evaluated two medical therapies (metformin alone, insulin glargine followed by metformin), and an adult surgery study evaluated laparoscopic gastric banding surgery versus metformin. The study populations included participants with impaired glucose tolerance (IGT) or recently diagnosed type 2 diabetes discovered at the time of study screening, except the pediatric population, which could have been diagnosed with type 2 diabetes up to 6 mo before screening. The main study endpoints were obtained from hyperglycemic clamps, performed at the seven participating clinical study centers. The institutional review board (IRB) at each center approved the protocol. Written informed consent or assent was obtained from each participant, consistent with the Helsinki Declaration and each center’s IRB guidelines.

Hyperglycemic Clamp Method

The RISE Consortium utilized a two-stage hyperglycemic clamp for the measurement of the main study endpoints. Participants were studied following an overnight fast, using blood samples from an arterialized retrograde warmed hand vein to sample for ongoing plasma glucose readings. Stage 1 was a hyperglycemic clamp targeting a steady-state plasma glucose of 200 mg/dL (7.8 mmol/L) by 120 min, using an initial 20 mg% dextrose (D20W) bolus followed by ongoing adjustment to the D20W infusion rate guided by bedside plasma glucose measurements made on (YSI) 2300 Plus or YSI 2700 devices (Yellow Springs Instruments, Yellow Springs, OH). A duration of 120 min was selected for stage 1; we allowed an extension up to an additional 20 min if the target plasma glucose was not stably achieved by 100 min following the start of the procedure. Terminal sampling for stage 1 took place from minutes 100 to 120, or as late as minutes 120 to 140 if the extension was needed; stage 1 was terminated by 140 min, even if the targeted steady-state plasma glucose had not been achieved. Stage 2 of the clamp procedure was intended to measure maximal secretory responses, achieved by raising plasma glucose to >450 mg/dL (25.0 mmol/L) for at least the final 15 min of a 30-min hyperglycemic stage, followed by a 5-g intravenous bolus injection of l-arginine (13). The duration of hyperglycemia for stage 2 could be extended to up to 45 min if needed to achieve the 15-min interval of target level hyperglycemia. Baseline and follow-up clamp procedures for each participant were matched for timing, achieved stage 1 steady-state plasma glucose, and achieved stage 2 pre-arginine plasma glucose.

The algorithm for achieving and maintaining target glycemia was formalized into an electronic spreadsheet that was distributed to all study sites. The spreadsheet provided calculations for D20W boluses and guided the selection of D20W infusion rates. Stage 1 method was derived from a series of ∼200 hyperglycemic clamp procedures performed under a two-site study protocol in individuals with early diabetes (13), updated based on the direct experience of the RISE investigators (16, 23–26). Stage 2 method was empirically derived, based on prior experience with achieving and maintaining the target level of glycemia (5, 6, 27). The calculations embodied in the spreadsheet are detailed in Table 1. The final version of the guidance spreadsheet is available at https://rise.bsc.gwu.edu/clamp. This tool provided sequential D20W infusion rate guidance, but decisions on actual infusion rates were made by study staff at the bedside after considering the recommended infusion rate and other potentially relevant factors. The selected D20W infusion rate was entered into the spreadsheet in real time for calculating the next sequential rate recommendation.

Table 1.

Glucose infusion protocol guiding the two-stage hyperglycemic clamp procedures

Calculations	Comment
Bolus	Results in mL of D20W
[(Gt − Gb) × 1.1 × Wt (kg)/180	No bolus if Gb >180 mg/dL
Stage 1: initial infusion rate	Results in D20W mL/h
[3.5 × Wt (kg) × 60/180]	See text
Stage 1: ongoing infusion adjustment	Results in D20W mL/h
[R + [(Gt − Gi)/Gt] × R × 1.1]	No change if |Gi − Gt|<2 mg/dL. No change if |Gi − Gt|<5 mg/dL and changing in correct direction
Stage 2: initial infusion rate
999 mL/h	After stage 2 bolus run pump at maximum. Reduction in rate allowed if Gi >650 mg/dL. D50W 50 mL supplemental bolus if Gi <400 mg/dL after 15 min

Stage 1 target glucose is 200 mg/dL; stage 2 target glucose is >450 mg/dL. Division by 180 approximates the glucose concentration in mg/mL of D20W. Bedside plasma glucose readings (in mg/dL) are input to guide decisions. Gb, basal glucose; Gi, glucose at each time point; Gt, target glucose; R, infusion rate of glucose.

Blood samples taken at prespecified time points for later measurement of glucose, insulin, and C-peptide concentrations were processed and shipped to the RISE Central Biochemistry Laboratory. These centrally measured values were later used for calculation of the primary study endpoints. The locally measured plasma glucose values that were used to guide the decisions made in the bedside procedure were distinct from the centrally measured values that were used for study endpoints.

Boluses: Boluses of D20W for step increments in plasma glucose concentrations at the start of each stage were calculated incorporating body weight, current plasma glucose, and the target plasma glucose (Table 1). Stage 1 bolus was intended to provide an immediate increase from fasting plasma glucose to the target of 200 ± 5 mg/dL (or to a modified target with follow-up procedures). Stage 2 bolus was intended to provide an immediate increase from the plasma glucose concentration at the termination of stage 1 to the target of 450 mg/dL or greater.

Stage 1: The initial rate for the D20W infusion was based on participant body weight (Table 1). With the first ∼60 procedures, this formula used a multiplier of 5; however, the data from this initial set of procedures indicated a consistent overshoot in the first 30 min, and subsequently, this multiplier was 3.5.

After the initiating D20W bolus and infusion rate selection, subsequent infusion rate changes were guided by the ongoing bedside plasma glucose readings, taken every 5 min (Table 1). The guidance template provided a recommended D20W infusion rate based on the current bedside plasma glucose and the trend in plasma glucose concentrations over the preceding 10 min. If the value was sufficiently close to target, or approaching the target with changes averaged over the past 10 min, no change was recommended (Table 1).

The target glucose for stage 1 was generally 200 mg/dL (7.8 mmol/L). Repeat studies within a participant employed the achieved steady-state glucose values from the baseline study and the multiplier for the initial infusion rate from their baseline study.

Stage 2: The physiology of maximal secretory responses requires only that the arginine bolus to be administered in the setting of ambient glucose >450 mg/dL (25 mmol/L), with no defined glucose maximum (27). The goal for this stage was therefore to achieve any plasma glucose concentration over 450 mg/dL. To accomplish this, we administered a second D20W bolus followed by a D20W infusion at the maximal infusion rate allowed by a single clinical infusion pump (999 mL/h). A third D20W bolus was allowed if needed to achieve the goal (Table 1), with no recommendation to exceed this infusion rate. Study personnel had the option of reducing the infusion rate if excessively high plasma glucose concentrations were observed, but no formal guidance was provided in this regard; in general, this choice was only made when bedside plasma glucose exceeded 650 mg/dL (36 mmol/L).

In parallel with the matching of plasma glucose concentrations undertaken for stage 1, achieved stage 2 bedside plasma glucose concentrations from the baseline study day were used as the targets for subsequent measurement days.

Training, Performance, and Surveillance

All participating study centers had prior experience with in vivo metabolic measures of varying complexity; nevertheless, central training was provided to standardize application of the hyperglycemic clamp procedure. This training included direct application of the spreadsheet tool during a live study procedure. Project coordinators and principal investigators all participated to provide a pool of knowledge across the study, using a “train the trainer” approach. Subsequent personnel changes required that the already trained local staff train new staff. Ongoing training and feedback regarding the use of the spreadsheet were performed via study-wide audio–video conferences and telephone calls.

Completed hyperglycemic clamp spreadsheets were collected centrally, immediately after study procedures were completed, and routinely surveyed by the study’s Procedures & Quality Control Committee. Evaluated features included 1) achievement of target glycemia with the initial bolus and across stage 1; 2) achievement of target glycemia for stage 2; 3) appropriate modification of target glucose for follow-up procedures to match achieved baseline studies; and 4) whether the procedure was functioning adequately across multiple study sites as intended. The review process entailed an assessment of individual clamp spreadsheets, generally within 4 wk of their central submission, but did not include any formal evaluations of aggregate data.

Statistical Analysis

The precision and accuracy of plasma glucose target achievement with this methodology were evaluated first using the achieved bedside plasma glucose values at key time points including 10, 30, and 60 min after the initial D20W bolus, and stage 1 steady state, in relation to the absolute target of 200 mg/dL (7.8 mmol/L). Second, we evaluated the deviation from this target over the time course of the procedure, using all bedside plasma glucose data points. Third, we evaluated stage 2 achieved plasma glucose values against the target of >450 mg/dL (25 mmol/L). Data were evaluated in aggregate, across the study centers, and then by divisions of adult versus youth, and by impaired glucose tolerance (IGT) versus diabetes status at the time of enrollment. These analyses employed simple descriptive statistics and groupwise analysis of variance. We also performed a parallel set of analyses within subgroups defined by whether stage 1 steady state was achieved on time (final steady-state sampling at T = 120 min), delayed (steady state between 125 and 135 min), or time-limited (steady-state T = 140 min).

The implementation of algorithm-guided D20W infusion rates was evaluated as the difference between spreadsheet-provided guidance and the selected infusion rates at each time point. These were compared in aggregate and by the subgroup divisions noted above.

RESULTS

In aggregate, 1,136 hyperglycemic clamp procedures were performed in 443 participants (91 youth and 352 adults; 302 classified as IGT at screening and 141 with diabetes at screening). Four hundred and twenty-seven procedures were performed at baseline, 367 at the 12-mo follow-up visit, and 342 at the third clamp visit (294 at month 15 for adult medication study participants, 48 at month 24 for the adult surgery study participants). Unless otherwise specified, the data were evaluated for all clamps combined.

Achieved Glycemic Targets

Achieved glucose targets were evaluated using the bedside plasma glucose measurements performed in the course of undertaking the procedure.

Hyperglycemic clamp stage 1.

The target plasma glucose value for all time points from T = 10 min onward for stage 1 of the procedure was 200 mg/dL. The accuracy and precision of the methodology in achieving this target at selected time points including steady state are presented in Table 2. Achieved glucose values over the course of stage 1 study procedures are presented in full in Fig. 1. Figure 1, A and B includes all clamp procedures performed at all study visits, including baseline and postrandomization procedures, generally including three such procedures per participant. Figure 1A illustrates adults versus youth, whereas Fig. 1B presents IGT versus diabetes based on categorization at the time of screening for RISE. Figure 1, C and D provides data from the adult medication study for single-clamp procedures performed at baseline (Fig. 1C) and the end of 12 mo of medication intervention (Fig. 1D), according to randomized treatment groups.

Table 2.

Precision and accuracy at specified time points across the glucose clamp procedures

	N Clamps	All	Range of Means Across Study Centers	P Value Comparing Study Center Means	Range of SD Across Study Centers	P Value Comparing Study Center Variances	Adults (n = 891 Clamps)	Youth (n = 245 Clamps)	P Value Comparing Age Groups	IGT (n = 764 Clamps)	Diabetes (n = 372 Clamps)	P Value Comparing Glycemic Status
Number of participants	443						352	91		302	141
Glucose 10 min, mg/dL	1,127	188.3 ± 18.2	181.4, 201.6	<0.001	12.6, 25.1	0.3699	186.9 ± 17.5	193.3 ± 20.0	<0.001	187.2 ± 18.0	190.6 ± 18.4	0.003
Glucose 30 min, mg/dL	1,127	201.1 ± 17.6	197.1, 207.8	0.010	12.3, 23.9	0.9432	200.2 ± 18.2	204.1 ± 14.6	0.002	198.8 ± 18.6	205.8 ± 14.0	<0.001
Glucose 60 min, mg/dL	1,129	214.6 ± 14.8	208.8, 234.2	<0.001	8.3, 17.8	0.0853	214.2 ± 14.6	215.8 ± 15.8	0.130	213.8 ± 14.7	216.1 ± 15.0	0.013
Stage 1 mean glucose, mg/dL	1,136	201.8 ± 7.9	199.2, 208.3	<0.001	3.6, 10.8	0.7478	201.5 ± 8.1	202.9 ± 7.2	0.013	200.7 ± 7.9	203.9 ± 7.5	<0.001
Stage 1 mean % diff from 200 mg/dL	1,136	2.5 ± 2.8	1.6, 4.2	<0.001	1.4, 3.8	0.7773	2.5 ± 2.8	2.7 ± 2.6	0.231	2.3 ± 2.7	3.0 ± 2.8	<0.001
Steady state mean glucose, mg/dL	1,136	198.2 ± 8.4	195.4, 201.0	<0.001	3.7, 13.6	<0.001	197.9 ± 8.5	199.3 ± 7.7	0.020	197.9 ± 8.0	198.8 ± 9.1	0.078
Steady-state mean % diff from 200 mg/dL target	1,136	3.1 ± 3.0	1.5, 4.0	<0.001	1.1, 3.9	0.0129	3.2 ± 3.1	2.7 ± 2.8	0.014	3.1 ± 2.8	3.1 ± 3.5	0.852
Stage 2 mean glucose	1,134	593.0 ± 90.6	514.7, 677.2	<0.001	53.3, 99.3	<0.0001	602.0 ± 91.8	559.8 ± 77.3	<0.001	583.1 ± 92.5	613.3 ± 82.9	<0.001

Summary data shown are means ± SD. P values represent ANOVA comparisons across groups (for the comparison across centers, a seven-group comparison) except the comparison of SD across centers, which applied Levene’s test for homogeneity of variance. Stage 1 mean glucose includes all values measured from T = 10 through the final steady state sampling time point. Stage 2 mean glucose is the average of the two readings immediately preceding the arginine infusion. IGT, impaired glucose tolerance at screening; SD, standard deviation.

Figure 1.

Achieved glucose values. Data are means ± SD for values at each time point; for some time points, the error is smaller than the size of the marker. Statistical evaluations of the achieved values relative to the 200 mg/dL target (dashed line) for selected time points are presented in Table 2. Sample sizes vary, please refer to text. A: presents data for all studies combined, separating studies performed in youth and adults. B: presents data for all studies combined, separating studies performed in individuals with IGT versus diabetes at screening. C and D: present data from the adult medication study only, presenting baseline (C) and 12-mo on-treatment data (D). SS1, SS2, and SS3 represent the three measurement time points that together comprised the steady state (SS); the actual timing of these was generally 100 through 120 min after the start of the procedure (T = 0 min); see text for further details. The initial negative time points represent sampling in the fasting state prior to commencing glucose administration at time 0. IGT, impaired glucose tolerance.

Averaged over all studies performed, the mean plasma glucose achieved at T = 10 min was below target (188.3 mg/dL; Table 2). This difference from target at T = 10 min was ∼9% of the mean value on average; this measure of accuracy improved over the time course of the procedure to ∼3% of the mean value. The mean plasma glucose values achieved at each evaluated time point were statistically different across the study centers, but modest in magnitude and with these cross-center differences encompassed within the accuracy achieved for the procedures overall (i.e., within 10% of each other in the early time points and within 5% of each other by the final stages of the procedure). The precision of the achieved plasma glucose was similar across study centers at the early time points but was significantly different at the final 20-min steady-state interval (with SD ranging from 3.7 to 13.6 across centers at that time; Table 2). Comparing clamps performed in youth versus adults (Table 2 and Fig. 1A), the achieved plasma glucose values in the clamps performed in youth were slightly but significantly higher than those performed in adults. Similarly, when evaluated according to groups defined by baseline presence of IGT or diabetes (Table 2 and Fig. 1B), achieved plasma glucose values were higher overall in clamps performed in those with type 2 diabetes; these differences were statistically significant only at the early time points of the procedure. The pattern of accuracy of the achieved plasma glucose values across these subgroups was parallel to what was observed in the aggregate evaluation, with variance ∼10% of the mean at the early time points and <5% of the mean at later time points. These subgroup analyses did not reveal any informative pattern of differences in precision (Fig. 1, A and B).

The majority of the clamp procedures (916/1,136, 80%) were completed with stage 1 steady-state plasma glucose achieved as intended at 100 through 120 min. Steady state was delayed in 72 (6%) procedures, with final sampling performed between 125 and 135 min, and steady state was time-limited in 148 (13%) procedures with final steady-state sampling performed at 140 min even if the target value had not been fully achieved. The accuracy and precision of the clamp procedures in the on-time and delayed clamps mirrored the overall results described above; the time-limited clamps exhibited material differences from the others only in the achieved glucose values at steady state (on-time 198.0 ± 7.4, delayed 198.9 ± 12.6, and time-limited 194.4 ± 11.6 mg/dL; P < 0.001). The mean plasma glucose achieved over the course of stage 1 was not materially different across these groupings, nor materially different from the 200 mg/dL target (mean stage 1 plasma glucose: on-time 202.3 ± 7.8, delayed 200.5 ± 6.6, time-limited 199.0 ± 8.4 mg/dL).

Three features are of note on the plots in Fig. 1. First, as also evident in the aggregate averages presented in Table 2, across all clamps on average, the first bolus did not fully achieve the initial target glucose of 200 mg/dL at T = 10 min. As evident in Fig. 1, C and D, this was not unique to any treatment group or this time point, but it is accentuated in the liraglutide + metformin treatment arm at the end of 12 mo of treatment. Second, there was a systematic rise from this initial reading to a peak, generally around the 60-min time point. The achieved plasma glucose at T = 60 was 214.6 ± 14.8 mg/dL, representing a 7.6 ± 5.8% overshoot on average. The subsequent time points of the procedures were characterized by a gradual descent to the steady-state target, which was well achieved as described above (Table 2). Third, modest differences in the average plasma glucose across the time course leading up to steady state were observed with the on-treatment studies performed at month 12. Differences in preprocedure fasting plasma glucose were evident (Fig. 1D) followed by differences in the mean achieved plasma glucose across the time course of the procedure (liraglutide + metformin 194.1 ± 7.1, glargine followed by metformin 202.4 ± 9.3, metformin alone 201.4 ± 6.4, and placebo 204.4 ± 19.4; P < 0.001). Despite these differences in the time course of the procedures, the system produced comparable and on-target steady-state plasma glucose values (mean achieved steady-state plasma glucose 196.7, 198.4, 196.5, and 199.2 for liraglutide + metformin, glargine followed by metformin, metformin alone, and placebo, respectively; P = 0.33 for the four group comparison) (Supplemental Table S2; all Supplemental material is available at https://doi.org/10.5281/zenodo.4284819).

In the placebo-treated participants, the time course and achieved glucose targets were notably stable across study visits (Supplemental Table S3). For example, the 30-min plasma glucose values at baseline (201.4 ± 12.4), month 12 (207.8 ± 46.0), and month 15 (204.1 ± 9.8 mg/dL) were not significantly different (P = 0.46), and the overall mean steady-state plasma glucose were practically identical (199.5, 199.2, and 198.4 mg/dL for baseline, month 12, and month 15, respectively).

Hyperglycemic clamp stage 2 targets.

The achieved plasma glucose values from stage 2 immediately before the arginine infusion are presented in Table 2 (and also in Supplemental Tables S1, S2, and S3). Notable features in the main comparisons in Table 2 include achieved plasma glucose targets well in excess of the minimum target in all groups evaluated. There were differences across centers in the mean achieved plasma glucose values and in the variability around stage 2 achieved values. There were also differences between adult/youth and IGT/diabetes groupings. These differences, while statistically significant and numerically relevant, nevertheless simply represent variation between groups above the minimum threshold.

Adherence to the Algorithm

Selection of D20W infusion rates using the algorithm for infusion rate guidance outlined in Table 1 pertains primarily to stage 1 of the procedure, where a glucose concentration of 200 mg/dL is targeted, requiring ongoing assessment and modification of the D20W infusion rate. In contrast, stage 2 of the procedure requires much less manipulation. Figure 2 presents the differences between guidance infusion rate provided by the algorithm and the actual infusion rate selected by the study staff performing the procedure, showing the full-time course across stage 1 of the procedures. Again, these are presented according to prespecified subgroups of interest, namely, youth versus adult (Fig. 2, A and B) and IGT versus diabetes (Fig. 2, C and D). Table 3 presents descriptive statistics for these differences at selected time points across the study procedures.

Figure 2.

Differences in selected versus guidance glucose infusion rates in stage 1 of the clamp procedure. Sample sizes vary, please refer to text. A and C: present the calculated guidance provided by the algorithm. B and D: present differences from guidance, calculated as (selected rate − guidance rate). A and B: show these data separated by studies performed in youth and adults. C and D: show these data separated by studies performed in individuals with IGT or diabetes status at screening. IGT, impaired glucose tolerance.

Table 3.

Differences between algorithm guidance and applied infusion rates

	N Clamps	All (Means ± SD)	Range of Mean Differences Across Study Centers	P Value Comparing Study Center Means	Range of SD of Differences Across Study Centers	P Value Comparing Study Center Variances	Adults (n = 888 Clamps)	Youth (n = 244 Clamps)	P Value Comparing Age Group	IGT (n = 764 Clamps)	Diabetes (n = 373 Clamps)	P Value Comparing Glycemic Status
Number of participants	443						352	91		302	141
T = 10	1,124	−11.2 ± 21.0	−22.6, −2.0	<0.001	7.5, 23.0	<0.0001	−11.6 ± 20.9	−10.0 ± 21.4	0.309	−12.5 ± 20.8	−8.6 ± 21.1	0.003
T = 20	1,120	2.0 ± 11.1	−5.0, 7.5	<.0001	5.9, 13.4	0.0243	2.7 ± 11.3	−0.6 ± 10.1	<0.001	1.5 ± 10.6	2.8 ± 11.9	0.065
T = 30	1,124	2.8 ± 13.2	−0.7, 8.9	<0.001	5.5, 31.8	0.0003	2.6 ± 10.6	3.5 ± 20.1	0.347	2.6 ± 12.5	3.3 ± 14.7	0.387
T = 60	1,128	0.7 ± 12.8	−5.6, 2.2	0.099	5.0, 18.8	0.6135	0.9 ± 12.9	−0.1 ± 12.5	0.264	0.9 ± 14.1	0.3 ± 9.8	0.416
T = 90	1,129	−3.0 ± 8.6	−4.8, 0.1	<0.0001	5.2, 10.1	0.0019	−3.3 ± 8.4	−2.1 ± 9.2	0.051	−3.5 ± 9.0	−2.0 ± 7.6	0.005
SS 1	1,131	−3.4 ± 15.9	−7.7, 0.0	0.027	5.8, 29.2	0.4715	−3.7 ± 15.4	−2.5 ± 17.6	0.325	−3.4 ± 17.0	−3.4 ± 13.5	0.968
SS 2	1,132	−3.6 ± 27.1	−8.5, 0.9	0.042	4.4, 45.6	0.8224	−3.6 ± 29.4	−3.7 ± 16.0	0.990	−3.9 ± 31.5	−3.1 ± 14.3	0.623
SS 3	1,129	−5.6 ± 24.9	−17.5, 1.7	<.0001	6.4, 45.2	0.7838	−4.4 ± 26.2	−9.8 ± 18.8	0.003	−5.8 ± 28.5	−5.1 ± 15.1	0.622

Summary data shown are means ± SD. Data were calculated as [guidance rate] − [applied rate], both in mL/min. P values represent comparisons across groups using ANOVA. IGT, impaired glucose tolerance at screening; SD, standard deviation; SS, steady state.

These data prompt the following observations: First, the selected D20W infusion rate was systematically lower than the guidance rate at the first postbolus time point (T = 10), which is the first time that a bedside plasma glucose reading was available; this time point produced the largest variance from guidance seen across the procedures. Second, the subsequent pattern was one of gradually lower selected infusion rates compared with the guidance, particularly after the 60-min time point and as the steady state was approached. This was most evident across the three steady-state time points, where the selected infusion rates were systematically below the algorithm-guided infusion rate recommendation. The magnitude of differences observed was small overall, with mean differences of only 0.7–5.6 mL/h after T = 20, although with relatively large standard deviations around these means (Table 3). A range of differences across centers was seen, generally within twofold of the overall mean but achieving statistical significance for the seven-group comparison. There was also variation across centers in the standard deviation of these differences, but these cross-center differences were statistically significant primarily for the early time points of the procedures. In studies performed in youth versus adults, there were minor differences in the deviation from guidance, but overall these did not reach statistical significance. A similar pattern of minor differences was seen in the IGT versus diabetes comparisons. In all instances, the greatest magnitude of difference between guidance and selected infusion rates was at T = 10 min, and at that time point, the cross-center comparison and the IGT versus diabetes comparison achieved statistical significance.

DISCUSSION

The Restoring Insulin Secretion (RISE) Consortium implemented a methodology combining technical and management tools to standardize the application of the hyperglycemic clamp procedure across multiple study centers with varying prior experience in the performance of such procedures. This approach enabled uniform study-wide achievement of the 200-mg/dL glucose target for stage 1 of the procedure with minimal variance from the target after an initial period of instability. This was achieved with minimal but systematic deviation from the algorithm-provided D20W infusion rate guidance by the operators as they fine-tuned the progress of the procedures. The system was robust to variations across study centers and over time and also to variations between physiologically different groups (youth vs. adults, IGT vs. diabetes). On-treatment measurements exhibited some differences in the time course but not the final results achieved at steady state by the procedure. Therefore, despite important differences in underlying physiology in the patient groups studied, the applied methodology was successful in achieving uniform performance of the hyperglycemic clamp procedure and in particular was able to achieve uniformity across multiple study centers.

Our methodology allowed the multiple centers in the study to achieve the glucose targets with good precision and accuracy. We observed high overall precision for the 200-mg/dL target of stage 1 (3%–9% variance across the procedure, ∼3% variance at the steady-state end point) and repeatability assessed in the placebo group (which was tested in three of the seven centers) showed <3% variation in achieved steady-state target over time with no difference across centers. We observed statistically significant variation in the achieved stage 1 steady-state plasma glucose values across centers, but this variation was small in magnitude, within the variance of the system overall, and not meaningfully different across centers. stage 2 (targeting glucose >450 mg/dL) was methodologically simpler, and this target was easily exceeded using the described methodology.

These results pertain to precision of the clamp methodology in terms of achieving the protocol’s glucose targets but do not directly relate to reproducibility of any endpoints derived from the clamp procedure itself (e.g., first-phase insulin secretory response). While we cannot therefore comment on sample size effects of extending beyond a single performance site, our observations do pertain to decisions regarding the involvement of multiple study centers. As noted, the cross-center variability in achieved glucose targets that resulted from our methodology was of similar magnitude to the overall measurement variability. This suggests that the statistical trade-off in adding centers is modest; however, the factors involved in implementing and achieving cross-center standardization are nontrivial, and therefore, decisions related to the benefits of adding study centers and/or expanding the available sample size need to be considered against the added management burden. On balance, our data support the feasibility of expanding to multiple centers, including providing methodological guidance, when the required sample size or other population requirements dictate the involvement of multiple centers.

In addition to providing central training at the initiation of the study, we implemented ongoing training and routine oversight of the incoming data with feedback to the teams performing the procedures. While central training at study initiation was performed for prior multicenter studies using hyperinsulinemic euglycemic clamp methods (7–12), repeat training and feedback based on ongoing assessment of study performance were only rarely undertaken (8) (personal communications with study investigators). A similar central training-focused approach was applied for one study using hyperglycemic hyperinsulinemic clamps (13). Bunck and colleagues implemented central training and data surveillance in their study using hyperglycemic hyperinsulinemic clamps (14, 15) (and personal communication) as did Cree-Green and colleagues in their study using hyperinsulinemic euglycemic clamps [(8) and personal communication]. Similar to the current report, Bunck and colleagues also implemented a spreadsheet-guided clamp procedure, using a centrally distributed spreadsheet (personal communication). Technical characteristics of the clamps performed in these prior studies have not been published, nor has any evaluation of the cross-center comparison of achieved results. The current report underscores the success of our combined technical and management approach to cross-center standardization.

The glucose clamp methodology applied in RISE was empirically derived using data from a set of ∼200 hyperglycemic clamp procedures performed in a study population parallel to the population targeted by the RISE Consortium (13). This was overall parallel to the methodology outlined by DeFronzo and Andres (1) in that the current deviation from target is combined with an empiric adjustment factor to calculate an adjustment to the glucose infusion rate. Both methods use time averaging to smooth the adjustments. The DeFronzo method implements a concurrent set of adjustments for time-averaged whole-body changes in the glucose concentrations (the “glucose space” correction); in general, this produces very modest fractional adjustments to the otherwise obtained adjustment, and we elected not to incorporate a correction for glucose space. The DeFronzo methodology has been widely implemented, and between-clamp repeatability has been repeatedly evaluated (28–30). However, evaluations of the technical performance of the method in terms of precision and accuracy appear to be limited to the original publication (1), which evaluated methodological performance at a single measurement center in 11 individuals aged on average 29 yr and within 10% of ideal body weight. The technical performance in that group was strong, with high precision (∼8%–10% on repeat studies) and accuracy (∼10% repeat-study variance) reported for time-averaged values in six subjects (1). Formal evaluations of technical performance in populations with different metabolic physiology, varying ages, or undergoing treatments have not been reported, and no reports of performance in multicenter applications have been published. Thus, the data we present are unique and broadly applicable to studies of individuals with metabolic dysfunction.

Precision and accuracy of other glucose clamp methodologies have been described. An empiric computer-controlled glucose clamping method was developed and evaluated by Matthews and Hosker (31), who reported a CV of 2.0% over the steady-state interval, averaged over 151 participants. Accuracy was not evaluated numerically. Algorithmic approaches to computer-based control of glucose infusions under clamp conditions have been reviewed in detail (32). Estimates of accuracy and precision are not generally available for these methods. The main exceptions are the LifeSciences “Biostator” (33, 34), the Nikkiso models STG-22 and STG-55 (35–37), and more recently, the Profil “ClampArt” method (38, 39). These approaches employ fully automated “artificial pancreas” systems that rely on continuous glucose monitoring rather than interval measurements of blood glucose. The algorithm for the Biostator is available (34). The Biostator method does not incorporate glucose space corrections. The algorithm for the STG systems is proprietary and unavailable (35, 37), as is the algorithm for the ClampArt method although this method appears to be derived from the Biostator approach (39). Precision of the Biostator was reported in the early application of this system to glucose clamps, with CV values 5%–7% (34), and later an updated version of the system had a CV of 4.7% (33). Accuracy is reported in these publications simply as the mean or median achieved glucose value without formal quantification of deviation from target at various time points across the procedure. Precision of the STG systems has been reported in the context of intensive care unit use, where the goal was maintenance of target glycemia, with a reported SD of 12.6 mg/dL at the target value of 150 mg/dL (suggesting a CV of ∼8% in this application) (37). The technical performance of the ClampArt system was evaluated in the performance of euglycemic clamps at two centers. They observed precision (CV 5.5%) and accuracy (mean absolute relative difference from target 5.3%) comparable to what we have reported here, without the operator-driven adjustments that provided fine-tuning of our system’s performance.

Limitations

The above results represent the integrated effectiveness of the algorithm plus local operators plus the cross-center management including surveillance and retraining methodology and therefore do not simply reflect the technical performance of any one component in isolation. Our system relied on interval sampling of plasma glucose, with adjustments in D20W infusion rates approximately every 5 min. The incorporation of a continuous glucose sensing system might provide improved control, as has been reported for the STG and ClampArt systems. Our approach to study surveillance included evaluation of individual studies as they were submitted to the central data repository but did not include any summary evaluations of achieved glucose targets at interim time points that might have highlighted the deviations from targets evident at T = 10 and T = 60 before the end of the study. The observation of a systematic pattern of operator intervention informs future adjustments of the correction factors and suggests that different correction factors applied over the course of the clamp might provide improved control overall.

CONCLUSIONS

The methodology and approach to standardization of the hyperglycemic clamp that was applied across seven centers in the three RISE studies were successful in achieving uniformity in attaining clamp glucose targets across centers, as well as in different age groups, glucose tolerance categories, and following interventions. The approach produced repeatable measurements with high precision and accuracy and was robust to differences in operator performance across centers, to important differences in underlying physiology and the effects of treatments. This approach provides a model for a comprehensive method that can be applied in future multicenter studies using glucose clamp methodologies.

GRANTS

RISE is supported by grants from the National Institutes of Health (U01DK-094406, U01DK-094430, U01DK-094431, U01DK-094438, U01DK-094467, P30DK-017047, P30DK-020595, P30DK-045735, P30DK-097512, UL1TR-000430, UL1TR-001082, UL1TR-001108, UL1TR-001855, UL1TR-001857, UL1TR-001858, and UL1TR-001863), the Department of Veterans Affairs, and Kaiser Permanente Southern California. Additional financial and material support from the American Diabetes Association, Allergan Corporation, Apollo Endosurgery, Abbott Laboratories, and Novo Nordisk A/S are gratefully acknowledged.

DISCLAIMERS

The content of this publication is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. A complete list of Centers, investigators, and staff can be found in appendix.

DISCLOSURES

At the time of publication, K. J. Mather was a full-time employee of Eli Lilly and Company. Prior to employment at Eli Lilly and Company, K. J. Mather served as a Principal Investigator for this NIH-funded study (DK094438). As such, data collection occurred before and independent of this employment. Data analysis and preparation of the manuscript were independent of Eli Lilly and Company. S. E. Kahn has served as a paid consultant on advisory boards for Novo Nordisk and steering committee for a Novo Nordisk sponsored clinical trial. S. A. Arslanian has served as a paid consultant on advisory boards for Novo Nordisk, and a participant in a Novo Nordisk sponsored clinical trial. S. E. Kahn has served as paid consultants on advisory boards for Novo Nordisk and Eli Lilly and Company. T. A. Buchanan has received research support from Allergan Corporation and Apollo Endosurgery. None of the other authors has any conflicts of interest, financial or otherwise, to disclose.

AUTHOR CONTRIBUTIONS

K.J.M., K.J.N., T.A.B., S.E.K., S.A.A., S.C., K.M.U., and S.E.E. conceived and designed research; K.J.M., A.H.T., K.J.N., T.A.B., S.A.A., S.C., K.M.A., M.C-G., K.M.U., and R.C. performed experiments; A.H.T., K.J.M., and S.E.E. analyzed data; K.J.M., M.C-G., A.H.T., A.H., K.J.N., T.A.B., S.E.K., S.A.A., S.C., K.M.A., K.M.U., S.E.E., and R.C. interpreted results of experiments; A.H.T. prepared figures; K.J.M., A.H.T., and A.H. drafted manuscript; K.J.M., A.H.T., A.H., K.J.N., T.A.B., S.E.K., S.A.A., S.C., K.M.A., M.C-G., K.M.U., and S.E.E. edited and revised manuscript; K.J.M., A.H.T., A.H., K.J.N., T.A.B., S.E.K., S.A.A., S.C., K.M.A., M.C-G., K.M.U., and S.E.E. approved final version of manuscript.

ENDNOTE

The final version of the guidance spreadsheet and a generalized version of the Manual of Procedures for performing the RISE clamps are available at https://rise.bsc.gwu.edu/clamp.

APPENDIX: RISE CONSORTIUM INVESTIGATORS

University of Chicago Clinical Research Center and Jesse Brown VA Medical Center

(Chicago, IL)

David A. Ehrmann, MD*

Karla A. Temple, PhD, RD**

Abby Rue**

Elena Barengolts, MD

Babak Mokhlesi, MD, MSc

Eve Van Cauter, PhD

Susan Sam, MD, MSc

M. Annette Miller, RN

VA Puget Sound Health Care System and University of Washington

(Seattle, WA)

Steven E. Kahn, MB, ChB*

Karen M. Atkinson, RN**

Jerry P. Palmer, MD

Kristina M. Utzschneider, MD

Tsige Gebremedhin, BS

Abigail Kernan-Schloss, BA

Alexandra Kozedub, MSN, ARNP

Brenda K. Montgomery, RN, BSN, CDE

Emily J. Morse, BS

Indiana University School of Medicine and Richard L. Roudebush VA Medical Center

(Indianapolis, IN)

Kieren J. Mather, MD*

Tammy Garrett, RN**

Tamara S. Hannon, MD

Amale Lteif, MD

Aniket Patel, MD

Robin Chisholm, RN

Karen Moore, RN

Vivian Pirics, RN

Linda Pratt, RN

University of Colorado Denver/Children’s Hospital Colorado

(Denver, CO)

Kristen J. Nadeau, MD, MS*

Susan Gross, RD**

Philip S. Zeitler, MD, PhD

Jayne Williams, RN, MSN, CPNP

Melanie Cree-Green, MD, PhD

Yesenia Garcia Reyes, MS

Krista Vissat, RN, MSN, CPNP

UPMC Children’s Hospital of Pittsburgh

(Pittsburgh, PA)

Silva A. Arslanian, MD*

Kathleen Brown, RN, CDE**

Nancy Guerra, CRNP

Kristin Porter, RN, CDE

Yale University

(New Haven, CT)

Sonia Caprio, MD*

Mary Savoye, RD, CDE**

Bridget Pierpont, MS**

University of Southern California Keck School of Medicine/Kaiser Permanente Southern California

(Los Angeles, CA)

Thomas A. Buchanan, MD*

Anny H. Xiang, PhD*

Enrique Trigo, MD**

Elizabeth Beale, MD

Fadi N. Hendee, MD

Namir Katkhouda, MD

Krishan Nayak, PhD

Mayra Martinez, MPH

Cortney Montgomery, BS

Xinhui Wang, PhD

George Washington University Biostatistics Center

(RISE Coordinating Center; Rockville, MD)

Sharon L. Edelstein, ScM*

John M. Lachin, ScD

Ashley N. Hogan, MPH

Northwest Lipid Research Laboratories

(Central Biochemistry Laboratory; Seattle, WA)

Santica Marcovina, PhD*

Jessica Harting**

John Albers, PhD

Belmar Pharmacy

(Drug Distribution Center; Lakewood, CO)

Dave Hill

NIH/NIDDK

(Bethesda, MD)

Peter J. Savage, MD

Ellen W. Leschek, MD

*Principal Investigator

**Program Coordinator