OBJECTIVE To investigate the effects of acute hypoglycemia on working memory and brain function in patients with type 1 diabetes.
RESEARCH DESIGN AND METHODS Using blood oxygen level–dependent (BOLD) functional magnetic resonance imaging during euglycemic (5.0 mmol/L) and hypoglycemic (2.8 mmol/L) hyperinsulinemic clamps, we compared brain activation response to a working-memory task (WMT) in type 1 diabetic subjects (n = 16) with that in age-matched nondiabetic control subjects (n = 16). Behavioral performance was assessed by percent correct responses.
RESULTS During euglycemia, the WMT activated the bilateral frontal and parietal cortices, insula, thalamus, and cerebellum in both groups. During hypoglycemia, activation decreased in both groups but remained 80% larger in type 1 diabetic versus control subjects (P < 0.05). In type 1 diabetic subjects, higher HbA1c was associated with lower activation in the right parahippocampal gyrus and amygdala (R2 = 0.45, P < 0.002). Deactivation of the default-mode network (DMN) also was seen in both groups during euglycemia. However, during hypoglycemia, type 1 diabetic patients deactivated the DMN 70% less than control subjects (P < 0.05). Behavioral performance did not differ between glycemic conditions or groups.
CONCLUSIONS BOLD activation was increased and deactivation was decreased in type 1 diabetic versus control subjects during hypoglycemia. This higher level of brain activation required by type 1 diabetic subjects to attain the same level of cognitive performance as control subjects suggests reduced cerebral efficiency in type 1 diabetes.
Acute episodes of hypoglycemia are a rate-limiting adverse effect in the treatment of type 1 diabetes. When severe, they can lead to seizures and coma (1). Even mild to moderate hypoglycemia is known to impair cognitive functions, such as working memory (2,3). Working memory is used to actively maintain and manipulate information over a brief period of time and to allocate attentional resources among competing subtasks (4,5). Traditionally, working-memory performance is thought to depend primarily on a network of brain regions, including portions of the frontal and parietal lobes, thalamus, precuneus, cerebellum, and insula (6,7).
Using blood oxygen level–dependent (BOLD) functional magnetic resonance imaging (fMRI), we evaluated how diabetes impacts these neural processes under euglycemic and hypoglycemic conditions when subjects were presented with a working-memory task (WMT). Diabetes is known to negatively affect working memory (8). This task evaluates functional effects that might reflect changes in brain structure and/or presage decreases in cognitive performance. A better understanding of the brain’s metabolic and physiological mechanisms underlying the cognitive functions implicated in working memory could lead to improved treatment strategies to help maintain cortical function in patients with diabetes during hypoglycemia (9).
BOLD fMRI is a well-established method for examining regional brain activation in response to physiological, pharmacological, sensory, or cognitive tasks (10). Studies that have examined brain activation in response to sensory stimulation or cognitive challenges using BOLD fMRI during hypoglycemic conditions in nondiabetic subjects (11–13) have shown that hypoglycemia reduces regional brain BOLD activation. This reduction in BOLD response during hypoglycemia has been attributed to low glucose levels causing decreases in neuronal activity, glucose oxidative metabolism, cerebral blood flow, neurovascular coupling, and/or neuronal recruitment (12).
Whether cognitive function in patients with type 1 diabetes is affected by hypoglycemia in the same manner as in nondiabetic individuals remains unclear because few studies using functional neural imaging have directly compared diabetic and nondiabetic subjects during the performance of cognitive tasks (14,15). If brain glucose transport or metabolism are altered in type 1 diabetes, as has been suggested in recent studies by our group (16) and others (17), then one would expect that the BOLD activation response during hypoglycemia may differ between diabetic patients compared with nondiabetic control subjects. On the basis of these findings, we hypothesized that 1) patients with type 1 diabetes would have greater BOLD activation during the performance of a WMT during hypoglycemia when compared with nondiabetic control subjects, 2) cognitive performance would deteriorate during hypoglycemia in both groups, and 3) among type 1 diabetic patients, better glycemic control (lower HbA1c) would correlate with BOLD activation responses to the WMT during hypoglycemia. We also conducted exploratory analyses to examine deactivation patterns in the default-mode network (DMN), the regions of the brain that are more active during rest (18), because of other research by our group examining the effects of diabetes on deactivation patterns during cognitive tasks and previous research suggesting that DMN function may be altered in diseases that affect cognition, such as Alzheimer’s disease (19).
RESEARCH DESIGN AND METHODS
The study sample consisted of 16 patients with type 1 diabetes and 16 healthy control subjects from an ongoing study of brain function during hypoglycemia in type 1 diabetes. Data from five type 1 diabetic and five control subjects were included in a previous publication (20). Demographic and clinical characteristics of the subjects are presented in Table 1. The patients’ number of self-reported hypoglycemic episodes (plasma glucose <3.9 mmol/L with concomitant symptoms) in the month preceding the initial visit averaged eight episodes (range 2–28). Patients with autonomic neuropathy (assessed by standard criteria ) (21), painful peripheral neuropathy, urinary albumin levels >300 mg/day, or proliferative retinopathy by review of medical records, physical exam, or self-report were excluded from the study. Other exclusion criteria were a history of psychosis or schizophrenia; cocaine, heroin, or alcohol dependence; and any contraindications to MRI, such as metallic implants, pregnancy, or claustrophobia.
Following approval from the institutional review boards of both the Joslin Diabetes Center (where patients were recruited) and the McLean Hospital (where the MRI was performed), patients provided the following information during screening: self-report of their lifetime experience of severe hypoglycemic events leading to unconsciousness (22); psychiatric history; handedness; medical history; and current medications.
Of 18 control participants who were eligible for the study, 2 were excluded from analysis because of excessive head motion during the fMRI by applying exclusion criteria of translations in excess of 3 mm in x, y, or z directions or rotations in excess of 3° around the x, y, or z axes. No diabetic subjects were excluded for head motion.
The experimental protocol is described elsewhere (20) and briefly reviewed below. On the day before the study, patients with type 1 diabetes had a continuous glucose monitor (CGM System Gold; Medtronic, Northridge, CA) inserted. If the continuous glucose monitor showed glucose <3.3 mmol/L, the study was postponed to a later date. The experiment used the insulin clamp technique with four successive time periods corresponding to different plasma glucose levels: baseline (30 min); euglycemic clamp (40 min, target glucose 5.0 mmol/L); declining glycemia (40 min, plasma glucose reduced from 5.0 to 2.8 mmol/L); and hypoglycemic clamp (30 min, target glucose 2.8 mmol/L). Anatomical MRI was performed during baseline, and fMRI was performed during the euglycemic and hypoglycemic periods while the WMT was administered (Fig. 1).
Insulin clamp technique.
An intravenous catheter was inserted into an antecubital vein for the administration of insulin and glucose, and a second catheter was inserted into a distal forearm or hand vein for the withdrawal of blood samples. A heated gel pack was used to warm the hand to arterialize the venous blood. After the baseline period, regular human insulin was infused at 12 pmol/kg per min for 110 min. The plasma glucose levels were maintained at the desired level by infusion of 20% dextrose using a negative-feedback algorithm, as previously described (23–25). After the euglycemic clamp period (40 min), the glucose infusion rate was reduced to allow the plasma glucose level to decline by 2.2 mmol/L (from 5.0 to 2.8 mmol/L) over the next 40 min, followed by the 30-min hypoglycemic clamp period. During the entire clamp protocol, glucose levels were measured every 5 min, and counterregulatory hormones (epinephrine, cortisol, growth hormone, and glucagon) were measured every 10 min. At the end of the protocol, the insulin infusion was discontinued, the glucose infusion was increased to restore euglycemia, and the subjects were given a meal and discharged.
Hormone and substrate assays.
Plasma glucose was measured using the glucose oxidase method. Serum insulin and growth hormone and plasma epinephrine were measured by enzyme-linked immunosorbent assay. Plasma glucagon and cortisol were measured by radioimmunoassay.
The task stimuli were administered using Presentation software (Neurobehavioral Systems, Albany, CA). The images were projected on a backlit screen that was visualized from within the magnet bore by a mirror mounted on the head coil. Response times and errors were collected using a magnetic resonance–compatible hand-held four-button fiberoptic response pad (FORP; Current Designs, Philadelphia, PA) connected to the PC by an optical cable interface. The stimulus presentation was synchronized with the fMRI acquisition sequence at the beginning of each trial by the interface that responded to scanner-generated trigger signals. The WMT was administered 15–20 min after the beginning of the euglycemic period and again at 15–20 min after the beginning of the hypoglycemic period (Fig. 1). Subjects performed practice tests before the study to achieve familiarity with the test procedure, and we used randomly assigned counterbalanced forms of the test during euglycemia and hypoglycemia to minimize learning.
During the WMT, adapted from Rypma et al. (26), participants viewed a string of six digits for 1,200 ms followed by a 2,000-ms unfilled retention interval (blank screen). Then, one digit was displayed on the computer monitor for 1,500 ms, and the subject had 1,300 ms to decide whether the digit was a member of the previously seen string. Using the same timing parameters, a matched control task was used in which subjects viewed six percent symbols (%) on the screen, followed by the unfilled retention interval, followed by a right- or a left-pointing arrow. The participants were asked to press the response key corresponding to the direction of the arrow. In the rest task, participants fixated on a plus (+) sign for 30 s. Each of these three tasks (WMT, matched control, and rest) was presented four times in a block design for a total of 6 min of testing. There were 5 memory-scanning trials per block for a total of 20 memory-scanning trials.
MRI acquisition methods.
All magnetic resonance images were acquired with a 3.0 Tesla Siemens Trio scanner (Siemens, Erlangen, Germany), using a circularly polarized birdcage radiofrequency head coil tuned to the proton frequency. Global field uniformity was adjusted at the beginning of each scanning session using Siemens’ automated shimming routines. A three-dimensional, T1-weighted anatomical image was acquired using a magnetization-prepared, rapid-acquisition gradient echo sequence (TR/TE = 2,100/2.74 ms; spatial resolution = 1 × 1 × 1.3 mm3; matrix size = 256 × 256 × 128) and used for functional image registration. BOLD fMRI images covering the whole brain were acquired in the axial plane using an echo planar imaging (EPI) sequence (TR/TE = 3,000/30 ms; 26 slices acquired in interleaved scanning order; slice thickness = 5 mm; field of view = 200 × 200 mm2; matrix size = 64 × 64), with an in-plane resolution of 3.125 × 3.125 mm2. One multislice EPI volume covering the whole brain was acquired every 3 s for a total of 120 volumes for the 6-min fMRI run.
Image processing and analyses were performed using the FSL software package (Analysis Group, FMRIB, Oxford, U.K.) running on a Mac-Pro Quad-Core Intel-Xeon computer (Apple, Cupertino, CA). All brain images were registered to the standard MNI-152 brain (Montreal Neurologic Institute) to allow multisubject analyses in standardized space and identification of all regions of interest. The first two EPI volumes of each functional run were discarded to allow for T1 equilibration. Prestatistical processing of EPI time series consisted of motion correction, slice scan-time correction, nonbrain removal, spatial smoothing with a 6-mm full-width at half-maximum three-dimensional Gaussian filter, linear trend removal, and a temporal high-pass filter with a cutoff of 120 s. Functional data were overlaid on the MNI-152 brain.
Regional brain activations and deactivations in response to the WMT were examined by performing a first-level general linear model (GLM) multiple regression analysis for each subject and glycemic condition. Model predictors for BOLD signal time courses were the boxcar time courses for each task stimulus paradigm convolved with a γ function to account for hemodynamic response. Predictors for the temporal derivatives of the task paradigm were added to the design matrix to improve the fit to the data by allowing for potential misspecification of the hemodynamic delay (27). The model for the baseline rest condition predictor was a constant function.
To compare activation and deactivation patterns between groups and conditions, higher-level multisubject mixed-effects GLM analyses compared multiple regression correlation coefficients associated with each model predictor for each brain voxel for each subject under each condition. Activations were computed by contrasting the WMT task to its control task. Deactivations were computed by contrasting the WMT task to the rest period. These higher-level analyses were performed using FSL’s local analysis of mixed effects stage 1. Resulting Z score (i.e., Gaussianized T/F) statistic images were thresholded using clusters determined by Z > 2.3 and a corrected cluster significance threshold of P = 0.05 (28). These statistical parametric maps yield the clusters of adjacent brain voxels of regional activation or deactivation from which we computed a single brain volume of activation or deactivation at the set statistical significance threshold (P < 0.05).
To investigate the variation of the regional percent BOLD signal change, we applied a region-of-interest (ROI) analysis in the superior parietal lobule (SPL) region, which had significant activation differences during hypoglycemia. To define the ROI, we created a binary mask of the SPL region provided by the probabilistic Harvard-Oxford Cortical Structural Atlas (29) thresholded at the 50% probability level. The ROI analysis was performed using FSL’s FEATQUERY tool to extract the average time courses of parameter estimates of the percent BOLD signal change from all voxels contained in the SPL mask.
To investigate the association of WMT activation and deactivation to glycemic control in type 1 diabetes, we performed the mixed-effects GLM group analyses while including HbA1c values as a covariate in the design matrix. To compute the regional coefficients of correlation of activation to HbA1c, the average percent BOLD activation values were extracted from the regions of significant correlation in type 1 diabetes during hypoglycemia using the FEATQUERY tool.
Standard statistical tests, including t tests for paired and unpaired data as appropriate, and ANOVA were used to compare glucose and counterregulatory hormone levels between diabetic and nondiabetic subjects during euglycemia and hypoglycemia. All data are presented as means ± SEM, unless otherwise specified, and statistical tests were conducted using a two-sided α level of 0.05.
Plasma glucose and counterregulatory hormones.
Average plasma glucose levels during the euglycemic period for nondiabetic control and type 1 diabetic subjects were 5.0 ± 0.4 mmol/L and 5.0 ± 0.6 mmol/L, respectively. During the hypoglycemic period, average glucose levels for the two groups were 2.8 ± 0.2 mmol/L and 2.7 ± 0.1 mmol/L, respectively. There was no significant difference between groups and no significant interaction between the glycemic condition and group. Mean counterregulatory hormone levels during baseline and euglycemia and peak levels attained during hypoglycemia are shown in Fig. 2. Glucagon and epinephrine secretion were lower in the type 1 diabetic patients during hypoglycemia compared with control subjects.
Type 1 diabetic and control subjects did not differ in accuracy or reaction time during the WMT at either euglycemia or hypoglycemia. During euglycemia, control subjects achieved 85 ± 2% correct, with a reaction time of 1,133 ± 54 ms, whereas type 1 diabetic patients had 85 ± 4% correct, with a reaction time of 1,200 ± 57 ms. At hypoglycemia, control subjects had 84 ± 5% correct, with a reaction time of 1,122 ± 57 ms, whereas type 1 diabetic patients had 85 ± 4% correct, with a reaction time of 1,204 ± 58 ms. There were no significant within-subject differences for either group when comparing the performance between glycemic conditions and no interaction between group and condition.
Brain activations in response to the WMT for each subject group during each glycemic condition are shown in Fig. 3. The regions of greater BOLD activation for the WMT relative to the control task (P < 0.05, corrected for multiple comparisons using the cluster-based threshold method) (28) are indicated in a red-to-yellow color scale overlaid on the gray scale standard MNI-152 T1-weighted anatomical brain atlas. For both subject groups during both glycemic conditions, the WMT activated brain regions located in bilateral medial superior frontal gyrus, left precentral gyrus, bilateral SPL, bilateral middle frontal gyrus, bilateral anterior cingulate cortex, bilateral insula, left supramarginal gyrus, bilateral thalamus, bilateral inferior occipital cortex, and bilateral cerebellum.
During euglycemia, the overall extent of activation was similar in the diabetic patient and control groups, with activated brain volumes of 613 and 498 cm3, respectively. During hypoglycemia, the extent of regional BOLD activation decreased relative to euglycemia in both subject groups; however, it decreased less in patients with diabetes than in control subjects (P < 0.05). The regional patterns of activation also differed between groups. In type 1 diabetes, activation decreased mostly in the insula, whereas for control subjects, activation decreased mostly in the cerebellum. Regions with higher activation in type 1 diabetes during hypoglycemia include the bilateral SPL, bilateral posterior supramarginal gyrus, bilateral anterior and posterior cingulate gyrus, left inferior frontal gyrus and posterior middle temporal gyrus, right hippocampus, and the cerebellum. Thus, during hypoglycemia, activation volumes were 80% larger in type 1 diabetic versus control subjects, with activation volumes of 431 and 239 cm3, respectively (P < 0.05). The extended activation volume in type 1 diabetic relative to control subjects during hypoglycemia was mainly a result of maintenance of a higher percent BOLD activation during the WMT, as illustrated in Fig. 4, which compares the time courses of the percent BOLD signal estimates averaged over the SPL region for each subject group under each glycemic condition.
Regional brain deactivations in response to the WMT for each subject group during each glycemic condition are shown in Fig. 5. Regions of lesser BOLD activation (or more deactivation) for WMT relative to the resting task (P < 0.05, corrected for multiple comparisons using the cluster-based threshold method) (28) are indicated in a dark blue–to–light blue color scale overlaid on the standard MNI-152 T1-weighted anatomical brain atlas in gray scale, along with activations in red to yellow. During euglycemia, WMT deactivated brain regions located in bilateral medial frontal cortex, posterior cingulate cortex, and superior lateral occipital cortex for both groups. The extent of deactivation was similar in the type 1 diabetic and control groups, with only a 10% difference between groups. Overall, deactivated brain volumes were 45 and 41 cm3, respectively. During hypoglycemia, the extent of regional BOLD deactivation decreased relative to euglycemia in both subject groups; however, it decreased more for patients with type 1 diabetes (P < 0.05) (i.e., they were unable to significantly deactivate these regions).
The regional patterns of decreased deactivation also differed between groups (Fig. 5). For patients with type 1 diabetes, deactivation decreased mostly in the medial frontal and posterior cingulate cortices, whereas for control subjects, deactivation decreased mostly in the posterior cingulate cortex. Control subjects also had small deactivation volumes in the cerebellum (right crus II) during hypoglycemia only. Thus, during hypoglycemia, the extent of deactivation was almost 70% smaller in type 1 diabetic relative to control subjects, with deactivation volumes of 5 cm3 located in the superior lateral occipital cortex for patients and 17 cm3 located mainly in the medial frontal and superior lateral occipital cortices for control subjects (P < 0.05).
Finally, we examined the relationship of HbA1c levels with activation patterns in type 1 diabetes by entering HbA1c as a covariate into the GLM. We found that lower HbA1c levels were associated with higher activation in the right parahippocampal gyrus and the amygdala (Fig. 6) during hypoglycemia. We also found that HbA1c levels were not associated with deactivation.
This study demonstrates that patients with type 1 diabetes show a different pattern of brain activation in response to a WMT than do nondiabetic control subjects during hypoglycemia. Specifically, we found that for patients with type 1 diabetes during hypoglycemia, WMT-related activation responses were increased in several cortical regions, including the parietal and frontal cortices, hippocampus, and cerebellum. Task-induced deactivations, typically observed in the DMN during cognitive effort, were significantly suppressed during hypoglycemia in bilateral medial-frontal and posterior cingulate cortices for type 1 diabetic patients compared with control subjects. Activation and deactivation patterns were similar across groups during euglycemia. Behavioral performance on the WMT was similar across groups and conditions. Finally, HbA1c was inversely correlated with WMT activation during hypoglycemia in the right parahippocampal gyrus and amygdala, two areas that have been reported to activate in memory-disordered populations as a form of compensatory recruitment (30–32).
The regional BOLD activations we observed in response to the WMT were compatible with those found in other fMRI studies of similar WMTs (5,26,33–35). Although the main regions that help govern working memory are the dorsolateral and medial prefrontal cortices and anterior cingulate cortex, other regions, such as the parietal lobe (36) and cerebellum (37), are known to play supplementary roles. These regions were activated more in type 1 diabetic patients than in control subjects during hypoglycemia, suggesting that supplementary brain regions may have been recruited to help preserve cognitive performance. The failure to suppress activation in the DMN also is consistent with an interpretation that type 1 diabetic patients need to recruit more brain resources for cognitive preservation (11). Similar patterns of increased activation and decreased deactivation have been observed with mild cognitive impairment (30) and in older individuals at risk for Alzheimer’s disease (38), suggesting, in these cases, a compensatory response to accumulating pathology.
Cognitive performance was not altered by hypoglycemia in either subject group. Although some studies have shown similar results for less challenging WMTs (39), others have shown severe impairment during hypoglycemia for highly challenging WMTs involving reasoning (2). Of importance, a number of studies have used the same Sternberg WMT used here to evaluate differences in brain activation patterns across different populations (40,41). In our study, different brain activation patterns, despite similar cognitive performance across groups, suggest unique strategies of brain recruitment used across groups to augment performance during the glycemic challenge.
Our results showing hyperactivation of brain regions in type 1 diabetic patients with low HbA1c and hypoactivation in patients with higher HbA1c could reflect upregulation of glucose transport in the brain, as seen in patients with good glycemic control (42,43). Type 1 diabetic patients may engage more brain regions to maintain the same performance to compensate for cerebral inefficiency attributed to reduced brain resources (44). These results also are consistent with the patterns observed in many physiologic systems in which a period of compensatory hyperfunction precedes functional decline and ultimate organ-system failure.
In an earlier report from our group, we demonstrated that type 1 diabetic patients showed reduced gray-matter density in the parahippocampal gyrus associated with higher HbA1c (45). One possible explanation may be that gray-matter loss in this temporal region prevents its participation in the brain’s response to moderate acute hypoglycemia. Wessels et al. (46) also observed abnormal brain activity patterns in patients with diabetes retinopathy, which is more likely to be associated with elevated HbA1c levels. However, our studies differed in both design and data analytic methodology, making a direct comparison difficult.
The differences in the hypoglycemic BOLD response between patients and control subjects found in this study may be attributed to a variety of mechanisms, including preservation of global brain glucose uptake in diabetes as a result of adaptation to hypoglycemia (12,47), differences in neurovascular coupling, resting cerebral blood flow, neuronal activity linked to oxidative metabolism, increased brain glycogen stores (48), or a tendency to use nonglucose substrates to support higher neuronal activity (49). In addition, changes in brain glucose transport or metabolism, resulting in increased brain glucose levels, might occur as a result of recurrent hypoglycemia. We reported such a finding along with accompanying increases in glutamate that were correlated with decreases in memory and executive function (16). It may follow that abnormal glutamate metabolism contributed to altered neurovascular coupling in patients in our study. Although these adaptive mechanisms in type 1 diabetes may in the short-term allow compensation for altered brain activity patterns during a hypoglycemic challenge, they may presage long-term maladaptive or adverse consequences.
Although our study demonstrates greater activation during hypoglycemia in type 1 diabetes along with reduced deactivation of the DMN, the small sample size may limit the implications of our results. However, the regional BOLD activations we observed in response to the WMT were compatible with regional activations found in other fMRI studies of similar WMTs (5,26,33–35). The reductions in BOLD response observed during hypoglycemia were also compatible with other reports showing reduced brain BOLD activation in primary or association cortex in response to sensory, motor, or cognitive tasks in nondiabetic subjects (11,12,14,47). Also, this study was unable to resolve whether the alterations in brain activation were secondary to chronic hyperglycemia or recurrent hypoglycemia. Future studies can help resolve this issue.
In summary, patients with type 1 diabetes activate more brain regions than control subjects during hypoglycemia by maintaining activity from euglycemia to hypoglycemia in task-relevant regions and by failing to suppress activation in the DMN. This suggests that type 1 diabetic patients may need to recruit more brain resources to preserve cognitive performance. The pattern of hyperactivation of both the DMN and task-relevant regions is consistent with findings in disease states with impaired cognition, such as mild cognitive impairment and Alzheimer’s disease (38). There has been persistent concern about the consequences of recurrent hypoglycemia on brain structure and cognitive function. There are minimal long-term effects of recurrent hypoglycemia on cognition into middle age (50), but it is not clear whether this resiliency will last throughout the aging process. Future research should evaluate our findings as an early manifestation and warning of future clinically relevant cognitive decline. This research may guide the development of treatment regimens to enhance symptom recognition or to stabilize the neurochemical response to hypoglycemia to reduce the impact of glucodeprivation on cognitive function.
This study was supported in part by National Institutes of Health grants 5R01-DK073843-03, DK-60754, DK-62218-01A1, and P30-DK-36836 (Joslin Diabetes and Endocrinology Research Center); the Herbert Graetz Fund; and grant 5M01-RR001032-32 to the Beth Israel Deaconess General Clinical Research Center.
No potential conflicts of interest relevant to this article were reported.
N.R.B., G.M., A.M.J., and K.W. researched data, contributed to discussion, and wrote, reviewed, and edited the manuscript. R.L.M. researched data, contributed to discussion, and reviewed and edited the manuscript. V.F. researched data and reviewed and edited the manuscript. P.F.R. contributed to discussion and reviewed and edited the manuscript. D.C.S. researched data, contributed to discussion, and wrote, reviewed, and edited the manuscript.
Parts of this article were presented in abstract form at the 70th Scientific Sessions of the American Diabetes Association, Orlando Florida, 25–29 June 2010, and at the 71st Scientific Sessions of the American Diabetes Association, San Diego, California, 24–28 June 2011.
The authors thank Judi Lauerman, RN, Brigham and Women's Hospital, Boston, Massachusetts, and Karen Branch, RN, Joslin Diabetes Center, Boston, Massachusetts, for their expert nursing assistance.
- Received April 14, 2011.
- Accepted August 25, 2011.
- © 2011 by the American Diabetes Association.
Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. See http://creativecommons.org/licenses/by-nc-nd/3.0/ for details.