The trial was registered at clinicaltrials.gov
(NCT00438568) and conducted over a 4-year period. Our study was approved by the Human Subjects Review Committees of the University of Washington and the Veterans Affairs Puget Sound Health Care System and was conducted in the Veterans Affairs Clinical Research Unit. Written informed consent was obtained from all participants. A total of 104 older adults enrolled in our study (64 participants with aMCI and 40 participants with probable AD who had Clinical Dementia Rating scores of 0.5–1 and Mini-Mental State Examination scores of >15). Sample composition (combined aMCI and AD) and power estimates to determine sample size were based on a previous 3-week trial of intranasal insulin.13
Forty participants (15 participants received placebo, 13 participants received a 20-IU dose of insulin, and 12 participants received a 40-IU dose of insulin) who completed the main study also completed the PET substudy. Twenty-three participants (8 participants received placebo and 15 participants received insulin) who completed the main study also completed the lumbar puncture substudy. Diagnoses and eligibility were determined by consensus of expert physicians and neuropsychologists following cognitive testing, evaluation of medical history, physical examination, and clinical laboratory screening using modified Petersen criteria for the diagnosis of aMCI13,14
and National Institute for Neurological and Communicative Disorders and Stroke–Alzheimer’s Disease and Related Disorders Association criteria for AD. For participants with aMCI, cognitive scores were compared with an age- and education-adjusted estimate of the participant’s premorbid ability (Shipley Vocabulary test). Participants whose delayed memory scores deviated at least 1.5 SDs from this estimate were considered for the diagnosis of aMCI, which was then determined by expert consensus using all available data, following published criteria.14
Participants were free from psychiatric disorders, alcoholism, severe head trauma, hypoxia, neurologic disorders other than aMCI or AD, renal or hepatic disease, diabetes mellitus, chronic obstructive pulmonary disease, and unstable cardiac disease. Participants, study partner, and all study personnel involved in data collection were blinded to treatment assignment. Treatment groups did not differ significantly in terms of education, body mass index, general cognitive status as assessed by the modified Mini-Mental State Examination, sex, diagnosis, whether they received cholinesterase inhibitor treatment, or whether they carried the apolipoprotein E (APOE) ε4 allele. The participants who received 40 IU of insulin were younger than the placebo-assigned participants (P = .02), whereas no differences were observed between the placebo group and the group who received 20 IU of insulin (age was included as a covariate in all analyses). Enrollment data are presented in , and demographic information is presented in .
Figure 1 Patient enrollment flowchart for our trial, which examines the effects of intranasal insulin administration on cognition, function, cerebral glucose metabolism, and cerebrospinal fluid biomarkers in adults with amnestic mild cognitive impairment or Alzheimer (more ...)
Demographics of Intent-to-Treat Sample of 104 Adults With aMCI or Mild to Moderate AD
A nurse unaffiliated with the trial used a table of random numbers to randomly assign participants to receive a daily dosage of 20 IU of insulin (ie, 36 participants received 10 IU of insulin twice a day), 40 IU of insulin (ie, 38 participants received 20 IU of insulin twice a day), or placebo (ie, 30 participants received saline twice a day) for 4 months. Participants were stratified by whether or not they were carriers of the APOE ε4 allele. Saline or insulin (Novolin R; Novo Nordisk, Princeton, New Jersey) was administered after breakfast and dinner with a ViaNase nasal drug delivery device (Kurve Technology, Bothell, Washington) designed to deliver drugs to the olfactory region to maximize transport to the central nervous system. This device released a metered dose of insulin into a chamber covering the participant’s nose, which was then inhaled by breathing regularly for 2 minutes until the prescribed dose was delivered.
Parallel versions of the cognitive and functional protocol were administered at baseline, months 2 and 4 of treatment, and 2 months after treatment. Testing occurred in the morning after a standard meal. Participants were instructed to skip their morning dose on the day of testing and thus had received their last dose more than 12 hours prior to cognitive testing. The coprimary outcome measures were the delayed story recall score and the Dementia Severity Rating Scale (DSRS) score, both of which had previously demonstrated the beneficial effects of insulin.13
The protocol consisted of the following measures: (1) The delayed story recall score13
was determined after a story containing 44 informational bits was read a single time to participants who were then asked to recall the story immediately and after a 20-minute delay. (2) The DSRS score15
was determined after a questionnaire was completed by the study partner; this questionnaire was used to rate the change in the participant’s cognitive, social, and functional status over a specified period of time, with higher scores indicating greater impairment. (3) The Alzheimer Disease’s Assessment Scale–cognitive subscale (ADAS-cog)16
includes measures of memory, praxis, orientation, and language, with higher scores indicating greater impairment. (4) The Alzheimer’s Disease Cooperative Study–activities of daily living (ADCS-ADL) scale17
was completed by the study partner and used to rate the participant’s ability to perform daily activities within the past month, with lower scores indicating greater impairment.
After a 12-hour fast, an intravenous catheter was inserted, and the L4–5 interspace was infiltrated with lidocaine, 1%, for anesthesia. Using a 24-gauge Sprotte spinal needle (B. Braun Medical, Bethlehem, Pennsylvania), 30 mL of CSF was withdrawn into sterile syringes, aliquoted into prechilled polyethylene tubes, frozen immediately on dry ice, and stored at −70°C until testing.
Levels of Aβ1–42, tau protein, and P181-tau in CSF were measured with the multiparameter bead-based immunoassay (INNO-BIA AlzBio3; Innogenetics NV, Gent, Belgium). The CSF Aβ40 level was measured by sandwich enzyme-linked immunosorbent assays using 6E10-coated plates (Signet Laboratories, Dedham, Massachusetts) in conjunction with biotinylated anti-Aβ40 as previously described.18
The limit of detection was 15 pg/mL.
PET WITH FDG
Positron emission tomographic imaging was obtained using a GE Advance PET scanner (GE Medical Systems, Milwaukee, Wisconsin). The participants were kept in a quiet, dimly lit room with their eyes open. Following an injection of 10 mCi of FDG, participants were monitored for 35 minutes, after which they were then moved to the scanner table, and a transmission scan and a brief emission scan were obtained in 2-dimensional data acquisition mode for 25 and 5 minutes, respectively, for postinjection attenuation correction. The final emission scan was obtained for 15 minutes in 3-dimensional data acquisition mode. Attenuation-corrected transaxial image sets were reconstructed using a filter-backprojection method. Reconstructed attenuation-corrected 3-dimensional emission images yield an in-plane spatial resolution of approximately 5 mm.
SAFETY AND COMPLIANCE
Study partners supervised participants in the administration of intranasal treatment. The blood glucose level was measured daily for the first week and weekly thereafter; no group changes were observed during the study period. Compliance was monitored by quantifying unused medication and via self-report. Safety data were reviewed semiannually by a data safety monitoring board. Adverse event reporting followed standard guidelines.
Data were analyzed with SAS version 9.2 (SAS Institute, Cary, North Carolina). For the intent-to-treat sample, coprimary and secondary cognitive and functional outcome scores (coprimary, delayed story recall and DSRS scores; secondary, ADAS-cog and ADAS-ADL scores) received identical analytic treatment. All scores were log-transformed to normalize distributions. To test the primary hypothesis that 4 months of insulin treatment would improve delayed memory recall and daily function, the a priori analytic plan called for each of the insulin groups to be compared with the placebo group. However, to provide the most conservative approach, scores were first subjected to mixed-model repeated-measures analysis of covariance, including all treatment groups (placebo, 20 IU of insulin, and 40 IU of insulin), as the between-subjects factor, and time (baseline to month 4), as the repeated factor, using the general linear models procedure, type III sums of squares. After a significant (P
< .05) time × treatment group interaction reflecting a different pattern of change, each of the 2 insulin groups was compared separately with the placebo group using repeated-measures analyses of covariance. Effect sizes (Cohen f
) were calculated for all significant effects. Age was included as a covariate in all analyses. Diagnosis (aMCI or AD), sex, APOE
ε4 carriage status (yes or no), baseline modified Mini-Mental State Examination score, and years of education were also included as covariates. Nonsignificant covariates were dropped from the model, and the SAS least squares means option was used to calculate means adjusted for significant covariates. Significant relationships with covariates were explored with the Pearson correlation coefficient (continuous variables) or with follow-up analyses of variance (class variables). Missing values (3% of all cognitive and functional outcome data) were treated with multiple imputation.19
Exploratory analyses were undertaken using the above-mentioned analytic strategy to determine whether changes in primary outcome measures were apparent 2 months after treatment initiation and whether they remained apparent 2 months after treatment cessation.
For exploratory biomarker analyses, because only a subset of participants elected to undergo lumbar puncture, and because no differences were observed between the 2 insulin groups, these groups were combined into a single insulin-treated group to maximize power. Biomarkers were then analyzed with the repeated-measures analysis of covariance strategy, and owing to the small sample size, exploratory Spearman rank correlations were calculated to examine relationships among changes in biomarkers and outcome measures.
Only participants who completed our study underwent a posttreatment PET-FDG scan. For FDG-PET analysis, pretreatment and posttreatment scans were coregistered for each participant and anatomically standardized to Talairach and Tournoux stereotactic coordinates.20,21
Pixel intensity was normalized to cerebellar and pontine values, for which the regional CMRGlc is known to be less affected in the course of AD.22
Interval and regional CMRGlc changes within groups were assessed using voxelwise 1-sample t
statistics (pretreatment and posttreatment pair), and probability integrals were converted to z
Interval changes in the regional CMRGlc were then compared between (1) the group receiving 20 IU of insulin vs the placebo group and (2) the group receiving 40 IU of insulin vs the placebo group. Based on the number of voxels and smoothness of the statistical map, a type I error rate was controlled at .05 to account for multiple comparison.23
The resulting statistical maps were visualized in 3-dimensional stereotactic surface projections. To confirm voxelwise analyses, an independent analysis using stereotactically predefined volumes of interest was also performed.23
In this analysis, stereotactically predefined regions shown to be abnormal in AD (medial and lateral frontal, parietal, and temporal association cortices and precuneus/posterior cingulate) were applied to the standardized and normalized image data sets, and averaged counts within each region were normalized to those of reference regions as described previously.23
The volume-of-interest values in each region were subjected to repeated-measures analysis of variance using the identical strategy and the same covariates described above for cognitive measures, with an additional within-subjects factor of side (right or left).