Insulin Resistance Is Associated With Cognition Among HIV-1-Infected Patients: The Hawaii Aging With HIV Cohort
JAIDS Journal of Acquired Immune Deficiency Syndromes:
1 December 2006
Valcour, Victor G MD; Sacktor, Ned C MD; Paul, Robert H PhD; Watters, Michael R MD; Selnes, Ola A PhD; Shiramizu, Bruce T MD; Williams, Andrew E MA; Shikuma, Cecilia M MD
From the *University of Hawaii, Honolulu, HI; Johns Hopkins University School of Medicine, Baltimore, MD; and Brown Medical School, Providence, RI.
"In summary, this analysis indicates that calculated levels of IR are associated with prevalent cognitive dysfunction, and that an increasing degree of IR is associated with a poor performance on neuropsychological summary scores. The effect size seems to be strengthened among older cohort participants. Because IR is a frequent HIV-1 complication and because effective treatment strategies for IR are available, this area deserves further investigation."
Objective: To determine if insulin resistance (IR) is associated with lower cognitive performance among HIV-1-infected adults and to determine if advanced age magnifies risk.
Design: Cross-sectional analysis within the Hawaii Aging With HIV Cohort.
Methods: We calculated the homeostasis model assessment of insulin resistance (HOMA-IR) among 145 cohort participants. Values were compared to concurrent neuropsychological test performance and cognitive diagnoses.
Results: Hypertension, body mass index (BMI), and non-Caucasian self-identity were directly related to insulin resistance (IR); however, age, CD4 lymphocyte count, and rates of treatment with HAART were not. In logistic regression analyses and stratifying cognition status on a 3-tiered scale (normal, minor cognitive motor disorder (MCMD), and HIV-associated dementia (HAD)), we identified an increased risk of meeting a higher diagnostic category as HOMA-IR increased (OR, 1.12; 95% CI: 1.003 to 1.242 per unit of HOMA-IR, P = 0.044). In linear regression models and among nondiabetic participants, an increasing degree of IR was associated with lower performance on neuropsychological summary scores.
Conclusions: IR is associated with cognitive dysfunction in this contemporary HIV-1 cohort enriched with older individuals. Metabolic dysfunction may contribute to the multifactorial pathogenesis of cognitive impairment in the era of HAART.
Despite major reductions in AIDS morbidity, the prevalence of HIV-1-associated dementia (HAD) has not declined appreciably since the introduction of highly active antiretroviral therapy (HAART) in the developed world.1 Several classically recognized HAD markers, such as plasma HIV RNA, seem to be less predictive in this setting,2,3 and factors such as mitochondrial dysfunction, oxidative stress, and diabetes mellitus (DM) are being investigated as potential co-contributors to neurological injury in HIV-1.4,5 These intermediates are associated with the use of HAART itself, leading to the theoretical possibility that HAART-related toxicities may partially offset neurocognitive gains seen with HAART.
Disorders in glucoregulation (DM and insulin resistance [IR]) require scrutiny because they are known to be associated with cognitive dysfunction among HIV-seronegative adults6,7 and because therapeutic strategies exist. Such disorders were once uncommon among HIV-1-infected individuals; however, various cross-sectional studies have now documented impaired glucose tolerance in up to 46% of nondiabetic HIV-1-infected individuals, with few discrepant reports.8 The Multicenter AIDS Cohort Study (MACS), for example, identified a 3-fold increase in DM with HAART.9 The prevalence of glucoregulatory disorders in HIV-1 increases with advancing age, rendering middle-aged and older HIV-1 patients particularly at risk.10
We previously reported an increased risk for HAD among individuals with self-reported DM in the Hawaii Aging with HIV Cohort (HAHC).11 This risk was most applicable to older participants because over 80% of known diabetic cases in this cohort were age 50 or older. Based on a model of cerebral reserve, we hypothesized that older individuals would be more vulnerable to HAART-related toxicities and therefore more likely to exhibit cognitive findings in relation to metabolic abnormalities.4 In the same report, we observed an association between fasting glucose levels and cognitive status among nondiabetic HIV-1 individuals,11 leading us to consider the possibility that IR, per se, rather than factors associated with end-stage DM, may contribute to the associations identified. Fasting glucose levels alone are poor markers of insulin resistance. In the current study, we measured insulin levels to test the hypothesis that increased IR among HIV-1-infected patients is associated with an increased risk of HAD and with decreased overall neuropsychological testing performance in the era of HAART. We further evaluate this relationship among patients without diabetes to determine if abnormalities in glucoregulation relate to cognition in the absence of established diabetes.
In this analysis, we demonstrate an association between IR and prevalent cognitive dysfunction, particularly among middle-aged and older participants in the HAHC. These data suggest new vulnerabilities with important applicability to the growing group of aging HIV-1 patients and merit further investigation. Our findings do not seem to be simply related to the end effects of established DM, as stratification to evaluate nondiabetic patients separately identified similar associations.
Although our results meet statistical significance, the effects sizes are generally modest. This may relate to heterogeneity and fluctuation of HIV-1-related cognitive disorders in the current era, whereby multiple factors may contribute to the cognitive decline, but the independent contribution of any one factor could be small.1,4 There are also limitations in our measurements that could contribute to attenuated effect sizes. We used the best available static measures of IR in the absence of oral glucose tolerance tests; however, these measures capture only about two-thirds of the variability associated with impaired glucoregulation as measured by SSPG.20 It is conceivable that the real impact of IR on cognition is larger than measured, particularly if the effects prove to be cumulative.
There seems to be a greater strength of association between IR and cognitive performance among older participants as previously hypothesized.21 We observed such a relationship between peripheral neurological disease and aging.22 Age-related brain senescence and neurodegeneration may impact frequency and severity of HIV-1-related cognitive dysfunction in older patients, and HIV-1 can be considered to decrease further the brain's capacity to respond to insults in this setting.
This post hoc analysis was nested within our longitudinal cohort using only specimens that were obtained during fasting. We identified no apparent differences between participants who were fasting or those who were not. Nearly half of the individuals evaluated in this study self-identified as non-Caucasian. This raises some concerns that using neuropsychological normative data acquired predominantly from Caucasian individuals may bias our results. We did not have sufficient power to evaluate the relationships presented herein within only Caucasian participants; however, adjustment for educational level did not appreciably affect our outcomes. We included self-reported ethnicity in our models. Due to the age selection structure of the parent cohort, we were not able to evaluate associations among individuals in their 40s.
Because self-reported DM is known to under-estimate true rates by as much as 10% in HIV-seronegative populations,23 we included as diabetic all participants with a fasting glucose reading exceeding 125 mg/dL. We were not able to evaluate factors that are closely related to IR, such as alteration in body habitus, blood pressure abnormalities, and lipid abnormalities, because these variables are so closely associated with IR and we did not have sufficient cases to stratify data.
Our conclusions are consistent with a growing literature among HIV-seronegative adults identifying IR and the metabolic syndrome as risk factors for cognitive dysfunction.24-26 Elevated fasting insulin levels and HOMA-IR are associated with lower cognitive performance with specific applicability to cognitive impairment with subcortical features,7 a pattern similar to that described in classical HAD. The accumulation of evidence implicating IR has resulted in therapeutic trials of insulin sensitizers in early AD.27
A number of potential mechanisms may explain our findings. Because IR is associated with a proinflammatory state and oxidative stress,28 and because oxidative stress is known to be associated with HAD,29 this pathway deserves further evaluation. Induced plasma hyperinsulinemia during euglycemic clamp studies among healthy HIV-negative adults results in elevated levels of CSF F2-isoprostanes and CSF cytokines (IL-1α, IL-1ß, IL-6, and TNF-α).30
Furthermore, tumor necrosis factor-alpha (TNF-α) has been specifically associated with IR in HIV-1 patients31 and is intimately associated with apoptosis32 and the neuropathogenesis of HAD.33 Although speculative, monocyte trafficking could be affected as the gene encoding for MCP-1 has been recently identified as insulin responsive.34 Another model implicates IR as a means of altering central nervous system (CNS) insulin levels,6 resulting in lower insulin levels in the CNS and decreased insulin-related facilitation of cognition. Among HIV-negative patients, insulin administered to the CNS by a novel intranasal delivery systems also improves memory performance.35 The proposed alteration in central insulin levels may also mediate changes in amyloid processing through insulin degradation enzyme (IDE), a metalloprotease that is highly expressed in the brain and liver and responsible for degradation of both amyloid beta and insulin in the CNS.36-38
In summary, this analysis indicates that calculated levels of IR are associated with prevalent cognitive dysfunction, and that an increasing degree of IR is associated with a poor performance on neuropsychological summary scores. The effect size seems to be strengthened among older cohort participants. Because IR is a frequent HIV-1 complication and because effective treatment strategies for IR are available, this area deserves further investigation.
Among the 273 (145 older and 128 younger) participants enrolled in the HAHC target groups between October 22, 2001, and June 30, 2005, 169 provided fasting specimens on at least one occasion. Baseline characteristics did not differ between individuals who were fasting and those who were not including important HIV clinical variables, such as CD4 count (P = 0.962), duration of HIV infection (P = 0.729), percent with detectable viral loads (P = 0.225), and percent on HAART therapy (P = 0.453). There were no apparent differences in maximal educational attainment (P = 0.674), sex (P = 0.065), age (P = 0.835), or ethnicity (P = 0.623). Participants were excluded from this analysis if they met substance dependence criteria at the time of entry, had a positive urine drug screen for psychostimulants on the day of neuropsychological testing, or had a history of stroke. Two individuals known to be taking insulin (both older) and one individual with an insulin value of 70 µU/mL (younger) were also excluded. Although there is no generally accepted upper limit for insulin levels, all other readings were <45 and we judged that this reading should be censured. The individual with an outlying insulin level met MCMD criteria at the time of insulin assessment, and one of the patients taking insulin met HAD criteria and the other was categorized as NP abnormal. The final sample for this analysis included 145 observations (74 older and 71 younger).
For statistical analyses relating to our main outcome variables, we introduced IR as a continuous variable to maximize information in our data. However, to display relationships between demographic, metabolic, and HIV parameters to insulin resistance, we first stratified our data by IR tertile (Table 2). Individuals in the highest IR tertile by HOMA-IR had a greater body mass index (BMI), were more likely to have hypertension and DM, and were more frequently self-identified as non-Caucasian. There were no significant differences in important HIV parameters or in rates of treatment with HAART among tertiles. There were no apparent differences in age by tertile. This lack of age effect was confirmed in a linear regression model with age and HOMA-IR entered as continuous variables (P = 0.183).
Using logistic regression models with insulin as a continuous variable and considering cognition on a 3-tiered scale (NL, MCMD, and HAD), increasing levels of serum insulin were associated with an increased risk of meeting a higher degree of impairment (OR, 1.04; 95% CI: 1.003 to 1.073, per unit of insulin, P = 0.034). The mean insulin levels were 12.3, 14.6, and 17.1 µU/mL for NL, MCMD, and HAD, respectively (Fig. 1, left figure). A similar analysis of HOMA-IR revealed a similar association for increased risk (OR, 1.12; 95% CI: 1.003 to 1.242, per unit of IR, P = 0.044). The mean IR levels were 3.0, 3.8, and 4.3 for NL, MCMD, and HAD, respectively (Fig. 1, right figure).
We hypothesized that the effect of IR may be strongest in older individuals given our model of increased vulnerability in this population. We explored the relationship between IR and cognition in our older subgroup (age >49 years old at entry into the study). Here, the odds ratios were 1.090 (95% CI: 1.032 to 1.151) per unit of insulin (P = 0.002, mean insulin level of 10.5, 15.9, and 19.0 µU/mL for NL, MCMD, and HAD, respectively) and 1.30 (95% CI: 1.095 to 1.552) per unit of IR (P = 0.003, mean HOMA-IR of 2.41, 4.32, and 4.82 µU/mL, respectively). We explored the role of ethnicity, current CD4 count, HAART, BMI, age, and duration of HIV-1 infection in the HOMA-IR model among older individuals; however, only ethnicity contributed to the model. The odds ratio related to IR among older HIV patients after adjustment for ethnicity was 1.268 (95% CI: 1.061 to 1.516; P = 0.009). We did not identify any statistically significant relationships among younger individuals (<40 years old upon entry into the study) for either insulin or insulin resistance in univariate models (P = 0.770 for insulin and P = 0.661 for HOMA-IR).
We wanted to demonstrate that the association between insulin resistance and cognition existed among participants without known diabetes. Here, we identified a similar level of risk. The odds ratio for increasing HOMA-IR adjusted for ethnicity was 1.169 (95% CI: 1.006 to 1.358, P = 0.042). We identified a weak effect for insulin alone (OR, 1.039; 95% CI: 0.999 to 1.080, P = 0.055)
A final set of models were designed to maximize information in our data. Here, we evaluated the association between HOMA-IR as a continuous variable and the neuropsychological summary scores. Increases in HOMA-IR was associated with lower performance on the NPZ-8 neuropsychological summary score (Table 3), and a trend was noted between HOMA-IR and the global deficit score. When data were re-evaluated considering only nondiabetic participants, an association was noted in all summary scores.
The Hawaii Aging With HIV Cohort (HAHC)
The HAHC was designed to evaluate the fundamental neuroepidemiology of aging with HIV-1 infection. Details of enrollment and clinical characterization are published elsewhere.12 Briefly, participants are enrolled if age is over 49 years (older group) or between 20 and 40 years (younger group) and major exclusion criteria, including head injury, learning disability, major neurological or psychiatric disease, or brain opportunistic disease, are not present. Participants are enrolled regardless of HAART status. Using careful recruitment strategies, we enrolled a cohort representative of patients living with HIV-1 in Hawaii.13
Annual evaluations include a macroneurological examination, neuropsychological testing, medical intake with record of comorbid illnesses, demographic data, risk behavior inventory, HIV laboratory parameters, and medication histories. The neuropsychological testing battery is designed to assess multiple cognitive domains in a standardized fashion while remaining feasible and practical for a large cohort that includes older individuals (Table 1). The battery measures performance in most major cognitive domains and is a selective expansion of that used in the Northeast AIDS Dementia Cohort (NEAD).14 All participants signed IRB-approved consents before participation.
Raw scores on neuropsychological tests were transformed to standardized z scores using published age- and education-matched normative datasets.12 A consensus conference of HIV neurologists, neuropsychologists, and a geriatrician determined research-based cognitive diagnoses based on American Academy of Neurology criteria utilizing all available participant data.15 Individuals with neuropsychological deficits that were too mild to meet HAD or MCMD criteria and participants with deficits of any magnitude but without evidence of functional compromise were categorized as neuropsychological testing abnormal (NP abnl). For the purposes of this analysis, we collapsed the NP abnl and normal groups.
Three composite neuropsychological indices were examined in the present study, including the NPZ-8, the Global Deficit Score, and the mean number of cognitive domains that were impaired. Neuropsychological testing scores provide an objective measure of cognitive performance and summary (composite) scores allow a more comprehensive assessment than that provided by independent tests. The NPZ-8 score is defined as the average of z scores for Timed Gait, Grooved Pegboard Dominant hand, Grooved Pegboard Nondominant hand, Trail Making Test Parts A and B, the Digit Symbol subtest from the WAIS-R, and the Choice and Sequential Reaction Time trials from the CalCap test battery. This common composite score was first reported to relate to cognitive improvement after zidovudine treatment.16 Our version substitutes the Choice and Sequential reaction times from the CalCap test for the Finger Tapping tests. This score is limited by disproportionate emphasis on tests of psychomotor and motor speed, which may be less applicable in the era of HAART.17 For participants over age 60, the NPZ-8 did not include the CalCap tests due to limited normative data for this age group.
We also used 2 additional contemporary summary scores. The Global Deficit Score was devised by Heaton and others to provide a weighted scheme that summarizes all available neuropsychological testing.18 Applying a similar technique to our testing battery and using a cut point of 0.85, the summary score provides a sensitivity of 0.91 and a specificity of 0.89 for the diagnosis of HAD at entry into the HAHC. Similarly, a cut point of 0.40 provides a sensitivity of 0.76 and a specificity of 0.74 for diagnosis of MCMD. In the third measure, we analyzed the mean number of cognitive domains impaired based on cognitive domains listed in Table 1. Here, impairment in one domain was defined as greater than or equal to a score of -1 standard deviation on any test in that domain.
Definition of Insulin Resistance
We identified all participants in the cohort who had been fasting at the time of specimen collection either at entry or at any annual visit. Fasting was defined as having nothing by mouth, except water, for at least 8 hours before blood draw. We highly encouraged, but did not require, participants to fast at the time of their visits. When individuals were fasting on more than one occasion, we included only the first fasting specimen and compared it to the concurrent cognitive data for that visit. Serum glucose determinations were completed in real-time at a clinical reference laboratory (Diagnostic Laboratory Services, Honolulu, HI) following a standardized protocol for specimen processing in accordance with Adult AIDS Clinical Trials Group (AACTG) guidelines. Briefly, blood was drawn in Serum Separator Tubes, allowed to clot, and then spun within 1 hour of bleeding to separate serum. Stored serum (frozen to -70°C), obtained at the same blood draw, was later sent to Quest Diagnostic Laboratory (Baltimore, MD) in batches where insulin levels were determined by chemiluminescence. Estimates of IR were calculated using the homeostasis assessment model of insulin resistance (HOMA-IR).19 HOMA-IR and fasting insulin levels are robust estimates of insulin-mediated glucose uptake as measured by steady-state plasma glucose (SSPG) during the insulin suppression test. In HIV-1-infected patients, the univariate correlation coefficients between SSPG and both insulin levels and HOMA-IR range from 0.61 to 0.68.20 For the purpose of this analysis, all participants with a known diagnosis of DM or with a fasting blood glucose level exceeding 125 mg/dL were considered to be diabetic.
We tested hypotheses concerning categorical endpoints using logistic regression models and hypotheses concerning continuous endpoints using standard regression models. Statistical analyses were carried out on SAS, v9.1 (SAS, Inc., Cary, NC) using PROC LOGISTIC and PROC GLM procedures. Cognitive diagnosis served as the main predictor variable and was stratified on a 3-tiered scheme of NL, MCMD, and HAD in logistical regression models. We also evaluated insulin and insulin resistance as continuous variables as predictors of neuropsychological performance on a continuous scale, using linear regression models. In regression analyses, ethnicity was entered as a dichotomous variable of Caucasian/non-Caucasian and HAART status was included as on/off antiretroviral therapy consisting of at least 3 potent antiretrovirals.