The role of HCV coinfection in HIV infected persons is also unclear. In the FIRST study, HCV coinfection was associated with a higher risk of DM in the antiretroviral-naive HIV infected population who were < 50 years old,⁸ while no increased risk was found in the Swiss HIV Cohort Study⁹ or an urban cohort of HIV infected persons in New York city.¹⁰ While it is generally accepted that protease inhibitor use, or care in the CART era is associated with an increased risk of DM,⁴ at least two studies do not support this assertion.^{10, 11}

For the purpose of our study, all subjects enrolled in VACS were eligible to be included in the analysis. HIV infection was defined based on ICD-9 codes and verified by an antibody test with a Western blot confirmation. The primary outcome was diabetes at baseline, which was defined as having any of the following: 1) Glucose ≥ 200 mg/dl on two separate occasions; 2) ICD-9 codes (two outpatient OR one inpatient) PLUS treatment with an oral hypoglycemic or insulin for ≥ 30 days; 3) ICD-9 codes (two outpatient OR one inpatient) PLUS glucose ≥ 126 mg/dl on two separate occasions; 4) Glucose ≥ 200 mg/dl on one occasion PLUS treatment with an oral hypoglycemic or insulin for ≥ 30 days. We created this multifacted definition to avoid biases that might be introduced by an exclusive focus on ICD-9 diagnostic codes. Diabetes medication prescription and ICD-9 codes have previosuly been used as diagnostic criteria for diabetes in the veterans.¹⁸ We compared our definition with presence of at least one inpatient or at least 2 outpatient codes for diabetes. Our definition had a sensitivity of 86.6%, specificity of 97.5%, agreement of 95.5% with a kappa value of 0.85 (95% CI 0.83-0.86), suggesting excellent agreement beyond chance. Lab data were obtained closest to the baseline date. Over 94% of the subjects had a biochemical profile performed within 360 days prior to the baseline visit. HCV coinfection was determined based on at least one inpatient or 2 outpatient ICD-9 codes. Quantity and frequency of alcohol use was determined using the AUDIT-C instrument. Illicit drug use was determined by subject self report of presence of at least one inpatient or 2 outpatient ICD-9 codes for drug abuse or dependence. Height and weight were measured as a part of routine clinical care and extracted from subject medical records at the enrolment date. Hypoglycemic medication and antiretroviral use and duration at baseline was determined and calculated from the subjects’ electronic medical records retrieved locally and through the PBM database.

We conducted analyses on all subjects stratified by HIV infection. We also conducted analyses including alanine and aspartate aminotransferase levels with and without HCV in the model, as well as separate comparative analyses on subjects with and without alcohol and drug use to explain some of the observed differences in the risk of DM. Subjects were considered to have received CART if they received ≥ 3 drugs from ≥ 2 classes of antiretrovirals. Individual class use was calculated as the cumulative number of days each class of antiretrovirals (nucleaside reverse transcriptase inhibitors [NRTI], non-nucleoside reverse transcriptase inhibitor [NNRTI], protease inhibitor [PI]) was used in any given subject. We analysed antiretroviral use as “any CART”, “use of each class per year”, as well as combinations of classes for ≥ 1 year to understand fully the role of antiretroviral therapy upon the risk of DM within the HIV-infected group.

In a univariable logistic regression model, HIV was associated with a lower risk of DM at baseline (OR 0.64, 95% CI 0.56-0.73). (Table 2) In the overall group, increasing age, male gender, minority race, and BMI were associated with an increased risk of DM. Increasing amount of alcohol use and a history of drug abuse or dependence were associated with a lower risk of DM. The effect of increasing age, minority race and BMI were more pronounced in the HIV infected group compared with HIV uninfected controls. HCV coinfection was associated with a higher risk of DM in the HIV infected, but not in the HIV uninfected group. (Table 2)

In multivariable analysis, HIV was still associated with a lower risk of DM (OR 0.84, 95% CI 0.72-0.97). (Table 3) Other factors associated with DM were similar to the univariable analysis. Increasing age and BMI had a more pronounced effect on the risk of diabetes in the HIV infected compared to HIV uninfected persons. HCV remained a significant predictor only in the HIV infected group.

In the HIV infected group use of CART was associated with a higher risk of DM (OR 1.11, 95% CI 1.05-1.17, result not otherwise shown). While each class of CART was associated with a higher risk of DM in the univariable model, only NRTI and NNRTI use was associated with a higher risk in the multivariable model. (Tables 2 and ​and3).3). When we analyzed various combinations of CART classes for varying durations (≤ 1 year vs. > 1 year), cumulative use of NRTI or PI or both was associated with a significantly higher risk of DM compared with the group of subjects with cumulative exposure to both classes for ≤ 1 year. (data not shown) Increasing CD4+ lymphocyte count was associated with an increased risk in univariable analysis, but not in the multivariable analysis. There was no association of DM with HIV RNA levels (data not shown).

We found that HCV infection is associated with a higher risk of DM in the HIV infected group, and demonstrated a similar trend in the uninfected group in multivariable analysis (although this trend did not reach statistical significance, the effect size was similar). This confirms the suggestion from some previous studies that HCV affects the risk of DM in a complex manner, and such risk is influenced significantly by other more traditional risk factors.^{10, 21} In our analysis, the risk conferred by HCV is not altered by the presence of liver damage as measured by elevated alanine and aspartate aminotransferase levels. (data not shown) Whether HCV and HIV act synergistically at a cellular level or through other factors to increase the risk of DM is not known. Insulin resistance and higher levels of inflammatory cytokines are also seen in patients with chronic HCV (but not with HCV/HIV coinfection as referenced in the previous paragraph), and may be one common pathway leading to a higher risk of DM.^22-25

We found that use of CART was associated with a significantly higher risk of DM in the HIV infected group. This is essentially a confirmation of multiple previous studies that have demonstrated induction of insulin resistance and a higher risk of DM with the use of CART. The precise role of each class, or each drug in the CART regimens is extremely difficult to determine since such therapy is always used in combination, and often changes in individual subjects. We studied the role of CART in several different ways, including cumulative exposure, current exposure and past exposure for each class (as a part of CART) as well as CART itself. The results were generally consistent indicating a higher risk of DM with use of NRTI and NNRTI. However, the association with PI use was not significant in the multivaraible model. Mitochondrial toxicity associated with nucleoside reverse transcriptase inhibitors (e.g. stavudine, zidovudine and didanosine), ^26-29 likely plays a role in the risk associated with NRTI class of drugs. Since PIs are almost never used alone, it is possible that the risk being attributed to NRTIs is at least partly due to the use of PIs. It is notable that the mean duration of NRTI use was nearly twice that of PI use, and duration of NNRTI use was significantly less than duration of NRTI or PI use. In addition, it is also possible that patients with higher risk for DM or with existent DM are more likely to be treated with a NNRTI containing regimen instead of a PI containing regimen. The risk of DM due to each class of drugs may be related to a cumulative dose effect, and individual class exposure cannot currently be separated from concurrent use of a second or third class of the CART regimen.

Our finding of a lower risk of DM associated with increasing alcohol use and drug use is intriguing. Increasing quantity/frequency of alcohol use was associated with increasing protection except in HIV infected persons who consumed > 60 drinks per month. Increasing alcohol use is associated with increasing liver damage, and may be expected to increase the risk of diabetes. We conducted separate analyses including liver damage (defined as alanine or aspartate aminotransferase levels > 5 times upper limit of normal) with and without HCV in the models and found no significant association with liver damage. It is also plausible that increased alcohol consumption and drug abuse or dependence may lead to poor nutrition and lower BMI which may indirectly dilute the association with DM. However, we found no significant association between quantity and frequency of alcohol use and BMI. Another possibility is that people with alcohol and drug abuse may not seek medical care and the opportunity to diagnose DM may have been missed. We did find that non-drinkers were older, while moderate to heavy drinkers were more likely to be younger. These data suggest that while some of the decreased association with alcohol is due to the alcohol consuming population being younger, there are other likely mechanisms that modulate this effect.

There are many strengths of our study. This study was conducted in a large, prospective cohort of HV infected persons and appropriate controls. VACS has validated measures of alcohol and drug abuse, as well as well defined algorithms for identifying comorbidities. We use multiple sources of data (surveys, electronic medical records, national data) to ensure as complete and accurate data collection as possible. However, certain limitations should also be noted. Most important, we analyzed prevalent diabetes, not incident diabetes. While we find a negative association between HIV infection and diabetes at baseline, it is entirely possible that incidence rates may be very different. Data on BMI and laboratory data were gathered in the course of routine clinical care. There are few women included in this analysis. Little is known about alcohol consumption patterns among women infected with HIV. Family history of DM is an important risk factor and was not determined in our study. Most of our study subjects were enrolled between 2002-2004. With the approval of newer antiretrovirals and updated recommendations about initial regimens, it is possible that the risk associated with CART may have changes. Finally, it has been argued that the veterans in care are a non-representative sample for the US population in general. However, with the exception of this gender difference, HIV infected veterans are similar to many other HIV infected persons, being more likely to be people of color, having contracted HIV via injection drug use or heterosexual exposure, and to be of lower socioeconomic status.² According to the National Institutes of Health, the estimated prevalence of diagnosed and undiagnosed DM among people age 20 or older is 11.2% in men and 10.2% in women. (http://diabetes.niddk.nih.gov/DM/PUBS/statistics/, accessed February 20, 2009) Prevalence of DM is 9.8% among non-Hispanic whites and 14.7% among non-Hispanic blacks. Since only 2.5% of the HIV infecteed and 7.9% of the HIV uninfected subjects in our study were women, readers are cautioned to interpret any gender differences in our study with caution.