Incidence and Risk Factors for New-Onset Diabetes in HIV-Infected Patients

The Data Collection on Adverse Events of Anti-HIV Drugs (D:A:D) Study

  1. Stephane De Wit, MD, PHD1,
  2. Caroline A. Sabin, PHD2,
  3. Rainer Weber, MD3,
  4. Signe Westring Worm, MD4,
  5. Peter Reiss, MD, PHD5,
  6. Charles Cazanave, MD6,
  7. Wafaa El-Sadr, MD, MPH7,
  8. Antonella d'Arminio Monforte, MD, DMSC8,
  9. Eric Fontas, MD9,
  10. Matthew G. Law, PHD10,
  11. Nina Friis-Møller, MD, PHD4 and
  12. Andrew Phillips, PHD2
  1. 1Centre Hospitalier Universitaire Saint-Pierre, Brussels, Belgium
  2. 2Royal Free and University College, London, U.K
  3. 3University Hospital Zurich, Zurich, Switzerland
  4. 4University of Copenhagen, Copenhagen, Denmark
  5. 5Academic Medical Center, Amsterdam, the Netherlands
  6. 6Bordeaux 2 University, Bordeaux, France
  7. 7Columbia University, Harlem Hospital, New York, New York
  8. 8University of Milan, Milan, Italy
  9. 9Centre Hospitalier Universitaire Nice, Hôpital de l'Archet, Nice, France
  10. 10National Centre in HIV Epidemiology and Clinical Research, Sydney, Australia
  1. Corresponding author: Stéphane De Wit, MD, PhD, Department of Infectious Diseases, St. Pierre University Hospital, 322, rue Haute, B-1000 Brussels, Belgium. E-mail: stephane_dewit{at}stpierre-bru.be

Abstract

OBJECTIVE—The aims of this study were to determine the incidence of diabetes among HIV-infected patients in the Data Collection on Adverse Events of Anti-HIV Drugs (D:A:D) cohort, to identify demographic, HIV-related, and combination antiretroviral therapy (cART)-related factors associated with the onset of diabetes, and to identify possible mechanisms for any relationships found.

RESEARCH DESIGN AND METHODS—D:A:D is a prospective observational study of 33,389 HIV-infected patients; diabetes is a study end point. Poisson regression models were used to assess the relation between diabetes and exposure to cART after adjusting for known risk factors for diabetes, CD4 count, lipids, and lipodystrophy.

RESULTS—Over 130,151 person-years of follow-up (PYFU), diabetes was diagnosed in 744 patients (incidence rate of 5.72 per 1,000 PYFU [95% CI 5.31–6.13]). The incidence of diabetes increased with cumulative exposure to cART, an association that remained significant after adjustment for potential risk factors for diabetes. The strongest relationship with diabetes was exposure to stavudine; exposures to zidovudine and didanosine were also associated with an increased risk of diabetes. Time-updated measurements of total cholesterol, HDL cholesterol, and triglycerides were all associated with diabetes. Adjusting for each of these variables separately reduced the relationship between cART and diabetes slightly. Although lipodystrophy was significantly associated with diabetes, adjustment for this did not modify the relationship between cART and diabetes.

CONCLUSION—Stavudine and zidovudine are significantly associated with diabetes after adjustment for risk factors for diabetes and lipids. Adjustment for lipodystrophy did not modify the relationship, suggesting that the two thymidine analogs probably directly contribute to insulin resistance, potentially through mitochondrial toxicity.

Mortality and morbidity from HIV and its complications have dramatically declined since the advent of combination antiretroviral therapy (cART). However, metabolic disorders have emerged (impaired glucose tolerance and diabetes as well as lipid disorders), leading to an increase in cardiovascular disease (CVD).

Data from the D:A:D (Data Collection on Adverse Events of Anti-HIV Drugs) Study (a complete list of the members of the D:A:D Study Group can be found in the online appendix [available at http://dx.doi.org/10.2337/dc07-2013]) suggest that the risk of a myocardial infarction is more than doubled among HIV-infected patients with diabetes (1). Insulin resistance among treated HIV-infected patients is multifactorial: in addition to the common contributors to insulin resistance (e.g., obesity, genetic influences, and physical inactivity), antiretroviral drugs and lipodystrophy (which may be a consequence of treatment, particularly with thymidine analogs) are involved. The aims of this analysis were to estimate the incidence of new-onset diabetes among patients with no history of diabetes at entry to D:A:D, to identify demographic, HIV-related, and cART-related factors that were associated with the onset of diabetes, and to identify possible mechanisms for any relationships found.

RESEARCH DESIGN AND METHODS—

The D:A:D Study is a large, prospective observational study formed by the collaboration of 11 cohorts of HIV-infected patients. The primary aim of the study was to establish whether an association exists between the use of cART and an increased risk of CVD. The 11 cohorts currently contribute data on 33,389 HIV-positive patients followed at 212 clinics in Europe, the U.S., Argentina, and Australia. The D:A:D study methodology has been described in detail (1). Patients eligible for inclusion were all being actively followed up at the time of initiation of the D:A:D protocol, irrespective of antiretroviral treatment status. Patients were followed prospectively, and data were obtained during visits scheduled as part of regular medical care. Patient follow-up started between December 1999 and April 2001 (D:A:D Phase I) and April 2001 and January 2005 (D:A:D Phase II). At enrollment and at least every 8 months thereafter, standardized data collection forms were completed, including sociodemographic characteristics, clinical data (AIDS events and known risk factors for CVD), laboratory markers (CD4 cell counts, HIV RNA load, and total cholesterol, HDL cholesterol, and triglyceride levels), and treatment variables (antiretroviral treatment and drugs modifying lipid levels or risk of CVD). Use of ritonavir includes both full and boosting doses. Data are transformed into a standardized format, transferred to the coordinating center (Copenhagen HIV Programme, Hvidovre Hospital, Copenhagen, Denmark) as anonymized computerized files, and merged into a central dataset.

End point definition

Diabetes has been determined as a secondary D:A:D end point. All prospectively documented cases of diabetes were verified by the submission of a D:A:D event monitoring case report form. New-onset diabetes was defined as either definite diagnosis if fasting plasma glucose >7.0 mmol/l (126 mg/dl) was measured on two consecutive occasions or possible diagnosis in the case of a physician-reported date of diabetes onset and initiation of antidiabetic therapy.

Statistical methods

The rate of new-onset diabetes was defined as the number of cases of diabetes divided by the total person-years of follow-up (PYFU). PYFU were counted similarly to the count for the primary analyses of the D:A:D Study (1) from the date of enrollment to the date of a first diagnosis of diabetes, death, 1 February 2006, or 6 months after the patient's last clinic visit, whichever occurred first. Factors associated with new-onset diabetes were identified using Poisson regression models. We assessed the univariable relationships between duration of exposure to cART (defined as any combination including a protease inhibitor and/or nonnucleoside reverse transcriptase inhibitor[NNRTI]) and the rate of new-onset diabetes. We then investigated whether any identified relationship applied to all drugs similarly or whether it varied according to the type of antiretroviral drug received. These analyses were then adjusted to take account of possible demographic and clinical risk factors for diabetes, including age (fitted as a time-updated covariate), sex, transmission group (homosexual, injecting drug use, heterosexual, or other/not known), race (white, black, other, or not known), BMI (categorized as <18, 18–26, 26.1–30, and >30 kg/m2), smoking status (current, ex-smoker, never, and not known), patient's nadir CD4 count, duration of HIV infection before enrollment in D:A:D, and calendar year. We further adjusted the analyses to take account of changes in lipids (total cholesterol, HDL cholesterol, and log2-transformed triglycerides) and development of fat loss (lipoatrophy) or fat gain (lipohypertrophy). Each of these variables was included as a time-updated covariate in a separate multivariable model with treatment exposure and other demographic/clinical risk factors. All analyses were adjusted for cohort.

RESULTS—

Overall, 952 of the 33,389 patients in D:A:D had a diagnosis of diabetes at entry to the study, giving a baseline prevalence of 2.85% (95% CI 2.67–3.03). The characteristics of the remaining 32,437 patients are shown in Table 1. In these patients, over 130,151 PYFU, diabetes was diagnosed in 744 patients with a diabetes incidence of 5.72 per 1,000 PYFU (95% CI 5.31–6.13). Of these diagnoses, 474 (63.7%) were definite and 270 (36.3%) were possible.

The incidence of new-onset diabetes increased with cumulative exposure to cART (Fig. 1A). This was significant in univariable analyses (unadjusted relative rate per year of exposure to cART of 1.06 [95% CI 1.03–1.09]; P = 0.0001) and after adjustment for other potential risk factors for diabetes (1.11 [1.07–1.15]; P = 0.0001).

When we analyzed whether the relationship with drug exposure was similar for all antiretroviral drugs, several findings emerged (Table 2). The strongest relationship with new-onset diabetes was with exposure to stavudine (adjusted relative risk [RR] per year of exposure of 1.19 [95% CI 1.15–1.24], P = 0.0001; unadjusted rates are shown in Fig. 1B). Although exposures to zidovudine and didanosine were also associated with an increased risk of new-onset diabetes, exposure to ritonavir and nevirapine were both associated with reduced risk. No other antiretroviral drug was significantly associated with the incidence of diabetes after adjustment for exposure to these drugs. Other demographic and clinical factors associated with increased risk of new-onset diabetes were older age, male sex, greater BMI, heterosexual or injection drug user risk group, black African and other ethnicities, and earlier calendar year. Current cigarette smoking was associated with a reduced risk. After adjustment for these factors, there were only weak and nonsignificant relationships between new-onset diabetes and the patient's nadir CD4 count (RR per 50 cells/mm3 higher: 0.98 [0.96–1.00]; P = 0.06) and duration of HIV infection at enrollment in D:A:D (RR per additional year: 0.98 [0.96–1.00]; P = 0.09).

In a series of multivariable analyses adjusted for these demographic and clinical factors as well as for exposure to the five drugs, time-updated total cholesterol, HDL cholesterol, and lipodystrophy (either peripheral loss or central fat gain) were associated with new-onset diabetes. A 1 mmol/l higher total cholesterol level was associated with a 9% increased rate of diabetes (RR 1.09 [95% CI 1.03–1.15]; P = 0.001), a 1 mmol/l higher HDL cholesterol level with a 49% reduction in the rate of diabetes (0.51 [0.40–0.66]; P = 0.0001), a 2-fold higher triglyceride level with an 81% increase in diabetes rate (1.81 [1.67–1.95]; P = 0.0001), and fat loss or fat gain with 28 and 57% increases in the risk of diabetes, respectively (1.28 [1.04–1.57]; P = 0.02; and 1.57 [1.28–1.92]; P = 0.0001). Although it was not possible to fit models that included both total cholesterol and triglyceride levels because of the correlation between the two, models that included fat loss/gain as well as HDL cholesterol and triglyceride levels (Table 3) confirmed that HDL cholesterol (0.75 per 1 mmol/l higher [0.58–0.96]; P = 0.02), triglycerides (1.64 per 2-fold higher [1.50–1.80]; P = 0.0001), and fat gain (1.36 [1.09–1.68]; P = 0.006) (but not fat loss) were all independently associated with new-onset diabetes. Adjustment for these variables did not substantially modify the relationships between the five drugs and diabetes, although the relationship with stavudine was reduced from RR 1.19 to 1.13.

CONCLUSIONS—

Our results show a significant relationship between new-onset diabetes and exposure to cART, with this effect being mainly related to exposure to stavudine. However, exposures to zidovudine and didanosine were also associated with an increased risk, whereas exposures to ritonavir and nevirapine were both associated with a reduced risk of diabetes. Our findings are consistent with those of two other cohort studies that showed a similar relationship between exposure to stavudine and incidence of diabetes (2,3). We found a lower incidence of new-onset diabetes than in the Multicenter AIDS Cohort Study (MACS) (5.72 per 1,000 PYFU vs. 47 and 17 per 1,000 PYFU among individuals receiving or not receiving cART) (2). This difference could be related to different size and demographic compositions of both cohorts as the MACS involved white males exclusively, who were likely to be exposed to the typical North American diet and who were older and had higher BMI than the D:A:D participants. In addition, in the MACS a single elevated fasting blood glucose measurement was sufficient to establish a diagnosis of diabetes, a less stringent criterion than that used in our study. Importantly, in the MACS, fasting glucose levels were obtained as part of the assessments at predefined cohort visits, whereas in our cohort the validation of diabetes as an end point is dependent on the actual screening policy for glucose intolerance and diabetes, which is in place in each of the 212 treatment units contributing data to D:A:D, which undoubtedly differs between centers.

The association between diabetes/insulin resistance and stavudine/zidovudine might be explained through an indirect mechanism, i.e., lipoatrophy, a state that is associated with insulin resistance. The increased lipolysis observed in patients with lipoatrophy actually reflects their adipose tissue being insulin resistant. Lipolysis leads to increased circulating free fatty acids, which may reinforce insulin resistance in the liver and skeletal muscles (4).

However, clinical evidence for a direct effect of thymidine analog nucleoside reverse transcriptase inhibitors (NRTIs) on insulin sensitivity is also emerging. Antiretroviral therapy-naive subjects randomly assigned to stavudine- and didanosine-based therapy had a significant increase in homeostasis model assessment of insulin resistance at 1 month, whereas there was no change in those randomly assigned to abacavir and lamivudine (5). Exposure to stavudine and didanosine is associated with greater lipoatrophy, illustrating the link between drug-related lipoatrophy and insulin resistance (26). Stavudine exposure leads to depletion of mitochondrial DNA content, which may result in mitochondrial dysfunction. A recent study performed in healthy volunteers demonstrated that a 1-month exposure to stavudine reduces insulin sensitivity in parallel with a 58% reduction in muscle mitochondrial DNA, suggesting that mitochondrial dysfunction and insulin sensitivity were linked (7). Several studies in HIV-uninfected individuals have suggested that mitochondrial dysfunction precedes the onset of diabetes in insulin-resistant offspring of patients with type 2 diabetes. Alterations of several genes involved in mitochondrial oxidative phosphorylation have been identified in muscle samples of patients with type 2 diabetes and impaired glucose tolerance (810).

The other risk factors for diabetes identified in our study included male sex, older age, greater BMI, and black race, largely consistent with other studies in both the HIV-uninfected and HIV-infected populations (11). Current smoking status appeared to be marginally protective, contradicting some studies but consistent with results from the Multiple Risk Factor Intervention Trial (MRFIT) for the reduction of CVD (12). The lower incidence of diabetes in recent calendar years could be inherent to the study design, i.e., follow-up of a closed cohort in which the patients at risk (i.e., susceptible to develop an end point) experience this relatively soon after enrollment, reducing the subsequent risk in the cohort. It is also possible that the movement away from “old” drugs, particularly stavudine, toward alternative agents that are not associated with lipoatrophy development and have a lesser or no effect on mitochondria may partially explain this finding.

Our data do not show a significant relationship between cumulative exposure to protease inhibitors and new-onset diabetes. This result is in line with a recent study in protease inhibitor–exposed women, which showed that cumulative exposure to protease inhibitors was not associated with incidence of diabetes, whereas cumulative exposure to NRTIs was (3). Antiretroviral regimens that include protease inhibitors for the treatment of HIV-1 have been associated with new-onset diabetes and insulin resistance (13,14). Reversal of hyperglycemia after protease inhibitor withdrawal, onset of hyperinsulinemia before measurable body composition changes in protease inhibitor recipients, and improvements in insulin sensitivity after substitution of protease inhibitors by the NNRTI nevirapine or the NRTI abacavir all suggest a direct effect of protease inhibitors on reducing insulin sensitivity in HIV-infected patients (15,16). In our study, we focused on the cumulative effect of exposure to antiretroviral drugs. Data from other studies suggest that the effect of indinavir on insulin resistance is an acute onset effect rather than a cumulative or long-term effect and that this effect is reversible after drug discontinuation. The acute effect seen with indinavir in human volunteers on insulin resistance is limited in size and not on the order of magnitude of the relationship seen with lipodystrophy. The Swiss HIV Cohort Study showed recently that new-onset diabetes was independently associated with current exposure to indinavir, lamivudine-stavudine, didanosine-stavudine, and didanosine-tenofovir; other protease inhibitors and NRTIs also showed trends (17). In a preliminary analysis, we tested whether current use of indinavir or other protease inhibitors was associated with the risk of developing diabetes, finding that current indinavir exposure was an additional risk factor for diabetes in our dataset. Additional analyses are planned to quantify this effect. The apparent slightly protective effect of ritonavir should be viewed cautiously; it could reflect the increasing use of more recent ritonavir-boosted protease inhibitor regimens with less impact on insulin sensitivity.

Total cholesterol, HDL cholesterol, and triglyceride levels were all associated with new-onset diabetes after adjustment for demographic and clinical factors as well as for stavudine exposure, but adjustment for lipid parameters only slightly reduced the relationship between stavudine and diabetes. This relationship could be due to a common pathophysiological mechanism leading to both lipid disorders and diabetes. Alternatively, lipolysis and increased serum free fatty acids have been documented in HIV-positive individuals. Excess free fatty acids in the circulation may reduce insulin sensitivity through inappropriate lipid storage in muscle and liver, resulting in impaired glucose utilization and insulin-mediated inhibition of glucogenolysis and gluconeogenesis (6,1820).

Clinically observed lipodystrophy was also significantly associated with new-onset diabetes in accordance with previous studies showing that abnormal body fat distribution in HIV-positive individuals is strongly associated with insulin resistance and/or glucose intolerance, with excess trunk or visceral fat being, as in the general population, a risk factor for insulin resistance among those with HIV infection. In addition, insulin resistance is itself independently associated with fat loss in HIV-positive individuals (4,21)

Our data confirm previous findings on the relationship between new-onset diabetes and exposure to stavudine (but also zidovudine and didanosine), increased total cholesterol, decreased HDL cholesterol, and increased triglycerides. Interestingly, relationships for those parameters remained significant after adjustment for all other available risk factors. The large size of this cohort provides greater power to detect findings that other cohorts may be insufficiently powered to detect. However, it is possible that some of these results may reflect chance findings. There are several limitations of the study that should be considered: first, cohort studies such as ours cannot formally determine causality. However, they do permit an association between drug exposure and incidence of diabetes to be established and our findings are both consistent with those of other cohorts and biologically plausible. Second, factors such as treatment interruptions or changes in adherence are not taken into account because of the complex treatment patterns in this population. Finally, although we recognize that standardized assessments of lipodystrophy would have been preferable, this is unrealistic to achieve in a large multicohort collaboration such as this.

The relationship between new-onset diabetes and exposure to stavudine (and other NRTIs) and lipodystrophy is a striking finding, particularly as adjustment for lipodystrophy did not modify the relationship between stavudine and diabetes. It is plausible that stavudine and other NRTIs directly contribute to insulin resistance and diabetes, apart from any indirect effect by way of lipodystrophy development. It should also be noted that our binary categorization of lipodystrophy may provide a relatively blunt tool with which to perform statistical adjustment; adjustment for the degree to which lipodystrophy is present (rather than just its presence or absence) may explain a higher proportion of the effect of stavudine. However, this level of information on lipodystrophy is rarely collected.

Figure 1—

Rates (per 1,000 PYFU) of new-onset diabetes stratified by years of exposure to cART (A) and stavudine (B).

Table 1—

Characteristics of patients at enrollment in the D:A:D study

Table 2—

Results from multivariable analyses of demographic and clinical factors associated with new-onset diabetes

Table 3—

Results from multivariable analyses to assess the association between exposure to five antiretroviral drugs and new-onset diabetes after adjustment for time-updated metabolic parameters

Acknowledgments

This work was supported by the Oversight Committee for The Evaluation of Metabolic Complications of HAART, a collaborative committee with representation from academic institutions, the European Agency for the valuation of Medicinal Products, the Food and Drug Administration, the patient community, and all pharmaceutical companies with licensed anti-HIV drugs in the U.S. market: Abbott, Boehringer Ingelheim, Bristol-Myers Squibb, Gilead Sciences, GlaxoSmithKline, Merck, Pfizer, and Hoffman-LaRoche. This work was also supported by a grant (CURE/97-46486) from the Health Insurance Fund Council, Amstelveen, the Netherlands to the AIDS Therapy Evaluation Project Netherlands (ATHENA); by a grant from the Agence Nationale de Recherches sur le SIDA (Action Coordonnée no. 7, Cohortes) to the Aquitaine Cohort; by the Commonwealth Department of Health and Ageing and a grant from the Australian National Council on AIDS, Hepatitis C and Related Diseases’ Clinical Trials and Research Committee to the Australian HIV Observational Database (AHOD); by grants from the Fondo de Investigación Sanitaria (FIS 99/0887) and Fundación para la Investigación y la Prevención del SIDA en Espanã (FIPSE 3171/00) to the Barcelona Antiretroviral Surveillance Study (BASS); by the National Institute of Allergy and Infectious Diseases, National Institutes of Health Grants 5U01AI042170-10 and 5U01AI046362-03 to the Terry Beirn Community Programs for Clinical Research on AIDS (CPCRA); by grants from the BIOMED I (CT94-1637) and BIOMED II (CT97-2713) programs and the fifth framework program (QLK2–2000-00773) of the European Commission and grants from Bristol-Myers Squibb, GlaxoSmithKline, Boehringer Ingelheim, and Roche to the EuroSIDA study; by an unrestricted educational grant from Glaxo Wellcome, Italy, to the Italian Cohort Naive to Antiretrovirals (ICONA); and by a grant from the Swiss National Science Foundation to the Swiss HIV Cohort Study (SHCS).

Footnotes

  • Published ahead of print at http://care.diabetesjournals.org on 11 February 2008. DOI: 10.2337/dc07-2013.

    Additional information for this article can be found in an online appendix at http://dx.doi.org/10.2337/dc07-2013.

    The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

    See accompanying editorial, p. 1267.

    • Accepted February 4, 2008.
    • Received October 19, 2007.

References

| Table of Contents

This Article

  1. Diabetes Care vol. 31 no. 6 1224-1229
  1. Online-Only Appendix
  2. All Versions of this Article:
    1. dc07-2013v1
    2. 31/6/1224 most recent