Risk Factors for Incident Diabetes in a Cohort Taking First-Line Nonnucleoside Reverse Transcriptase Inhibitor-Based Antiretroviral Therapy.

Efavirenz is the preferred nonnucleoside reverse transcriptase inhibitor (NNRTI) in first-line antiretroviral therapy (ART) regimens in low- and middle-income countries, where the prevalence of diabetes is increasing. Randomized control trials have shown mild increases in plasma glucose in participants in the efavirenz arms, but no association has been reported with overt diabetes. We explored the association between efavirenz exposure and incident diabetes in a large Southern African cohort commencing NNRTI-based first-line ART. Our cohort included HIV-infected adults starting NNRTI-based ART in a private sector HIV disease management program from January 2002 to December 2011. Incident diabetes was identified by the initiation of diabetes treatment. Patients with prevalent diabetes were excluded. We included 56,298 patients with 113,297 patient-years of follow-up (PYFU) on first-line ART. The crude incidence of diabetes was 13.24 per 1000 PYFU. Treatment with efavirenz rather than nevirapine was associated with increased risk of developing diabetes (hazard ratio 1.27 (95% confidence interval (CI): 1.10-1.46)) in a multivariate analysis adjusting for age, sex, body mass index, baseline CD4 count, viral load, NRTI backbone, and exposure to other diabetogenic medicines. Zidovudine and stavudine exposure were also associated with an increased risk of developing diabetes. We found that treatment with efavirenz, as well as stavudine and zidovudine, increased the risk of incident diabetes. Interventions to detect and prevent diabetes should be implemented in ART programs, and use of antiretrovirals with lower risk of metabolic complications should be encouraged.

INTRODUCTION

Access to antiretroviral therapy (ART) has considerably reduced morbidity and mortality associated with human immunodeficiency virus (HIV) infection. However, long-term ART is associated with adverse metabolic effects including dysglycemia and new onset diabetes mellitus.[1,2] With the prevalence of noncommunicable diseases, including diabetes, increasing in low- and middle-income countries (LMICs),[3] patients on ART in LMICs face a dual burden of disease.[4]

A number of antiretroviral drugs are known to cause diabetes, including the nucleoside reverse transcriptase inhibitors (NRTIs) stavudine (d4T) and zidovudine (AZT),[2] and the older protease inhibitors (PIs) indinavir[5] and ritonavir.[6,7] Efavirenz, which is now the preferred nonnucleoside reverse transcriptase inhibitor (NNRTI) for first-line ART in LMICs,[8] is associated with slight increases in blood glucose in randomized controlled trials,[9-13] and, in one study conducted by our group.[14] However, there is no good evidence that efavirenz is associated with an increased risk of developing diabetes.

The aim of our study was to investigate the association between efavirenz use and the incidence of diabetes mellitus in a South African cohort of patients on first-line ART.

METHODS

Study Population and Data Source

The study population comprises South African HIV-infected adults enrolled in a private sector HIV disease management program, Aid for AIDS (AfA). The AfA program collects demographic, laboratory, and clinical data on individuals who registered for HIV benefits. Claim data were captured by AfA from the medical insurance fund claim database. These include laboratory, hospitalization, pharmacy, and medical practitioner claims which were submitted to the scheme for processing either: at the time of the service by the provider (eg, pharmacy, hospitalization) for direct reimbursement or after the service date by the member where the member had already paid the claim. Reimbursement was subject to established AfA protocols, including protocols for ART initiation, change of ART regimen, and the treatment of certain opportunistic infections. No copayment was required for ART, viral load (VL) and CD4 monitoring, and doctor visits.

Despite being a private sector program, AfA standardized guidelines for HIV management, are similar to the World Health Organization (WHO) guidelines for LMICs.[8] Patients were eligible for ART initiation if their CD4 cell count was below 350 cells/μl or they had WHO stage 3 or 4 illness irrespective of the CD4 count. The recommended initial regimen was a combination of 2 NRTIs and an NNRTI. VL and CD4 counts were monitored every 6 months.

Data linkage to the South Africa death registry allowed ascertainment of deaths and date of death, as previously described.[15,16]

Variables and Definitions

We extracted sex, date of birth, weight, height, Republic of South Africa Identity Number, and date of joining the AfA program from the form completed by the doctor on registering the patient with AfA.

We extracted longitudinal results for CD4 count and VL, and all medication claims for antiretrovirals and concomitant medicines. We created a list of diabetogenic drugs using a pharmacology reference textbook[17] and a review[18] (see Appendix 1). We categorized patients as exposed to diabetogenic drugs if they submitted claims for a diabetogenic drug on 2 or more occasions.

We defined the ART start date as the date on which antiretroviral drugs were first dispensed in the AfA program. The ART starting regimen was the regimen dispensed on this date. The baseline CD4 count, VL, and weight were the values measured closest to the date of ART initiation, within the 12-month window before the ART start date. The primary exposure variable of interest was the NNRTI component of the first-line antiretroviral regimen.

Inclusion and Exclusion Criteria

For this study we included AfA-registered patients who initiated a first-line NNRTI-containing ART regimen from January 2002 to December 2011 and were 19 years or older when starting ART. We excluded patients already on antidiabetic medication before starting ART, and patients with missing South Africa identification numbers (as the identification number was used to determine if death occurred by linkage with the South African death registry).

Endpoint

Incident diabetes was defined, using claims data, as the date on which any of the antidiabetic agents available in South Africa (insulins, metformin, sulfonylureas, alpha glucosidase inhibitors, thiazolidinediones, dipeptidyl peptidase-IV inhibitors, glucagon-like-peptide-1 receptor agonists and meglitinides) were initiated. Patients who started and stopped antidiabetic medication in the months around a pregnancy (from 6 months before delivery until 1 month after delivery) were assumed to have diabetes in pregnancy and were therefore not included as incident diabetes.

Imputation of Missing Data

Some data were missing for baseline CD4 count, baseline VL, baseline weight and height (see Table 1). We imputed 5 datasets with the multiple imputation by chained equations (MICE) module in STATA version 13. The imputation model included the following variables: sex, baseline weight, height, age, CD4 count, and VL, death, incident diabetes, and exposure time. CD4 count and VL were actively imputed, and body mass index (BMI) was passively imputed (using actively imputed baseline weight and height). We checked the results of the imputation model by comparing the imputed data with the actual data.[19,20]

TABLE 1

Cohort Description

Analysis

We collated and prepared the data for statistical analysis using a relational database (Microsoft SQL Server 2008). We used STATA Version 13 (StataCorp LP, College Station, TX) and R version 3.2.1 (R Development Core Team) for statistical analyses.[21] We compared the incidence of diabetes in patients receiving efavirenz-containing regimens versus nevirapine-containing regimens with a Kaplan–Meier plot and a log-rank test.

We explored the association of efavirenz exposure with the hazard of developing diabetes using a multivariate Cox-proportional hazards model. We adjusted for the following variables: age, sex, baseline BMI, baseline CD4 count, baseline VL, exposure to diabetogenic drugs. For the primary analysis ART was included in the model as time-updated NRTI backbone (AZT-containing, d4T-containing, or other NRTI combination) and time-updated NNRTI (efavirenz or nevirapine); and patients were censored when they died, left the medical insurance scheme, switched to PI-based ART, or reached the end of the study period. We performed a secondary analysis exploring incident diabetes within the first ART regimen only. For this analysis we included 2 additional reasons for censoring: NRTI or NNRTI substitution. We explored the effect of including calendar year in the multivariate model. We performed the following sensitivity analyses: we controlled for the competing risk of death, we constructed a model excluding patients virologically suppressed at baseline.

The proportional hazards assumption was verified by testing interaction effects of analysis time with baseline variables (α = 0.05), and graphically in each imputed dataset via log–log plots, amongst others. We performed model selection using the Akaike Information Criterion (AIC) after multiple imputation.[19,22-24]

Ethics

The study protocol was reviewed and approved by the University of Cape Town Faculty of Health Sciences Human Research Ethics committee.

RESULTS

Between January 2002 and June 2011, 62,467 patients commenced ART in the AfA program, of whom 56,298 patients met our inclusion and exclusion criteria (Figure 1) and were included in the analysis. The demographic and clinical characteristics of patients are given in Table 1. Results from the multiple imputations for missing covariates are shown in Table 2. Median follow-up was 1.56 years (interquartile range (IQR): 0.71–2.79 years), 21.7% of patients were followed up for 3 or more years.

FIGURE 1

Participant selection and exclusion criteria.

TABLE 2

Results of Imputation for Baseline Variables With Missing Data

We identified new onset diabetes in 1500 (2.66%) patients over 113,297 patient-years of follow-up (PYFU), giving a crude incidence of 13.24 cases per 1000 PYFU (Figure 2). There were 17 pregnancy-associated diabetic events, which were not included as cases of incident diabetes. Exposure to diabetogenic medicines occurred in 19,137 (41.01%) of patients taking efavirenz and 3643 (37.82%) of patients taking nevirapine. The Kaplan Meier analysis of incident diabetes in efavirenz versus nevirapine-containing regimens is shown in Figure 2.

FIGURE 2

Kaplan–Meier analysis of incident diabetes.

The results of the Cox proportional hazard regression analyses, including univariate and multivariate analyses and model selection, are shown in Table 3. Efavirenz-containing ART was associated with a higher risk of developing new-onset diabetes than nevirapine-containing ART, adjusted hazard ratio (HR) 1.27 (95% confidence interval (CI): 1.10–1.46). Zidovudine and stavudine-containing NRTI backbones, older age at baseline, elevated baseline BMI, and exposure to diabetogenic medication were also associated with an increased risk of developing diabetes. We found no association between baseline CD4 count and an increased risk of diabetes.

TABLE 3

Univariate, Multivariate, and Model Selection Results for the Cox Regression Model of Associations With Incident Diabetes

The results of the Cox proportional hazard regression analyses where we censored patients at the time of first drug switch are shown in Table 4; reasons for censoring for this analysis are shown in Supplementary Table 1. The estimated HRs of the variables remained similar after model averaging, confirming the stability of our findings.[25] These findings did not differ from the regression model with updated regimen (censoring at the point of change to second line regimens). Adding calendar year to the model did not change the associations observed (Supplementary Table 2). Findings did not change when we accounted for competing risk (Supplementary Table 3). Excluding patients with suppressed baseline VLs attenuated the effect of efavirenz on incident diabetes somewhat: adjusted HR 1.16; 95% CI: 0.99–1.36 (Supplementary Table 4).

TABLE 4

Univariate, Multivariate, and Model Selection Results for the Cox Regression Model of Associations With Incident Diabetes (Censored at First Ever Drug Switch)

DISCUSSION

We found efavirenz use to be associated with a significantly higher incidence of diabetes than nevirapine in a large cohort of South African patients. To the best of our knowledge, this is the first cohort study to show an increased risk of diabetes from efavirenz use in first-line ART. We also found that the NRTIs stavudine and zidovudine were associated with an increased incidence of diabetes. These findings have important implications for LMICs, which are facing a burgeoning diabetes epidemic, as efavirenz is the preferred NNRTI in first-line ART, zidovudine is recommended in second line ART, and many people are still taking stavudine, even though it is no longer recommended by the WHO.[8]

HIV-infected patients have an estimated 4-fold greater relative risk of developing diabetes than the HIV-uninfected population.[26] We found a crude incidence of diabetes of 13.24 per 1000 PYFU, which is at the upper end of the range reported from cohort studies in high income countries (4.2–14.1 per 1000 PYFU).[2,26-28] Factors contributing to the increased risk of diabetes in people with HIV include insulin resistance due to the chronic inflammatory response to HIV infection,[29] which persists despite effective ART[29,30] and the effects of certain antiretroviral drugs. Our finding that the NRTIs stavudine and zidovudine were both associated with an increased incidence of diabetes has previously been reported.[2,31] NRTI's inhibit the enzyme DNA polymerase-γ, responsible for mitochondrial replication. The dysregulation of mitochondrial function in different compartments of the body results in various clinical manifestations of NRTI toxicity, including insulin resistance and diabetes.[32-34] In a prior cross sectional study we found an increased risk of dysglycemia in South African patients taking efavirenz compared with those taking nevirapine, but there were insufficient numbers of cases of diabetes for analysis.[14] A small case–control study from Botswana suggested an association between efavirenz use and diabetes.[35] Randomized controlled trials showed significantly higher serum glucose concentrations in participants in the efavirenz arms than the following comparator antiretroviral drugs: nevirapine,[13] abacavir,[13] atazanavir,[11] atazanavir–ritonavir,[9] and raltegravir.[12]

The mechanism by which efavirenz mediates insulin resistance and diabetes is unknown. Possible mechanisms include mitochondrial toxicity[36] and toxic effects on adipocytes and increased rates of lipoatrophy.[37,38] Efavirenz causes hepatic mitochondrial toxicity[36] and induces hepatocyte endoplasmic reticulum stress leading to activation of the unfolded protein response, and apoptosis.[39,40] Efavirenz mediates mitochondrial toxicity via various mechanisms. Firstly, efavirenz directly inhibits Complex I of the electron transport chain, resulting in a markedly reduced mitochondrial transmembrane potential, thus compromising oxidative phosphorylation and ATP generation.[41-43] Secondly, efavirenz reduces complex IV (COIV) mRNA (a marker gene of mitochondrial function), and impairs mitochondrial function in adipocytes.[44] Furthermore, efavirenz-associated mitochondrial dysregulation in adipose tissue causes impaired adipogenesis, increased lipolysis, and release of free fatty acids and inflammatory cytokines.[44] The increased release of fatty acids due to adipocyte mitochondrial toxicity are thought to impair muscle and liver insulin sensitivity, leading to insulin resistance and diabetes mellitus.[45-51] In addition to its mitochondrial toxicity, efavirenz has been shown to reduce the secretion of adiponectin (an insulin-sensitizing, antidiabetic adipokine) by adipocytes.[44] We hypothesize that impairment of mitochondrial bioenergetics and toxicity to adipocytes contributes to the development of diabetes in patients on efavirenz. By contrast, nevirapine does not appear to exert mitochondrial toxicity.[37,44]

Our group have demonstrated a positive correlation between plasma efavirenz concentrations and both fasting and 2-hour glucose concentrations after oral glucose tolerance tests in South African patients.[52] People of African origin are more likely to be genotypic “slow metabolizers” of efavirenz, which results in elevated efavirenz plasma concentrations, than people of European descent (20% and 3%, respectively).[53] Therefore efavirenz may have a larger diabetogenic effect in Africans, which may explain why, in contrast with our findings, studies from high income countries have not found an association between efavirenz and diabetes.

Our finding that increasing age, and male sex were associated with an increased risk of diabetes is consistent with findings of other studies.[2,29,54-59] We could not show any association between baseline CD4 count and diabetes, which is similar to that reported by a French cohort,[27] but other studies have found an increased risk of diabetes with lower CD4 counts.[29,54] We found an association between the lowest stratum of baseline VL and an increased relative risk for developing diabetes, but no association with higher VL strata. The French cohort reported no association between VL and diabetes.[27] In contrast, other studies have found an association between high VLs and diabetes.[29,58] The association we observed between low baseline VLs and increased risk of diabetes may be attributed to the inclusion of patients already on undisclosed ART on entry to the AfA program.

Our study has limitations. We had missing baseline data, notably of BMI. However, we imputed missing data, which are known to be superior to using complete case analysis.[60,61] ART exposure may have occurred before commencing ART within the AfA program and there were 18.8% of patients with a suppressed VL at baseline, which is likely due to undisclosed ART use. We identified incident diabetes based on initiation of diabetes therapy as the results of plasma glucose or glycated hemoglobin are not captured in the database. We will therefore have missed cases of diabetes that were only treated with lifestyle modifications. As patients were not routinely screened for diabetes at the time of ART initiation, some patients may have entered the cohort with undiagnosed diabetes. We could not adjust for weight changes during follow-up in our analyses, as weight is only recorded at baseline. A strength of our study is that we adjusted for the concurrent use of diabetogenic medication in the multivariate model, which was associated with a 53% increase in the relative risk of diabetes.

In conclusion, we found that exposure to efavirenz, stavudine, and zidovudine was associated with an increased incidence of diabetes. Although the increased risk of diabetes with these antiretrovirals was relatively modest, the large African patient population exposed to ART for prolonged periods means that our finding has important public health implications. While screening for diabetes should be increased in people on long-term ART, consideration should be given to using antiretrovirals with less risk of metabolic complications. Further studies to confirm the association of efavirenz and risk of diabetes should be conducted in LMICs, and the molecular mechanisms of efavirenz-induced dysglycemia need further investigation.