Incidence and clinical outcomes of diabetes mellitus in HIV-infected adults in Thailand: a retrospective cohort study

The burden of diabetes prevalence is rising especially in low and middle-income countries [1]. In 2017, the International Diabetes Federation estimated that 425 million people worldwide, or 8.8% of adults 20–79 years, were affected by diabetes type 1 and type 2 (8.3% in Thailand) and that half (50.0%) of all people 20–79 years with diabetes are unware of their disease [2].

Known risks factors of diabetes mellitus (DM) include older age, male sex, family history of DM, alcohol use, adiposity, dyslipidemia, hypercholesterolemia, and hypertriglyceridemia [1]. Indeed, hyperlipidemia, insulin resistance and lipodystrophy are commonly observed in HIV-infected patients on antiretroviral treatment (ART) and several studies have suggested that HIV infection and/or antiretroviral drugs may increase the risk of DM [3‐6]. The Thai national AIDS program database, which includes data on DM incidence, related mortality and prevalence of related complications, provided a unique opportunity to assess the importance of DM in a South-East Asian HIV infected adult population.

Since Fiscal Year (FY) 2005 (October 1, 2004 to September 30, 2005), Thailand has provided free health services, including ART and laboratory monitoring to HIV-infected patients through the Thailand National AIDS Program (NAP), under the National Health Security Office (NHSO). At each visit, or at the end of each month for hospitalizations, data from HIV-infected patients covered by these schemes are entered into the NAP database. We estimated the incidence of DM diagnosis, investigated associated risk factors and clinical outcomes using the data from adults registered with the NAP under the Universal Coverage Scheme (UCS), which covers about three-quarters of the Thai population [7].

This study was a retrospective cohort study and an analysis of secondary data of the NHSO. A total of 199,707 HIV infected adults (≥18 years old) with no history of DM registered between FY2007 to FY2013 and follow up until September 30,2014 (end of FY2014) for care with the NAP and received care under UCS in 1035 hospitals throughout Thailand.

We extracted the following patient characteristics from the NHSO database: sex, date of birth, weight, height, fasting plasma glucose (FPG), triglycerides, hepatitis C virus infection, absolute CD4 cell count (including nadir) and date of ART initiation [7]. A patient not showing up for at least 7 months after last visit was considered lost to follow up (LTFU). The number of PYFU was calculated from date of NAP registration (baseline) to censoring date, i.e. 7 months after last visit date, date of death (within 7 months after last visit), date of first DM diagnosis, or September 30, 2014, which ever occurred first.

For this study, DM was considered diagnosed at the first date of at least two of the following records: 1) FPG ≥126 mg/dl following the 2013 American Diabetes Association criteria [8] or 2) confirmed diagnosis codes E11-E14 (which excludes type-1 diabetes mellitus) of the 2010 WHO International Classification of Diseases (ICD-10) [9], or 3) confirmed receipt of anti-diabetic drugs [8]. Hypertriglyceridemia was defined by triglycerides ≥200 mg/dl [4]. DM incidence rates were estimated by the number of new diagnoses divided by the total number of PYFU. The 95% confidence intervals (CIs) were calculated using the quadratic approximation to the Poisson log likelihood [10].

We fitted competing risks survival regression models (Fine-Gray) [11‐13], treating death without DM as a competing event (except in case of death after DM diagnosis), to assess the association of new DM diagnosis occurrence with sex, baseline age (categories: 18–34, 35–44, 45–59 or ≥ 60 years [14]) and Body Mass Index (BMI) (≥25 or < 25 kg/m²); and baseline and time-updated triglycerides (< 200 or ≥ 200 mg/dl), baseline absolute CD4 cell count (< 200 or ≥ 200 cells/mm³), and time-updated ART initiation (yes/no) [4, 14]. All models were adjusted for the existence of at least a previous record of FPG (dichotomous, time-updated variable) and time-updated absolute CD4 < 200 cells/mm³. We imputed missing data by multiple imputation with chained equations (MICE procedure, Stata Corp, College Station, Texas, USA) based on logistic regression for binary variables if they were ≤ 20% missing data (time-updated absolute CD4 values if there was a missing data more than 9 months after the previous known value) [15]. We conducted multivariable analyses using a backward selection approach starting with factors significantly associated with time to DM in the univariable analysis (p ≤ 0.20), excluding variables with more than 20% missing values (baseline BMI, baseline and time-updated triglycerides, baseline absolute CD4 data) and sensitivity analyses [16] were conducted including each of these variables. The cumulative incidence function (CIF) at time t was defined as the probability of a new diagnosis from baseline using Gray’s sub-distribution hazard technique [17, 18].

A Poisson distribution was assumed to calculate the 95% confidence intervals (CIs) of prevalence. We used the 2010 WHO ICD-10 [9] codes for microvascular diabetic complications: ophthalmic complications code E11-E14 with additional code “.3”, diabetic retinopathy codes H36.0*, renal complications codes E11-E14 with additional external cause code “.2” or code N18, neurological complications codes E11-E14 with additional code “.4” (diabetic amyotrophy, autonomic neuropathy, mononeuropathy, polyneuropathy and autonomic), and macrovascular diabetic complications: ischaemic heart diseases (ICH) codes I00-I25, cerebrovascular disease (CVD) codes I60-I69, peripheral circulatory complication codes E11-E14 with additional code“.5” (diabetic gangrene, peripheral angiopathy or diabetic ulcer), and amputation (The 2010 WHO ICD-9-CM procedure code 84.11, 84.12, 84.14, 84.15 and 84.17 [19]).

We analyzed the interaction between time-updated ART initiation and time-updated DM for the risk of death by year using a constant proportional hazard in univariable and multivariable Cox’s survival regression analyses (p ≤ 0.05). We tested the proportional hazards assumption of each model based on Schoenfeld residuals [20] and considered time-updated DM diagnosis as a time-varying variable [21, 22].

Analyses were performed using Stata software, version 13.1 (Stata Corp, College Station, Texas, USA). Patient identifiers were encrypted by NHSO prior data management and analysis. The analysis plan was approved by the Ethical Committee of the Faculty of Medicine, Chiang Mai University, Thailand on March 18, 2014 (114/2014, Research ID: COM-2557-02140).

The incidence of new DM diagnosis during the follow up of 199,707 of HIV infected adults over 8 years in Thailand (763,666 PYFU) was 11.0 per 1000 PYFU with slight variations over time. The study population included 45.8% females and was relatively young (median 36.2 years), with a lean baseline body shape (median 20.2 kg/m²) and a low absolute CD4 cell count (median 120 cells/mm³). In multivariable analyses, risk factors associated with DM diagnosis were male sex and older age, which are known risk factors in the general population. In addition, initiation of antiretroviral treatment was associated with DM, as previous reported [14, 23‐25].

Our overall DM incidence estimate was similar to or slightly lower than that reported in several studies conducted in population with similar characteristics (sex, age, CD4, ART): 11.35 per 1000 PYFU in a study of 6816 HIV-infected patients registered with the South Carolina Medicaid system [3] in the USA (43% women, median age 39.0 years, and 80.4% on ART); 14.1 per 1000 PYFU (95% CI, 11.6 to 17.0) in the French Aproco-Pilote cohort [6] (78.5% men, median age 37.0 years, median CD4 cell count 280 cells/mm³, all on ART, men 14.6 per 1000 PYFU versus women 12.6 per 1000 PYFU); 13.1 per 1000 PYFU in the National Taiwan University Hospital study [26] (86.0% men, median age 34.0 years, median CD4 cell count 92 cells/mm³, all patients on ART). The incidence rate of DM and the risk factors for DM in our study were similar to 13.7 per 1000 PYFU, the pooled incidence rate estimated in a meta-analysis [27].

Male sex and older age, which are known factors of DM [1] may account for the differences in estimated incidence reported by previous studies of HIV infected patients, for instance 5.0 per 1000 PYFU in a younger population composed of 76% women in Thailand [28], or 26.0 per 1000 PYFU in older HIV-infected men (median 46 years) participating in the US Multicenter AIDS Cohort Study [29]. ART initiation was identified as a risk factor of DM in our study, similarly to that reported in the Data Collection on Adverse events of Anti-HIV Drugs study (relative risk of 1.1 per year of exposure) [5]. It is possible that, even after adjustment for age and treating death as a competing event, improved survival on ART is a confounder for this association, i.e. patients who survive longer are more likely to develop DM. Also, the risk may vary according to specific antiretroviral drugs [28].

The incidence of DM in HIV-infected adults aged 35 to 59 years in our study was 14.9 per 1000 PYFU, higher than in the Thai general adult population (7.8 to 11.4 per 1000 PYFU) [23, 24], suggesting a contribution of HIV infection or ART in the risk of DM.

At the end of study period, the overall estimated prevalence of DM was lower than in the Thai general population (5.0% versus 8.9%) [30]. This may have resulted from several differences between our population and the general population: lower median age, lower BMI, or regular medical follow-up and FPG assessments in our cohort. Also, DM-related complications seemed less prevalent [7, 30, 31], likely for similar reasons. Generally, microvascular complications develop 5 years after diagnosis [32, 33]. However, with a median follow-up of 2.7 years in our study the prevalence of microvascular complications was 11.5% and the most common was diabetes nephropathy. HIV infection or/and antiretroviral drugs may also have affected the kidneys or accelerated microvascular complications.

Our study showed that the risk of death in patients diagnosed DM tended to be higher after 4 years of follow-up, a finding similar to that reported by a previous study in Thailand [34]. Although causes of death were not available, we hypothesized that the increasing risk of death with DM may have been related to cardiovascular and renal complications as well as HIV infection itself.

A limitation of our study is that the NAP database has been primarily designed for overall monitoring of the program with missing data not systematically tracked as in clinical studies. Importantly, the NAP database represents a unique source of information that likely reflects the actual DM burden in the HIV infected population. Also, the lack of systematic FPG assessments in some patients may have led to an underestimation of DM incidence. This issue is now addressed by a specific program [35] which promotes systematic DM screening in at risk patients [14].

HIV infected adults in Thailand had a slightly higher incidence of DM than that estimated in general population with similar common risk factors. Antiretroviral treatment may contribute to the risk of DM but to a lesser extent than the common risk factors of DM.