Introduction

Diabetes is a chronic non-communicable disease that increases the risks of infection, hospitalization, and death and remains a major cause of global morbidity and mortality [1, 2]. In sub-Saharan Africa, the age-adjusted burden of diabetes is estimated to be 15.5 million people (4.4%) and is projected to reach 40.7 million people by the year 2045 [2]. As antiretroviral therapies (ART) have increased life expectancy for people living with HIV (PLHIV), diabetes may play a larger role in chronic care and management. Among PLHIV, diabetes is common and associated with age and long-term HIV or ART exposure [3, 4].

In sub-Saharan Africa, where the burden of diabetes and HIV are high, there is limited data on the prevalence of diabetes among PLHIV and the impact of diabetes on clinical outcomes. Available data suggest PLHIV with diabetes may be at increased risk of active tuberculosis (TB) and death compared to PLHIV without diabetes [5,6,7].

A major challenge for diabetes care is early diagnosis [2, 8, 9]. In South Africa, the country-wide diabetes prevalence is 5.4% and more than twice that in urban settings [10, 11]. However, prevalence surveys show 31–85% of people with diabetes are unaware of their disease status [9, 11, 12]. Clinics have established infrastructure for HIV screening and are uniquely positioned to facilitate diabetes diagnosis and treatment, but diabetes screening rates are thought to be low [13, 14].

We examined the burden of diabetes, pre-diabetes, and associated risk factors among adults testing positive for HIV in South Africa. We estimated the number needed to screen (NNS) overall and using age and/or body mass index (BMI) to guide testing of patients at high risk for diabetes, and assessed relationships between diabetes and key clinical outcomes.

Methods

Study population

Beginning in September 2013, we enrolled adults presenting for outpatient HIV screening at the iThembalabantu People’s Hope Clinic, a large public clinic in the Umlazi township of Durban, South Africa. Eligible participants were ≥ 18 years, ART-naïve, English or Zulu speaking, had not received anti-fungal therapy within three months, and were not pregnant. For these analyses, we included only HIV-positive participants from January 2017 to February 2019 when point-of-care hemoglobin A1c (HbA1c) testing for diabetes was performed routinely.

Ethics approval and consent to participate

All study participants provided written informed consent in English or Zulu. The study was approved by the University of KwaZulu-Natal Biomedical Research Ethics Committee (BF052/13) and the University of Washington Institutional Review Board (49563).

Study procedures

Prior to HIV testing, a research assistant collected demographic (age, sex, ethnicity, education, measures of socioeconomic status) and health information (cigarette and alcohol use, history of diabetes, kidney disease, liver disease) using a standardized in-person questionnaire. Research assistants also measured the participant’s blood pressure, height, and weight. Serial HIV rapid testing and referral for ART was performed at the clinic by routine clinic staff as per local standard of care [15].

Among PLHIV, a research nurse obtained blood samples, screened participants for TB, and performed a fingerpick HbA1c test (A1cNow®+, PTS Diagnostics, Indianapolis, IN). A1cNow® + is an immunoassay that has been certified by the National Glycohemoglobin Standardization Program (NGSP), which works to harmonize HbA1c tests [16]. Participants were referred to a clinician for additional testing and care according to national guidelines [17].

At three, six, and 12 months after enrollment, a research assistant reviewed pharmacy and medical records and attempted to telephone the participant at least three times. At each time point they documented available HIV outcomes including HIV viral load, hospitalizations, and death. Hospitalized participants had additional chart review to determine the cause. Opportunistic infections were documented including herpes simplex lesions, oral/esophageal candida, pneumonias, cytomegalovirus, cryptosporidiosis, bacterial meningitis, toxoplasmosis, and herpes zoster. Tuberculosis diagnoses were confirmed in the South African TB Registry. Death was determined through clinical record review, call to a named contact of the participant if they could not be reach directly, and review of the South African National Death Registry for participants who could not be contacted, had no recent medical records, and were not confirmed to be attending another clinic. Participants were considered retained in care if they had a 12 month study visit or were known to have transferred to another study clinic. They were considered lost to follow-up if they did not have a 12 month clinic or study visit and were not known to have transferred to another clinic or died.

Study definitions

We defined diabetes as HbA1c ≥ 6.5%, which is consistent with World Health Organization (WHO) and American Diabetes Association (ADA) guidelines for lab-based high-performance liquid chromatographic HbA1c testing [18, 19]. We considered pre-diabetes to be HbA1c of 5.7–6.4% and hyperglycemia to be inclusive of both diabetes and pre-diabetes using HbA1c ≥ 5.7% [19]. Measured cofactors for diabetes included BMI (overweight 25–29.9 kg/m2; obese ≥ 30 kg/m2), mean arterial pressure (MAP, [(2*diastolic blood pressure) + systolic blood pressure]/3), and anemia (hemoglobin < 12 g/dL in women or < 13 g/dL in men) [20]. We used the nine-question validated Household Food Insecurity Access Scale (HFIAS) to ascertain food insecurity and defined it as the presence or absence of any food insecurity [21]. HIV viral load suppression was defined as ≤ 50 copies/mL 4–8 months after ART initiation. Tuberculosis was defined as acid-fast bacilli smear positive, GeneXpert positive, culture positive, or empiric treatment initiation.

Statistical analysis

We performed statistical analyses using SAS version 9.4 (Cary, NC). We used the Agresti-Coull method to calculate 95% confidence intervals (CI) for the prevalence of diabetes and pre-diabetes [22]. We tested for correlates of diabetes at baseline and the intermediate study outcomes of ART initiation, missing ≥ 1 clinic visit, and missing ≥ 1 ART pharmacy refill using the χ-square or Fisher’s exact test for categorical variables and ANOVA for continuous variables.

We estimated the NNS to yield one case of diabetes or hyperglycemia, the proportion identified, and proportion missed overall and for a range of age and BMI cut points. We generated receiver operating characteristic (ROC) curves using logistic regression to estimate the area under the curve (AUC) and identified the value at which sensitivity and specificity were maximized using Youden index [23].

We estimated the risk of key HIV clinical outcomes by diabetes status in univariate analysis and after adjusting for age, sex, BMI, ART initiation, opportunistic infections, and anemia. We used log-binomial regression or Poisson regression with robust standard errors to estimate the relative risk (RR) of (1) not achieving viral suppression, (2) retention in care, and (3) ART default or lost to follow-up. We used Cox proportional hazard models to estimate the relative difference in time to (1) first hospitalization, (2) tuberculosis diagnosis, or (3) death.

Results

Description of study population

Of the 7,877 adults who presented to iThembalabantu Clinic for HIV screening and were provisionally enrolled into the study, 3,104 tested positive for HIV. Hemoglobin A1c testing following a positive HIV test was performed as part of routine study procedures for 1,207 participants. The majority (52%) of participants were diagnosed with HIV at the time of enrollment. Those with a diagnosis prior to study enrollment reported their diagnosis occurred a median of nine months prior (IQR 0–43.5 months) and 12% reported previously receiving ART. Overall, 44% of participants were men, mean age was 31.3 ± 9.5 years, and median CD4 count was 353 cells/mm3 (IQR 209–545 cells/mm3). Anemia was common among both men (31%) and women (54%). Only 1.3% (n = 7) reported a prior diagnosis of diabetes. A viral load measurement was available for 806 participants collected 4–8 months after ART initiation (mean 6 months, standard deviation 0.9) at which point 83% (n = 668) were virally suppressed. At 12-months, 51% of participants were reached by phone call.

Correlates of diabetes and pre-diabetes at time of HIV testing

Overall, 2.2% (95% CI 1.4–3.2%, n = 26) of participants screened positive for diabetes and 17.7% (95% CI 15.6–19.9%, n = 213) for pre-diabetes (Table 1). The mean age of participants who screened positive for diabetes was higher than those with pre-diabetes or normal HbA1c (p < 0.001).

Table 1 Baseline characteristics of people living with HIV screened for diabetes in South Africa
Full size table

Of those with pre-diabetes or diabetes, 73% were overweight or obese compared to 55% of those with normal HbA1c (p < 0.011). Participants with diabetes also had higher MAP (mean 104.4 mmHg) compared to those with pre-diabetes (94.8 mmHg) or normal HbA1c (92.7 mmHg) (p-value < 0.001).

Diabetes and hyperglycemia screening at HIV testing

Overall, the NNS to identify one newly diagnosed PLHIV at this urban clinic with diabetes was 47 people and the NNS for hyperglycemia was five people (Table 2). The NNS for one new case of hyperglycemia was 14 people among those ≥ 45 years, though only 39% of cases could potentially be identified relying solely on this metric. Using a BMI ≥ 25 kg/m2, the NNS to identify one instance of hyperglycemia is 36 people and 73% of cases could be identified.

Table 2 Prevalence of elevated hemoglobin A1c and the number needed to screen at the time of HIV testing according to national screening guidelines
Full size table

Age and BMI were predictive of diabetes with AUCs of 0.76 and 0.61, respectively (Fig. 1A). Using age alone to identify hyperglycemia, the combined sensitivity and specificity were maximized at 39.0 years with a specificity (i.e. the proportion without diabetes who would not be eligible for screening) of 78.5% and sensitivity (i.e. the proportion with diabetes who would be eligible for screening) of 65.4%. The predictive accuracy of BMI alone for hyperglycemia was maximized at 25.5 kg/m2 with a specificity of 46.8% and sensitivity of 73.1%.Taking age and BMI together the AUC was 0.76. At 45 years and 25 kg/m2, the NNS is 36.5 people (Fig. 1B) and 76.9% of diabetes cases would be captured (Fig. 1C).

Fig. 1
figure 1figure 1

Accuracy (A), number needed to screen (B), and percent of cases captured (C) using age (years) and body mass index as discrete screening criteria for hemoglobin A1c ≥ 6.5%. A Distance from the receiver operating characteristic curve to the no discrimination line is maximized for HbA1c ≥ 6.5% at age alone at 38.6 years (specificity: 77.1%, sensitivity: 62.5%, positive likelihood ratio: 2.72), BMI alone at 31.6 kg/m2 (specificity: 70.8%, sensitivity: 60.4%, positive likelihood ratio: 2.07). B Number need to screen to identify one instance of HbA1c ≥ 6.5%. The lower leftmost black square represents screening the entire population (NNS = 28.5) and the upper rightmost light grey square is the most restrictive screening algorithm represented (NNS = 12.7). C Proportion of participants with HbA1c ≥ 6.5% captured. The lower leftmost white square represents screening the entire population (100% captured) and upper rightmost black square is the most restrictive screening algorithm represented (43.8% captured)

Full size image

Diabetes and clinical outcomes

The majority of participants initiated ART; 94.8% of those with normal HbA1c initiated ART, 93.9% with moderately elevated HbA1c, and 95.2% with high HbA1c (p = 0.811). Among those who initiated ART, there was no difference in the proportion who defaulted, missed ≥ 1 pharmacy refill, or missed ≥ 1 clinic visit by HbA1c level.

In adjusted analyses, there was no difference in the risk of achieving viral suppression (adjusted RR 0.66, 95% CI 0.10–4.36) or receiving a TB diagnosis (adjusted hazard ratio 1.28, 95% CI 0.40–4.1) (Table 3). Median time to death was 182 days (IQR 162–187, n = 37) among participants without diabetes and 139.5 days (IQR 98–181, n = 2) among those with diabetes. The adjusted mortality rate was 1.79-times higher among those with diabetes than those without (95% CI 0.41–7.87).

Table 3 Associations of diabetes with HIV outcomes 12 months after HIV diagnosis
Full size table

Among those with hyperglycemia, there was no difference in the risk of lack of viral suppression, TB, hospitalization, death, retention in care, or loss to follow-up compared to those without hyperglycemia.

Discussion

In this cohort of newly diagnosed HIV-positive adults in Durban, South Africa, the prevalence of pre-diabetes was high and diabetes was low with few people reporting a prior diabetes diagnosis. Significant risk factors for diabetes included older age, higher BMI, and higher MAP. Diabetes was not associated with 12 month study outcomes. Our data support screening all PLHIV for diabetes or targeting those with a BMI ≥ 25 kg/m2 if universal screening is not practical.

The prevalence of diabetes among PLHIV in this cohort was 2.2%, which is similar to observations from other studies of PLHIV in urban South Africa and other sub-Saharan African countries [4, 24,25,26,27,28]. The prevalence of diabetes in the general population in South Africa is estimated to be slightly higher with recent estimates ranging from 5.4% to 10.1% [2, 9, 29]. Risk factors for diabetes in this population included those that are well-established among HIV-uninfected adults and PLHIV in sub-Saharan Africa, such as older age and hypertension [19, 30].

Missed opportunities for diagnosing diabetes remains a major concern in this population. In our cohort, the proportion of people aware of their diabetes was lower than in national prevalence surveys [11, 12]. The Society for Endocrinology, Metabolism, and Diabetes of South Africa recommends diabetes screening for PLHIV at the time of ART initiation or regimen change while South Africa’s national HIV guidelines do not specify recommendations for routine diabetes screening [17]. Inclusion of diabetes screening recommendations in HIV guidelines may help improve uptake.

As in the general population, age and BMI could be used to further increase the yield of diabetes screening by restricting screening to those more likely to have hyperglycemia among PLHIV. We were unable to directly assess if PLHIV were likely to demonstrate elevated HbA1c at relatively lower ages and BMIs compared to HIV-negative adults because of a lack of comparison group. However, data from this cohort suggests screening PLHIV with a BMI ≥ 25 kg/m2 could reduce the NNS while capturing most diabetes cases. The AUC suggests the optimal age threshold is lower than guideline of ≥ 45 years for HIV-negative adults, which is supported by research showing PLHIV are more likely to develop diabetes at younger ages [31, 32].

Our study contributes to the literature suggesting the risk of active TB among PLHIV is not substantially elevated in the presence of diabetes, in contrast to the threefold elevated risk of TB among people without HIV infection [34]. Of three studies examining the association of diabetes with active TB among HIV-infected adults, two found an association with higher odds of TB (n = 232, aOR 4.7, 95% CI 1.1–20.8; n = 521, aOR 2.4, 95% CI 1.0–5.9), and one found no effect (n = 382, aOR 0.14, 95% CI 0.01–1.81) [5,6,7]. In this cohort, ascertainment of TB cases over the 12 month study period was thorough and statistical significance was not achieved.

The presence of multiple morbidities complicates patient care, especially when the diseases differ in pathogenesis and management. Regular clinic visits for ART maintenance are an opportunity for non-communicable disease care. In South Africa, PLHIV on maintenance therapy can have better managed diabetes than the general population, but evidence is mixed [14, 28, 35,36,37]. In this cohort, there was no significant difference in ART initiation or maintenance, and time to death was elevated but not statistically significantly among with diabetes, possibly due to lack of statistical power. In Brazil and the United States, risk of death among PLHIV was 2–3-times higher for diabetic patients [38, 39]. As testing and diagnosis of diabetes and other non-communicable diseases increase for PLHIV, it will be important to consider how to optimize care for patients and clinics and to study the cost-effectiveness of different approaches to integrated care.

Our study had several strengths and limitations. Our study cohort was large and representative of the township population at risk for HIV. We comprehensively ascertained HIV outcomes by calling the participant and/or family member, and reviewing hospital records, pharmacy records, and national registries. We estimated RRs, which are more consistent, conservative, and interpretable than odds ratios, and used time-to-event analyses where relevant to not introduce bias. However, the prevalence of diabetes was low limiting our ability to detect true associations. Also, we relied on a single point-of-care HbA1c test to screen for diabetes, whereas guidelines recommend two lab-based HbA1c tests several weeks apart [19, 40,41,42,43,44]. Point-of-care HbA1c tests are not standardized, require regular calibration, and therefore are not recommended as diagnostic tests by the WHO or ADA. In prior validation diagnostic studies, A1cNow® + had excellent correlation (> 90%) with gold standard high performance liquid chromatography HbA1c, 82% specificity, 100% sensitivity, but moderately overestimated HbA1c levels (mean difference 0.2–0.3% units) [45, 46]. The coefficients of variation (2.7–2.9%) were reasonable for a point-of-care test, although not below the recommended level of < 2% [47]. Additionally, HbA1c may moderately overestimate diabetes among PLHIV especially with concurrent iron deficiency and therefore our estimate of the prevalence of diabetes may be high, although we adjusted for anemia in outcome analyses [19, 40,41,42,43,44]. Lastly, all major risk factors for diabetes were not ascertained in this cohort; therefore, we could not test for associations with all potential risk factors.

Conclusions

In summary, in urban South Africa hyperglycemia was common among adults testing positive for HIV in an outpatient setting. Diabetes prevalence was lower than the general population, but the majority reported no prior diabetes diagnosis indicating an opportunity for disease detection. The NNS to identify one new case of hyperglycemia was low overall and could be further reduced using age or BMI. Further research is needed to refine diabetes screening guidelines among PLHIV using other methods of diabetes testing, assess the impact of diabetes rapid diagnosis and treatment on HIV outcomes, and optimize integrated chronic disease care.