Tuberculosis, Human Immunodeficiency Virus, and the Association With Transient Hyperglycemia in Periurban South Africa

Abstract Background Diabetes mellitus (DM) increases tuberculosis (TB) risk. We assessed the prevalence of hyperglycemia (DM and impaired glucose regulation [IGR]) in persons with TB and the association between hyperglycemia and TB at enrollment and 3 months after TB treatment in the context of human immunodeficiency virus (HIV) infection. Methods Adults presenting at a Cape Town TB clinic were enrolled. TB cases were defined by South African guidelines, while non-TB participants were those who presented with respiratory symptoms, negative TB tests, and resolution of symptoms 3 months later without TB treatment. HIV status was ascertained through medical records or HIV testing. All participants were screened for DM using glycated hemoglobin and fasting plasma glucose at TB treatment and after 3 months. The association between TB and DM was assessed. Results Overall DM prevalence was 11.9% (95% confidence interval [CI], 9.1%–15.4%) at enrollment and 9.3% (95% CI, 6.4%–13%) at follow-up; IGR prevalence was 46.9% (95% CI, 42.2%–51.8%) and 21.5% (95% CI, 16.9%–26.3%) at enrollment and follow-up. TB/DM association was significant at enrollment (odds ratio [OR], 2.41 [95% CI, 1.3–4.3]) and follow-up (OR, 3.3 [95% CI, 1.5–7.3]), whereas TB/IGR association was only positive at enrollment (OR, 2.3 [95% CI, 1.6–3.3]). The TB/DM association was significant at enrollment in both new and preexisting DM, but only persisted at follow-up in preexisting DM in patients with HIV-1 infection. Conclusions Our study demonstrated high prevalence of transient hyperglycemia and a significant TB/DM and TB/IGR association at enrollment in newly diagnosed DM, but persistent hyperglycemia and TB/DM association in patients with HIV-1 infection and preexisting DM, despite TB therapy.

Mortality in South Africa is characterized by concurrent infectious and noncommunicable diseases.The 2015 mortality report confirms this transition, with tuberculosis (TB) and diabetes mellitus (DM) ranked the first and second leading causes of death, while human immunodeficiency virus type 1 (HIV-1) was ranked fifth [1].Although the role that HIV-1 plays as a significant driver of the TB epidemic is well recognized, the emerging and rapidly growing burden of DM, another TB risk factor, presents another challenge to TB control.
An increasing body of research shows an association between DM and TB [2,3].This association is becoming more apparent due to the epidemiological transition as DM is growing rapidly in settings where HIV-1 and TB epidemics persist.Diabetes increases the risk of developing TB and is also associated with adverse treatment outcomes, including death [3][4][5][6].
Tuberculosis increases insulin resistance and stress-induced hyperglycemia that may revert to normal during treatment [7,8].Therefore, testing for DM in persons with recently diagnosed TB may lead to misclassification of transient hyperglycemia as DM, and overestimation of the diabetes/TB association.Testing for DM in TB patients is recommended, with confirmatory tests after 2-3 months of TB treatment initiation [9,10]; however, the optimal time for screening and implications for clinical management are unknown.
The objective of this study was to assess the association between hyperglycemia and TB, at TB diagnosis, and after 3 months of TB treatment.

Study Design and Sampling
A 3-month cohort study on consecutive patients with respiratory symptoms presenting to the clinic from July 2013 to August 2015.Patients were eligible if they provided consent, were ≥18 years of age, and had received <48 hours of TB chemotherapy.Those critically ill and in need of emergency clinical care were ineligible due to inability to provide informed consent.Based on the 4.5% and 1.2% prevalence of diabetes in TB patients and non-TB patients, respectively [13], assuming 80% power and 5% significance level, the required sample size was 798 (n = 399 per group) [18].

Case Definitions
All participants were tested for TB according to South African guidelines [19].Samples were analyzed in a centralized national health laboratory.TB cases had a positive GeneXpert result.Non-TB participants were those with a negative GeneXpert result, who after examination by a physician had resolution of respiratory symptoms without TB treatment after 3 months.HIV status and antiretroviral therapy (ART) were ascertained from participants' medical records.For participants with unknown HIV status, voluntary HIV testing was offered, and if found to have HIV infection, they were provided counseling and ART initiation.All participants were tested for DM using fasting plasma glucose (FPG) and glycated hemoglobin (HbA1c).DM was defined as self-reported DM, FPG ≥7.0 mmol/L, or HbA1c ≥6.5%.Impaired glucose regulation (IGR) was defined as FPG 5.5 to <7.0 mmol/L or HbA1c 5.7% to <6.5% [20].

Measurements
After TB diagnosis, sputum microscopy was repeated at months 3 and 5 in patients with pulmonary smear-positive TB according to South African guidelines [19].For DM diagnosis, venous blood was drawn after an overnight fast for FPG.At 3-month follow-up, both DM tests were repeated in all participants.All blood samples were processed on the day of collection at a centralized national health laboratory using standardized operating procedures of the Roche/Hitachi Cobas C311 system analyzer assay [21].Weight, height, and waist circumference were measured [22].The body mass index (BMI [kg/m 2 ]) was categorized as follows: underweight, <18.5; normal, 18.5-24.9;overweight, 25-29.9;obese, ≥30) [22].The cutoff point for high waist circumference was ≥94 cm (males) and ≥88 cm (females) [22].Hypertension was defined as a single measured blood pressure (BP) of systolic BP >140 mm Hg or diastolic BP >90 mm Hg [20], or a preexisting diagnosis.

Questionnaire
Socioeconomic, demographic, and chronic medical and medication history for HIV-1, DM, and hypertension were collected using a researcher-administered questionnaire.

Statistical Analysis
Medians and interquartile ranges (IQRs) and proportions summarized continuous and categorical variables.The χ 2 and Fisher exact tests assessed associations between categorical variables, respectively.The Mann-Whitney test was used to compare medians between 2 groups and the Kruskal-Wallis test for >2 groups.A multivariable logistic regression model for TB/DM association was manually built, using forward selection, controlling for potential confounding variables.To retain statistical power in the regression analysis, multiple imputation was used to impute HIV-1 serostatus for 50 participants with unknown HIV-1 status.We conducted sensitivity analysis comparing complete case and imputed analyses for multivariate analysis results on the association between TB and IGR/ DM (Supplementary Table 1).Statistical significance was set at P < .05.All data were analyzed using Stata version 13.0 software (StataCorp, College Station, Texas).

Ethical Considerations
This study was approved by the University of Cape Town Human Research Ethics Committee (HREC REF: 403/2011).

Study Sample at Enrollment
Nine hundred eighty-six participants were recruited, and 48 participants (4.9%) were excluded as TB could not be confirmed or excluded.A further 88 participants did not complete diabetes screening at enrollment.For the analysis, 850 participants were included: 412 TB cases and 438 non-TB participants (Figure 1).

Study Sample at Follow-up
Of the 850 patients enrolled, 639 returned for 3-month follow-up with 211 patients lost to follow-up (108 TB,103 non-TB) (Figure 1).Data comparing participants lost to follow-up to those followed up are presented in Supplementary Table 2.

Glycemic Levels in TB Patients at Enrollment and Follow-up
Among TB patients with newly diagnosed DM, median HbA1c decreased at follow-up (5.7% vs 5.4%; P < .0001),although FPG in this group slightly increased (4.6 vs 4.7 mmol/L; P < .0064)at follow-up (Table 2).Among those with a preexisting diagnosis of DM, glycemic levels were sustained at high levels with no significant changes at follow-up (Table 2).

Association Between Hyperglycemia (DM and IGR) and TB at Enrollment and Follow-up
Irrespective of HIV-1 status, the overall TB/DM association was positive and significant at both enrollment (OR, 2.4 [95% CI, 1.3-4.3])and at follow-up (OR, 3.3 [95% CI, 1.5-7.3])but only when DM was defined by a positive result for either of the diagnostic tests or a previous DM history (Table 3).This significant association was not observed at either time point when DM diagnostic tests were used in isolation (Table 3).The overall association between TB and IGR (by FPG or HbA1c) was positive at enrollment (OR, 2.3 [95% CI, 1.6-3.3];Table 3) but not at follow-up (OR, 0.8 [95% CI, .5-1.4]).On further analysis by DM diagnostic test, the overall TB/IGR association at enrollment was significant when using the HbA1c test (OR, 1.6 [95% CI, 1.1-2.3])but not by FPG (OR, 0.9 [95% CI, .5-1.5]) (Table 3).

DISCUSSION
We previously reported the prevalence of DM in patients with newly diagnosed TB and the cross-sectional TB/DM association at enrollment [12].In this study, we investigated whether hyperglycemia identified in the patients with newly diagnosed TB at enrollment persisted after 3 months of TB treatment.We also assessed association between TB and hyperglycemia at both time points.This is the first study in South Africa, a setting of high TB and DM burden, to document transient hyperglycemia in TB patients with preexisting DM and newly diagnosed DM.Overall, we report significant association between TB and DM at both enrollment and follow-up and significant association IGR by FPG 0.9 (.5-1.Data are presented as odds ratio (95% confidence interval).Bold text represents significant associations.Reference group for the associations: patients with no DM or IGR.All odds ratios were adjusted for sex, age, household size, income, hypertension (baseline), previous miner, previous prisoner, marital status, work status, and HIV-1 status.
between TB and IGR at enrollment but not follow-up.However, when these results were further analyzed by DM category (newly diagnosed vs preexisting) and by HIV status within these categories, differing patterns emerged.Given the socioeconomic conditions of our study setting, the patients with newly diagnosed DM in our study may represent the proportion of undiagnosed DM due to factors associated with limited access to DM screening and having health facilities overwhelmed by TB and HIV [17].

Newly Diagnosed DM
Hyperglycemia, transient in the majority of participants with newly diagnosed DM and IGR, was predominantly accounted for by the latter at enrollment, and normalized at follow-up.Similar to our findings, other studies have shown frequent hyperglycemia in patients with TB at initiation of TB treatment, followed by normalization during treatment [23].This may be due to inflammation in response to active TB [25][26][27], driven by complex interactions between hormones and cytokines [24].Consistent with the literature, the association between DM/ IGR and TB in newly diagnosed DM participants was only significant at enrollment but not at follow-up.With respect to the different diagnostic tests separately, none of the associations were significant.

Preexisting DM
Patients with preexisting DM had poorly controlled disease at both diagnosis of TB and 3 months later, as reflected by both FPG and HbA1c at these timepoints.Unlike newly diagnosed DM participants, the significant TB/DM association at enrollment (OR, 3.7 [95% CI, 1.5-9.1]) in this group, which persisted at follow-up (OR, 4.0 [95% CI, 1.6-10.1]),reflects poor glycemic control between the study timepoints.Hyperglycemia may have been exacerbated by acute TB.The persistence of raised glycemic levels reflects inadequate management of these patients, possibly due to poor follow-up, and highlights the complexity of clinical care in this subgroup.
A Korean retrospective study showed that TB is usually diagnosed 1 year following the diagnosis of DM [27].Therefore, the observed relative odds of preexisting DM among patients with TB compared to those without TB in this study (OR, 2.8 [95% CI, 1.5-5.3])may approximate the relative risk to develop TB among patients with diabetes.Although the temporal relationship between TB and DM remains contentious, irrespective of the causal direction, comorbidity of TB and DM increases the risk of adverse TB treatment outcomes including treatment failure, mortality, and drug resistance [3,5,13].The observed poorly managed DM in these individuals with TB also highlights the increased risk of adverse TB treatment outcomes among these patients.

TB, DM, and HIV
When stratified by HIV-1 status, the TB/DM association (when DM is defined by the different diagnostic tests separately) was not significant at either time point.Interaction between HIV, TB and DM is still not well understood and different studies report varying results.
It is complicated to interpret the interplay between TB, HIV, and DM as a wide range of factors may influence what is observed.This includes the effect of ART, HIV-1 infection as an independent risk factor for both DM and TB, or the choice of diagnostic test.Cotrimoxazole, administered to people living with HIV-1, can lead to hypoglycemia [28].Conversely, ART, particularly regimens containing protease inhibitors, increases insulin resistance, thus increasing the risk of diabetes [29].
Other studies suggest that HbA1c may underestimate the presence of hyperglycemia in people living with HIV-1 and that this may be due to nucleoside reverse transcriptase inhibitor use [30,31].In our study, HbA1c detected a higher prevalence of hyperglycemia.In a 2015 review by English et al, it was suggested that an HbA1c test is likely to be affected by iron deficiency anemia and may result in spurious increases of HbA1c level [32].On the other hand, non-iron deficiency anemia may lead to a reduction in HbA1c levels [32,33].A recent study observed lower HbA1c mean levels in severely anemic patients; however, due to a limited sample, no further analysis was performed to explore this relationship [34].The effect of anemia on the direction of the association is therefore unclear.
We explored the potential effect of unmeasured confounding on our results by performing sensitivity analysis (Supplementary Table 3) [35].This showed that in IGR and newly diagnosed DM, the association with TB was rendered nonsignificant at baseline and follow-up.As such we note that weaker residual confounding may explain away our observed estimates.However, the association between TB and preexisting DM was significant both at enrollment and follow-up with ORs ranging between 3.7 and 9.3.To explain away these associations, an unmeasured confounder would need to be associated with both TB and preexisting DM with risk ratios ranging between 2.9 and 6.9, but weaker confounding would not.
There were strengths and limitations to this study.A limited number of studies have evaluated hyperglycemia in individuals with TB, particularly in Africa where there is a high prevalence of comorbidity with other diseases such as HIV-1.TB cases were diagnosed according to South African guidelines [19], with the Gene Xpert analyzed in a centralized national health laboratory.For enrollment and follow-up, DM measurements were performed using 2 recommended tests.We relied on medical records to document ART use, as such we could not reliably ascertain duration of ART use in our multivariate analysis.
Our study follow-up time was limited to 3 months and we were not able to analyze the effect of hyperglycemia on TB outcomes.Because critically ill patients were excluded in the study, it may have biased the study population to appear healthier.
The proportion of participants lost to follow-up in this study, and to the health system, was relatively high (n = 212 [24.9%]) (Supplementary Table 2).Those followed up were older, mostly unemployed, and likely to have known DM and hypertension.Our results at follow-up may thus be slightly biased as they represent an older population prone to chronic conditions.Reasons for loss to follow-up include migration to other parts of the country and transfer to other health facilities in the province.To reduce bias and loss of statistical power, we imputed HIV status for patients with unknown HIV status.Therefore, loss to follow-up is less likely to have biased the associations observed in the study.The loss to follow-up observed in this study reflects how patients are lost in healthcare systems in this setting.This therefore highlights the importance of improving retention of patients in care for optimal management of all chronic diseases.

CONCLUSIONS
This is the first study in the South African context of high TB/HIV and rapidly increasing DM burdens to describe changes in glucose levels among patients with TB during treatment.This study showed that hyperglycemia was common in TB patients with DM.This confirms the need for confirmation DM tests in TB patients during and/or after the course of TB treatment.The association between DM and TB persisted at follow-up in participants with preexisting DM, particularly those infected with HIV-1.This highlights an important need for improved co-management of TB, DM, and HIV to limit the risks of adverse outcomes.

Table 2 . Glycemic Levels Among Participants With Tuberculosis
Data are presented as no.(%) unless otherwise indicated.Number of participants is shown at the end of each category.Values in bold indicate statistical significance.

Table 1 .
Continued Institute, which receives funding from Cancer Research UK (grant number FC001010218), Research Councils UK (grant number FC0010218), and the Wellcome Trust (grant number FC0010218).He also receives support from the National Institutes of Health (NIH) (grant number U1 AI115940), NIH (grant number WILK116PTB), and European and Developing Countries Clinical Trials Partnership (grant number SRIA 2015-1065).M. K. is supported by the South African Centre for Epidemiological Modelling and Analysis, the International Epidemiology Databases to Evaluate AIDS, and the NIH (grant number U01AI069924).Potential conflicts of interest.The authors report no potential conflicts of interest.All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest.Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.