The prevalence and determinants of active tuberculosis among diabetes patients in Cape Town, South Africa, a high HIV/TB burden setting

Highlights • First in Africa to screen all DM patients for TB (irrespective of symptoms using Xpert) and HIV.• Study found high prevalence of TB in DM patients, over half of whom were asymptomatic.• HIV prevalence was 13% in DM patients, showing importance of HIV screening this older population.• DM patients with HIV or hemoptysis at greater TB risk, and should be targeted for active TB case-finding.• Given high subclinical active TB prevalence, screening should be irrespective of symptoms in this group.


Introduction
The Sustainable Development Goals (SDGs) [1] have made ending the global TB epidemic a priority and to achieve these new strategies need to be employed. Early detection of TB can lead to prompt treatment, increasing survival, and diminishing transmission risk. Interventions to identify TB in those at high risk for developing the disease are needed.
HIV-1 infection is the strongest recognized risk factor for TB, however other risk factors contribute to the TB burden, including smoking [2] alcohol abuse [3], vitamin D deficiency [4], and diabetes mellitus (DM) [5]. The growing DM epidemic has been associated with rapid urbanization and changing lifestyles in low and middle-income countries (LMIC), where the greatest increase in the number of people with DM and the greatest number of deaths due to DM are found. In a recent estimate, DM prevalence in sub-Saharan Africa (SSA) is projected to increase from 14.2 million cases in 2015 to more than double by 2040 [6]. This is emerging against the backdrop of a persisting high burden of infectious diseases including TB and HIV.
Previous studies have demonstrated that DM increases the risk of TB [7][8][9][10], with DM associated with a 3-fold risk of TB compared to those without DM (95% CI 2.27-4.27) [5]. A systematic review on bi-directional screening of TB and DM cases concluded that active screening of DM patients leads to increased detection of TB cases [11]. TB prevalence among DM patients has been found to vary from 3% [12] to 36% [13]. While large-scale screening for TB amongst DM patients has been demonstrated to be feasible in routine and low-resourced healthcare settings [14,15] the cost-effectiveness and validity of such strategies are unclear.
Studies investigating the prevalence of TB among persons with DM in the SSA context are limited [16][17][18]. In Tanzania, 1.3% of screened adults with DM had TB, 7-fold greater than the general population [16]. A modelling study estimated that up to 15% of TB cases in South Africa (SA) might be attributed to DM [19]. With TB incidence in SA remains one of the highest in the world with 834 cases per 100,000 people [20] and these studies demonstrate the potential to improve case detection through systematic screening of persons with DM, however the evidence to support this is lacking. A recent study of 672 patients with DM in Soweto, SA found a 4% prevalence of TB symptoms but no active TB cases amongst those with respiratory symptoms. Of note, only patients with clinical symptoms of TB were screened; and the influence of HIV infection was not explored [21]. Our study therefore aimed to investigate the prevalence of active TB, irrespective of the presence of symptoms, in this high HIV/TB co-infection setting.

2.
Subjects, materials and methods

Study design and recruitment
Between September 2014 and October 2015, we conducted a cross-sectional study at a primary care clinic in Khayelitsha, a peri-urban township in SA. With a predominantly Black African population of 391,749 [22], it has one of the highest burdens of TB and HIV globally [23]. In 2012, the HIV antenatal preva-lence in Khayelitsha was estimated at 34% (95% CI 31.0-36.6) (unpublished data from the 2012 Western Cape Department of Health Antenatal Survey) and the TB case notification rate was 1400 per 100,000 people annually, with 70% of TB cases being HIV-1 co-infected [24]. A community-based study in 2011 demonstrated a 13.1% (95% CI 11.0-15.1) prevalence of DM in Cape Town, specifically in the township communities [25].
To evaluate the prevalence of active TB in a routine clinical setting, consecutive adult DM patients attending a DM clinic for chronic disease management at a community health clinic were approached. Participants were recruited from a diabetes clinic, where their diagnosis had been previously made. As such, all patients attending this clinic had a pre-existing DM diagnosis, All adult patients (18 years at recruitment) attending this diabetes clinic were eligible to participate and their DM status was confirmed on review of patient records and treatment schedule. Patients were eligible to participate if they had a pre-existing DM diagnosis, were receiving DM treatment, and were aged 18 years at the time of recruitment. Participants were enrolled after written consent was given and then given a scheduled appointment to return, fasted, for sputum and blood collection.
The study received ethics approval from the University of Cape Town Human Research Ethics Committee (HREC Ref: 377/2015).

Data collection
Questionnaire: Research community health workers, bilingual in both English and Xhosa, the predominantly spoken languages in Khayelitsha, administered a questionnaire to collect socio-demographic and behavioral information using the WHO's STEPwise approach to surveillance of chronic disease risk factors (STEPS) [26]. Participants were then scheduled to return within a week. A validated TB screening tool (adapted from the symptom-based Practical Approach to Lung Health and HIV/AIDS in SA [27]) was used to assess the presence of TB symptoms, including: cough, night sweats, fever, hemoptysis, and weight loss. In the same screening tool, information on TB contact history, previous history of TB, HIV status and self-reported chronic disease co-morbidity history were also collected. ART (Antiretroviral therapy) use was extracted from the clinic's electronic database. Participants who did not know their HIV status were encouraged to undergo HIV testing, which consisted of a rapid point-ofcare HIV antibody based test [28]. Measurements: At the return visit, following an overnight fast of 8-12 h, blood samples were collected for HbA 1c (glycated hemoglobin) and fasting plasma glucose (FPG) measurement. Participants were categorized as having ''controlled" (HbA 1c <7% (53 mmol/mol)) and ''poorly controlled," (HbA 1c >7% [53 mmol/mol]) DM. Waist circumference, weight and height were measured and body mass index (BMI kg/m 2 ) calculated as follows: underweight: <18.5; normal: 18.5 BMI < 25; overweight: 25 BMI < 30; obese: 30. Vital signs were also measured including respiratory and pulse rates, temperature, and blood pressure. Hypertension was defined as a systolic blood pressure (BP) 140 mmHg or diastolic BP 90 mmHg, a pre-existing d i a b e t e s r e s e a r c h a n d c l i n i c a l p r a c t i c e 1 3 8 ( 2 0 1 8 ) 1 6 -2 5 diagnosis of hypertension, or taking medication for hypertension. TB screening and diagnoses were conducted using the national TB management guidelines [28]. Participants were classified as having subclinical TB if diagnosed with active TB but with an absence of any clinical symptoms. All participants underwent spontaneous or induced sputum collection. Sputum samples were processed according to national TB program guidelines [28]. This included GeneXpert processing and drug resistance testing for all participants, with additional smear microscopy if the GeneXpert result was positive or if participants were HIV-1-infected. An active TB case was defined as persons who tested positive for M. tuberculosis by either GeneXpert, smear microscopy, or TB culture in the presence or absence of clinical symptoms. An accredited national laboratory (National Health Laboratory Service) that adheres to standardized protocols and follows quality assurance measures processed collected samples. Patients diagnosed with active TB and/or HIV were referred for further clinical care and treatment.

Statistical analysis
The baseline characteristics were described using descriptive statistics and univariate analyses. Associations between participants with and without prevalent active TB were tested using Pearson's v 2 test and Wilcoxon rank-sum test for categorical and continuous variables, respectively. The prevalence of active TB, including the confidence interval, was calculated.
Possible risk factors associated with active TB were further analyzed using multivariate logistic regression. Significance testing was two-sided at p-values 0.05. The multivariate model was built using purposeful selection where risk factors were selected based on clinical and statistical significance to control for possible confounders [29]. Odds ratios (OR) and 95% confidence intervals were also estimated. Model validation was performed to identify any outliers and influential observations and the fit of the model to the data was evaluated using Pearson's goodness of fit test. Potential effect modification was assessed using interaction variables. A sensitivity analysis was conducted to investigate the potential impact of unknown HIV status. All data analyses were conducted using STATA version 13 (StataCorp, College Station, TX, USA).

Results
We approached 492 DM patients to participate in the study, of whom 52 were excluded, resulting in a final sample size of 440 ( Fig. 1). Reasons for exclusion include: consent refusal (n = 6), missed attendance of follow-up appointment (n = 25), and indeterminate/contaminated sputum results (n = 21). Demographic (Table 1) and baseline clinical (

Discussion
The main findings of this study were that the prevalence of active TB among DM patients at 3.0% was 4.3-fold greater than the national estimate of 696 per 100,000 [20]. DM patients with HIV-1 infection were 11 times more likely to have active TB than those without HIV; hemoptysis was significantly associated with prevalent active TB and there was a lack of association between glycemic status and active TB. The prevalence rate of active TB in our study was within the range of 0.1-6.2%; reported prevalence rates from previous screening studies conducted in similar high TB burden settings such as Tanzania and Ethiopia [16,17]. However, our study differs from many others in that all DM patients were screened for active TB irrespective of reported symptoms. In similar screening studies, sputum samples were collected from only DM patients clinically suspected of having TB [15,16,30]. Previous evidence has shown that symptom screening may not be an effective method in detecting TB in those with HIV [31,32] and DM [33]. Our findings support this, as 54% of TB cases had no TB symptoms. This was similar to a study conducted in a comparable setting where a high prevalence of sub-clinical TB was found among HIV patients [30], with more than half of these cases developing symptoms 3 days to 2 months later. This highlights the importance of early active TB case detection in sub-clinical patients, especially those co-infected with HIV and those with co-morbid with DM. Additionally, we noted that classic TB symptoms were not significantly different between active TB and non-TB cases. This alludes to the poor specificity of these symptoms in persons with DM, with symptoms such as fatigue and weight loss also representing symptoms of DM.
Other studies have reported the risk of active TB to be higher in DM males compared to females [30,[34][35][36][37]. In our study, whilst the odds of having active TB was greater in females than males, this difference did not achieve statistical significance.
In our study, elevated HbA 1c and FPG (using cut-off values of 7% [53 mmol/mol] and 7 mmol/L respectively) were not statistically significant risk factors for prevalent active TB. This was comparable to other studies evaluating HbA 1c that found no association [38,39]. However, other studies have reported an association between TB and HbA1 c levels [30,40], particularly in patients with HbA1 c >9.0% (75 mmol/mol) [30]. In one cohort study, there was a linear relationship between FPG and TB risk [41]. These differences may be due to the different measures (HbA 1c , FPG) and cut-off values used to define glycemic control. For example, using quintiles to explore the association between HbA 1c and TB may reveal a higher prevalence of active TB at much higher HbA 1c levels that may be masked using a binary categorization. However, due to the small sample of TB cases, our study was underpowered to explore this. In addition, FPG may be more likely to reflect stress-induced hyperglycemia than HbA 1c . Given these inconsistencies between studies, proxies of optimal DM management have been explored as an alternative method to further evaluate the association between poor glycemic control and TB risk. In one longitudinal study conducted in a low TB burden setting, whilst there was no association found, those with the highest and lowest numbers of consultations or clinic follow-ups had the strongest association with TB risk, compared to other significant risk factors [38]. This finding suggests that measures of DM control that take into consideration poor or excessive attendance at the DM clinic could be valuable measures to better assess DM control and subsequent TB risk.
While the prevalence of HIV among DM patients screened for this study (12.7%) was similar to the national SA prevalence rate at 12.6% [42], the prevalence of patients with both DM and TB was nearly 5 times greater than the national  and study population rate. DM participants with TB were more likely to have HIV compared to DM participants without TB as the odds were nearly 10 times higher compared to patients without TB. It was surprising that smoking and alcohol dependence, well-known risk factors for TB [43,44], were not identified as such in our study. As smoking rates are higher in males, the higher proportion of female participants in this study is likely to have reduced the statistical power of our study to detect this association. Overall our study revealed a high prevalence of smoking (18%), which was similar to the SA national prevalence 17.6% [45], emphasizing the need to address modifiable risk factors as part of an integrated preventative approach to improving chronic disease outcomes. Several limitations were identified in this study. Selection bias may have been present as inclusion required participants to return for a scheduled  appointment. However, as only 25 participants did not return, this is unlikely to significantly influence the results. Additionally, our participants were recruited from a DM clinic and thus our results may underestimate the prevalence of TB in undiagnosed or non-adherent DM patients. Since the study was underpowered, we were unable to explore the effect of additional factors such as glycemic control, duration of DM diagnosis and medication use on TB risk; consequently, these variables were not collected. Reported symptoms were not confirmed or validated. There is a possibility that patients might have misinterpreted or misperceived definitions of TB symptoms that might have resulted in an over-or under-estimation of reported symptoms leading to respondent/recall bias. Apart from cough and fever, the temporality and duration of these classical TB symptoms were also not reported. For TB contact, date of contact was not obtained and so it was not possible to ascertain the importance of duration since contact on risk of active TB. Potential confounders such as vitamin D and consultation rates, were not measured. As such there is potential for error due to these unmeasured confounders. Lastly the wide confidence intervals of HIV and hemoptysis in the multivariate analysis produce uncertainty and a lack of precision about the true magnitude of association with active TB.
Nonetheless, the prevalence of active TB in DM patients was much higher than the national estimate in the general population, demonstrating that screening approaches targeting DM patients are potentially more efficient than screening the general population. This is particularly so in DM patients with HIV-1 infection who are at an even greater risk of TB.
The importance of HIV-1 as a significant risk factor for TB has been further confirmed in DM patients in this study. This finding also shows the growing trend of multi-morbidity, and the need to focus on the care and management of both chronic infectious and non-communicable diseases, including early diagnosis and treatment of HIV in (often older) patients with diabetes.
Whilst the presence of hemoptysis was associated with a higher risk of prevalent active TB, the majority of active TB cases were asymptomatic at the time of diagnosis. This finding poses a serious threat to TB control as conventional symptom screening could delay TB diagnosis in this target population, as a significant proportion of DM patients with TB would be missed on routine screening. The low sensitivity of symptom screening highlights the need for accurate pointof-care diagnostic tools such as the GeneXpert to detect asymptomatic TB cases.
For this target population, further studies are required to investigate who should be screened to improve screening feasibility needs to be considered. Should we consider routine screening for TB amongst all DM patients in high TB prevalence settings? Or should screening focus on DM-HIV patients only? A significant decrease in TB incidence and mortality is required if the SDG and WHO goals to eliminate TB are to be met. Further research is therefore required to determine the most cost-effective and accurate TB screening algorithms to increase early case detection, as well as evaluating the impact of implementing TB screening on TB outcomes in this high-risk population.