Divergence in Accuracy of Diabetes Screening Methods in Tuberculosis Patients: A Cross- Sectional Study from Brazil and Peru

Background: To evaluate the accuracy of distinct diabetes mellitus (DM) screening methods in persons with active Tuberculosis PWTB. Methods: Levels of fasting plasma glucose (FPG) and glycated hemoglobin (HbA1c) at the time of tuberculosis diagnosis at the study clinics were assessed from two distinct retrospective cohorts of PWTB from Brazil (n=116) and Peru (n=136) to evaluate accuracy for detecting pre-DM and DM cases. Additionally, we investigated the association of clinical and sociodemographic factors with tuberculosis and pre-DM or DM in each country. Results: When comparing PWTB from Brazil and Peru, Peruvian individuals presented higher FPG levels at baseline (median [IQR] 91 [81–106 vs 95 [88.4–102.1]]; p=0.02), while those from Brazil had signicant higher levels of HbA1c (median [IQR] 6.3 [5.7-7.15] vs 5.1 [4.9-5.4]; p<0.01). Additional analysis using the receiver operating characteristic curve revealed that the markers showed distinct accuracy to identify dysglycemia among PWTB in each country. Conclusion: Our ndings indicate that there are signicant differences in the total accuracy of the glycemic screening methods evaluated between PWTB from two highly endemic countries from South America, highlighting the need to revisit the diagnostic criteria of DM/PDM in individuals with tuberculosis. when appropriate. Quantitative variables were expressed as median with interquartile range (IQR) and compared using the Mann-Whitney U test, for two groups, or the Kruskal Wallis test with Dunn’s multiple comparisons posttest for more than two groups. Analyses of stratied or matched categorical data were performed with Cochran-Mantel-Haenszel test. The Kappa (K) statistic test was calculated to assess agreement between FPG or HbA1c as diagnostic test for DM/PDM in both countries. Kappa statistic results were interpreted using the Landis and Koch criteria [23]. All tests were pre-specied, two-tailed and differences were considered statistically signicant with p ≤ 0.05. Data analysis was performed using SPSS 24.0 (IBM statistics), Graphpad Prism 7.0 (GraphPad Software, San Diego, CA) and JMP 13.0 (SAS, Cary, NC, USA).


Background
Tuberculosis (TB) is a major public health concern at global level. In 2019, The World Health Organization (WHO) estimated 10 million new cases and 1.4 million deaths caused by TB [1]. Notably, among risk factors associated with TB morbidity and mortality, it is a common knowledge that glycemic disorders such as diabetes mellitus (DM) and prediabetes (PDM), exhibit critical in uence over immune pathology and systemic in ammation, thus increasing risk of worse clinical outcomes [2][3][4]. Moreover, it is known that DM triples active TB risk and exacerbates TB clinical presentation resulting in early mortality rates (death within 100 days of starting anti-TB treatment) and increasing the odds of unfavorable TB treatment outcomes [2][3][4][5][6]. On the converse, TB may lead to infection-related hyperglycemia, often decompensating glycemic control in diabetics [4,5]. Hence, the relationship between dysglycemic states and TB is bidirectional, establishing a mutual harmful association.
TB and DM convergence results in a signi cant disease worldwide, especially in low and middle-income countries [3,7]. Indeed, despite all TB therapeutic strategies and preventive support improvement, Brazil remains among the 20 countries with the highest disease burden in the world and number one in the Americas, followed by Peru [1,8]. While Brazil is responsible for the largest number of absolute cases for its large population, Peru has the second highest incidence rate in the Americas (123/100.000 versus 45/100.000 in Brazil) [8]. In Brazil, 7.6% of incident TB cases are associated to DM [9]. As an effort to decrease the impact caused by TB-DM syndemic, the WHO as well as the Peruvian and the Brazilian National TB Program (NTP and PNCT, respectively) recommend screening for DM those presenting with active TB and frequent screening for TB symptoms in persons with DM (PWDM) [1,10,11].
In the clinical practice, fasting plasma glucose (FPG) and glycated hemoglobin (HbA1c) are the most frequent tests used to diagnose DM, followed by the oral glucose tolerance test (OGTT) [5,12]. OGTT is considered the gold standard for DM diagnosis, but it has some expressive limitations as a mass screening test, due to the uncertainty of the fasting state, the time-consuming nature of the test, poor reproducibility of the results and need of a standard 75g glucose load, which is a challenge in primary care settings in resource-limited countries [5,13,14]. Nevertheless, the accuracy of HbA1c and FPG can be limited by issues such as accuracy variation among subjects of different ethnicity [15]. Of note, previous investigations assessing accuracy of these laboratory parameters for screening of DM in PWTB exposed con icting results in different populations, especially in those from South America [13,16]. This could result from blood tests being affected by ethnicity and genetic variations in uencing hemoglobin concentrations in peripheral blood [17], strengthening the hypothesis that unique features determined by genetic background might have an impact on the choice of the most reliable DM screening method.
In the present study, we investigated the performance of HbA1c and FPG in identifying PDM and DM cases among PWTB from two South American countries with high burden for both TB and DM, Brazil, and Peru. Differences in associations of clinical and sociodemographic characteristics with TB and dysglycemia between the countries were also examined.

Study design and population in Brazil
We conducted a retrospective cross-sectional study of data retrieved from a prospective cohort conducted between May 2010 and June 2011 in the Instituto Brasileiro para Investigação da Tuberculose (IBIT), Fundação José Silveira, Salvador, Brazil, which aimed to examine the association between glucose metabolism disorder and pulmonary TB [18]. For the present study, the inclusion criteria were patients ≥ 18 years of age, diagnosed with TB according to the Brazilian Manual of Recommendations for TB Control [10] and who were not undertaking anti-TB drugs for more than 5 days, whereas incomplete medical report was considered as the exclusion criteria. Sociodemographic and clinical characteristics were collected by trained physicians during each patient visit and recorded in standardized electronic forms, which are part of the Brazilian National TB control program. Moreover, as part of the laboratory investigation, three sputum smears stained by Ziehl-Neelsen and examined by microscopy at the IBIT's microbiology referral laboratory, processed by the modi ed Petroff's method and cultured on Lowenstein-Jensen medium. Diagnosis of DM or PDM was performed at the time of TB diagnosis (baseline visit) in agreement with American Diabetes Association (ADA) guidelines [19], and was based on fasting plasma glucose (FPG), glycated hemoglobin (HbA1c) and oral glucose tolerance test (OGTT) as previously described [18]. Further details about the study site and data management are described in previously published studies [18,20].

Study design and population in Peru
We retrospectively analyzed data of 136 individuals diagnosed with pulmonary TB, from a larger prospective cohort study conducted over February and November 2017 with patients from North Lima, Peru [2]. TB diagnosis was performed by the National TB Program (all patients had microbiologically con rmed TB) and inclusion criteria comprised patients with age ≥ 18 years and who were not receiving anti-TB treatment or had started in no more than 5 days. Information on sociodemographic and clinical evaluation was retrieved from the medical records. All enrolled participants provided sputum samples for acid-fast bacilli (AFB) smear, which were stained by Ziehl-Neelsen and examined by microscopy. Then sputum specimens were cultured by Lowenstein-Jensen medium and BD MGIT 960 System (liquid culture). Smear and cultures were graded according to AFB and colonies numbers following standard guidelines [21]. Evaluation of glycemic markers, FPG, HbA1c and OGTT, was performed to establish DM or PDM diagnosis according to ADA criteria [19]. Such procedures were performed at the baseline clinical visit when TB was diagnosed. Measurement of HbA1c in blood was conducted using TRI-stat™ platform (Trinity Bio-tech, Ireland). FPG and OGTT were performed following standard methods. Supplementary information about procedures and patient management, are well described in previous publications [2,6].
Epidemiological characteristics of TB and DM in Brazil and Peru.
Overall distribution of TB and DM cases in both countries was described accordingly with the data gathered from the international reports published by the WHO [8] and International Diabetes Federation [22]. Such distribution is shown in Fig. 1.

Data analysis
Descriptive analysis was performed with categorical variables presented as frequency and percentages and compared using the Fisher's exact test (between 2 groups) or Pearson's chi-square test (more than 2 groups), when appropriate. Quantitative variables were expressed as median with interquartile range (IQR) and compared using the Mann-Whitney U test, for two groups, or the Kruskal Wallis test with Dunn's multiple comparisons posttest for more than two groups. Analyses of strati ed or matched categorical data were performed with Cochran-Mantel-Haenszel test. The Kappa (K) statistic test was calculated to assess agreement between FPG or HbA1c as diagnostic test for DM/PDM in both countries. Kappa statistic results were interpreted using the Landis and Koch criteria [23]. All tests were pre-speci ed, two-tailed and differences were considered statistically signi cant with p ≤ 0.05. Data analysis was performed using SPSS 24.0 (IBM statistics), Graphpad Prism 7.0 (GraphPad Software, San Diego, CA) and JMP 13.0 (SAS, Cary, NC, USA).

TB and DM burden in Brazil and Peru
The overall distribution of TB and DM cases using the most updated report containing estimated data from both countries, performed in 2019, is shown in Fig. 1 [8,22]. Brazil led the burden of TB in South and Central America, with a total of 96,000 TB cases reported and a rate of 46 cases per 100,000 inhabitants, while Peru had a total of 39,000 TB cases estimated and a rate of 119/100,000. Regarding DM distribution, more than 16.8 million cases were reported among Brazilians, with a rate of 80 cases per 100,000 estimated in 2019 [1] and among Peruvian approximately 22 cases per 100,000 [24]. Furthermore, the cohort explored in the present study was located in state of Bahia, represented in the map with an incidence of TB-DM of 9 cases per 100 inhabitants. Similarly, the Province of Lima, from where our Peruvian cohort was followed, had an incidence of 10 cases per 100 inhabitants of TB-DM comorbidity in 2019. Therefore, our cohorts are originated from relevant endemic areas with relatively high burden of both TB and DM.

Characteristics of the study populations
Brazilian participants were signi cantly older than the Peruvians, (median age [IQR]: 45  vs. 29 [IQR: 23-45], respectively; p < 0.01) ( Table 1). Distribution of sex was similar between the cohorts (Table 1). With respect to clinical presentation, we found that Brazilian participants presented more frequently with cough (p < 0.01), fever (< 0.01) and weight loss (p = 0.03), whereas individuals from Peru more often had hemoptysis (p < 0.01) and lack of appetite (p = 0.02) ( Table 1 and Supplementary Fig. 1 consumption were more frequent in Peruvians (p < 0.01) ( Table 1). Metformin use was more frequently documented in Peruvian participants than that in those from Brazil (11.4% vs 1.3%, p < 0.01). Finally, Peruvians more often were BCG vaccinated than Brazilians (93.3% vs 75.4%, p < 0.01). Additional comparisons are depicted in Table 1.  Fig. 2A). However, FPG levels in this group showed that most individuals with dysglycemia presented values under the reference baseline, whereas, in Peru, 70.5% of patients with DM/PDM exhibited values above the limit for dysglycemia with this marker (Fig. 2A). Comparison of FPG median values also displayed differences between countries (p < 0.05), with higher levels found in the Peruvian cohort (Fig. 2C). On the other hand, HbA1c levels had an inverse distribution in both populations (Fig. 2B), in which Brazilians presented the vast majority of values above the reference for DM/PDM diagnosis. In addition, the median values of HbA1c showed to be signi cantly increased among Brazilians in comparison with Peru (p < 0.001) (Fig. 2D). These results suggest that the glycemic screening methods showed a distinct behavior according to the country.
Glycemic screening methods among TB cases Next, we aimed to evaluate FPG and HbA1c level distribution according to the following groups: TB, TB-DM and TB-PDM. (Fig. 3). Using a demographic density analysis with histograms, we found a similar distribution of FPG values and HbA1c percentage on both countries, where TB normoglycemic participants demonstrated a peak curve in lower FPG levels, followed by TBPDM (Fig. 3A). Of note, our ndings revealed that the TBDM group displayed wider distribution in both country curves (Fig. 3A).
To better understand the differences in the glycemic markers according to TB-DM comorbidity between the two countries, we compared the levels of FPG and HbA1c in TB, TB-PDM and TB-DM groups in Brazil and Peru (Fig. 3B). In all clinical groups, levels of FPG were higher among the Peruvians, whereas Brazilians displayed higher HbA1c values, except for the TBDM group (Fig. 3B).
Distinct accuracy of glycemic screening methods in dysglycemia diagnosis among TB South American individuals.
In order to extend our investigations concerning the discriminative performance of glycemic markers to diagnose dysglycemia in TB patients, a receiver operating characteristic (ROC) curve analysis using values of the glycemic markers was employed in each country (Fig. 4). HbA1c exhibited a good performance to identify dysglycemia the Brazilian cohort, with an area under curve (AUC) of 0.98 (CI: 0.93-1.00) (Fig. 4A), whereas FPG demonstrated superior performance among Peruvians (AUC: 0.90; CI, 0.84-0.95) (Fig. 4B). We next compared the different cohorts with regard to the performance of the FPG or HbA1c tests in distinguishing dysglycemia and found substantial differences between the AUC (p = 0.0089 and p < 0.0001 correspondingly) (Fig. 4C). This nding reinforces the idea that there are important discrepancies in the overall accuracy of the FPG and of HbA1c tests to identify individuals with prediabetes or diabetes between Brazilians and Peruvians.
Finally, additional comparisons were made to narrow down the concordance between HbA1c and FGP to detect diabetes or prediabetes in the study populations. We noted that the FPG test had a good degree of agreement in the Peruvian cohort for identi cation of dysglycemia cases (k = 0.77) or PDM (k = 0.6), and very good degree for identi cation of DM individuals (k = 0.96). Interestingly, for Brazilian cohort the degree of concordance for identi cation of dysglycemia or PDM was poor (k = 0.20 and k = 0.08, correspondingly) and for DM it was moderate (k = 0.11) (Fig. 4D). In contrast, the HbA1c test exhibited a very good agreement in the Brazilian cohort to detect dysglycemia (k = 0.83), DM (k = 0.98) or PDM (k = 0.84). Meanwhile, in the Peruvian cohort, the agreement for dysglycemia was moderate (k = 0.44), for DM it was very good (k = 0.97) and for PDM it was just fair (k = 0.35).

Discussion
The ndings presented here depict a detailed investigation of the performance of two tests widely used for assessment of dysglycemia among PWTB from Brazil and Peru. The studied countries represent an expressive portion of TB cases and around of 50% of Latin American cases in the world, with also elevated prevalence of DM and PDM [1,8,22]. As discussed, PWTB with concurrent uncontrolled glycemic status have an increased risk of increased morbidity and poor treatment outcomes [25,26]. Therefore, there is an urgent need for systematic screening of dysglycemia in PWTB, in order to reach those at highest risk of disease complications. Understanding the peculiarities of diagnosing dysglycemia in TB cases in different populations is key to develop focused control strategies adapted to local epidemiological trends.
Our analyses revealed that the accuracy of HbA1c and FPG to diagnose dysglycemia in PWTB differs between Brazil and Peru, with a better performance of HbA1c in Brazil and FPG in Peru. Findings from studies in other countries reinforce the idea that there is heterogeneity in performance of DM screening tests in PWTB. A study from India has demonstrated that HbA1c performed better than FPG with an AUC of 0.754 (0.682-0.828) for newly diagnosed DM among subjects with TB [13]. Furthermore, a Pakistan study showed that the proportion of participants falsely classi ed as positive was higher for FPG, although the performance of HbA1c and FPG had no differences in terms of diagnosing new DM cases [17]. In a recent study, FPG levels were able to detect more cases of PDM in PWTB from Peru than HbA1c [2]. In a Chinese study, FPG performed better than HbA1c in identifying newly diagnosed DM and PDM [27]. Such peculiarities in test performance to diagnose dysglycemia reported in different countries may be determined by genetics, environmental factors or a combination of both and should be explored in future mechanistic studies.
Previous studies have shown that HbA1c values vary according to ethnicity, even in individuals without dysglycemia [28,29]. Moreover, abnormalities of erythrocyte indices are considerable confounders in the analysis of HbA1c [30]. Possible explanations for this could be hemoglobin-related factors such as red cell turnover, variations in hemoglobin glycation, differences in the passage of glucose mediated by GLUT1 transporter into the erythrocyte, higher prevalence of hemolytic conditions such as glucose-6-phosphate dehydrogenase de ciency or sickle cell trait, among others [29,31]. Genetic factors can also modulate HbA1c levels with heritability of 47-59% [32]. Interestingly, a genetic risk score, based on the 14 single nucleotide polymorphisms (SNPs) found that some people had a higher genetic risk of higher levels of fasting plasma glucose [33]. Regarding DM in PWTB, a multicentric prospective study compared the accuracy of random plasma glucose, point-of-care HbA1c, FPG, urine dipstick, risk scores and anthropometric measurements, using HbA1C as a golden standard. They also found heterogeneous performance of laboratory markers to diagnose DM across countries [12].
More than identifying the reasons for HbA1c variation among populations, it is essential to know what the clinical importance of these ndings is. It is still under discussion how this variation in test performance can affect DM outcomes, but some studies have found no difference in long term risk for cardiovascular disease, nal-stage renal disease and retinopathy [29,34,35]. To our knowledge, no studies to date have evaluated the in uence of HbA1c accuracy variation in TB outcomes. A previous study reported higher HbA1c values being predictive of unfavorable outcomes in PWTB [36], but there was no comparison between populations with distinct genetic backgrounds.
The ADA guidelines for DM recommends that any of three diagnostic test mentioned in this study can be used to diagnose DM [19]. The Brazilian guidelines use the same criteria as ADA and the Peruvian guidelines have not included HbA1c as a diagnostic tool because of its low availability in the public health system and lack of standardization in the country [37,38]. Our ndings suggest that the use of HbA1c should be used with caution in the diagnosis, as the cutoff is still inconsistent and may vary in dissimilar populations. It is possible that different thresholds need to be used in different populations.
More research needs to be done in evaluating the impact of variability in DM screening test accuracy in TB treatment outcome. Moreover, additional studies are required to de ne the most reliable screening methods for each population. As an example, Grint D, et al tested a combination of two tests that increased DM diagnosis accuracy in PWTB from Peru and Indonesia, but not in other countries [12]. Meanwhile, standard diagnostic thresholds should be used with caution, particularly in the population with greater variability of tests results. Clinicians should also be attentive for factors that are predictors of high glycemia, such as older age, hypertension, and increased body mass index [28] and repeat or associate a different method when there is high pre-test probability of dysglycemia.

Availability of Data and Materials
The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.

Con icts of Interest
The authors declare no con ict of interest.