Nutritional assessment among adult patients with suspected or confirmed active tuberculosis disease in rural India

Objectives Our study goal was to evaluate a set of nutritional indicators among adults with confirmed or suspected active tuberculosis disease in southern India, given the limited literature on this topic. Study objectives were to assess the: I) double burden of malnutrition at individual and population levels; II) relative performance of anthropometric indicators (body mass index, waist circumference) in diabetes screening; and III) associations between vitamin D and metabolic abnormalities. Design Cross-sectional study. Setting Hospital in rural southern India. Participants Among adult patients (n = 834), we measured anthropometry, body composition, and biomarkers (vitamin D, glycated hemoglobin, hemoglobin) of nutritional status. Subsets of participants provided blood and sputum samples. Results Among participants, 91.7% had ≥ 1 malnutrition indicator; 34.6% had both undernutrition and overnutrition indicators. Despite the fact that >80% of participants would be considered low-risk in diabetes screening based on low body mass index and waist circumference, approximately one-third had elevated glycated hemoglobin (≥ 5.7%). The lowest quintile of serum 25-hydroxyvitamin D was associated with an increased risk of glycated hemoglobin ≥ 5.7% (adjusted risk ratio 1.61 [95% CI 1.02, 2.56]) compared to the other quintiles, adjusting for age and trunk fat. Conclusions Malnutrition and diabetes were prevalent in this patient population; since both can predict poor prognosis of active tuberculosis disease, including treatment outcomes and drug resistance, this emphasizes the importance of dual screening and management of under- and overnutrition-related indicators among patients with suspected or active tuberculosis disease. Further studies are needed to determine clinical implications of vitamin D as a potential modifiable risk factor in metabolic abnormalities, and whether population-specific body mass index and waist circumference cut-offs improve diabetes screening.


Introduction
Mycobacterium tuberculosis (M. tb) causes the greatest number of deaths worldwide, compared to any other single infectious agent [1]. According to the World Health Organization (WHO), there were 1.7 billion people with latent tuberculosis (TB) infection, 10.0 million incident cases of active TB disease, and 1.2 million TB-related deaths in 2018 [1]. In India, there were nearly 2.7 million incident cases of active TB disease and 449,000 TB-related deaths in 2018 [1]. The burden of disease from active TB disease disproportionately affects low-and middle-income countries [2,3]. Over 95% of reported TB cases were in 119 low-and middle-income countries [2]." Active TB disease is associated with malnutrition risk factors, including undernutrition and overnutrition-related clinical sequelae (e.g. diabetes) [2][3][4][5][6][7]. Undernutrition and diabetes, respectively, are prevalent among patients with active TB disease, and increase the risk of progressing to active TB disease as well as worse TB treatment outcomes, including greater risk relapse and drug resistance [5][6][7]. Putative mechanisms at this complex nexus of undernutrition, diabetes, and active TB disease include that: inadequate nutrients adversely affect cellmediated immunity, which is necessary in the human host response against M. tb; [8] active TB disease may alter metabolic processes, including increasing energy requirements and loss of appetite [7].
Nutritional screening, assessment, and management are recommended for all patients with TB during diagnosis, treatment, and management of active TB disease [7,9]. However, there have been a limited number of comprehensive evaluations of nutritional status among patients with active TB disease in many resource-limited settings. Our study goal was to assess a panel of nutritional indicators among a patient population with confirmed or suspected active TB disease. Our three study objectives were based on prior literature. First, we assessed the prevalence of double burden of malnutrition (under-and over-nutrition indicators) at both the individual and population levels. Second, we evaluated the relative performance of anthropometric indicators (body mass index [BMI], waist circumference [WC]) in diabetes screening, compared to HbA1c. Last, we assessed the associations between vitamin D and metabolic abnormalities.

PLOS ONE
Nutritional assessment among adult patients with suspected and confirmed active tuberculosis disease PLOS ONE | https://doi.org/10.1371/journal.pone.0233306 May 22, 2020 2 / 18 and Digestive and Kidney Diseases or the National Institutes of Health. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: E.A.Y., J.L.F., P.M.B., W.B., and D.G.R. have no conflicts of interest. S.M. is an unpaid board member of and holds equity stake in a diagnostic start-up focused on developing assays for low-cost and point-of-care measurement of certain nutrients from a drop of blood using results from his research as a faculty member at Cornell University. M.J.G. receives funding to his institution from Gilead Sciences for an investigatorinitiated study unrelated to the current work. There are no further patents, products in development or marketed products to declare. This does not alter our adherence to PLOS ONE policies on sharing data and materials.

Ethical conduct of research
The study protocol was designed in accordance with the Declaration of Helsinki principles, and approved by the Institutional Review Board at Cornell University and Institutional Ethics Committee at Arogyavaram Medical Centre. All study participants provided informed written consent to participate prior to data collection. Study participants had minimal risks or harms, largely associated with additional biological sample collection, aside from that associated with their routine clinical care.

Study population
This cross-sectional study included a convenient consecutive sample of patients at a hospital (Arogyavaram Medical Centre) in Andhra Pradesh, India. The clinic is part of a hospital with inpatient and outpatient facilities, which regularly provides services for patients with suspected or confirmed active TB disease. Hospital physicians referred their patients with suspected active TB disease to study staff. Study participants (n = 834) were recruited and sequentially enrolled during their hospital visits between September 2014 and May 2016. The total sample size was based on all participants who were enrolled during the window of data collection dates.

Data collection
Trained research assistants administered structured interviews to collect sociodemographic and clinical data in the local language, Telugu. Sociodemographic covariates included: age, sex, educational level, monthly household income, and cigarette use. A study physician conducted a complete examination, including blood pressure measurements. Data were collected based on the study protocol, including data management plans (e.g. quality control via skip patterns in electronic data collection forms, data confidentiality). Subsets of study participants provided sputum and blood samples, based on their hospital visit and the recommendations of their physicians. Trained phlebotomists collected blood samples using standard clinical protocols. From this convenient sample, available blood samples were assayed for serum 25-hydroxyvitamin D (25[OH]D; n = 156) and glycated hemoglobin (HbA1c; n = 236). For active TB disease assessment (n = 363), patients provided a sputum sample at the time of his or her initial hospital visit, and a second sputum sample on the following morning.

Laboratory analyses
Each sputum sample was assessed for active TB disease by the detection of standard acid-fast bacilli (AFB) with Ziehl-Neelsen staining and conventional light microscopy. Blood samples were assayed for HbA1c (%) by high-performance liquid chromatography (D-10; Bio-Rad Laboratories, Hercules, California, United States [US]). Serum 25-hydroxyvitamin D (25[OH] D; nmol/L) was assayed by chemiluminescence immunoassay (LIAISON; DiaSorin Inc., Stillwater, Minnesota, US). We also participated in the D External Quality Assurance Scheme (www.deqas.org) program; compared to the National Institute of Standards and Technologies target values, our median percentage difference was -8.8 . Complete blood counts were assessed by an automated hematology analyzer (BC-2800; Mindray Bio-Medical Electronics Co., Ltd., Shenzhen, People's Republic of China).

Anthropometry
Anthropometric measurements were recorded, based on World Health Organization (WHO) recommendations and other commonly utilized standard methods [10,11]. Height, WC, midupper arm circumference (MUAC), and skinfold thickness were measured to the nearest 0.1 centimeter; weight was assessed to the nearest 0.1 kilogram. Total body and trunk fat (%) were assessed by bioelectrical impedance analysis (BC-418 MA; Tanita Corporation, Tokyo, Japan; 8 electrode).

Definitions
Anthropometric measurements (BMI, WC) were considered as continuous and categorized variables. BMI was categorized per standard WHO cut-offs (underweight < 18.5 kg/m 2 ; normal weight � 18.5 and < 25.0 kg/m 2 ; overweight � 25.0 and < 30.0 kg/m 2 ; and obese � 30.0 kg/m 2 ) [12]. Additionally, we considered BMI cut-offs for Asian populations (� 18.5 to < 23.0 kg/m 2 ; � 23.0 to < 27.5 kg/m 2 ; � 27.5 kg/m 2 ), based on a WHO expert consultation [13]. WC was considered as a continuous variable and tertiles based on distribution in this population. Elevated WC was also defined based on the International Diabetes Federation (IDF) cut-offs (men �90 cm, women �80 cm) recommended for individuals in South Asia [14]. We calculated limb fat as the sum of body fat (kg) in all four limbs.
Hemoglobin (g/L) was considered as a continuous and categorical variables (including quintiles). Biologically implausible values were considered missing. Anemia and severe anemia were defined by hemoglobin cut-offs, per WHO recommendations (S1 Table) [15]. Hemoglobin was adjusted for smoking by subtracting 0.3 g/L among any individuals who self-reported as currently smoking [15]. We also considered other factors that affect anemia (e.g. pregnancy, residential elevation above sea level, smoking) [15]. However, no study participants selfreported pregnancy; and all study participants resided near the hospital study site, which had an elevation that did not require altitude adjustment [15].
HbA1c was categorized by common cut-off values (� 6.5%, � 5.7% to < 6.5%, and <5.7%) that are recommended as part of the clinical diagnostic criteria of diabetes and pre-diabetes [19,20]. Active TB disease was defined as at least one positive AFB sputum smear result. This included patients with one positive AFB result (regardless of first or second sputum sample), as well as two positive AFB results, according to the standard active TB disease diagnostic guidelines in India Educational level was categorized based on self-reported completion of formal coursework (i.e., no formal education or illiterate; primary [grades [1][2][3][4][5], secondary [grades 6-12, including higher secondary], and any higher education [college, graduate, post-graduate]). Self-reported monthly household income was dichotomized as < 5000 Indian rupees (INR) or � 5000 INR) [25]. This cut-off was rounded from the cut-off of 4860 INR, based on the estimated monthly consumption expenditure for a family of five residing in a rural area (2011-2012 prices) and the national poverty line of India (Government of India, Planning Commission, 2014 Report) [25]. Cigarette use was categorized as current, previous, or never.

Statistical analyses
Continuous variables were assessed for normality using the Shapiro-Wilk test statistic. For descriptive statistics, continuous variables were reported as medians (interquartile ranges [IQRs]); categorical variables were reported as percentages. Subgroup comparisons were based on tests for continuous (i.e., Kruskal-Wallis) and categorical variables (i.e., likelihood ratio test).
We assessed associations between anthropometric indicators and HbA1c with univariate and multivariate linear and binomial regressions. Associations between anthropometric screening indicators (BMI, WC) and HbA1c � 6.5% were assessed by multivariate log-binomial regressions; key covariates (age, sex) were accounted for in these models. The predictive performance of anthropometric screening indicators (i.e., BMI, WC; categorical variables) for elevated HbA1c (� 6.5%) were assessed by sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and receiver operating characteristic (ROC) curve analysis [26]. Observed area under the ROC curves (AUCs) were compared to AUC for the null hypothesis (0.5), based on the contrast matrices of differences between the areas under the ROC curves (ROCCONTRAST in SAS statistical software) [27]. We also examined the predictive performance of anthropometric screening indicators, considering a cut-off of HbA1c � 5.7% as the outcome of interest.
For the associations between vitamin D and metabolic abnormality outcomes, the selection of potential confounders was based on approaches suggested by Rothman and Greenland [28]. In brief, we identified known or suspected risk factors based on a priori literature review [29,30]. For each association of interest, we included the confounders in the final adjusted model based on a 10% change-in-estimate criterion [28]. For binary outcomes, binomial regressions were utilized when models converged. Modified Poisson estimates were utilized if binomial models failed to converge [31]. The analytical subsets were as follows (for each respective outcome): WC (n = 150), HbA1c (n = 149), and blood pressure (n = 99); missing-data indicators were utilized for covariates with missingness [28].
Statistical analyses were conducted with SAS statistical software (version 9.4; SAS Institute Incorporated, Cary, North Carolina). All comparisons were two-sided; and considered statistically significant with an α value of 0.05.

Patient involvement
Prior to data collection, patients provided input regarding the structured interview questions; some of their feedback was integrated in the final data collection forms. These patients were individuals who visited and resided near a hospital in rural southern India, and the source population of our study.

Sociodemographic and clinical characteristics
The median age of study participants was 48 years (IQR 35, 60; Table 1), including individuals ranging from 18 to 87 years of age. Nearly three quarters of study participants were male (n = 610; Table 1). Over half of study participants (54.0%) had a self-reported monthly household income less than 5000 INR (Table 1). Among study participants with AFB assessment (n = 363), 24.2% had active TB disease (Table 1).

Anthropometry
Per the WHO categories for BMI, 46.8% of men and 40.2% of women were considered underweight (BMI < 18.5 kg/m 2 ; Table 2). Only 9.0% of men and 19.6% of women were considered overweight or obese (BMI � 25.0; Table 2). Based on the alternative WHO BMI cut-offs recommended for Asian populations, [13] 83.1% of men and 69.2% of women were underweight

Diabetes screening performance of anthropometric indicators
Sensitivity, specificity, PPV and NPV of anthropometric indicators (BMI [standard and alternative WHO categorizations], and WC [IDF cut-offs]) of HbA1c � 6.5% are in Table 5.  Table 5).
We also considered the predictive performance of anthropometric indicators for elevated HbA1c, using a cut-off of � 5.7% as the outcome of interest (S3 Table). With BMI (standard WHO categories) as the indicator, the AUCs were: overall 0. 49     25(OH)D < 50 nmol/L was associated with HbA1c (%; p = 0.04), adjusting for age and fat free mass ( Table 6). The lowest quintile of serum 25(OH)D was associated with an increased risk of HbA1c � 5.7% (aRR 1.61 [95% CI 1.02, 2.56]) compared to the other quintiles, adjusting for age and trunk fat (Table 6). However, the other associations that were assessed between 25(OH)D (continuous, <50 nmol/L) were not associated with HbA1c (continuous, � 6.5% or � 5.7%; p>0.05; Table 6). 25(OH)D was not associated with elevated WC (above the IDF cut-off values) in multivariate linear and binomial (or Poisson) regression models (p>0.05; S4 Table). The considered associations between 25(OH)D (continuous, <50 nmol/L) were not associated with blood pressure, including systolic (continuous, elevated), diastolic (continuous, elevated), and abnormally high blood pressure (p>0.05; S5 Table).

Discussion
Our results showed a high prevalence of malnutrition among a patient population with recently confirmed or suspected active TB disease. Over 90% of participants had at least one malnutrition indicator; one in three participants had both undernutrition and overnutrition indicators. Despite the fact that more than 80% of participants would be considered low-risk in diabetes screening, based on low BMI and WC, approximately one-third had elevated HbA1c (� 5.7%). Common cut-off values for anthropometry (overnutrition indicators) had suboptimal predictive performance in detecting elevated HbA1c among an adult outpatient population with a high prevalence of low BMI. Findings suggest the need for population-specific cut-offs for BMI and WC, given that each remained respectively associated with HbA1c. Lastly, vitamin D status and HbA1c were inversely associated.

Undernutrition and elevated HbA1c among patients with suspected or confirmed active tuberculosis disease
The prevalence of underweight (45.0%; BMI <18.5 kg/m 2 ) in our study population was substantially higher than prior South Asia (<25.0%) and global (<12.5%) estimates for adults [32], though other estimates among patients with active TB disease in India have ranged Our results confirmed a high prevalence of malnutrition among patients with confirmed or suspected active TB disease; over nine of every ten study participants had at least one indicator of malnutrition. Moreover, approximately one-third of study participants had indicators of both undernutrition and overnutrition, suggesting these patients are affected by the double burden of malnutrition at the individual-level.

Relative performance of anthropometric indicators in diabetes screening
Higher BMI (overweight and obesity) and WC are well-established modifiable risk factors of type 2 diabetes mellitus (T2DM) [14,20,[39][40][41][42]. The respective associations between BMI and WC with elevated HbA1c have been confirmed in several studies among populations in North America [41,43] and Asia (India [44,45], China [46]). One US study among Mexican Americans found an 11 times increased risk of non-insulin dependent diabetes mellitus among those with WC in the highest quartile, compared to the lowest quartile [43]. Despite the high prevalence of low BMI in our study population, BMI and WC similarly were associated with elevated HbA1c.
International and national public health entities, including the WHO [39] and IDF [14], recommend common cut-off values of elevated BMI and WC that identify individuals at risk for T2DM. However, the observation of higher diabetes prevalence among populations with lower mean BMI has instigated the question of whether population-specific cut-off values of anthropometric indicators would be more appropriate for diabetes screening in some populations [13,[47][48][49][50], such as India.
Previous studies have shown a wide heterogeneity of predictive performance of anthropometric indicators in diabetes and pre-diabetes screening. For example, in one US study among 12,814 adults (African American, white), areas under the ROC curves were similar for BMI (African American men 0.69, white men 0.70; African American women 0.66, white women 0.72) and WC (African American men 0.70, white men 0.70; African American women 0.69, White women 0.73) in predicting diabetes [51].
A growing body of evidence suggests research gaps and limitations in the predictive performance of commonly used anthropometric indicator cut-offs for T2DM screening. First, the heterogeneity of body fat distribution is hypothesized to affect T2DM risk [52], which could cause common cut-points of anthropometric screening indicators for diabetes to perform worse (more false negatives or positives) in some populations. Central obesity as well as visceral fat have been more strongly associated with insulin resistance and T2DM, relative to overall obesity [53][54][55]. Studies have demonstrated that individuals with similar BMI sometimes differ substantially in body fat distribution and percentage [50,56,57], metabolic syndrome [58], T2DM [59,60]. As an example, the predisposition for central fat accumulation among Asian populations has been observed to differ from Caucasians [61], which could explain differential T2DM risk among individuals with the same BMI [13].
Second, studies have begun elucidating the biological basis for these observed patterns. At the cellular level, functional metabolic differences between adipocytes (brown, white, and beige [brown in white]) have been characterized [62][63][64]. Critically, studies have shown metabolically active brown adipocytes associated with improved T2DM indicators and lower BMI [63,65]; in contrast, white adipocytes were associated with visceral fat, which has been linked to insulin resistance [66].
Successfully addressing the diabetes epidemic requires considering effective screening among populations with different body composition patterns. Overall in our study, common cut-off values for anthropometric screening indicators had suboptimal predictive performance in detecting elevated HbA1c among an adult outpatient population with lower adiposity. Findings suggest the need for population-specific cut-offs for BMI and WC, given that: 1) each remained respectively associated with elevated HbA1c; and 2) standard cut-offs misclassify the HbA1c status of many study participants.
Although our results reveal several research gaps, the importance and challenges of determining appropriate population-specific cut-off values of anthropometric indicators have been acknowledged in previous literature [47][48][49][50] and by a WHO expert consultation [13]. Future research questions include: What are the appropriate cut-offs across racial and ethnic subgroups, based on representative samples with external validity? How do different fat distributions (including differing body fat percentage and adipocyte type) affect the risk of T2DM incidence and severity? What are the cellular mechanisms involving different adipocyte types that contribute to T2DM development and progression?

Vitamin D as a modifiable risk factor of metabolic indicators
Previous literature has found an inverse association between vitamin D status (25[OH]D concentration) and HbA1c [67,68], which was consistent with our result. In a nationally representative study among adults in the US, the prevalence of high HbA1c (>6.0%) linearly decreased across vitamin D quintiles (p<0.01) [68]. Separately, systematic reviews have shown vitamin D supplementation was associated with HbA1c in some studies [69,70]. Although other studies observed null results [71][72][73], many differed widely in methodology, including vitamin D dosage (duration, frequency, dosage).
Our finding that serum 25(OH)D was inversely associated with WC, which has been corroborated by other studies [74][75][76]. At the cellular level, other key findings that support epidemiological findings include the: a) isolation of vitamin D receptor (VDR) as well as hydroxylating enzymes of vitamin D in adipose tissues; b) storage and release of vitamin D in adipocytes [77][78][79][80][81][82]. Many questions remain, in order to elucidate the etiology and mechanisms of 25(OH)D in the context of adiposity and energy homeostasis [83], including the: extent of differences in vitamin D metabolism (e.g. VDR signaling, vitamin D activation:inactivation ratio, interactions with lipid-mediated regulatory processes such as via peroxisome proliferator-activated receptor gamma) across adipocyte types and heterogeneous body composition.
While our study demonstrated a null association between low 25(OH)D concentration and high systolic blood pressure, prior literature has supported an inverse association between 25 (OH)D and the renin-angiotensin-aldosterone-system (RAAS), which regulates hypertension [84][85][86]. One hypothesized mechanism is that elevated vitamin D inhibits renin and angiotensin expression, which dampens the RAAS activity and subsequently decreases blood pressure [85][86][87][88]. Additionally, an overview of systematic reviews of vitamin D supplementation randomized controlled trials found that among nine meta-analyses, two showed protective effects of vitamin D supplements on blood pressure and six had null findings [89].

Strengths and limitations
In our study, there were several strengths, including the: sample size, assessment of multiple BMI and WC categories (based on widely used cut-off values and population distribution [quantiles]), evaluation of microcytosis and hypochromia.
This study had several limitations, including the: cross-sectional study design (with a single timepoint assessment); potential residual confounding; external validity (generalizability of findings, especially among healthy populations); assessment of additional causes of low hemoglobin; biological samples obtained per standard of care (only from participants with a clinical indication and not collected at random or from all participants); and limited biomarker data (such as diagnoses of human immunodeficiency virus [HIV] and no peripheral smears to determine iron deficiency anemia). Active TB disease and diabetes have bi-directional impacts that we were not able to assess, particularly as we only measured HbA1c and not diabetes mellitus [90]. Iron deficiency with and without anemia has been associated with increased HbA1c [91], and this potential interaction needs be evaluated in this study population.

Conclusions
In summary, our findings confirmed that malnutrition and elevated HbA1c were prevalent among this patient population with suspected and confirmed active TB disease in rural India. Dual screening and management of under-and overnutrition-related indicators are needed among patient populations with confirmed and suspected active TB disease, in order to facilitate improved TB control efforts. Further studies are needed to determine any clinical implications of the potential role of vitamin D as a modifiable risk factor in metabolic abnormalities, as well as whether population-specific BMI and WC cut-offs are needed among specific populations (e.g. metabolically unhealthy normal or underweight).
Supporting information S1