The monocyte-to-lymphocyte ratio: Sex-specific differences in the tuberculosis disease spectrum, diagnostic indices and defining normal ranges

Background The monocyte-to-lymphocyte ratio (MLR) has been advocated as a biomarker in tuberculosis. Our objective was to evaluate its clinical value and associations. Methods Blood counts, inflammatory markers and clinical parameters were measured in patients with and those screened for tuberculosis. Complete blood counts (CBCs) from a multi-ethnic population aged 16 to 65 years were evaluated; a sub-group with normal hematological indices was used to define the range of MLRs. Results Multivariate analysis in proven tuberculosis (n = 264) indicated MLR associated with low serum albumin, high white cell counts and a positive culture; values were higher in sputum smear-positive pulmonary tuberculosis (S+PTB). Analysis in S+PTB (n = 296) showed higher MLRs in males and those with high neutrophil counts, low serum albumin and high C-reactive protein. The diagnostic value of MLRs was assessed by comparing notified patients with TB (n = 264) with denotified cases (n = 50), active case-finding in non-contacts (TB n = 111 and LTBI n = 373) and contacts of S+PTB (n = 149) with S+PTB found at screening (n = 75). Sensitivities and specificities ranged from 58.0–62.5% and 50.0–70.0% respectively for optimal cut-off values, defined by ROC curves. In CBCs obtained over one month, ratios correlated with neutrophil counts (ρ = 0.48, P<0.00001, n = 14,573; MLR = 0.45 at 8–8.9 x 109/L) and were higher in males than females (P<0.0001). The MLR range (mean ± 2SD) in those with normal hematological indices (n = 3921: females 0.122–0.474; males 0.136–0.505) paralleled LTBI MLRs. Ratios did not predict death (n = 29) nor response to treatment (n = 178 S+PTB with follow-up CBCs). Ratios were higher in males than female in the 16–45 years age group, where immune differences due to sex hormones are likely greatest. Conclusions Severe tuberculosis and male sex associated with high MLRs; the same variables likely affect the performance of other biomarkers. The ratio performed poorly as a clinical aid.


Introduction
Tuberculosis (TB) was declared a global health emergency by the World Health Organization in April 1993 and remains a leading cause of death and disease [1]. The microbiological diagnosis of TB has advanced considerably through molecular testing [2]. The use of two proteins found in Mycobacterium tuberculosis but not in Bacille Calmette-Guérin (BCG) in interferongamma release assays (IGRAs) has improved the specificity of diagnosing latent tuberculosis infection (LTBI) [3].
Many non-specific biomarkers have been proposed to identify those with a positive IGRA who will develop active TB [4], those with sub-clinical disease (i.e. those without symptoms often with a normal chest x-ray without raised inflammatory markers) [5], those likely to die from TB [6], those whose genotype might suggest a protective response to Mycobacterium tuberculosis [7] and transcriptomic measures of risk, diagnosis and treatment response [8].
The monocyte-to-lymphocyte ratio (MLR) has experienced a revived interest as a biomarker. Monocytes are released from the bone marrow into the bloodstream in sepsis and other infections and migrate to foci of infection where they mature into macrophages [9]. Lymphocytes are responsible for the cellular and antibody responses to infection [10]; for instance, the low number of CD4+ T lymphocytes in HIV infection are responsible for the susceptibility to infections. Originally identified in TB patients, a rabbit model suggested that both high and low MLRs might be markers for TB progression [11]. In a mouse model, high MLRs were associated with impaired protection of BCG against TB [12]. An elevated MLR was thought to correlate with different stages of TB disease [13], identify those with HIV infection most likely to develop active TB [14,15], contacts likely to develop active TB [16] and neonates at risk of TB [17]. In vitro studies with a mycobacterial growth inhibition assay suggested that monocytes and lymphocytes from those with a higher MLR were less able to inhibit the growth of BCG [18]. We, therefore, examined the MLR in patients with TB and LTBI in order to test the hypothesis that such a simple measurement from a complete blood count (CBC) could be of clinical value. Early results were reported in the form of abstracts [19,20].
TB patients had given informed written consent in the clinical trial Blood Tests in Tuberculosis III, approved by the East London and City Health Authority Research Ethics Committee (P/ 03/285), registered under the number NIHR 4177. Those screened for TB had given informed written consent in the UK PREDICT TB Study, whose procedures and protocol were approved by the Brent NHS Research Ethics Committee (10/H7017/14). Both gave access to their clinical data and investigations which were used in this study. For those who were screened for participation in these studies or had provided these data taken during normal clinical care but had not been entered into these trials, waivers were given according to the processes set by the UK Research Institute/Medical Research Council/NHS Health Research Authority (http://www. hra-decisiontools.org.uk/ethics/). In addition, for non-participants of the trials, the study was confirmed as an audit and related to quality improvement of clinical care by the local Research and Development Office.

Setting
A university district general hospital in an inner-city London borough with a high incidence (falling from 62 to 22 per 100,000 during the period of study) of TB in a multi-ethnic population.

Participants
Patients were eligible if they attended TB Clinics between 2005 and 2018 and had a complete blood count (CBC) before the start of any treatment (Fig 1). Children under 16 years of age  Table 1 and S1 were excluded. Active TB was defined by a) a positive culture or by b) a combination of consistent radiology, histology and other tests sufficient for the clinician to notify the patient as having TB and to start a course of treatment without a change in this opinion during treatment nor within the following three years. Contacts were defined by their index case disease type, had spent > 8 h in close proximity to the index case, had attended the TB screening, had a positive IGRA or tuberculin skin test (TST) and were referred for consideration of preventive treatment. LTBI (latent tuberculosis infection) was defined in the screened populations as having a positive IGRA (QuantiFERON > 0.35 IU/mL or positive T-SPOT.TB) or a TST � 15 mm for those with a BCG scar and � 10 mm for those without a BCG scar, and without evidence of active TB. All participants constituted consecutive series (Fig 1).
Control groups included those originally notified as having TB who subsequently proved to have another diagnosis (denotified TB), those with LTBI and a anonymized CBCs analyzed during one month in 2018, in whom those with TB or over 65 years on the grounds of likely co-morbidities were excluded. Only data where the hemoglobin, total red cell count, packed cell volume, mean cell volume, mean cell hemoglobin, platelet count, total white cell count, neutrophils, eosinophils, and basophils were all within normal ranges were used (S2 Table in S2 File).

Variables
TB notification data (age, sex (self-identified), ethnicity (self-identified), country of birth, previous TB, BCG scar, HIV status, diabetes, problem alcohol use, site of disease (patient or index)) were recorded as required by Public Health England for TB as a notifiable disease. Chest radiograph zones out of six and cavitation, sputum smear, days to culture, drug susceptibility test (DST) results, sputum culture at 2 months and end of treatment, and outcome were noted. Deaths were classified as being a) due to TB, i.e. TB was the main cause of death when any other illness at the time was either not present or merely a co-morbidity, such as diabetes; b) TB contributed, i.e. where a clear main cause of death (e.g. pulmonary embolus) was not due to TB, but where TB may have contributed (e.g. rifampicin treatment having induced a hypercoagulable state); or c) clearly unrelated to TB (e.g. a road traffic accident). At screening, reason for attendance, TST and IGRA results were recorded. Blood samples were taken before the start of any treatment, at 2 months and the end of treatment.

Statistical analysis
Data were analyzed with GraphPad Prism version 7 and Microsoft Excel. Univariate analyses were performed with Student's t-test for continuous predictors and linear regression with categorical predictors. A cut-off of P < 0.1 was employed to define those predictors to be entered into multiple regression analysis (https://stats.blue/Stats_Suite/multiple_linear_regression_ calculator.html). Initial examination included all such predictors, which were then removed individually and added back until the best fit model was determined where all coefficients (excluding those for the constant) were significantly greater than zero.
Receiver operating characteristic (ROC) analysis addressed the optimal cut-off MLR and area under the curve (AUC) values (John Hopkins web-based calculator for ROC curves, http://www.rad.jhmi.edu/jeng/javarad/roc/JROCFITi.html [21]). Cut-off MLRs above the 95 th and below the 5 th centile of the LTBI and the selected anonymized control groups defined diagnostic indices. Chi-squared analysis assessed the significance of MLRs related to these cutoffs. A power calculation showed that to detect a 10% difference between the sensitivities of the MLR with an 80% power at the 5% level required 199 patients in each group if the higher sensitivity were 20%, or 293 if 30%, but >3000 would be required if the differences between males and females were 5% and the sensitivities lay between 30 and 70% [22].

TB patients
356 patients were notified as having TB over the study period. Twenty were children and 13 had no pre-treatment blood test; 50 were subsequently denotified (Fig 1). The firm alternative diagnoses included cancer (5), pneumonia (2), abscesses (3), Crohn's disease, glomerulonephritis, lupus and autoimmune encephalitis, non-tuberculous mycobacteria, self-healed tuberculosis and LTBI. Demographic details of the remaining 264 patients are given in Table 1.
MLRs showed a right-sided skew which was improved by log transformation (Fig 2). Those with culture-positive tuberculosis showed more values in the higher ranges than either those who had other forms of TB or had LTBI.
Clinical variables were examined for any significant association with the MLR ( Table 2). Interactions were examined and significant associations were found for smear and culture (coefficient 0.36 ± 0.12, p = 0.008). The best fit model was with albumin, white cell count and culture ( Table 2 and S1 File). Addition of log-transformed CRP showed a significant interaction with albumin and review of the CRP data indicated many had a CRP of <5 mg/L, which could not be log. transformed.
In view of the significant difference in proportion of males and females with TB, the sexes were also analyzed separately. An improved model was found for males using predictors of albumin, culture, HIV co-infection and albumin-to-globulin (AG) ratio, the latter two predictors replacing white cell count (adjusted R-squared 0.395 compared to 0.328; total sum of squares 11.533 compared to 11.974; F = 26.478 compared to 28.003; S1 File). The negative coefficient for HIV in males was associated with low monocyte counts (range 0.1 to 0.5). Surprisingly, the AG ratio did not show a significant interaction with albumin nor with globulin and its coefficient of association with log MLR was positive. No additional improvement above albumin, culture and white cell count could be found for females, although for this group there was a significant interaction between culture and albumin values in females (P = 0.026; S1 File).

MLRs in smear-positive pulmonary tuberculosis (S+PTB)
Patients with S+PTB had shown the highest MLRs (Table 2). We therefore wished to explore whether the preliminary findings in the tuberculosis cohort could be refined by further examining this group (Table 3). In particular, we wanted to know whether the association with the white cell count pertained mainly to the neutrophil count and whether the time to culture, an indicator of bacterial load, was a significant predictor for the MLR. This group would also permit further exploration of other variables that might be associated with higher MLRs, considering the lesser overlap of values from those with LTBI compared to other patients within the TB spectrum. Demographic details of the make-up of this group are given in Table 1 and S1 Table in S2 File. We also expected that this group might shed more light as to why higher MLRs are found in males than females (Fig 3).
Serum albumin remained an important negative predictor of the MLR. The association with neutrophil count was greater than with the white cell count. However, the time to culture was not a predictor. The best fit model for log MLRs included the predictors of sex, albumin, log neutrophil count and log CRP, although there was a significant interaction between albumin and log CRP (P = 0.001; S1 File).

PLOS ONE
distribution of values is more apparent in the female S+PTB population. Multiple linear regression analysis for females indicated a best fit model for predictors of log. neutrophil count, albumin-to-globulin ratio, smear and cavities, whereas for males only log. CRP and albumin were valuable and interacted significantly with each other (P = 0.026; S1 File).

Diagnostic indices
Screened persons with LTBI were identified from records of those offered preventive treatment (Fig 1). Contacts of patients with S+PTB (n = 149, 146 without TB) were considered separately (Table 1) as being most likely to develop active disease as a result of recent infection. For more distant infection, only screened individuals with a positive IGRA were included (n = 373, 372 without TB). The MLR for those with LTBI, whether with recent or distant exposure, showed an almost normal distribution with data skewed towards higher ML values ( Fig  2A). Log-transformation of the data showed a normal distribution. Assessing receiver-operating characteristics (ROC), the areas under the curve (AUC) varied according to the comparison made (Fig 4). The shortest distance of the curve to the top left-hand corner was used to define the optimal diagnostic indices ( Table 5).

Could higher cut-off values of the MLR be useful?
One of the problems in evaluating the literature on the MLRs is to know whether the values that are given might be expected in a "normal" population or do indeed reflect the pathology of TB. Fig 2 indicates

PLOS ONE
corresponded to those with a neutrophil count of 8-8.9 x 10 9 /L, the laboratory upper limit of normal being 8 x 10 9 /L (S1 Fig). The MLR showed a distribution with a right-sided skew, which was partly normalized by log transformation. A box plot indicated that there still remained a skew toward higher values. Plotting histograms for monocytes and lymphocytes indicated that the former showed an almost normal distribution, but higher monocyte counts were responsible for most of the skewing. A QQ plot confirmed this impression (S2A Fig). Indeed, high monocyte values have been associated with a number of clinical conditions [23,24]. On the assumption that other hematological indices might be affected in disease, all samples with any value outside the normal range except for monocytes and lymphocytes were first removed, leaving 4,254 CBCs. If samples were selected with normal hematological indices apart from the monocyte and lymphocyte counts, then upper limit would have been 0.515 (95% CI 0.511 to 0.518) for females and 0.595 (95% CI 0.584 to 0.607) for males and few TB Excluding CBCs to those with a lymphocyte count > 4 x 10 9 /L and monocytes to > 1 x 10 9 /L still showed skewing on a QQ plot and a significant difference from the LTBI population. A monocyte count > 0.8 x 10 9 /L has been associated with significant clinical disease in an emergency room [24]. Using a lymphocyte count > 4 x 10 9 /L and monocytes to > 0.8 x 10 9 /L defined a population close to a normal distribution as well as with the closest approximation to the LTBI population and this was therefore used to define the cut-off limits (S2D, S2G and S3 Figs). CBCs from the TB clinics were removed from this last population. There remained a significant difference in the log . MLRs between the sexes (mean (SD): female -0.619 (0.148); male, -0.582 (0.143), P < 0.0001). Defining cut-off values as log. mean ± 2SD, for females the upper   In patients with all forms of TB within the 3-year cohort, six patients had an MLR below the lower cut-off limits, two with culture-positive TB; two had diabetes, one HIV co-infection (culture-positive), one an alcohol problem, one with malnutrition (culture-positive) and one with no co-morbidities. Using the upper MLR cut-off values, the associations observed in the univariate analysis (Table 2) were confirmed. More patients with culture-positive TB from any  Table 5 for values of the area under the curve (AUC) and optimal diagnostic indices.
https://doi.org/10.1371/journal.pone.0247745.g004  3). Three patients with miliary TB had MLRs of 0.10 (lymphocytes 5.1 x 10 9 /L), 0.36 and 1.00 (bone marrow depression with lymphocytes and monocytes both with an absolute value of 0.2 x 10 9 /L). Three patients with problem alcohol use and disseminated TB had high MLRs due to lymphopenia (all three had a lymphocyte count of 0.6 x 10 9 /L).
During the same month as the anonymized samples, 8 patients with notified tuberculosis had 11 blood tests (Fig 1). In the various TB clinics, 93 individuals were seen, several on more than one occasion: none with LTBI or cured TB had an MLR above the upper limit for their sex, although one patient with sarcoidosis had an MLR below the lower limit. Those with a positive TST or IGRA did not have higher MLRs than those who were negative.

Prognostic value of the MLR in those with LTBI
In contacts of index cases with S+PTB, there was one patient with an MLR of 1.0 who on review was considered to have active TB, based on atelectasis in left lower zone, TST conversion from 6 to 15 mm induration and a CRP of 28 mg/L and whose chest radiograph cleared following treatment for TB (Table 2). Two other patients, who developed culture-positive TB, had MLRs of 0.26 and 0.22 at screening.
There were 11 contacts of S+PTB who showed a completely negative response to tuberculin at their first visit with a positive TST at 6-8 weeks follow-up. Eight of nine with an IGRA were positive (one rising from 0 to 0.5 IU/mL). All these with evidence of recent infection with TB had an MLR within the normal ranges and none developed TB (all received preventive treatment).
In the IGRA+/TST+ screened population without known contact with TB, 8 had a high MLR (S4 Table in S2 File). One had initially been notified as a case of TB being a contact of an uncle, who had active but not pulmonary TB; this person had several risk factors for TB, minor gastro-intestinal symptoms and was given a trial of standard treatment which was stopped at 2 months (2 months of rifampicin and pyrazinamide being a proven preventive treatment for LTBI). The other 7 did not develop TB, having received preventive treatment.

Deaths and MLR
There were 29 deaths (1.9%) in notified TB cases from 2005-2017. Three were due to TB, three where TB contributed to death and the remainder were unrelated to TB. MLRs above the sex-specific limits were found in 18 (62%) but were not significantly higher or more numerous in those where TB played a role (5 males and 1 female) compared to those where there was another cause of death. MLRs measured within a day or two of death did not improve on the predictive value above samples taken at the time of TB diagnosis.

Effect of treatment on MLRs
Two-week follow-up MLRs were available for 178 with S+PTB. Five showed no change, 57 were higher and 115 were lower; MLRs were more likely in males than females to fall to the normal range (χ 2 = 6.5, P = 0.011). Looking only at results obtained at the start of treatment, at 2 months and at the end of treatment, there was a significant fall in the MLR by 2 months (paired t-test: t = 7.5, P < 0.0001, df 136) and 6 months (t = 5.3, P<0.0001, df 117). Looking at individual patients, a persistently high MLR could not be used to predict who would remain culture-positive at 2 months or have a poor outcome. Four patients had extended inpatient stays and multiple blood tests taken due to their severity of disease. Three showed a reduction in MLR from high to normal values by days 2, 6 and 21 respectively, whilst another showed an MLR which remained high for 247 days. No additional effect was seen in patients with drugresistant TB.

Effect of age on MLRs
In order to determine whether the sex difference might be related to sex hormones or differences in inflammatory responses with age, MLRs were divided into age groups 16-45 years (women should be pre-menopausal), 46-65 years and >65 years. In order to ensure that the level of disease burden was comparable, only those with S+PTB were examined. The 16-45 years age group showed the greatest difference in MLR between the sexes (P = 0.0035), although the 46-65 years age group was also significantly different (P = 0.047; Fig 4).

Discussion
In TB, MLR was associated with a low serum albumin, high white cell count and a positive culture, as well as with high CRP. Males also showed an association with HIV and the serum albumin-to-globulin ratio. In S+PTB, this was refined to a high neutrophil count (females only), with albumin and a high CRP (a significant interaction in males) and male sex. Females with S +PTB also showed a link between MLR and smear and lung cavities. The sensitivity and specificity of MLRs were poor for the diagnosis of TB or as a screening tool to detect active TB in contacts and IGRA+ individuals. High MLRs did not predict the development of TB in screened populations, nor treatment effectiveness, although MLRs usually fell with treatment. MLRs correlated with neutrophil counts in the one month CBC population. High MLRs were more often due to monocytes, especially in males. The sex difference in MLRs was most marked in the 16-45 years age group.

Sex-specific differences
Males had higher MLRs than females. Males have more S+PTB, whereas the sex ratio in extrapulmonary TB is equal [25,26]. Lung cavities are more common in males than females [27]. Many of the reported data on MLRs either had a predominance of males in the TB group [13,28] or did not give sex-specific data [18]. The immune response in males is less effective than in females despite greater TLR2 and TLR4 expression, more Th17 cells and CD4+ Treg cells, CD8+ T cells and NK cells [29]. Monocytosis is a feature of infection [23] and high values are found especially in males [24]. Thus, the higher MLRs in males and their association with more severe forms of TB and with lung cavities might be expected as a sex-specific immune response. Such individuals are more readily diagnosed and regularly bias studies in TB towards a male-predominant population.

Clinical value
The early use of the MLR in TB noted that "the . . . ratio cannot . . . be used as a criterion for diagnosis", the "monocyte-lymphocyte ratio . . . is not a sensitive index of activity in the tuberculous patient" (activity being measured by the extent of disease, tachycardia, fever, clinical judgment of prognosis and follow-up over two years) and that "a considerable number of patients with tuberculosis will have a normal monocyte-lymphocyte ratio" [30]. These results from the pre-antibiotic era of TB treatment have been confirmed in a modern setting by this study. The association of lymphopenia and a lower MLR with TB in a rabbit model [11,31] was rarely observed in our study, even in those with miliary disease as the human comparator of the experimental method.
A low or high MLR has been associated with a hazard ratio of 2.47, 1.5 and 1.22 for the development of TB during follow-up of adults with HIV infection before anti-retroviral therapy [14], in HIV-infected post-partum women [15] and in infants [17] in South Africa, respectively. However, the sensitivities of cut-off values were low (< 10%), such that the MLR had little predictive value for an individual person. A higher hazard ratio (4.5) was observed in contacts who went on to develop TB, but a combination of a tuberculin skin test � 14 mm (sensitivity 7.5%) with monocytes >7.5% of all white blood cells (sensitivity and specificity about 75%) was reported to have a hazard ratio of 8.78, noting that many of the diagnoses of TB were a decision to treat in young children [16]. Using the same criteria from previous studies [16,28] in our group of contacts, the sensitivities were greater but the specificities were poorer than with our cut-offs and the positive predictive values were lower (S5 and S6 Tables in S2 File).

Limitations
This audit was performed at a single site. However, this ensured that data collection was uniform and follow-up data consistently available. Few patients had HIV co-infection and those with HIV were taking antiretroviral therapy and most had normal lymphocyte counts (four were < 1 x 10 9 /L). Children were excluded, but this avoided a bias in terms of empirical treatment for TB and a population in which blood tests are fewer and for highly selected reasons. Those over 65 years of age, excluded in the control group, are likely to have disease co-morbidities, especially malignancy, ischemic heart disease and metabolic syndrome [32,33], which are associated with higher MLRs. Older persons also experience "inflammaging", having a higher inflammatory response than younger adults [10,34,35].

Generalizability
The use of a real-life group of TB patients and contacts from many ethnic backgrounds encourages extrapolation to the global HIV-uninfected TB population. Analysis of groups by smear and culture status, ensured that both real-life results and data uncontaminated by variations in the diagnosis of "clinical TB" could be obtained.

Implications
This study has shown the importance of sex as a variable in the evaluation of biomarkers. Such differences touch upon lipid and metabolic differences in cardiovascular disease [36], cancer biomarkers [37] and studies distinguishing general differences between the proteome [38] and transcriptome [39,40] between the sexes. Both estrogens and androgens affect immune responses [29,41]. Our data suggested that the difference in MLRs might be due to sex hormones rather than an X-linked cause or sex-related differences in inflammaging due to the differences seen among the age groups, most especially those aged 16-45 years where differences in sex hormones are most apparent (Fig 5). Thus, diseases with a female or male predominance, such as autoimmune diseases or cancer respectively, may need to address the relative efficacy of diagnostic markers and prognostic indicators between the sexes.
Biomarkers and the sexes are especially problematic in tuberculosis. The male predominance is found almost entirely in S+PTB [25,26] and therefore patient selection will be affected by the number with this form of disease. The relative ease of diagnosis of this form of TB again means that follow-up studies of, say, those with LTBI who go on to develop TB will mean that prognostic biomarkers will be biased towards males.
Interleukin-6 (IL-6) is important in the mobilization of monocytes from the bone marrow [9]. Transcriptomic studies examining the genetics of high MLRs suggested a link with both IL-6 and interferon type I [18]. However, a closer link has been shown with an IL6/IL6R/ CEBP gene module, also associated with monocyte expansion, high CRP and a positive sputum smear [42]. These associations are consistent with the findings of the multivariate analyses in this study.
Transcriptomic data investigating the pathogenesis of active TB disease show sex-specific responses [43,44]. Estrogens, via the alpha receptor, downgrade IL-6 expression in human monocytes [45]. The 3-gene transcriptomic signature of Sweeney, which performed better than 15 other signatures for discriminating latent from active TB [46], has genes whose expression are all affected by sex. Dual-specificity phosphatase 3 (DUSP3) deletion protects female but not male mice from endotoxin shock due to a dominance of M2 (non-classical and antiinflammatory) macrophages [47]. Guanylate-binding protein 5 (GBP5) is a marker of IFNγ-

PLOS ONE
induced classical monocytes [48], which are more frequent in males, and has been found in those who develop innate immunity after an IFNγ response [49]. Kruppel-like factor 2 (KLF-2), in association with the male-dominant cytokines IL-6 and monocyte chemoattractant protein-1, promotes inflammation in response to tumor necrosis factor (induced by mycobacterial lipoarabinomannan) [50]. If biomarker studies select a majority of males and/or those with S +PTB, their value should be less in females and those with more limited disease.

Conclusion
The MLR is neither diagnostic nor typical of TB. It shows a bias towards males with S+PTB and other clinical markers of severe disease. These findings might account for its association with S+PTB and its previous popularity as a biomarker in tuberculosis and suggests that other biomarkers may face similar biases.