Prediabetes and risk of active tuberculosis: A cohort study from Northern Taiwan

Abstract

Background:

Diabetes mellitus (DM) is a well-established risk factor for active tuberculosis (TB) infection. Despite the worldwide rapid increase in the prevalence of prediabetes, its impact on the risk of active TB remains largely unknown. This study aimed to investigate the relationship between prediabetes and risk of active TB in a large cohort study.

Methods:

A total of 119,352 participants were screened from a community-based health screening program in Northern Taiwan. Diabetes mellitus and prediabetes were defined by baseline fasting plasma glucose (FPG) and prescription of anti-diabetic drugs. Incident cases of active TB were identified from the National Tuberculosis Registry. Kaplan-Meier curves and Cox regression analysis were employed to estimate the hazard ratios for prediabetes and DM compared with normoglycemia. Spline regression was performed to investigate the dose-response relationship between FPG level and risk of TB disease. At baseline, 27,404 (22.96%) participants had prediabetes and 10,943 (9.17%) participants had DM. After an average follow-up of 7.2 years, 322 TB cases occurred. The adjusted hazard ratio of developing active TB disease was 0.73 (95% confidence interval [CI] 0.55-0.97) for prediabetic and 1.48 (95% CI 1.11-1.98) for diabetic participants compared with normoglycemic individuals. Spline regression revealed a U-shaped association between FPG level and risk of active TB disease, with the lowest risk at FPG around110 mg/dL. Sensitivity analyses were conducted to exclude factors such as potential confounders (including body mass index), misclassification of glycemic level, and selection bias, and results showed that those factors could not explain the lower risk of active TB.

Conclusions:

Prediabetes was associated with a 27% reduced risk of active TB disease compared with normoglycemia. The biological mechanism of this inverse association and its implication for global nutrition transition and TB control should be further investigated.

Introduction

Tuberculosis (TB) is among the leading causes of death globally more than 10 million incident cases annually [1]. The top 30 high TB burden countries accounted for 87% of estimated incident TB cases, and most of which were low- and middle-income countries [1]. Numerous studies and systematic reviews have confirmed that diabetes mellitus (DM) is a major risk factor for active TB disease and latent TB infection [2–5]. In view of the dramatic surge in the prevalence of DM during the past decades, an increased burden of TB that would complicate its treatment is predicted [6].

Prediabetes is a stage before DM when blood glucose levels are higher than normal but have not yet reached the diagnosis criteria of DM. It is defined by impaired fasting glucose (IFG) and/or impaired glucose tolerance (IGT) [7]. Similar to DM, prediabetes has shown a rapid rise in prevalence in recent decades. The estimated number of adults with IGT was 352.1 million in 2017 and would increase to 532 million by 2045 [8]. Whether the persistent increase in prediabetes prevalence will affect the TB burden remains uncertain. Evidence from a few case-control studies reported inconsistent results about the association between prediabetes and risk of active TB [9, 10]. Notably, the results from these case-control studies were subject to measurement error as prediabetes was ascertained after the diagnosis of active TB. TB patients may have transient hyperglycemia due to systemic inflammation; hence, these studies may suffer from reverse causality bias. To the best of our knowledge, no longitudinal cohort studies have been conducted prospectively to investigate the association between prediabetes and the risk of active TB. This study aimed to estimate the association between prediabetes and incident active TB disease using prospective data from a large community-based health screening cohort.

Methods

Study design and population

This study used the information of 124,455 participants obtained from a community-based health screening service in Northern Taiwan from 2005 to 2008. The health screening service included physical examination, blood test, urine test, and questions about health behaviors, family history, and lifestyle. Participants with TB history and a diagnosis of TB within the first 28 days of follow-up were excluded (n=847). Participants who had missing information for covariates (3895 participants) or unusual values in weight and height (361 participants) were also excluded from the analysis. After the exclusion, 119,352 participants were followed up until 31 December 2013, an occurrence of active TB, or death (Figure 1).

Figure 1.

Study flow chart. ¹Tuberculosis; ²Fasting plasma glucose

Measurement of prediabetes and DM

Baseline fasting plasma glucose (FPG ≥ 126 mg/dl) or use of anti-diabetic drugs were used to define DM [11]. The period of anti-diabetic drug prescription had to be longer than 28 days within two years before the screening. The anti-diabetic drugs include sulfonylureas, biguanides, alpha-glucosidase inhibitors, thiazolidinediones, meglitinides, and insulin. We further defined prediabetes among non-diabetics using FPG (100 mg/dL ≤FPG<126 mg/dL) following the recommendation of the American Diabetes Association [11]. Participants without DM or prediabetes were defined as normoglycemic. The baseline FPG value was obtained from the screening service, and the prescription information was obtained from the National Health Insurance (NHI) database. The Taiwan NHI database contains complete outpatient visits, hospital admissions, prescriptions, diseases, and vital status covering around 99% of the total population in Taiwan.

Measurement of active TB

Incident cases of active TB were identified from the National Tuberculosis Registry. TB is a notifiable disease in Taiwan. Presumptive TB cases must receive sputum smear examinations at least twice and sputum cultures. All confirmed TB cases are registered in the National Tuberculosis Registry and receive standardized and free treatment [12]. In this study, patients with smear-positive or culture-positive would be defined as active TB cases.

Measurement of other covariates

In this analysis, potential confounders include the following TB risk factors, namely sex, age, marriage status, education level, body mass index (BMI), smoking status, alcohol use, and steroid use. Subjects were classified into three groups according to marital status: married/cohabiting, single, widowed/divorced/separated/other, and also education level: college or above, senior high, and junior or below. Non-smokers and former smokers were merged as the reference group for smoking; the same applied to subjects with alcohol use. Steroid use was defined as the prescription of systemic steroids for over 30 days in the year before the screening. Data of potential confounders were collected from the questionnaire responses of the health screening service and the NHI database.

Statistical analysis

Kaplan-Meier curves were generated to compare the TB-free survival period among the three groups classified by prediabetes and DM status [13]. Inverse probability weighting was performed to adjust for the possible confounders among the three Kaplan-Meier curves, and the reference target was the normoglycemia group [14]. Briefly, inverse probability weighting generates a pseudo-population in which the distribution of potential confounders is balanced across the exposure groups. Each individual is weighted by the inverse of the multiple propensity scores. Age, sex, smoking, alcohol use, BMI, steroid use, marital status, and education level were employed to estimate the multiple propensity scores.

This study used Cox regression models to estimate the hazard ratios (HRs) and the 95% confidence intervals for prediabetes and DM, using normoglycemic participants as the reference group [15]. The proportional hazards assumption was checked using a statistical test based on scaled Schoenfeld residuals. In the multivariable Cox regression models, we considered the following TB risk factors as potential confounders in this analysis: sex, age, marital status, educational level, BMI, smoking status, alcohol use, and steroid use.

To further explore the association of active TB with different FPG levels, patients without DM were further divided into four groups according to FPG level (≤ 90 mg/dl, 90<FPG≤100 mg/dl, 100<FPG≤110 mg/dl, and 110<FPG≤126 mg/dl), and patients with DM were divided according to the use of anti-diabetic drugs and FPG level (using anti-diabetic drug and FPG<126 mg/dl, using anti-diabetic drug and FPG ≥126, and not using anti-diabetic drug). Cox regression was employed to investigate the hazard ratio for each group compared with the reference group (90<FPG≤100 mg/dl). Lastly, this study used cubic spline regression with three internal knots to determine the dose-response relationship between FPG and active TB risk in total and in the non-diabetic population.

Subgroup analysis was conducted according to sex (female, male), age (>65, ≤65 years), and BMI (<25, ≥25 kg/m²) by adding interaction terms to the Cox regression model. The subgroup differences were examined using the likelihood ratio test. All analyses were conducted using SAS software version 9.4 (SAS Institute) and R software 3.4.0 (R Project).

Sensitivity analysis

To fully adjust for the well-known confounders, the continuous variables of age and BMI were modeled using alternative nonlinear approaches (categorical and spline regression) in the multivariable Cox models. To account for the possible confounding effect of metformin, we conducted an analysis excluding those in the normoglycemia and prediabetes groups with metformin use for longer than 28 days during follow-up [16,17]. Lastly, to exclude the impact of major illnesses on blood glucose, an additional analysis was conducted excluding those with major illnesses (systemic autoimmune diseases, organ-specific autoimmune disease, cancer, alcoholism, liver cirrhosis, and dialysis) (Supplementary Table S1).

To account for the change in FPG levels during follow-up, we performed a time-dependent Cox regression analysis using the subset of the cohort participants with multiple screening records from 2005 to 2008. These participants were followed up until two years after the last observed screening record, the occurrence of TB, or death. FPG level was defined as a time-dependent variable and divided into DM, prediabetes, and normoglycemia (Supplementary Table S8). The change of glycemic status during follow-up is listed in Supplementary Table S2.

Furthermore, to test the validity of a single FPG level at baseline, this study conducted a separate analysis focusing on the association between prediabetes and risk of ischemic heart disease using the same cohort. The positive association between prediabetes and risk of ischemic heart disease was considered a positive control [18].

Lastly, to exclude the influence of potential TB at baseline on the observed association, the length of the exclusion period of potential TB cases were extended to 60, 90, and 180 days in addition to 28 days in the main analysis.

A priori power estimation

Type I error was set at 0.05. The baseline incidence rate of TB and HRs for prediabetes compared with normoglycemia were estimated according to surveillance information and previous studies [9, 10, 19, 20]. The power would reach above 0.8 with all estimations of hazard ratios (1.1, 1.3, 1.5, and 2) when the TB incidence rate increased to 60 per 100,000 (Supplementary Table S3).

Results

Of 119,352 participants, 27,404 (22.96%) had prediabetes and 10,943 (9.17%) had DM at baseline (Table 1). The mean age was 51.82 years and around two-thirds of the participants were female (64.36%). The proportions of males, overweight, alcohol use, low education level, and steroid use among prediabetes and DM groups were higher than those of the normoglycemia group at baseline (Table 1).

Table 1.

Basic characteristics of study participants by diabetes mellitus (DM) status

Characteristic	Normoglycemia (n=81005)	Prediabetes (n=27404)	DM (n=10943)
Sex
Male	27147 (33.51%)	10784 (39.35%)	4610 (42.13%)
Female	53858 (66.49%)	16620 (60.65%)	6333 (57.87%)
Age (yr)	49.59	55.22	59.91
Body mass index
<18.5	2838 (3.5%)	380 (1.39%)	100 (0.91%)
≥18.5 to <25.0	52007 (64.2%)	12922 (47.15%)	4296 (39.26%)
≥25.0 to <30.0	22428 (27.69%)	11344 (41.40)	4760 (44.41%)
≥30.0	3732 (4.61%)	2758 (10.06)	1687 (15.42%)
Smoking
Former/never	68865 (85.01%)	23506 (85.78%)	9162 (83.72%)
Current	12140 (14.99%)	3898 (14.22%)	1781 (16.28%)
Alcohol use
Former/never	75727 (93.48%)	25073 (91.49%)	10037 (91.72%)
Current	5278 (6.52%)	2331 (8.51%)	906 (8.28%)
Marital status
Married/Cohabiting	68507 (84.57%)	23329 (85.13%)	8968 (81.95%)
Single	4760 (5.88%)	949 (3.46%)	307 (2.81%)
Widowed/Divorced/Separated/Other	7738 (9.55%)	3126 (11.41%)	1668 (15.24%)
Educationlevel
Junior high school or below	36702 (45.31%)	16346 (59.65%)	7855 (71.78%)
High school	25249 (31.17%)	6701 (24.45%)	1933 (17.66%)
College or above	19054 (23.52%)	4357 (15.90%)	1155 (10.55%)
Steroid use
No	80044 (98.81%)	27045 (98.69%)	10719 (97.95%)
Yes	961 (1.19%)	359 (1.31%)	224 (2.05%)

After an average follow-up of 7.2 years, 322 tuberculosis cases occurred (Table 2). In the adjusted Kaplan-Meier analysis, TB-free survival differed among the three groups (log-rank test: p<0.001). The DM group developed TB fastest, followed by the normoglycemia group and then the prediabetes group (Figure 2). The proportional hazards assumption of DM status was supported by the non-significant relationship between residuals and time. The crude incidence rates of TB for normoglycemia, prediabetes, and DM groups were 32.67, 32.42, and 82.73, respectively (Table 2). Compared with that for normoglycemic subjects, the age-adjusted HR of prediabetes and DM subjects were 0.66 (95% confidence interval [CI]: 0.50-0.88) and 1.27 (95% CI: 0.95-1.69), respectively. The multivariable-adjusted HR was 0.73 (95% CI: 0.55-0.97) and 1.48 (95% CI: 1.11-1.98) for prediabetes and DM subjects, respectively (Table 2).

Table 2.

Age- and multivariable-adjusted models of association between diabetes mellitus (DM) status andrisk of active TB

DM status	Number of cases	Incidence rate of TB	Age-adjusted model	Age, BMI-adjusted model	Multivariable-adjusted model
			adjusted HR1 (95% CI2)	adjusted HR (95% CI)	adjusted HR3 (95% CI)
Normoglycemia	193	32.67	Reference	Reference	Reference
Prediabetes	65	32.42	0.661 (0.498-0.877)	0.731 (0.550-0..973)	0.725 (0.545-0.965)
DM	64	82.73	1.266 (0.949-1.689)	1.492 (1.110-2.006)	1.478 (1.105-1.978)

¹Hazards ratio

²Confidence interval

³Adjusted for age, sex, smoking, alcohol use, body mass index, steroid use, and educational level, and marital status.

Figure 2.

Adjusted Kaplan-Meier curves for tuberculosis (TB)-free survival by diabetes mellitus (DM) status.

To examine the dose-response relation between FPG level and TB risk, the study population was stratified according to FPG levels. Compared with that of the reference group (FPG: 90-100 mg/dL), the adjusted HRs of the other three non-DM groups (FPG: <90, 100-110, and 110-126 mg/dL) were 0.94 (95% CI: 0.71-1.26), 0.90 (95% CI: 0.51-1.04), and 0.85 (95% CI: 0.56-1.44), respectively (Table 3). DM patients receiving anti-diabetic treatment with FPG <126 mg/dL had an adjusted HR of 0.85 (95% CI: 0.41-1.74), while poorly controlled DM patients (FPG≥126 mg/dL despite the use of anti-diabetics) had an adjusted HR of 2.00 (95% CI: 1.38-2.90) (Table 3).

Table 3.

Cox model of association between FPG and risk of TB according to diabetes mellitus (DM) status

Fasting plasma glucose	Number of cases	Person-years	Incidence rate of TB	Age-adjusted model	Multivariable-adjusted model
				adjusted HR1 (95% CI2)	adjusted HR (95% CI)
Non-DM (90≤FPG≤100 mg/dl)	119	337314.3	35.28	Reference	Reference
Non-DM (FPG<90 mg/dl)	78	278879.4	27.97	1.029 (0.773-1.370)	0.942 (0.707-1.255)
Non-DM (100<FPG≤110 mg/dl)	41	129487.6	31.66	0.691 (0.484-0.986)	0.730 (0.511-1.043)
Non-DM (110<FPG≤126 mg/dl)	20	45599.5	43.86	0.820 (0.510-1.318)	0.895 (0.555-1.442)
DM with drug (FPG <126 mg/dl)	8	13624.5	58.72	0.749 (0.365-1.538)	0.848 (0.413-1.741)
DM with drug (FPG ≥126 mg/dl)	38	33280.0	114.18	1.724 (1.193-2.490)	2.003 (1.384-2.900)
DM without drug (FPG≥126 mg/dl)	18	30456.6	59.10	1.156 (0.704-1.900)	1.256 (0.763-2.069)

¹Hazard ratio

²Confidence interval

Further dissection of the dose-response relationship using the cubic spline regression revealed a U-shaped association between FPG and active TB risk in the total population (Supplementary Figure S1). When only the non-diabetic population was examined, the U-shaped association between FPG and TB risk remained; those with an FPG of around 110 mg/dL had the lowest risk of active TB (Figure 3). In the subgroup analysis, a sex difference (P-value: 0.022) was found in the association between prediabetes and risk of TB (Table 4). The adjusted HR was 0.46 (95% CI: 0.24-0.89) for females and 0.82 (95% CI: 0.60-1.13) for males. On the other hand, there was no evidence ofeffect modification by age group (P-value: 0.487) or BMI level (P-value: 0.701).

Figure 3.

Dose-response relationship between fasting plasma glucose (FPG) level and risk of tuberculosis (TB) in patients without diabetes mellitus (DM)

Table 4.

Subgroup analysis of association between diabetes mellitus (DM) status and risk of active tuberculosis (TB) by sex, age, and body mass index (BMI)

Covariates	FPG1 status	Number of cases	Multivariable-adjusted model	Subgroup p value
			adjusted HR2 (95% CI3)
Overall	Normoglycemia	193	Reference
	Prediabetes	65	0.725 (0.545-0.965)
	DM5	64	1.478 (1.105-1.978)
Sex				0.022
Female	Normoglycemia	58	Reference
	Prediabetes	11	0.464 (0.243-0.887)
	DM	24	2.161 (1.329-3.513)
Male	Normoglycemia	135	Reference
	Prediabetes	54	0.821 (0.597-1.129)
	DM	40	1.247 (0.872-1.782)
Age				0.487
≤65 yr	Normoglycemia	101	Reference
	Prediabetes	27	0.636 (0.422-0.959)
	DM	23	1.302 (0.839-2.018)
>65 yr	Normoglycemia	92	Reference
	Prediabetes	38	0.821 (0.565-1.193)
	DM	41	1.641 (1.141-2.358)
BMI				0.701
<25 kg/m2	Normoglycemia	153	Reference
	Prediabetes	43	0.708 (0.503-0.995)
	DM	37	1.353 (0.940-1.948)
≥25 kg/m2	Normoglycemia	40	Reference
	Prediabetes	22	0.785 (0.466-1.323)
	DM	27	1.757 (1.074-2.874)

¹Fasting plasma glucose

²Hazard ratio

³Confidence interval

In the sensitivity analyses, different ways to adjust for age (Supplementary Tables S4) and the additional adjustment for metformin use, which may influence risk of TB [16, 17] did not change the observed association between prediabetes and TB compared with the main analysis (Supplementary Tables S5). Similarly, excluding patients with major illnesses did not change the results (Supplementary Table S6). Consistent with the previous meta-analysis, prediabetes was associated with an increased risk of ischemic heart disease in our cohort (adjusted HR 1.19, 95% CI: 1.08-1.32)[18]( Supplementary Table S7). In the time-dependent Cox analysis that allows the FPG level to change over time, prediabetes was also inversely associated with the risk of active TB, although the confidence interval became wider due to the much-reduced sample size (adjusted HR: 0.49, 95% CI: 0.16-1.47) (Supplementary Table S8). Applying different exclusion periods for TB cases occurring soon after the baseline did not change the results (Supplementary Figure S2).

Given the strong and inverse association between BMI and risk of TB, we further adjusted for BMI using different approaches (linear, categorical, and spline regression) when estimating the association between prediabetes and TB. The observed inverse association between prediabetes and TB remained unchanged. (Supplementary Table S9).

Discussion

Our cohort study did not reveal an increased risk of active TB among prediabetic subjects. Contrary to our hypothesis, we found that prediabetic subjects had a 27% reduced relative risk of active TB disease when compared with normoglycemic subjects. Among non-diabetic subjects, we observed a U-shaped association between the FPG level and risk of active TB disease with the lowest TB risk at FPG at around 110 mg/dL. The results were robust in a series of sensitivity analyses addressing biases from possible confounders, misclassification of glycemic status, and selection of study participants.

Limited numbers of studies have examined the association between prediabetes and TB. Observational case-control studies have investigated the prevalence of diabetes and prediabetes among TB patients. A case-control study in the US showed that4.1% of 166 participants with a self-reported TB history had intermediate diabetes, while 2.3% of 15,199 participants without a self-reported TB history had intermediate diabetes (adjusted OR:1.34[95% CI: 0.50-3.62]) [9]. In this study, only one-fourth of participants had results from glucose tolerance tests and the others self-reported their diabetes status. Therefore, recall bias might impact the categorization of the prediabetic population. Another case-control study in China found a higher prevalence of prediabetes among 6382 TB patients than among 6674 non-TB subjects (7.4% versus 6.6%), and the adjusted OR was 1.14(95% CI: 0.50-2.57) for prediabetic compared with normoglycemic patients [10]. Despite the large population size in this study, the odds of prediabetes among subjects with TB could be overestimated due to the stress hyperglycemia induced by active infection [21]. Overall, the positive association between prediabetes and TB from previous case-control studies may be biased. This current study provided the first evidence from a cohort study showing an inverse association between prediabetes and the risk of TB disease.

The inverse association between prediabetes and TB disease risk was beyond our expectation. We therefore conducted extensive sensitivity analyses to examine whether this inverse association could be due to potential biases in our study. First of all, the possibility of residual confounding bias was considered. In particular, prediabetes is closely associated with overweight and obesity and previous studies consistently showed a protective effect of overweight and obesity on TB disease. We extensively adjusted for BMI using various approaches and found that the inverse association between prediabetes and TB remained unchanged. Secondly, most of the participants only had a single FPG measurement at baseline, which may not capture their long-term glycemic status. However, when the time-dependent Cox model with repeated measurements of FPG level was conducted, the risk of TB for the prediabetes group was still lower than the normoglycemia group. Lastly, the inverse association might be due to major systemic illnesses which resulted in hypoglycemia and increased risk of TB. However, the inverse association persisted when patients with major illness were excluded from the analysis. In summary, our extensive sensitivity analyses suggested that biases alone may not completely explain the inverse association between prediabetes and TB observed in this study.

The biological mechanism behind the association between prediabetes and a lower risk of TB disease remains elusive. Abundant evidence showed that overt DM with overt hyperglycemia impairs both innate immunity and adaptive immunity [23,24]. Notably, insulin treatment reverses these adverse effects conferred by hyperglycemia [25]. In contrast, prediabetes is characterized by insulin resistance with hyperinsulinemia without overt hyperglycemia. The insulin resistance status in prediabetes is characterized by chronic inflammation with activated innate and adaptive immunity [26,27]. Consistent with our hypothesis, a study using the NHANES data reported higher interferon-γ responses in participants with latent TB infection and prediabetes than in participants with latent TB and euglycemia (OR: 1.9, 965 C.I.: 1.0-3.6) [28]. A multi-center prospective cohort study reported an association between diabetes and unfavorable outcome after anti-TB treatment. However, prediabetic TB patients have a reduced risk of unfavorable outcomes after anti-TB treatment (adjusted HR: 0.88) [29]. Lastly, Cegielski et al. found that anthropometric parameters including skinfold thickness and mid-upper arm muscle area were closely and inversely associated with risk of developing TB [30]. Future experimental and clinical research on how nutritional status affects the development and treatment of TB will help to control and prevent the disease.

This study has several limitations. First, the incident TB cases could be either newly infected or latent TB infection. Owing to the lack of information about latency status, it is unclear whether prediabetic patients had a lower risk of TB infection or lower risk of progression from latent TB infection. Second, this study may suffer from selection biases [3]. Supposing prediabetic patients were more likely to attend the screening service and so were potential TB patients, this would overestimate the causal association between prediabetes and TB. Third, there is a time gap between DM screening and the time TB diagnosis. Therefore the participants’ prediabetes or DM status might have changed over the period prior to the diagnosis of TB. We performed additional analyses in a subset of participants who received at least two FPG measurements in the study period. Using time-dependent Cox regression to account for the changing FPG status, the inverse association between prediabetes and TB remained. Lastly, we have few repeatedly measured data on BMI and other confounders during the study period in this cohort. We have few repeatedly measured data on BMI and other confounders during the study period in this cohort. However, according to the data from ~38, 000 participants receiving health examination of the MJ Health foundation in Taiwan [32], the percentage of BMI > 24kg/m² changed from 37.8 % to 41.98 % in male adults and 20.2% to 21.68 % in female adults during follow. According to the national survey of Health Promotion Administration of the Ministry of Health and Warfare of Taiwan, the percentage of cigarette smoker changed from 40.6% to 37.7 % in male adults and 4.8 % to 4.3 % in female adults and the average alcohol beverage consumption per person changed from 3.3 to 3.0 litters during follow-up period [33]. These data suggested that the changing status of these potential confounders were likely to be too subtle to change the conclusion of the study.

This observational study showed a 27% reduction in risk of active TB among prediabetic subjects compared with normoglycemic subjects, and the lowest risk was found at an FPG level of around 110 mg/dL. Consistent results were found in a series of sensitivity analyses in this study. This is the first longitudinal study investigating the association between prediabetes and risk of TB. This novel finding has critical implications on the highly discussed dual epidemic of diabetes and tuberculosis. Given the protective association of obesity and prediabetes with tuberculosis, the global nutrition transition may represent a double-edged sword to global tuberculosis control. More research is needed to fully understand the biological mechanism and the public health implications of the relationship between prediabetes and TB.

Supplementary Material

Supplementary Material

Supplementary Figure S1.Dose-response relationship between fasting plasma glucose level and risk of tuberculosis in total population. CI: confidence interval; HR: hazard ratio

Supplementary Figure S2. Hazards ratios for prediabetes by different exclusion days of potential tuberculosis (TB)

Supplementary Table S1. The ICD-9 codes of major illness include systemic autoimmune diseases, organ-specific autoimmune disease, cancer, alcoholism, liver cirrhosis, and dialysis

Supplementary Table S2. Change of glycemic status during follow-up

Supplementary Table S3. A Priori power analysis

Supplementary Table S4. Age-adjusted Cox models by continuous, categorical and spline adjustment

Supplementary Table S5.Cox model of the association between diabetes mellitus (DM) status and risk of active tuberculosis (TB) after excluding normoglycemic and prediabetic subjects with metformin use of two different definitions

Supplementary Table S6. Cox models of the association between diabetes mellitus (DM) status and risk of tuberculosis (TB) after excluding patients with major illnesses

Supplementary Table S7. Age-adjusted and multivariable Cox model of the association between diabetes mellitus (DM) status and ischemic heart disease

Supplementary Table S8.Time-dependent Cox model: association between diabetes mellitus (DM) status and risk of active tuberculosis (TB) with repeat measurements of FPG

Supplementary Table S9. Boyd mass index (BMI)-adjusted Cox models by continuous, categorical, and spline regression

Key features:

• In contrast to diabetic subjects, prediabetic subjects had a 27% reduced relative risk of active TB disease compared with normoglycemic individuals.

• Existing studies had multiple sources of biases; the present findings were robust in a series of sensitivity analyses addressing biases from possible confounders, misclassification of glycemic status, and selection of study participants.

• Given the vast obese and prediabetic population worldwide, this unique longitudinal cohort finding has significant implications on global TB control. The global nutritional transition may represent a double-edged sword to TB control worldwide.

Data Availability

The study protocol and statistical analysis for this project are available on request from the corresponding author.