Gradient association between pulmonary tuberculosis and diabetes mellitus among households with a tuberculosis case: a contact tracing-based study

Pulmonary tuberculosis (PTB) and diabetes mellitus (DM) remain high morbidity and mortality, especially when they are comorbid with each other. Screening for diabetes mellitus in tuberculosis is essential as the incidence and mortality of DM in the population with PTB are higher than in the general people. We aimed to examine the gradient association of tuberculosis on developing DM, the additional yield and the number needed to screen (NNS) to find a new diabetes case. A cross-sectional study was conducted on 801 tuberculosis cases and 972 household contacts in Guizhou, China, from April 2019 to October 2020. After screening for PTB among contacts, all participants were screened for DM and interviewed. Kendall’s tau-b test and proportional odds logistic regression analysis were applied to identify the gradient associations. Among the 1773 subjects, the additional yield of screening was 21.8%. The NNSs of the non-PTB group, the sputum-culture negative and positive groups were 50, 60 and 113, respectively. The gradient incremental establishment of DM and PTB were positively correlated. The general trend on the gradient of DM significantly increased with the gradient increase of PTB. Age 35 years and over, excessive edible oil intake and DM family history were identified as significant predictors of diabetes. Integrated screening for DM targeted to different gradients of PTB combined with associated factors is necessitated to achieve a higher additional yield.


Results
Based on experts' review, the content validity of the questionnaire met the standard requirement of rigorous research procedure with an alpha coefficient of reliability of 0.85.
Sampling and general characteristics. Overall, 809 index PTB cases were recruited. During the followup, six who did not respond and did not participate in the disease screening test were excluded from further analysis. Subsequently, 1,016 adult household contacts were visited, among whom 39 did not respond and therefore did not participate in the disease screening. Two PTB index cases and their five household contacts were also excluded due to unqualified questionnaire information. Eventually, 801 (45.2%) PTB index cases and 972 (54.8%) household contacts were recruited in the study (Fig. 1). Table 1 compares the background characteristics of the PTB index case group and their household contacts group. The PTB index cases were significantly older, more likely to be male, migrant-laborers, married or cohabitating, and having a lower income. The two groups were not significantly different in terms of religion, ethnicity or level of education attained. Table 2 tabulates the gradient of PTB against that of DM as well as summaries the additional yields and NNSs. In the sixth column of Table 2, the total additional yield was 21.8%, with 25.8% for non-PTB, and 20.0% for both culture (−) and (+) PTB subjects, adding a quarter to one-fifth of the existing known DM cases. Table 2 indicates how many people in each group needed to be screened to find one new DM case. The NNSs of the two PTB groups were approximately half that of the non-PTB one (50 and 60 versus 113).
We combined previously known DM and newly diagnosed DM to compute the correlation between the gradients of PTB and DM. Kendall's tau-b was 0.066 (P value < 0.001), indicating a very weak but statistically significant positive rank correlation between the gradients of DM and PTB. In other words, the general trend on the gradient of DM increased with the gradient increase of PTB.  Table S1). Table 3 summarizes the POLR models with covariates incrementally added. The successively added variables were gender, age, ethnicity, occupation, smoking, excessive edible oil intake and DM family history, labeled with the symbol "+" when added in each model, subsequently to the principal hypothesis variable, the gradient of PTB from the Models 2 to 8. The AIC in the last row indicates the level of the fitting. The AICs from all the models except model 9 contain the gradient of PTB severity as it is the principal hypothesis. Model 2 to 5 had demographic characteristics incrementally added. Model 6 and 7 added the subject's risk behaviors and model 8 added family history of DM. Eventually, to check whether the principal hypothesis still worked, we removed the PTB gradient in model 9 to compare specially with Model 8, which presented the lowest AIC (2658.426) to test which one would fit better when the principal hypothesis was the absence or presence of the PTB gradient. As seen, the AIC of model 8 was lower than that of model 9 (2661.744). Model 8 remained the one with the lowest AIC. Thus, our hypothesis is supported by the data analysis even though we had put some other explanatory variables.

Proportional odds logistic regression (POLR) analysis results.
Model 8 was chosen as the best model with the lowest AIC value as 2658.426. The effects of independent variables included in this model are displayed in Table 4 as the adjusted ordinal odds ratios. Table 4 provides the gradient effect of PTB and other covariates to the outcome on the risk of getting DM based on model 8 from Table 3. The difference between the presence and the absence of a particular variable, with other variables in common, is analyzed as LR test P values in the last column. For the first and the main variable, the risk of DM was significantly higher for both the culture (−) group (adjusted ordinal OR = 1.33, 95% CI 1.06, 1.68) and the culture (+) group (adjusted ordinal OR = 1.35, 95% CI 1.00, 1.82) compared to that of the non-PTB group. The P value of the LR test was 0.025, indicating that the PTB gradient significantly contributes to the model in predicting the severity of DM. Gender was not significant as the LR test P value was 0.195; nor   Gradients of PTB (principal hypothesis) www.nature.com/scientificreports/ was smoking (LR test P value = 0.495). However, the older the age, the higher the gradient of DM. Those who took excessive edible oil had 1.63 times the odds of developing diabetes compared with those who consumed an average amount of oil. In addition, people with a family history of diabetes have a 2.16 times higher risk of developing diabetes than people without a family history of diabetes.

Discussion
Summary of findings. PTB and DM are two essential health issues of worldwide concern. A deep exploration of the gradient influence of PTB on developing DM is beneficial for planning the corresponding measures to control these diseases for policymakers according to the different clinical characteristics of patients. Our study found that the incremental establishment of DM and PTB were positively correlated, which was confirmed by Kendall's tau-b test and POLR analysis. The general trend on the gradient of DM significantly increased with the gradient increase of PTB. The screening could add a quarter to one-fifth of the existing known DM cases, and the NNSs to find a new DM case for the two PTB groups were around half of that in the non-PTB group. In addition, age 35 years and over, having excessive edible oil intake, and presence of DM family history were identified as significant predictors of developing DM.

Risk of DM and public health implications.
The total prevalence of DM among all subjects (5.7%), which may represent a family with a PTB patient, is lower than that (7.6%) from a survey in the general population 15 .
The prevalence of DM among the two PTB groups, whether 7.8% or 9.2%, was slightly higher than the 6% reported in Angola in 2019 16 , 6.2% in Paraguay 17 and 7.5% in India among PTB cases 18 . On the other hand, it is lower than the 11.9% reported by a study in Eastern Nepal 19 . However, these studies might have subjects with different age and gender distributions. A common factor among all these studies was that the PTB patients and families were from a low socioeconomic group.
Although the association between PTB and DM was very weak, the PTB gradient significantly associated with the degrees of DM from the univariate analysis remained significant even when using POLR analysis to adjust for other explanatory factors, such as gender, age, ethnicity, occupation, smoking, excessive edible oil intake and DM family history. This finding is similar to a previous study in Myanmar 6 . Sputum culture-positive is significantly associated with developing DM. It might be linked to the predominant type of M. tuberculosis being the Beijing strain in Guizhou 20 . Beijing strain might contain two genes, ESAT-6-and CFP-10-like proteins 21,22 , which could induce the severe responses of the immunologic substances, such as chemokine 10 (CK10), tumor necrosis factor α (TNF-α) 23 , interferon γ (IFN-γ) and interleukin 17 (IL-17) 24 . These cytokines are the effector molecules, regulating the expression of corresponding genes by activating nuclear transcription factors 25 , thereby regulating the apoptosis of pancreatic β-cells and triggering DM 26 .
For the plausible explanation of this gradient association, some studies also reported a relationship between DM and PTB at both the epidemiological and molecular levels. The multidrug-resistant TB, which was the most successful pathogen, could subvert the host immune response and glycemic metabolism through multiple mechanisms, resulting in insidious diseases, such as DM and other diseases 6 . However, if the PTB patients who do not have DM could be identified as PTB and accept anti-TB treatment timely, they may not develop DM after 3 months of anti-PTB treatment 27,28 .
Like PTB, early detection of DM is of value in preventing complications. The additional yield from DM screening reflects its level of worthiness in families with a PTB member. The higher the additional yield is, the more significant the screening would be. The additional yield estimated in our study (21.8%) is much lower than that of two studies conducted, in India (43.0%) 17 and in Nigeria (59.0%) 10 . Likewise, the NNS in our study was higher than that in those countries (77 in our study, 11 in India and 18 in Nigeria). Our lower additional yield and higher NNS may reflect better access to DM screening in our population than in the countries; it might also be because of some differences in dietary habits and cultural customs 29 . Yet, the screening in this population is worthy of the area where PTB is common. DM screening among households of PTB cases could add a quarter to one-fifth of the existing known DM cases, which indicates an alarming signal that DM should be investigated as routine in PTB screening programs 14 .
The prevalence of prediabetes in this study is noteworthy, which expressed a considerably high prevalence as a quarter in all groups. This means that there will be a large proportion of DM in the future among these family members. According to a study in Germany, prediabetes was observed to demonstrate a predisposition of 29.3% to develop DM after around 10 years, although with 22.5% reverting to normal glucose regulation 30 . Furthermore, even in the prodromal stage, after controlling the development of diabetes, prediabetes is also related to some dysfunction or disability, such as increased chair time, decreased walking speed, and accelerated disability progression 31 . According to a Japanese study, compared with regular blood glucose, prediabetes was associated with a higher risk of all-cause death and death from cancer, but not death from cardiovascular disease. The results were materially unaltered when prediabetes was defined according to three kinds of diagnostic criteria 32 . The high prevalence of prediabetes in our subjects indicates that public health services should be appropriately allocated to the control of prediabetes.
Other associated factors. Those aged 35 years and above were 1.75-2.59 times more likely to develop DM than those below 35 years, which is consistent with a previous study 33 . Older age was reported as an associated factor for DM, which may be related to decreased immune status in older individuals, making them more susceptible to developing DM 1 . However, this result does not mean that the relevant health services targeted to young people should be ignored. On the contrary, about one-fifth of participants aged between 15 3 . In that study, the overall unadjusted incidence rates of type 2 diabetes increased by 7.1% annually. Participants with a family history of DM were 2.15 (95% CI 1.38, 3.13) times more likely to develop DM than those without, which may be related to familial predisposition due to holding similar genes. Growing evidence in some countries based on genetic research has emerged to support this hypothesis 35 . A study from China suggested that two regions, on chromosomes 6q21-q23 and 1q21-q24 respectively, present a significant linkage to type 2 diabetes/impaired glucose homeostasis in Chinese. Another study in China found that INAFM2 rs67839313_T was associated with increased type 2 diabetes risk and FPG levels in Chinese individuals 36 .
The positive relationship between excessive edible oil intake and DM is plausible as absorption of more lipids results in overweight, which has been shown to be a risk factor of DM 37 . In addition, in the present study, compared to Han ethnicity, people from the Buyi ethnic minority group were 1.37 (95% CI 0.98, 1.90) times more likely to develop DM, which might be due to the different lifestyle and cultural customs 38 . In our study, no gender difference was found, which is different from that in the previous studies in India 39 and Zambia 40 .
Overall, with the potential confounders incrementally included in the various models, the odds ratios of PTB gradient on DM risk were persistently significant. Thus, no potential confounders were evidenced.
Limitations. There are some limitations in this study. First, we included only index PTB patients and their household contacts. The level of association in this study may not fully reflect that level in the general population. Thus, we could not make a direct comparison of the DM prevalence between index PTB cases and the general population. Second, asymptomatic PTB subjects might be missed from the data due to the way of screening of symptoms and signs first, which is usually performed. Finally, this is a cross-sectional study, which is difficult to identify the exact chronological association of PTB and DM.

Conclusion
The general trend on the gradient of DM increased with the gradient increase of PTB. Integrated screening for diabetes targeted to different gradients of PTB combined with the associated factors is necessitated to achieve a higher additional yield.

Study design. A PTB contact tracing-based cross-sectional study was conducted in Guizhou Province from
April 1, 2019 to October 30, 2020. Eleven out of 88 counties/districts were randomly selected as the study sites.

Relevant definitions.
An index PTB case: a person who has been diagnosed with PTB by at least two positive results of sputum smear, or positive result of one sputum smear with Chest X-rays positive subsequent to 2 weeks of antibiotic medication, or positive result using Xpert MTB/RIF cartridges assay 41,42 , or one sputum sample cultured containing bacilli.
A household contact: a person who was in the same house with a PTB index patient > 6 h per week 43 from 3 months before the diagnosis and 14 days after the initiation of anti-tuberculosis treatment of the PTB case.
PTB gradients (applied for both the PTB index cases and the contacts): (1) non-PTB = negative Chest X-rays and negative results of sputum exam; (2) culture negative = positive smear and/or Chest X-rays but negative sputum-culture; and (3) culture positive = positive sputum-culture.
DM gradients: (1) non-DM: fasting plasma glucose (FPG) < 110 mg/dl; (2) prediabetes: glycosylated hemoglobin (HbA1c) value between 5.7 and 6.4% (39 ± 46 mmol/mol), or FPG at least 110 mg/dl but below 126 mg/ dl 44 according to the parameters set by the American Diabetes Association (2016); and (3) DM: HbA1c ≥ 6.5% (48 mmol/mol) 45,46 , or FPG ≥ 126 mg/dl, or random plasma glucose (RPG) ≥ 200 mg/dl or a previous diagnosis of DM 44  www.nature.com/scientificreports/ research locations were retrieved consecutively. Those who were on their retreatment regimen for PTB, pregnant, mentally retarded, or living alone were excluded from this study. During the eligible patients' monthly visits to the hospital for medications, TB health care staff would contact the patients to make an appointment with patients for home visits.
During the home visits, we surveyed 1-3 household contacts per household covering all-age in-home visits. Kish's method was employed to choose the respondents when the number of adult household contacts exceeded three 48 . Although we investigated the all-age household contacts to detect PTB, only those aged ≥ 15 years old were included in the analysis related to DM in this report.
The screening for PTB among the household contacts was conducted following the guideline recommendations of the WHO 49 . First, during home visits, contacts were initially screened for PTB-related symptoms, including cough, expectoration, bloody sputum, or hemoptysis for more than 2 weeks, or fever for more than 2 weeks, and other symptoms as night sweats, fatigue, and chest pain, etc. The participants with the positive symptom results were then referred to the local hospitals to do an Xpert MTB/RIF cartridges assay where available, or Chest X-rays, sputum examinations for acid-fast bacilli and sputum culture. For sputum testing, three expectorated sputum specimens were collected and detected for smear or culture, including spot sputum, night sputum, and morning sputum. PTB was diagnosed if the participant had at least one initial sputum sample containing acidfast bacilli; or positive chest radiograph not resolved after 2 weeks of broad-spectrum antibiotics treatment 14 .
DM detecting was performed through FPG/RPG or HbA1c. The newly diagnosed subjects with DM and/ or PTB were then transferred to the relevant hospitals to receive further treatments by the health system staff.
Sample size calculation. We used the two independent proportion formula to obtain the minimum sample size using a continuity correction.
where n 1 is the number of the PTB cases, including the cases with culture (−) and culture (+) among PTB index cases and among household contacts screened as positive PTB; n 2 is the number of the non-PTB subjects, including the household contacts screened as negative PTB; the ratio = n 2 /n 1 = 2.0; p 1 = 12.5%, the assumed prevalence of DM for the PTB groups, which was considered as 1.56 times higher than that of the non-PTB group; p 2 = 8.0%, the assumed DM prevalence of the non-PTB group, which is slightly higher than that of an epidemiological survey on DM recorded as 7.6% in Guizhou in 2010 15 ; the type I error rate (α) = 0.05; and the type II error rate (β) = 0.20. The formula resulted in n 1 = 553, n 2 = 1,106. With the consideration of 10% of non-response, ultimately, 1843 subjects were planned for recruitment. Statistical analysis. Data obtained from the questionnaires and medical record review were entered into EpiData version 3.1 (http:// www. epida ta. dk/), and R version 3.6.3 (https:// cran.r-proje ct. org/) was employed for the statistical analysis. Descriptive statistics was used for general characteristics of the participants, reporting frequencies, proportions, means, medians, and ranges where appropriate. Student's t-test or ANOVA test was employed for age, FPG/RPG and SBP/DBP comparison between/among groups as appropriate, which were summarized using the mean and standard deviation. When the data were not normally distributed, the nonparametric Mann-Whitney U test was performed on the continuous variables. Chi-square tests were used for categorical variables, such as prevalence, percentage, and ratio comparisons. When the expected cell size was less than five observations, Fisher's exact test was performed on discrete variables.
The additional yield of screening DM was calculated as the number of newly diagnosed DM/(previously known + newly diagnosed DM) × 100% 14 . The NNS to find a new DM case was calculated as one divided by the prevalence of newly diagnosed DM after excluding self-reported previously known cases of DM 50 .
Kendall's tau-b test was employed to assess the gradient association between PTB and DM. Once this relationship was established, we used proportional odds logistic regression (POLR) models to examine predictors of increased risk for DM.
Univariate analysis was used to identify the significant factors associated with DM. Then, with the DM level as the dependent variable, PTB gradient as the principal hypothesis, and the positive factors from the univariate analysis as the covariables, different POLR models were established. Akaike's Information Criterion (AIC) 51 was computed to reflect the level of model fitting for adding each variable in each model. The model with the lowest AIC value was considered as the best fitting.
Once we had chosen the best model, the Wald test was conducted to check whether the difference from the variable referent levels was significant. Each independent category was evaluated whether it significantly contributed to the model using the likelihood ratio (LR) test.