Latent tuberculosis infection among patients with and without type-2 diabetes mellitus: results from a hospital case-control study in Atlanta

Objective The purpose of this study is to compare the prevalence of latent TB infection (LTBI) among patients with type-2 diabetes mellitus (T2DM) to healthy controls without T2DM. To achieve this objective, we conducted a case-control study in a large hospital in Atlanta from 2016 to 2019. Results We enrolled 98 cases; 119 potential controls were screened, 84 of which had HbA1c  ≥  5.7% and one did not have QFT result, leaving 34 (28.6%) individuals enrolled as controls. LTBI prevalence was 9.2% among cases and 14.7% among controls (crude odds ratio 0.59, 95% CI 0.19–2.04). After adjusting for age and sex, the adjusted odds of LTBI among patients with T2DM was 0.45 (95% CI 0.13, 1.71) times the controls. We did not observe a statistically significant association between LTBI and T2DM. However, we reported a positive correlation between HbA1c level and nil count among individuals with LTBI (R2  =  0.55, p  <  0.01). In addition, we reported a high prevalence of LTBI among adults with T2DM and family members without T2DM. Supplementary Information The online version contains supplementary material available at 10.1186/s13104-021-05662-0.


Introduction
Emerging cross-sectional evidence suggests type 2 diabetes mellitus (T2DM) is associated with higher prevalence of latent tuberculosis infection (LTBI) [1,2]. Results from US national survey data reported 12% LTBI prevalence among adults with diabetes compared to 5% LTBI prevalence among adults without diabetes [3,4]. However, less is known regarding US regional differences in the relationship between LTBI and T2DM or the extent that the association is different in racial/ethnic subgroups [5].
Although T2DM is an established risk factor for tuberculosis (TB) disease, whether T2DM increases the risk of LTBI, or vice versa, remains a critical gap in knowledge [2,6]. Results from murine models of diabetes and TB suggest that phagocytosis and uptake of Mycobacterium tuberculosis (Mtb) by monocytes and macrophages are reduced in the context of diabetes [7]. Pro-inflammatory cytokine expression profiles associated with Mtb control [i.e., interferon (IFN)-γ] were also delayed in diabetic animal models [8]. These may support the premise that diabetes increases the risk of TB infection or reactivation from latency. Alternatively, LTBI may influence T2DM risk as increasing evidence indicates TB modulates human adipose tissue function and may impact host metabolic homeostasis [9]. For example, an animal model of nondiabetic Guinea pigs reported that an infection of Open Access BMC Research Notes *Correspondence: asalindri1@student.gsu.edu 1 Division of Epidemiology and Biostatistics, Department of Population Health Sciences, School of Public Health, Georgia State University, Atlanta, GA 30303, USA Full list of author information is available at the end of the article Salindri et al. BMC Res Notes (2021) 14:252 Mtb alone could result in glucose intolerance and incresead level of serum free fatty acid, two strong predictors of T2DM development [10].
To definitively establish whether LTBI increases diabetes risk or impact host metabolic outcomes will require large longitudinal studies and extensive followup time. Thus to gain preliminary insight into the LTBI-T2DM relationship, we conducted a case-control study to (a) compare the prevalence of LTBI among patients with T2DM to controls without T2DM, and (b) assess whether glycated hemoglobin (HbA1c) level is correlated with QuantiFERON Gold-in-tube test (QFT) quantitative measures (i.e., antigen, mitogen, or nil count).

Methods
We conducted a case-control study in a large hospital in metro Atlanta, Georgia, US, from 2016 to 2019. Eligible cases were human immunodeficiency virus (HIV)-negative adults (≥ 21 years) with newly diagnosed T2DM (diagnosed within the past 3 years) and no history of TB disease. Among cases, HbA1c (%) was obtained from the electronic medical chart by abstracting the plasma HbA1c value closest to the date of study enrollment. Eligible controls included adult family members/friends of cases with HbA1c < 5.7% (measured at the time of screening by a point-of-care HbA1c [11], Siemens DCA Vantage Analyzer) and no self-reported prior diagnosis of pre-diabetes, T2DM, HIV, or TB disease. HbA1c values were categorized according to the American Diabetes Association classification with HbA1c < 5.7% considered as "normoglycemic", HbA1c 5.7-6.4% as "pre-diabetes" and, HbA1c ≥ 6.5% as "diabetes" [12]. Enrolled participants had LTBI status determined by QFT Gold-In-Tube test at the time of study enrollement. QFT samples were prepared and processed following the manufacturer's (QIAGEN) guidelines. Results of the quantitative QFT measures were interepreted using three ciriteria: (a) mitogen-nil and tuberculin-nil values, (b) percentage of avian difference, and (c) percentage of tuberculin response [13], and classified as "positive", "negative", or "indeterminate" for TB infection. Cases and controls were excluded if they were using steroids or tumor necorosis factor (TNF)-α antagonist therapy at the time of screening, resided outside DeKalb or Fulton counties, or did not speak English. Participants clinical and demographic characteristics were obtained using study questionnaires and medical chart review.

Data analysis
We compared demographic and clinical characteristics of cases to controls using Chi-square and Fisher's Exact tests. Logistic regression was used to estimate the association between LTBI and T2DM quantified by odds ratios and 95% confidence intervals (CIs). Covariates included in the final adjusted model were selected purposively based on previously published literature and directed acyclic graph theory [14]. Among those with LTBI, linear regression was used to estimate the correlation between HbA1c and quantitative QFT measures. Linear regression models were used to measure the relationship between HbA1c and (a) nil count, (b) TB antigen-nil, and (c) mitogen-nil values. In addition to beta estimates and corresponding 95% CI, we also reported R 2 values or the coefficient of determination as a measure of how well the linear regression model described the observed data.

Discussion
We reported a high prevalence of LTBI among patients with newly diagnosed T2DM and healthy controls without T2DM in the metro Atlanta area. Our LTBI prevalence estimates (9.2% among cases, 14.7% among controls) were higher compared to the US national (5.0%) and US race-specific estimates among African Americans (5.3%) [15]. Inconsistent with our findings, previous studies using nationally representative data from the US reported that diabetes is associated with increased odds of LTBI by one-three-fold [3,4]. Another study from Atlanta conducted among recently arrived refugees also reported higher LTBI prevalence among participants with diabetes and pre-diabetes compared to euglycemic participants [1]. However, it is important to note that this hospital-based study, conducted primarily among African Americans, is consistent with recent US data that suggests the relationship between LTBI and T2DM differs across racial/ethnic groups [5].
Our findings also suggest that there may be regional differences in the relationship between LTBI and T2DM, which could be affected by background prevalences of both diseases or socioeconomic characteristics (i.e., an established risk factor for LTBI and T2DM) [2]. Of note, we reported more than 70% prevalence of undiagnosed prediabetes among screened family members/friends of cases. This finding is consistent with previously published studies reporting high prevalence of prediabetes/diabetes among Bold indicates that the finding is significant at α = 0.05 BMI body mass index; T2DM type-2 diabetes mellitus; IQR interquartile range; LTBI latent tuberculosis infection; TB tuberculosis; TST tuberculin skin test * p value obtained from Fisher's exact tests    household/family members of individuals with T2DM [16][17][18]. Our findings also support previous recommendations suggesting that friends or family members of patients with T2DM would benefit from T2DM screening in an effort to introduce early intervention to prevent T2DM development.
Among individuals with LTBI, we reported a positive correlation between HbA1c and the QFT negative control without antigens or mitogens (nil count). The nil value is used to determine if patient has a pre-existing nonspecific immune response which could lead to a falsepositive. A cross-sectional study among US adults with LTBI reported a higher IFN-γ antigen response among those with pre-diabetes compared to euglycemic adults, but found that the average nil value was similar among patients with diabetes, prediabetes, and euglycemia [19]. To date, the relationship between the quantitative IFN-γ responses specific to Mtb and T2DM is inconclusive. We did not observe a significant correlation between glycemic control with TB antigen or mitogen responses, although we only analyzed this relationship in 14 study participants. Similar to our findings, a cross-sectional study conducted in Indonesia from 2014 to 2015 reported no significant difference of median TB antigen-nil value across different HbA1c levels (HbA1c < 7.0% vs. 7.0-9.9% vs. ≥ 10.0%; p = 0.73) [20]. However, a 2014 cross-sectional study conducted in India reported that individuals with LTBI and T2DM had decreased mean of Mtb antigen-stimulated (net cytokines) levels including lower IFN-γ (10.5 pg/mL vs. 249.2 pg/mL), TNF-α (6.5 pg/mL vs. 328.1 pg/mL), interleukin (IL)-17A (14.2 pg/mL vs. 24.4 pg/mL), and IL-10 (95.6 pg/mL vs. 220.6 pg/mL) when compared to LTBI individuals without T2DM [21]. Further studies to better characterize the relationship between T2DM, hyperglycemia, and immune responses specific to LTBI are still warranted.

Conclusion
In conclusion, we reported a high prevalence of LTBI among adults with T2DM and family members without T2DM. Although we did not observe a significant association between LTBI prevalence and T2DM we did observe a positive correlation between HbA1c and nil count among individuals with LTBI. Larger prospective investigations across different regions and race/ethcity subgroups are warranted to determine the role of LTBI in pre-diabetes/T2DM risk. Further studies that measure LTBI prevalence among household members living with patients with T2DM are also needed to determine whether household/family members of individuals with T2DM could be considered as a priority target group for LTBI screening.

Limitations
Our study was subject to several limitations. First, we had a small sample size enrolled from a single hospital and new diabetes cases were diagnosed within the past three years. Thus, our results may not be widely generalizable to other settings or other new diabetes patients. However, we used 3 years cut-off to define newly diagnosed T2DM due to the long natural progression of T2DM [22,23] and the potential delay in receiving T2DM diagnosis among our study population, of which the majority came from lower socioeconomic levels [24,25]. Second, our study was designed to assess the association between LTBI and the risk of T2DM. The non-significant findings we reported in this manuscript may in part be due to a bi-directional relationship between LTBI and T2DM where T2DM increases the risk of LTBI. Third, because we were unable to compare at what age cases or controls were exposed to MTB, the higher prevalence of LTBI among controls may be a reflection of when during their lifecourse they were initially infected with LTBI. Last, we enrolled a smaller number of controls compared to cases, which may due to our control selection strategy (i.e., friends or family members vs. community-based recruitment). It is plausible that friends/ family members shared lifestyle or other modifiable T2DM risk factors including diet, physical activity, or smoking [18,26], which could lead to pre-diabetes (HbA1c ≥ 5.7%) or previous T2DM diagnosis and exclusion from this study. Previous studies highlighted that selecting friends/ family as controls may pose several epidemiologic challenges including potential case-control overmatch, similarity in responding study's questionnaires, and potential bias among cases in nominating their controls [27]. However, we believe that using friends or family members as controls is still a viable and cost-efficient option for a small pilot project like the present study.