Association between diabetes mellitus and active tuberculosis: A systematic review and meta-analysis

1Despite the decline in the mortality rate of active tuberculosis (TB) since 1990, TB is ranked as one of the leading causes of death. In 2015, there were an estimated 10.4 million incident TB cases worldwide. The “End TB Strategy” launched by the World Health Organization (WHO) in 2016, aims to end the global TB epidemic by 2035. Targets set in this strategy include 90% reduction in TB deaths and an 80% reduction in TB incidence by 2030, compared with 2015.

2The growing epidemic of diabetes mellitus (DM) is set to become one of the major global health challenges. The number of individuals with DM is projected to rise from 415 million in 2015 to 642 million by 2040. It is estimated that over a million TB cases among adults were affected by DM in 2012. The rising DM epidemic could contribute to an increase in TB burden. Although a few studies have failed to confirm a positive association between TB and DM, other studies reported a strong association between DM and active TB. Based on earlier published summary effect estimates, DM increases the risk of active TB by 3.11–fold and latent TB by 1.18–fold. DM also has a major effect on TB treatment outcomes; in particular, it is associated with delayed sputum culture conversion, increased risk of treatment failure, and increased risk of TB relapse and mortality. With the accumulation of recent evidence supporting the TB–DM association, there is a need for an updated understanding of the magnitude of the TB–DM association. This understanding is critical for implementation of comprehensive TB and DM control programs.

3In this study, we aimed to systematically review the published literature on the association between active TB and DM, and to statistically summarize the evidence on the strength of the association.

4This review follows Cochrane Collaboration guidelines and reports findings using the Preferred Reporting Items for Systematic reviews and Meta-analyses (PRISMA) guidelines.

5We searched Medline (from 1945 to December 22, 2015) and EMBASE (from 1980 to December 22, 2015) databases, for studies on the TB–DM association. For inclusiveness, our search strategy covered studies that examined any risk factor for TB. The literature search protocol is summarized in the S1 Box.

6Inclusion of studies was restricted to human studies that provided an estimate or allowed us to compute an estimate of the association, adjusted at least for one variable. No restrictions were made on study language, population, publication year, or region.

7We excluded studies: amongst animals or children, if qualitative in design, case reports, case series, reviews, anonymous reports, editorials or author commentaries, with no appropriate control arm or comparator group, where TB patients with DM were not separated from those with other co-morbidities, of TB outcomes rather than the association, that did not report adjusted estimates of the TB–DM association, and duplicate reports.

8We contacted authors of potentially eligible studies to provide the adjusted estimate for the association, if the adjusted estimate was not included in the publication. Studies whose authors did not respond were excluded. The flow diagram of study selection is shown in Fig 1. First author (RHA) screened all titles and abstracts, reviewed full-text articles, and assessed their eligibility for inclusion.

9Three authors (RHA, JAC, and LJA) designed the literature search terms and strategy. All potentially relevant full-text articles retrieved and reviewed to confirm eligibility. If eligible, data were extracted. All authors (RHA, FP, JAC, and LJA) contributed in assessing the eligibility of the included studies. The first author (RHA) extracted the data, which were re-extracted independently by at least one co-author (FP, JAC, or LJA). Discrepancies in data extraction were resolved by consensus or consultation with a third co-author. In addition to extracting baseline characteristics of included studies, we assessed methodological aspects, such as sampling strategy, characteristics of the study population, and TB and DM ascertainment. If a study reported more than one adjusted estimate or stratified adjusted estimates, we chose the most representative and relevant estimate with the most confirmatory DM (i.e. prioritizing HbA1c over fasting blood glucose (FBG)) and/or TB (i.e. prioritizing bacterial culture over X-ray) ascertainment strategy, or the estimate adjusted for the largest number of appropriate variables when the study reported multiple adjustment models. Five of the contacted authors have provided us with adjusted estimates. Two adjusted estimates were extracted from a previous review.

10We evaluated each study’s risk of bias (ROB) using nine domains for cohort and cross-sectional studies, and eleven domains for case-control studies. The ROB domains were adapted from Cochrane guidelines for systematic reviews and other validated quality assessment tools. The utilized ROB domains are presented in the S2 Box and were related to different quality criteria such as rigor of sampling strategy, TB and DM case definition and ascertainment, and DM timing in relation to TB. In case-control studies, convenience sampling of TB cases was considered as a probability-based sampling method as long as all cases in the sampling frame were selected. Based on the composite scores of the ROB domains, each cohort or cross-sectional study was classified as “potentially of low ROB” (score ≥7) or “potentially of high ROB” (score <7). Each case-control study was classified as “potentially of low ROB” (score ≥9) or “potentially of high ROB” (score <9) (S2 Box).

11For studies reporting stratified crude estimates, we calculated an overall adjusted estimate by only one stratification if there was an overlap with other strata. In such studies, we prioritized pooling crude estimates that stratified by co-morbidity, location, age, or sex, consecutively. When there were two or more levels of stratification without overlap, we pooled estimates for this sub-levels stratification. If a study stratified estimates according to DM type, we pooled the overall adjusted estimate regardless of the DM type.

12In sensitivity analyses, we pooled estimates of studies of “potentially low ROB”, studies unlikely to contain duplicate individual–level data, and studies in the general population rather than specific clinical populations. We further pooled estimates stratified by potential for duplicate data on same patients, country-income level, background TB incidence (≤50 or >50 cases per 100,000 person-year), region, and TB and DM ascertainment methodology. We obtained relevant data on TB incidence from the included studies or from the closest matching year made available by public databases.

13We pooled adjusted estimates stratified by study design and regardless of study design using random-effects model. Cochran’s Q statistic was used to test for evidence of heterogeneity. We estimated the I-squared (I2) as a measure of heterogeneity. We computed Tau-squared (τ2) to estimate the between-study variance of the true association between TB and DM.

14We assessed the presence of publication bias by examining the funnel plots using Egger’s t statistic to examine asymmetry (S2 Fig). We used the pooled effect estimate in four prospective studies to estimate the attributable risk fraction of DM in developing active TB among people with DM and to estimate the population attributable risk fraction of DM in developing active TB among the entire population in six high-TB-burden countries (India, Indonesia, China, Nigeria, Pakistan, and South Africa), that accounted for 60% of new TB cases in 2015. Calculations are presented in the S1 Text.

15All Statistical analyses were performed using STATA SE 14 (Stata Corporation, College Station, TX).

16We identified 19,963 publications, 44 of which were found relevant and included in this systematic review and meta-analysis (Fig 1). The included studies consisted of 58,468,404 subjects and 89,592 TB cases and they were set in 16 countries. Most studies were conducted in Taiwan (11 studies) and USA (11 studies), while only one study was in Africa. Two studies of different designs stratified patients according to DM type (1 or 2), one study was in patients with type 1 DM, while the rest of the studies were either among type 2 DM patients or the type of DM was not specified (presumably, type 2 DM as it is more prevalent). There were four prospective, 19 retrospective, 17 case-control, and three cross-sectional studies. One study was classified as “other” as the exact study design could not fit into the other categories. One of the prospective studies was among people aged ≥65 years and one was among renal allograft recipients. Four of the retrospective studies were among renal patients. Several studies were national in scope, thereby including the national population as the study sample size, such as for a study from Australia. Seven retrospective studies in Taiwan and one retrospective and one case-control-study in the United kingdom, and three cross-sectional studies. One study was classified as “other” as the exact study design could not fit into the other categories. One of the prospective studies was among people aged ≥65 years and one was among renal allograft recipients. Four of the retrospective studies were among renal patients. Several studies were national in scope, thereby including the national population as the study sample size, such as for a study from Australia. Seven retrospective studies in Taiwan were potentially duplicate studies using the same database with overlapping years. One cross-sectional study was set in 46 countries and one retrospective and one case-control-study in the United kingdom were potentially duplicate studies using the same database with overlapping years. One cross-sectional study was set in 46 countries.

17In prospective studies, estimates were adjusted at least for age except for one study that reported sex-specific crude estimates. For the latter we pooled these for the present review. In one prospective study the effect estimate was not adjusted for sex. In retrospective studies, all estimates were adjusted at least for age or sex. In case-control studies, estimates were adjusted at least for age and sex except in two studies where the estimate was adjusted for age but not for sex. All cross-sectional studies were adjusted at least for age and sex. Baseline characteristics of all included studies are shown in Tables 1 and 2.

18The strongest estimate of the TB–DM association was 7.83 (95% CI 2.37–25.09) in a case-control study from Russia followed by 7.58 (95% CI 2.94–19.49) in a prospective study from USA. The lowest effect estimate was 1.00 (95% CI 0.69–1.44) in a retrospective study from Canada followed by 1.16 (95% CI 0.58–2.32) in a case-control study from USA. All four prospective studies demonstrated a positive association (p<0.05). Fourteen of the 19 retrospective and 12 of the 17 case-control studies demonstrated a positive association (p<0.05) between DM and TB (Tables 1 and 2).

19In the four prospective studies, DM was associated with 3.59–fold (95% CI 2.25–5.73) increased risk of TB. The I2 was estimated at 77.9% indicating that most variation across studies was due to heterogeneity in effect size rather than chance. In 16 retrospective studies, DM was associated with 1.55–fold (95% CI 1.39–1.72) increased risk of TB (I2 = 77.1%). In the 17 case-control studies, DM was associated with 2.09–fold (95% CI 1.71–2.55) increased risk of TB (I2 = 79.5%). In the three cross-sectional studies, DM was associated with 1.70–fold (95% CI 1.28–2.24) increased risk of TB (I2 = 28.9%) (Table 3). In all studies regardless of study design, DM was associated with 2.00-fold (95% CI 1.78–2.24) increased risk of TB (I2 = 90.5%). Forest plots of meta-analysis according to study design are shown in Fig 2 with summary findings presented in Table 3.

20All of the four prospective studies were judged of “potentially low ROB”. Except one study, the 16 retrospective studies reporting relative risk (RRs), hazard ratio (HR), or rate ratio (RR), were judged of “potentially low ROB”. Of the 17 case-control studies, 13 were judged of “potentially low ROB” (S1 Table).

21In the sensitivity analyses presented in S1 Table, restricting the meta-analyses to studies judged of “potentially low ROB”, among only the general population, and with no potential for duplicate data on the same patients, DM patients were overall, that is by including all studies regardless of study design, at 2.00–fold (95% CI 1.77–2.27), 2.12–fold (95% CI 1.82–2.48), 1.63–fold (95% CI 1.45–1.82), respectively, increased risk of TB compared to the original overall estimate including all studies of 2.00–fold (95% CI 1.78–2.24) increased risk of TB (Table 3).

22Moreover, overall, DM patients in low- or middle-income countries (3.16, 95% CI 2.20–4.53), in settings with TB incidence >50 cases per 100,000 person-year (2.05, 95% CI 1.80–2.33), or in Asian continent (2.46, 95% CI, 2.04–3.02) were at higher risk of TB than DM patients in high-income countries, in settings with TB incidence ≤50, or in Europe or USA and Canada, respectively (S1 Table).

23In two prospective studies, with microbiologically defined TB, and with blood testing for DM, patients were at 3.67–fold (95% CI 2.01–6.70) increased risk of TB. Overall, DM patients with microbiologically defined TB were at 3.03-fold (95% CI 2.31–3.98) increased risk of TB compared to 1.58–fold (95% CI 1.46–1.71) and 1.73–fold (95% CI 1.36–2.20) based on medical records or self-reported TB, respectively. Overall, blood tested DM patients were at 3.10–fold (95% CI 2.02–4.74) increased risk of TB compared to 1.60-fold (95% CI 1.18–2.17) and 1.95-fold (95% CI 0.90–4.25) based on medical records or self-reported DM, respectively (S1 Table).

24Overall, DM patients with HbA1c ≥ 6.5%, FBG ≥ 120 mg/dl, or on insulin treatment, were at 1.87–fold (95% CI 1.19–2.93), 3.30-fold (95% CI 2.12–5.14), or 2.51–fold (95% CI 1.62–3.87) increased risk of TB, respectively (S2 Table).

25In the four prospective (p = 0.495), 16 retrospective that reported RR, RRs, or HR (p = 0.439), three retrospective that reported an OR (p = 0.864), and three cross-sectional (p = 0.696) studies, reporting on the association between DM and TB, the Egger’s t statistic for asymmetry in the funnel plot indicated no evidence for the presence of a small-study effect (S2a, S2b, S2c and S2d Fig, respectively). However, in the 17 case-control studies, the Egger’s t statistic for asymmetry in the funnel plot indicated evidence for the presence of a small-study effect (p = 0.005) (S2e Fig).

26With a summary RRs of 3.59 among DM patients in the four prospective studies, DM accounts for 72% of active TB cases among strictly DM patients (attributable risk fraction). In the six high-TB-burden countries (India, Indonesia, China, Nigeria, Pakistan, and South Africa), that accounted for 60% of new TB cases in 2015, 18%, 14%, 22%, 11%, 15%, and 15% of active TB cases in the entire population in these countries is attributed to DM, respectively. For the population attributable risk fraction, calculations are presented in S1 Text.

27In this systematic review and meta-analysis of studies on the TB–DM association, we identified a strong positive association, but with substantial heterogeneity in effect size between studies. Stronger associations were noted among DM patients where TB was confirmed microbiologically, where DM ascertainment was based on blood testing (rather than self-report or medical records only), and among DM patients with uncontrolled blood glucose. This most comprehensive review and meta-analysis included 44 observational studies, compared to nine and 13 studies in the previously published two reviews, one of which included a meta-analysis of prospective studies. With this increase in published evidence, our meta-analysis confirmed the result of the earlier meta-analysis and strengthened the evidence base for a strong association between DM and active TB. Our results demonstrated consistent evidence of a two- to four-fold increased risk of developing TB disease for DM patients compared to non-DM patients.

28As a result of aging and increasing prevalence of major DM risk factors; particularly obesity and tobacco use, it is projected that the number of individuals with DM will rise from 415 million in 2015 to 642 million by 2040. With the observed TB relative risk of 3.59 among DM patients in prospective studies, DM will therefore account for an increasing fraction of active TB cases in the entire population in the future. A frequent misperception is that chronic diseases such as DM are diseases of affluence, in fact they are also common amongst poorer populations, where infectious diseases such as TB remain prevalent. Estimates suggest that the DM burden is increasing fastest in those regions where TB remains endemic. From a public health perspective, it is of note that there are now more TB patients with concomitant DM than with HIV. Given these findings, it may be challenging to control TB, particularly in settings that experience the double burden of the twin–epidemic of TB and DM. Robust public health intervention programs should consider tackling the underlying factors of DM such as lack of physical activity and obesity. As well as, programs to screen DM patients for TB alongside controlling blood glucose in TB patients to improve TB/DM treatment outcomes and to control this “twin epidemic”. Our findings strengthen the evidence base for how DM can impact upon the achievements of the WHO “End TB Strategy”.

29The observed slight discrepancy in the summary estimate in the present and the previous meta-analysis is partially due to the fact that one of the prospective studies included in the previous meta-analysis was classified as retrospective in our review. Also, we used estimates from more confirmatory TB (bacterial culture confirmed rather than X-ray ascertained TB) and DM (HbA1c rather than FBG ascertained DM) ascertainment criteria that adjusted for the largest number of variables to pool strata-specific estimates, which in turn would produce more reliable association between TB and DM. In the previous meta-analysis, the estimate obtained from Kim et al., 1995 was age-adjusted for all types of TB using a fixed-effect model, whereas we adjusted for sex for bacteriologically proven TB only, using a random-effects model. Moreover, we identified two more prospective studies that contributed 44% of the weight in our summary estimate. However, excluding one of these two studies, as DM and one-third of TB cases were ascertained by “self-report”, the summary estimate was 3.18 (95% CI 1.95–5.18), which is more comparable to that reported in the previous meta-analysis. In the prospective studies, DM was mostly ascertained prior to the development of TB, suggesting that DM increases the risk of developing active TB, though some studies screened for DM at the time of TB diagnosis, and thus infection-related hyperglycemia could also explain some of the association.

30We conducted sub-group sensitivity analyses to assess the heterogeneity in effect size. Several factors appeared to have contributed to this heterogeneity, including sampling methodology, study subjects, year of study, geographical location, exposure and outcome ascertainment methodology, variability within the specific subpopulation studied, sex and age-group representation in the sample, and publication bias. However, with the relatively small number of outcome measures according to study design, it was not possible to quantify the contribution of these sources of variation to the heterogeneity in the association through a meta-regression analysis.

31All included studies were adjusted for at least age or sex, and estimates from majority of studies were also adjusted for different demographic and other potential confounders. This ensured that overall summary estimates were adjusted for at least the major confounding effects of age or sex. The strongest TB–DM association was observed from the four prospective studies. Data from almost three-quarters of included studies were representative of the general population. In studies reporting more than one adjusted estimate or strata-specific estimates, we included the estimates with more confirmatory ascertainment criteria for TB and/or DM, and that adjusted for the largest number of variables. This in turn produced summary estimates with lower potential of including false positive or false negative DM or TB cases. Overall, limiting the meta-analysis to studies judged as “potentially of low ROB” and excluding potentially duplicate studies did not change the direction nor magnitude of the association.

32In the present review, the overall summary estimate in settings with TB incidence >50 cases per 100,000 person-year showed stronger association compared to that in settings with TB incidence ≤50 (S1 Table). This is in line with the findings of the previous meta-analysis. Dobler et al., 2012, hypothesized that the reason for the stronger association in settings with higher TB incidence could relate to the quality of diabetes management, assuming healthcare services may be poorer or harder to access in higher TB incidence settings.

33We noticed a stronger association in blood-tested DM patients. DM patients with well-controlled glucose levels are less likely to be included when the definition of DM is based on blood glucose levels, which implies that hyperglycaemia rather than a DM diagnosis per se, increases the risk of TB. DM patients suffer from immune system impairments, resulting in a lack of energy supply to immune cells, that subsequently increases virulence of infectious microorganisms. These impairments weaken the immune system response to Mycobacterium tuberculosis. This is supported by the observed stronger association in patients with uncontrolled blood glucose level (FBG ≥120 mg/dl or HbA1c ≥6.5%) (S2 Table).

34There are several biological mechanisms that appear to alter the immune system and by which DM patients may develop TB. High levels of insulin were associated with a decrease in T helper 1 (Th1) immunity through a reduction in the Th1 cell to T helper 2 (Th2) cell ratio and interferon-c (IFN-c) to interleukin-4 (IL-4) ratio. Other studies showed that nonspecific IFN-c levels were significantly reduced in people with diabetes compared to people with no diabetes, and that levels of IFN-c were negatively correlated with levels of HbA1c. Neutrophils in people with diabetes were found with a lack in chemotaxis and oxidative killing potential compared to non-diabetic controls. Leukocyte bactericidal activity was found to be reduced in people with diabetes, especially those with poor glucose control. These observed immunologic alterations seen in people with diabetes have also been supported in experimental animal studies. Diabetic mice experimentally infected with Mycobacterium tuberculosis have higher bacterial loads compared to euglycemic mice with significantly lower production of IFN-c and interleukin-12 and fewer T cells.

35Several of the included studies had methodological weaknesses. Eight studies relied on “self-reported” DM and four studies relied on “self-reported” TB. Studies that utilized blood tests to define DM may also have reported stronger associations between DM and TB, since they can identify undiagnosed DM, which is common in many low–and middle–income countries. In studies that relied on “self-reported” DM, subjects with controlled blood glucose (euglycemic) would be “misclassified” as DM patients. This assumption is supported in our sensitivity analyses (S1 Table). There is an additional potential misclassification of TB and DM cases as studies often used routinely collected data without validation using laboratory tests. For instance, a single HbA1c measurement might misclassify individuals as either DM or non-DM patients. It is recommended that DM diagnosis should be confirmed with a repeat HbA1c test, unless clinical symptoms and plasma glucose levels >11.1mmol/l (200 mg/dl) are present. Missing adjustment for potential confounders is also a noteworthy limitation. In six studies, estimates were adjusted only for age and/or sex. Individual studies that controlled for the influence of age, sex, and smoking produced stronger estimates than those controlled for age and sex. Biased estimates on the TB–DM association may have occurred in studies among patients receiving dialysis or among subjects from specialty clinics or hospitals rather than the general population. In almost all case-control studies, sampling of cases and/or controls was based on non-probability sampling. Studies using hospital-based controls reported weaker estimates for the association.

36DM can affect different aspects of TB natural history and treatment outcomes, and therefore can impact TB transmission dynamics. An ongoing study has identified seven epidemiologically-relevant plausible effects for DM on TB natural history, and three for DM on TB treatment outcomes. Our study, however, was focused on one major aspect of the TB–DM synergy, that of the association between DM status and active TB disease diagnosis—we did not assess other aspects of this synergy such as effects of DM on TB infection acquisition, TB reactivation among those latently infected, TB infectiousness, or TB treatment outcomes. A recent review, for example, reported that DM increases risk of latent TB by 1.18–fold, though with substantial heterogeneity across studies, and other studies have demonstrated major effects for DM on TB treatment outcomes. Comprehensive and granular characterization and quantification of the diverse effects of DM on TB is essential for a proper understanding and estimation of the impact of DM on TB epidemiology.

37In most of included studies, type of DM was unclear, thereby limiting our ability to assess the association by DM type (1, 2, or both). Having said so, in the three studies that assessed the association of TB with type 1 DM, the effect size was comparable to that seen for type 2 DM studies. HIV/AIDS is a strong risk factor for TB, but only a fraction of included studies controlled for its effect in their assessment of the association. This may not affect appreciably our results, as HIV prevalence is very low in nearly all countries where the association was assessed. Age is another confounding factor for the TB–DM association, and nearly all studies controlled for this factor. Our study was focused on the overall effect of DM on TB disease, and we did not provide a pooled effect size stratified by age. Despite these limitations, our review and meta-analyses compiled and summarized important data and critically provided narrative information from a large number of studies that reported on the TB–DM association.

38Our systematic review and meta-analysis demonstrated consistent evidence of a substantially increased risk of TB disease among people with DM. This evidence was based on data from studies using different designs and reported from six continents. DM patients with uncontrolled blood glucose (measured by higher FBG or HbA1c) appeared to be at higher risk of active TB than patients with controlled DM. Efforts to halt the burgeoning DM epidemic would have an accompanied benefit of alleviating the global burdens of DM and TB. The burgeoning epidemic of DM is likely to impact upon the achievements of the WHO “End TB Strategy”. Our findings inform strategy planning of health service provision and implementation of effective prevention programs to control the “twin epidemic” of DM and TB.