Our findings suggest that a substantial proportion of incident tuberculosis in India is attributable to diabetes;14.8% of pulmonary tuberculosis and 20.2% of smear-positive – i.e. infectious – tuberculosis. They also suggest that diabetes is present in 18.4% of adults in India with pulmonary tuberculosis and in 23.5% of those with smear-positive tuberculosis, despite a national adult diabetes prevalence of 4.3%. This result is comparable to that of a recent study in Mexico, which found a diabetes prevalence of 35% among tuberculosis patients in a district with an adult diabetes prevalence of 5.3%[5].
Estimates of the urban/rural distribution of the annual risk of tuberculosis infection suggest that, on average, smear-positive tuberculosis incidence in India is 69.2% higher in urban compared with rural areas. Crowded living conditions in urban districts are one possible factor. However, the increased prevalence of diabetes in urban areas may also play a role – according to our calculations, diabetes is responsible for the urban incidence of smear-positive tuberculosis being 15.2% greater than that in rural areas, or approximately a fifth of the total difference. Our results suggest therefore that the increased diabetes prevalence associated with the rapid urbanization taking place in India has important implications for tuberculosis control.
Our findings are subject to the general caveats applied to attributable risk estimates, for example that we assume a causal association, that other risk factors for tuberculosis are equally distributed across those with and without diabetes, and that those made more susceptible to infection by diabetes are fully exposed to the tuberculosis risk. One underlying risk factor for tuberculosis that may not be equally distributed between those with and without diabetes in India is poverty. Consistent with this a recent case control study from India of risk factors for TB found a univariate odds ratio of 1.8 for previously diagnosed diabetes, which strengthened to 2.44 when controlling for other risk factors, including low socio economic status[19]. However, even allowing for an uneven distribution in other risk factors between those with and without diabetes our attributable risk estimates may well be conservative because our prevalence figures for diabetes are conservative. A large study measuring the prevalence of diabetes in urban areas in India reported that 12.1% of adults had diabetes[10], compared with an urban prevalence of 5.6% found by the study used in our calculations[3]. Recalculating the Attributable Fraction (Population) using this higher value suggests that in urban areas this could be as high as 33.3% (7.4% to 64.2%) for pulmonary tuberculosis and 42.5% (19.0% to 66.2%) for smear-positive tuberculosis. Additionally, we have not considered the contribution to tuberculosis risk from hyperglycaemia below the diabetic threshold. Published data on the association between non-diabetic hyperglycaemia and tuberculosis are rare. However, a recent case control study from Indonesia[24] reported an odds ratio for the risk of tuberculosis associated with impaired fasting glucose (4.2, 95% CIs 1.5–11.7) as similar to that for diabetes (4.7, 2.7 – 8.1). The prevalence of impaired fasting glucose and of impaired glucose tolerance tend to be similar to or higher than the prevalence of diabetes[8, 10], and thus the overall impact of hyperglycaemia may be even higher than our estimates presented here suggest. Population-level measures for managing hyperglycaemia may potentially be cost-effective simply in terms of their benefit to tuberculosis control.
Limitations and strengths
A consequence of using separate studies for the different estimates used in our calculations is an inability to account for the inherent biases of each contributing study. However, so long as each of the studies is independently valid, this does not invalidate our conclusions as long as the assumptions involved are clearly understood.
In deciding on which study to use for the relative risk estimates we searched thoroughly for studies describing the association between tuberculosis and diabetes, and have critically reviewed these studies elsewhere [25]. There is consistent evidence from a number of studies, with different designs and from geographically diverse areas that diabetes is associated with an increased risk of tuberculosis, with an overall increased risk around 1.5 to 8 times higher. However, there are several limitations in the published studies, concerning in particular sample size, the case definitions used for diabetes and tuberculosis, the assessment and control for potential confounders and the fact that most do not provide age specific relative risks or odds ratios[25].
We chose to use relative risk estimates from the study in Korea[17] for several reasons. Firstly, the lack of robust studies reporting age specific relative risk estimates on the association between diabetes and TB from India meant that we had to look elsewhere. Secondly, the study from Korea is the only genuine prospective cohort study on this topic in the past 20 years, and thirdly it is one of only two studies we found that provided age specific relative risk estimates. In addition, based on chest X-rays at baseline the study was able exclude reactivation of pulmonary TB and assess the association of diabetes with new cases. It is, however, important to acknowledge the study's shortcomings. In particular the definition of diabetes was based on unconventional glucose cut points (i.e. 150 mg/dl for fasting and 180 mg/dl post prandial – as opposed to 140 and 200 mg/dl respectively as recommended by WHO at that time). In addition, the diagnosis of diabetes was based on glucose measurement at one point in time, rather than repeated measurements to confirm the diagnosis. This is common to virtually all epidemiological studies of diabetes but is likely to result in significant misclassification of cases of diabetes due to a mixture of biological variation in blood glucose levels and measurement error. It is likely that this led to an underestimate of the association between diabetes and tuberculosis. The crude prevalence of diabetes was low, being 1.2% in men and only 0.2% in women, and there were only 3 women with diabetes (out of 320) who developed TB, and thus sex and age specific relative risk estimates were not available. A further limitation is that there are likely to be underlying confounding factors that we have not accounted for. One of these is smoking, which is implicated as a risk factor for both diabetes and tuberculosis. Further work could include adjusting the diabetes-associated risk of tuberculosis incidence for the effect of smoking.
Nonetheless the study was large and well-structured, similar age specific relative risk estimates were found by a group working in Mexico[5], and the physiological mechanisms underlying diabetes-associated susceptibility are unlikely to vary between populations.
Our finding that diabetes is more strongly related to smear positive than smear negative TB reflects the greater relative risks of diabetes for this form of TB found in the study from Korea (see table 1). This relatively greater association between diabetes and smear positive TB compared to smear negative pulmonary TB, has been found in most, but not all, studies that have addressed this issue [25]
The strengths of our study are that the estimates used are taken from reliable, published sources, chosen after a consideration of the available options, and we explore a new hypothesis using a straightforward and transparent method. Further, our study represents the first attempt we are aware of to quantify the population impact of diabetes on tuberculosis in India.
Implications
Currently, the future impact of tuberculosis control programmes is predicted from knowledge of the effects of chemotherapy and how it is modified by the HIV epidemic. The findings we report indicate that diabetes also has a considerable effect on tuberculosis epidemiology, and so it is important to adapt tuberculosis programme forecasts to incorporate additional risk factors.
The importance of the association between diabetes and tuberculosis is highlighted by the immediate relevance to the UN Millennium Development Goals, as it offers opportunities for reducing the death rate from tuberculosis, and improving its detection and treatment. It is widely recognised that HIV makes a substantial contribution to the global tuberculosis crisis. It is also known that cooperation to target HIV and tuberculosis simultaneously is crucial for the control of both diseases. In India, HIV accounts for 3.4% of adult tuberculosis incidence[2]; the proportion we estimate to be attributable to diabetes is 14.8%. The impact of diabetes on tuberculosis is therefore already considerable, and the predictions of a diabetes epidemic suggest this is likely to escalate.
In the past, an association between tuberculosis and diabetes was widely accepted. Indeed, half a century ago expert clinics were established for "tuberculous diabetics" and appeared to be successful in reducing the otherwise high mortality rate[26]. Today, however, the potential public health and clinical importance of this relationship seems to be largely ignored. For example, national clinical and policy guidance in the UK on the control of tuberculosis does not consider the relationship with diabetes[27]. The World Health Organization's new "Stop TB Strategy" refers to the problem of TB in "high-risk groups" including people with diabetes[28], but WHO has not yet made specific recommendations concerning the relationship between the two conditions. The recently published international standards for TB care give only cursory mention to diabetes [29, 30]. There are, however, some guidelines, such as those from American Thoracic Society[31], which explicitly recommend screening for latent tuberculosis in patients with diabetes and a low threshold of investigation for tuberculosis in people with diabetes with unexplained symptoms. There is a need for new research to guide policy and practice in this area. This includes the need for robust studies of the association between the two conditions, particularly from parts of the world such as India where diabetes is increasing rapidly and TB remains highly endemic. There is evidence that people with TB and diabetes have worse TB outcomes than those without diabetes[25]. For example, a study from Indonesia found that people with diabetes are more than twice as likely to remain sputum culture positive at the end of treatment[32]. The potential impact of diabetes on the success of TB treatment and hence appropriate treatment strategies for those with the two diseases deserves investigation in other parts of the world. Another area worthy of investigation is the potential cost effectiveness of screening people with diabetes for TB in highly endemic areas where diabetes is now common.