- Top of page
The double burden of infectious and chronic diseases in low-income countries is becoming a global health problem. Due to the nutrition and lifestyle transition, that is, increased intake of refined sugar and fat in combination with physical inactivity (Kuhnlein & Receveur 1996; Torun et al. 2002; Yajnik 2004), even among the poor, those already burdened by infectious diseases are now also at risk of chronic metabolic diseases.
In 2011, around 366 million people had diabetes, of whom 4 million died of diabetes-related causes (Roglic & Unwin 2010; Whiting et al. 2011). By 2030, the number of people with diabetes is projected to reach 552 million (Whiting et al. 2011). The increase is mainly driven by changes in low- and middle-income countries, which also harbour 95% of the global burden of TB. This is of particular concern, as there is evidence to suggest that diabetes increases the risk of TB (Jeon & Murray 2008; Young et al. 2009), which will further encumber already burdened TB control programmes.
Diabetes may also worsen the clinical manifestations and treatment outcomes of TB (Dooley & Chaisson 2009; Ruslami et al. 2010a; Baker et al. 2011). More specifically, a couple of studies have reported more severe manifestations of TB in people with diabetes (Wang et al. 2005, 2009; Restrepo et al. 2007; Faurholt-Jepsen et al. 2012) and prolonged time to TB culture conversion and cure (Güler et al. 2007; Dooley et al. 2009; Wang et al. 2009), which may cause more frequent recurrence of TB (Sasaki et al. 2003). Finally, although diabetes has been associated with increased mortality in retrospective studies (Patel et al. 1977; Oursler et al. 2002; Dooley et al. 2009; Wang et al. 2009), only one small study reports mortality data from a low-income country (Mboussa et al. 2003). An inherent limitation of the available studies is that TB to some extent causes stress hyperglycaemia (Başoğlu et al. 1999; Gearhart & Parbhoo 2006), which may affect the diagnosis of diabetes. Hence, the associations between diabetes and TB found in case–control studies, the associations between diabetes and severe clinical manifestations based on cross-sectional studies and between diabetes and poor treatment outcomes, could be affected by reverse causality.
Based on studies in Mwanza, Tanzania, we have recently reported that diabetes was strongly associated with incident TB (Faurholt-Jepsen et al. 2011). However, in contrast to previous studies, in this case–control study, an attempt was made to adjust for the acute phase response to reduce the significance of stress hyperglycaemia. Based on follow-up data, we now present data on the role of diabetes on treatment outcome (i.e. sputum culture conversion and mortality) among patients undergoing TB treatment.
- Top of page
Of the 1250 TB patients recruited for the nutritional intervention studies, 40.8% (n = 510) were women, mean (SD) age was 36.5 years (12.9), 50.6% (n = 633) were HIV co-infected, and 16.4% (n = 197) were diagnosed with diabetes as previously reported (Faurholt-Jepsen et al. 2011; PrayGod et al. 2011, 2012). Background characteristics are shown in Table 1.
Table 1. Background characteristics of 1250 patients with pulmonary tuberculosis (TB)
|Age, years [mean (SD)]||36.5 (12.9)|
|Female sex||510 (40.8)|
|Msukuma tribe||570 (45.6)|
|Other tribes||679 (54.4)|
|Alpha-1 glycoprotein, g/l [mean (SD)]||2.4 (0.9)|
|Sputum negative pulmonary TB (PTB−)||427 (34.2)|
|Sputum positive pulmonary TB (PTB+)||823 (65.8)|
|HIV infected||633 (50.6)|
Of 1102 (88.2%) patients with diabetes and TB culture data available, 735 (66.7%) had positive sputum cultures at baseline. Among these PTB+ patients, 5.5% and 1.0% continued to be culture positive after 2 and 5 months of TB treatment, respectively. There were no differences between participants with and without diabetes regarding the proportion and intensity of positive cultures at any time point (Table 2). However, among those with a positive culture at baseline, 10 and 19 patients with or without diabetes, respectively, had died before the 5-months follow-up visit. Thus, if these deaths were due to treatment failure and considered culture positive at 5 months, then the proportion of TB patients with a positive culture at 5 months would have been higher among those with diabetes comorbidity (12.6% vs. 5.0%. P = 0.01). Participants with diabetes more often had missing culture data – due to either missing sample or contamination – at both 2 (26.9% vs. 18.9%, P = 0.010) and 5 months (28.4% vs. 20.7%, P = 0.203).
Table 2. Tuberculosis culture grading and conversion among 735 tuberculosis (TB) patients with available baseline diabetes and TB culture data stratified by diabetes status
| ||Patients without diabetes||Patients with diabetes|| P |
|(n = 617)||(n = 118)|
|Baseline TB culture grading, n (column%)|
|+1 (incl. 1–19 colonies)||103 (16.7)||26 (22.0)||0.373|
|+2||87 (14.1)||15 (12.7)|
|+3||427 (69.2)||77 (65.3)|
|Follow-up TB culture grading, n (column%)|
|Negative||436 (94.2)||76 (96.2)||0.465|
|Total positive TB culture (non-conversion)||27 (5.8)||3 (3.8)|
|+1 (incl. 1–19 colonies)||25 (5.4)||1 (1.3)|
|+2||0 (0.0)||0 (0.0)|
|+3||2 (0.4)||2 (2.5)|
|Negative||440 (99.1)||76 (98.7)||0.741|
|Total positive TB culture (non-conversion)||4 (0.9)||1 (1.3)|
|+1 (incl. 1–19 colonies)||0 (0.0)||0 (0.0)|
|+2||1 (0.2)||0 (0.0)|
|+3||3 (0.7)||1 (1.3)|
Complete survival data up to at least 12 months were available for 1239 (99.1%) participants and were included in the mortality analyses. Median (interquartile range) follow-up time was 647 (492–833) days, and 172 (13.9%) died within the study period. Half of the study participants died in hospital, the other half died at home. Of the 172 (13.9%) deaths, 39 (3.1%) occurred within the first 60 days, 64 (5.2%) within 100 and 138 (11.1%) within 365 days. Within the first 60 days, 10 of 39 (25.6%) patients reported that the deceased had unknown diabetes status. Among the 1205 with available diabetes data, the proportion of deaths involving a patient with diabetes was 41.4, 34.0, and 24.4% after 60, 100 and 365 days, respectively.
In univariate Cox proportional hazards analysis, diabetes did not seem to be associated with mortality (crude RR 1.35, 95% CI 0.91; 2.00, P = 0.14), but this was due to a time-dependant association (score process test, P = 0.02) with excess mortality risk exclusively within the initial 100 days of treatment (crude RR 2.70, 95% CI 1.53; 4.77, P = 0.001). After the initial 100 days, diabetes was not associated with excess mortality (RR 0.80, 95% CI 0.45; 1.43, P = 0.45). Furthermore, PTB− compared to PTB+ (crude RR 2.47, 95% CI 1.83; 3.33, P < 0.001) and HIV infected compared to HIV uninfected (crude RR 3.04, 95% CI 2.16; 4.29, P < 0.001) were associated with mortality.
The association between major morbidities (i.e. diabetes, TB, HIV) and risk of mortality was assessed in (i) a Kaplan–Meier survivor function (Figure 1, diabetes and HIV only), (ii) in a multivariate Cox proportional hazards analysis (Table 3, model 1), and (iii) in a multivariate analysis further adjusted for age (centred on the median, 34 years) and sex (Table 3, model 2). There was an interaction between diabetes and HIV (model 1: P = 0.02; model 2: P = 0.07), which was due to a higher diabetes-associated risk of mortality among HIV uninfected, and, similarly, the highest HIV-associated risk of mortality among non-diabetes participants. In the adjusted model (model 2), diabetes was associated with a fivefold increased risk of mortality (RR 5.09, 95% CI 2.36; 11.02, P < 0.001) among HIV uninfected and within the initial 100 days of treatment. Diabetes was also associated with mortality in HIV infected individuals during the first 100 days, but not later. PTB− was associated with a twofold increased mortality risk compared to PTB+. Due to an interaction between age and HIV (P < 0.001), the model was centred on the median age (34 years). Thus, the HIV-associated mortality risk of RR 2.23 (95% CI 1.01; 4.90, P = 0.046) and 4.86 (95% CI 2.93; 8.05, P < 0.001) among participants with and without diabetes, respectively, reflects the risk for participants aged 34 years. Further, adjustment with AGP did not affect the estimates (model 3).
Table 3. Relative risk for death during tuberculosis treatment among 1239 patients with pulmonary tuberculosis (TB)
| ||Model 1a||Model 2b||Model 3c|
|RR||95% CI|| P ||RR||95% CI|| P ||RR||95% CI|| P |
|Diabetes (<100 days)d|
|HIV negative||5.48||2.56; 11.72||<0.001||5.09||2.36; 11.02||<0.001||4.95||2.28; 10.76||<0.001|
|HIV positive||1.95||1.01; 3.76||0.048||2.33||1.20; 4.53||0.012||2.18||1.11; 4.41||0.024|
|Diabetes (>100 days)d|
|HIV negative||1.60||0.74; 3.45||0.229||1.37||0.62; 3.06||0.435||1.41||0.64; 3.13||0.396|
|HIV positive||0.59||0.29; 1.11||0.100||0.63||0.32; 1.25||0.187||0.62||0.31; 1.24||0.180|
|Non-diabetes||3.14||2.06; 4.79||<0.001||4.86||2.93; 8.05||<0.001||4.86||2.93; 8.06||<0.001|
|Diabetes||1.12||0.55; 2.27||0.762||2.23||1.01; 4.90||0.046||2.14||0.97; 4.76||0.061|
|Tuberculosis culture status|
|Culture positive (PTB+)||Ref.||–; –||–||–||–; –||–||Ref.||–; –||–|
|Culture negative (PTB−)||1.88||1.36; 2.58||<0.001||2.02||1.44; 2.82||<0.001||1.91||1.33; 2.77||0.001|
Figure 1. Mortality among 1239 pulmonary tuberculosis patients with and without diabetes and HIV. TB, tuberculosis; DM-, non-diabetes; DM+, diabetes; HIV-, HIV uninfected; HIV+, HIV infected.
Download figure to PowerPoint
The analysis used in model 2 also revealed that men had a 53% higher mortality risk compared to women (RR 1.53, 95% CI 1.09; 2.16, P = 0.015). The role of the nutritional intervention was tested in a stratified analysis, but had no effect on the risk of mortality or any of the other covariates and was therefore not included in the final model. Also, treatment with ART known to cause metabolic changes (Carr et al. 1999) did not alter the estimates of the Cox regression (data not shown).
- Top of page
We have recently reported that diabetes is strongly associated with risk of pulmonary TB in Tanzania, and here, we show that diabetes is highly associated with mortality during the first 100 days of TB treatment. The main methodological issue in this and other studies on the role of diabetes in relation to risk of TB, severity of manifestations, treatment outcomes and survival is that often diabetes status is assessed after development and diagnosis of TB disease. The acute phase response caused by serious infections comprises hyperglycaemia (Başoğlu et al. 1999; Gearhart & Parbhoo 2006). It is therefore possible that some of the pulmonary TB patients with blood glucose values suggestive of diabetes (Alberti et al. 1998) have non-diabetes stress-induced hyperglycaemia rather than diabetes. Therefore, estimates of the associations between diabetes and risk of TB, severity of TB, treatment outcomes and mortality may be affected by reverse causality, that is, that pulmonary TB causes non-diabetes stress hyperglycaemia misclassified as diabetes. Recently, a systematic review reported decreasing levels of hyperglycaemia during TB treatment (Jeon et al. 2010), indicating the existence of stress hyperglycaemia at the start of TB treatment. Therefore, we measured serum concentrations of the acute phase reactant AGP and adjusted the survival analysis for elevated levels to see whether the diabetes-associated mortality was simply due to a higher inflammatory response. However, this adjustment had no effect on any of the mortality estimates.
If TB patients with more severe disease, accompanied by an acute phase response that leads to stress hyperglycaemia, are incorrectly diagnosed with diabetes, this leads to an overestimation of the true association between diabetes and mortality. In this situation, adjustment for elevated AGP is intended to reduce the significance of this reverse causality. If diabetes disease leads to more severe TB, which is accompanied by an acute phase response and cause mortality, then adjustment for elevated AGP is expected to remove or reduce the estimate of a true causal association between diabetes and mortality. However, after the adjustments, we still found that a diagnosis of diabetes was associated with increased risk of dying. This suggests that the association found is neither due to reverse causality nor more advanced TB disease. Thus, the current data do not explain why diabetes comorbidity was a major risk factor for mortality within the early phase of treatment and why diabetes was not a long-term risk factor.
We recently reported that diabetes had little, and possibly no clinical relevant, impact on the clinical presentation of TB, when the patients were enrolled in the study (Faurholt-Jepsen et al. 2012). Despite few baseline effects, our findings reflect that diabetes has a negative effect on treatment outcome. Also, our analyses suggest that treatment failure attributable to diabetes might mediate an effect of diabetes on mortality. In fact, several studies have reported that pulmonary TB patients with diabetes had more advanced TB disease than patients without diabetes and that diabetes was associated with delayed sputum conversion and reduced cure rate (Wang et al. 2005, 2009; Alisjahbana et al. 2007; Restrepo et al. 2007). Indeed, the latter could be explained by an effect of diabetes on absorption and metabolism of TB drugs, as a study from Indonesia found lower plasma levels of rifampicin in TB patients with diabetes during the continuation phase of treatment (Nijland et al. 2006). The effect on absorption could possibly be mediated through increased body weight and differences in hepatic induction. However, the study found no indication of a diabetes-associated rifampicin difference during the initial phase of TB treatment (Ruslami et al. 2010b). In our study, those with a diagnosis of diabetes had only slightly elevated serum acute phase reactants and neutrophil counts (Faurholt-Jepsen et al. 2012), but due to the high early mortality rate before the 2-month follow-up among patients with diabetes, it is possible that there was a larger proportion of non-converters in this group. From a study in Maryland, USA, it was reported that diabetes delayed time to conversion from 39 to 49 days, but with similar conversion rates after 60 days (Dooley et al. 2009). Also, in an Indonesian study, diabetes was associated with a positive culture at the end of treatment (Alisjahbana et al. 2007). As we did not follow sputum conversion during the intensive phase (<60 days), we cannot exclude the possibility that the apparent effect of diabetes on mortality was explained by early impairment of the treatment effect. Besides, while one of six patients was diagnosed with diabetes at baseline, almost half of the deaths occurring during the first 60 days involved a patient with diabetes. Thus, it is possible that death among those with diabetes was due to treatment failure. Furthermore, the diabetes status of 10 patients dying in this period was unknown, and we can only speculate whether diabetes was involved in these deaths.
There are divergent data on the role of diabetes on TB mortality, and few data are available from low-income countries (Baker et al. 2011), which harbour the largest burden of TB. Two studies from Indonesia and Saudi Arabia found no differences in mortality risk among TB patients with or without diabetes (Singla et al. 2006; Alisjahbana et al. 2007), whereas three smaller, retrospective studies, two from the USA and one from Taiwan, reported increased mortality risk among patients with diabetes (Oursler et al. 2002; Dooley et al. 2009; Wang et al. 2009). The study from Indonesia (Alisjahbana et al. 2007), a lower-middle-income country, was the only available prospective study and used similar methods, but based their diabetes diagnosis solely on FBG and excluded all patients with impaired fasting glycaemia. However, the study had very little mortality, and this was not associated with diabetes (Alisjahbana et al. 2007). As in our study, they did not find baseline differences on bacteriological examination between patients with or without diabetes, but diabetes comorbidity was associated with a higher treatment failure (not cured) at the end of treatment. Despite more severe TB among those with diabetes, only two deaths were reported from the study at the end of treatment, both of which had diabetes comorbidity.
The remaining four studies were all retrospective, with the largest study reporting no effect of diabetes on mortality among TB inpatients (Singla et al. 2006). In one study from Maryland, USA, the overall reported mortality was 9% with the odds of death twice as high among TB patients with diabetes and with increasing risk after adjustment for HIV status (Dooley et al. 2009). The patients were considered to have diabetes, if they had a history of elevated random non-fasting samples or from medical records; thus, the diagnosis was not obtained during the study. Also from another TB population from Maryland the reported deaths were 21% with the risk of mortality being almost seven times higher among patients with diabetes (Oursler et al. 2002). However, there is no indication of how the diabetes diagnosis was obtained, whether the patients were treated or suffered from insulin-dependent diabetes (type I). Also, it is surprising that the mortality in general was very high in the US studies, but the reason for this remains unclear. Finally, in a retrospective study from Taiwan with 217 in- and outpatients without HIV comorbidity, 11.1% were reported dead with the odds of dying being more than seven times higher among patients with diabetes comorbidity (Wang et al. 2009). The diabetes diagnosis was based on historical records or FBG suggestive of diabetes, whereas patients who did not have elevated FBG were chosen as controls. None of the previous studies have reported a time-dependent mortality risk among patients with diabetes comorbidity. If diabetes in general is associated with early TB mortality, timing of the diabetes diagnosis as well as subsequent treatment may be essential for survival. By splitting the data at 100 days, it seems as if this period is the most vulnerable for the TB patients. We also tested the data by splitting at 60 and 80 days, which similarly was accepted by the model (not shown), but we chose reporting the effect as we did to ensure future interventions based on data like ours would not end prematurely.
Ideally, patients with a test result compatible with diabetes should have a confirmatory diabetes test, after the acute phase response has faded. If confirmed, then lifestyle changes or medical intervention could be introduced. However, if indeed diabetes increases early mortality, then there is a need for diagnostic tools to distinguish between stress hyperglycaemia and diabetes and for studies to determine the effects of glycaemic control among diabetics during TB treatment.
In most low-income settings, people do not know whether they have diabetes or not. If diabetes (or hyperglycaemia) is associated with excess mortality, a routine glucose screening at the TB clinics may be necessary to identify those at increased risk of death due to diabetes. As we missed the blood glucose screening in 10 of 39 deaths in the first 60 days, it may be necessary to improve the timing and give the blood glucose test high priority immediately after TB diagnosis. In addition to the identification of patients at risk, it is equally important to improve the management of such patients; better supervision of TB treatment and more frequent routine visits during the initial treatment phase could be one approach, whereas a medical intervention to normalise the blood glucose levels should be tested under safe conditions. The use of intensive insulin therapy to control hyperglycaemia in severe illness has proven difficult to manage (Griesdale et al. 2009) and may not even be feasible in a resource poor setting.