## Abstract

Background A growing body of evidence supports the role of type 2 diabetes as an individual-level risk factor for tuberculosis (TB), though evidence from developing countries with the highest TB burdens is lacking. In developing countries, TB is most common among the poor, in whom diabetes may be less common. We assessed the relationship between individual-level risk, social determinants and population health in these settings.

Methods We performed individual-level analyses using the World Health Survey (n = 124 607; 46 countries). We estimated the relationship between TB and diabetes, adjusting for gender, age, body mass index, education, housing quality, crowding and health insurance. We also performed a longitudinal country-level analysis using data on per-capita gross domestic product and TB prevalence and incidence and diabetes prevalence for 1990–95 and 2003–04 (163 countries) to estimate the relationship between increasing diabetes prevalence and TB, identifying countries at risk for disease interactions.

Results In lower income countries, individuals with diabetes are more likely than non-diabetics to have TB [univariable odds ratio (OR): 2.39; 95% confidence interval (CI): 1.84–3.10; multivariable OR: 1.81; 95% CI: 1.37–2.39]. Increases in TB prevalence and incidence over time were more likely to occur when diabetes prevalence also increased (OR: 4.7; 95% CI: 1.0–22.5; OR: 8.6; 95% CI: 1.9–40.4). Large populations, prevalent TB and projected increases in diabetes make countries like India, Peru and the Russia Federation areas of particular concern.

Conclusions Given the association between diabetes and TB and projected increases in diabetes worldwide, multi-disease health policies should be considered.

## Introduction

When chronic non-communicable diseases proliferate faster than infectious diseases recede, previously uncommon disease interactions can take on population health significance. Recent systematic reviews1,2 suggest that type 2 diabetes mellitus (T2DM) increases individual risk of Mycobacterium tuberculosis (TB) disease. Country-level analyses suggest that TB prevalence is mediated by both social determinants and public health strategies.3,4 Yet, strikingly little work has been done to assess the relationship of TB and diabetes at the individual level in countries where TB prevalence is highest and diabetes prevalence is rising most rapidly. Given increases in diabetes and the persistence of TB in these areas, the relationship of individual risks, social determinants and population health impacts due to interactions between diabetes and TB should be assessed.

Classical descriptions of epidemiologic transition involve the replacement of infectious diseases of deprivation with non-communicable diseases of affluence,5 though currently many countries face growing dual disease burdens.6 Diabetes prevalence continues to rise rapidly in developing countries, driven by changes in diet and lifestyle.7 Though historically having fewer diabetic individuals than their share of the world’s population would imply, by 2025, India, China, Indonesia, Pakistan and Brazil alone are projected to carry nearly half the world’s diabetes burden.7 In many countries, TB epidemics continue, fueled by drug resistance,8 HIV/AIDS9,10 and social inequalities.4

Analyses that inform both TB and diabetes policies must estimate the individual-level relationship between the two conditions and translate it into projections of future population burden, accounting for mediating effects of social determinants and prevailing health policies. Though few such studies exist, it is suggested that the impact of a diabetes/TB interaction may play a substantial role in fueling the ongoing TB epidemic in India.11 Given existing data limitations, such studies have necessarily depended on the assumption that the strength of association between diabetes and TB risk is the same as that estimated almost exclusively from non-population-representative studies from higher income countries.

In order to appreciate the global population health significance of rising diabetes prevalence in the presence of persistent TB epidemics in settings where current TB burdens are high, we evaluated the relationship of diabetes and TB at both individual and country levels using population-representative data from largely lower income countries.

## Methods

An individual-level cross-sectional analysis examined the association between symptoms of TB and self-reported diagnosed diabetes conditioning on known TB risk factors across individuals from 46 countries. This analysis was repeated for countries grouped by geographic region to assess how the relationship as well as mediating influences of social determinants varied across regions. A country-level analysis examined the association between diabetes and TB prevalence across 163 countries. Its goals were to assess the impact of longitudinal increases in diabetes prevalence on changes in TB prevalence and incidence; and identify countries potentially ‘at risk’ for TB/diabetes interactions because of the high prevalence of both.

### Individual-level analysis

Data were derived from individual responses to the World Health Organization’s (WHO) 2002–03 World Health Survey (WHS).12,13 This analysis included individuals in countries that had sampling weights available and used the WHS long-form questionnaire, providing self-reported diabetes diagnosis and symptoms of TB. The main analysis employed data on individual adults (n = 124 545; representing approximately 400 million people) from all 46 countries for which these data are available (Table 1).

Table 1

Countries from the WHS included in the individual-level analysis

Country N (survey responses) Actual population representeda Per-capita GDP (PPP I$2005) Congo 1010 310 000 244 Ethiopia 46 307 000 546 Malawi 4585 3 111 000 608 Nepal 12 14 000 910 Burkina Faso 1529 1 052 000 939 Mali 349 445 000 944 Bangladesh 421 5 183 000 953 Chad 2106 1 211 000 987 Ghana 2897 4 931 000 1053 Zambia 2052 2 086 000 1066 Comoros 1523 103 000 1127 Kenya 3521 11 500 000 1274 Senegal 641 433 000 1401 Mauritania 1416 404 000 1566 Laos 4449 2 187 000 1573 Cote d’Ivoire 1768 3 353 000 1672 Vietnam 2974 35 400 000 1780 India 4380 82 700 000 1816 Pakistan 1481 14 300 000 1937 Georgia 2410 2 550 000 2649 Philippines 7518 32 800 000 2686 Sri Lanka 3833 6 075 000 3092 China 3812 3 780 000 3115 Morocco 1050 4 871 000 3212 Paraguay 4541 2 172 000 3640 Namibia 2894 555 000 4140 Swaziland 1005 129 000 4256 Ukraine 583 7 548 000 4324 Dominican Republic 2639 2 412 000 5192 Bosnia Herzegovina 776 2 028 000 5448 Tunisia 3349 3 427 000 5677 Ecuador 1549 2 880 000 5879 Kazakhstan 3957 8 654 000 6748 Uruguay 2763 2 129 000 7598 South Africa 1016 8 170 000 7733 Brazil 380 8 795 000 8017 Mauritius 2186 396 000 9070 Russia 3223 47 400 000 9549 Malaysia 4316 9 133 000 10 420 Mexico 23 283 37 800 000 10 815 Croatia 888 3 000 000 11 551 Estonia 884 1 208 000 12 921 Slovakia 1139 2 571 000 13 660 Czech Republic 650 7 736 000 17 635 Spain 5788 29 600 000 25 922 United Arab Emirates 955 793 000 40 712 Country N (survey responses) Actual population representeda Per-capita GDP (PPP I$2005)
Congo 1010 310 000 244
Ethiopia 46 307 000 546
Malawi 4585 3 111 000 608
Nepal 12 14 000 910
Burkina Faso 1529 1 052 000 939
Mali 349 445 000 944
Bangladesh 421 5 183 000 953
Chad 2106 1 211 000 987
Ghana 2897 4 931 000 1053
Zambia 2052 2 086 000 1066
Comoros 1523 103 000 1127
Kenya 3521 11 500 000 1274
Senegal 641 433 000 1401
Mauritania 1416 404 000 1566
Laos 4449 2 187 000 1573
Cote d’Ivoire 1768 3 353 000 1672
Vietnam 2974 35 400 000 1780
India 4380 82 700 000 1816
Pakistan 1481 14 300 000 1937
Georgia 2410 2 550 000 2649
Philippines 7518 32 800 000 2686
Sri Lanka 3833 6 075 000 3092
China 3812 3 780 000 3115
Morocco 1050 4 871 000 3212
Paraguay 4541 2 172 000 3640
Namibia 2894 555 000 4140
Swaziland 1005 129 000 4256
Ukraine 583 7 548 000 4324
Dominican Republic 2639 2 412 000 5192
Bosnia Herzegovina 776 2 028 000 5448
Tunisia 3349 3 427 000 5677
Ecuador 1549 2 880 000 5879
Kazakhstan 3957 8 654 000 6748
Uruguay 2763 2 129 000 7598
South Africa 1016 8 170 000 7733
Brazil 380 8 795 000 8017
Mauritius 2186 396 000 9070
Russia 3223 47 400 000 9549
Malaysia 4316 9 133 000 10 420
Mexico 23 283 37 800 000 10 815
Croatia 888 3 000 000 11 551
Estonia 884 1 208 000 12 921
Slovakia 1139 2 571 000 13 660
Czech Republic 650 7 736 000 17 635
Spain 5788 29 600 000 25 922
United Arab Emirates 955 793 000 40 712

aBased on sample weights from the WHS.

The main outcome was the presence of symptoms of active TB disease. Microbiological confirmation was not possible as the WHS did not include mycobacterial culture or sputum smear microscopy. We classified active TB as an affirmative response to two questions: ‘Over the last 12 months, have you had blood in your phlegm or have you coughed blood?’ and ‘Over the last 12 months, have you experienced cough lasting 3 weeks or longer?’. It has previously been shown that such questions have sensitivities between 65 and 70% and specificities between 55 and 75% for TB with other respiratory conditions also associated with an affirmative answer.14–16

The main predictor was the presence of T2DM. Because the WHS did not test fasting-plasma glucose, we classified individuals as having diabetes if they responded affirmatively to the question: ‘Have you ever been diagnosed with diabetes (high blood sugar)?’ It has previously been shown that the sensitivities and specificities of such questions are ∼65–85% and 95–99%, respectively, compared with biochemical markers or clinical records, with overall Kappas of 0.70–0.85.17–22 Though the cross-sectional nature of the analysis precluded statements about time ordering and causality, we interpreted a positive association between our outcome and predictor as evidence consistent with prior longitudinal studies showing that diabetes increased subsequent risk of developing active TB.1,23–26

The model included covariates that have previously been associated with TB: gender, age categorized in 10-year intervals; body mass index (BMI) categorized as ≤17, 17–20, 20–25, 25–30, ≥30; schooling in years categorized as <1, 1–7, 8–12, >12; current daily smoking and the number of years a person has been a daily smoker; drinking at least 1 drink per day in the previous week; urban or rural residence; number of individuals living in the household, number of individuals per room; socio-economic status defined by whether the house had floors made from tile, cement, brick or wood and walls made from cement, brick, stone or wood, whether the house had toilet facilities flushed to a piped sewage system or septic tank, per-capita household total expenditure in the previous month expressed in year 2003 international dollars calculated using purchasing power parity (PPP)27 and having health insurance.

We estimated univariable associations between TB and each covariate as unadjusted odds ratios (ORs). Given the large sample size, we included all covariates in the multivariable models to obtain adjusted ORs. We fit two multivariable models. The first included all covariates except diabetes to assess how ORs for known TB risk factors changed from the crude analyses when they were simultaneously included. The second included all covariates and diabetes to assess the final adjusted relationship of TB and diabetes.

The multivariable analysis was repeated for individuals from subgroups of the countries formed by geographic regions (Africa, Asia, Latin America and Europe). ORs were compared for consistency of relationships across regions.

To explore the effects of relative socio-economic status on the relationship between TB and diabetes, we augmented the multivariable model with indicator variables for an individual’s relative expenditure (household expenditure tertile) compared either with other individuals in the same country or with individuals in all countries in the same per-capita gross domestic product (GDP) quintile. We also included interactions between diabetes and the relative expenditure measures. Additionally, we assessed interactions between gender, age, smoking and alcohol consumption and measures of relative expenditure.

To assess potential bias due to non-response, we compared the association between TB and each covariate separately for the sample restricted to individuals who responded for all covariates and for all individuals who provided a response for the particular covariate. We assessed the possibility of misclassification of diabetes and TB due to self-report. Since self-reported diagnosed diabetes may not capture true diabetes status, we confirmed that other covariates had expected relationships with self-reported diabetes (e.g. positive association with higher BMI). Collinearity between variables in the multivariable model was assessed using correlation coefficients, with those highly collinear variables deleted one at a time in an analysis to assess the extent to which collinearity impacted ORs and confidence intervals (CIs).

All individual-level analyses used logistic regressions with individual sampling weights (13), country-fixed effects and robust standard errors clustered by country.

### Country-level analysis

Data included adult population size and diabetes prevalence,28,29 WHO estimates of TB prevalence and incidence,30 and per-capita GDP expressed in year 2005 international dollars using PPP.31 Included countries had total adult populations of 3.2 billion in 1995 and 3.6 billion in 2003.

We examined the relationship between TB prevalence and TB incidence, diabetes prevalence, and per-capita GDP longitudinally between 1990–95 and 2003–04. Longitudinal changes in TB incidence, TB prevalence and diabetes prevalence from 1990–95 to 2003–04 were characterized by logistic regression to quantify the association between increases in diabetes and in TB (prevalence or incidence) conditioning on a country’s GDP quintile in 1995. We also identified countries with relatively high TB and diabetes prevalence using a cut-off that also provided sufficient numbers of dual burden countries with WHS data for further individual-level sub-analyses (both values above the 45th percentile across all countries in 2003).

### Software

All analyses were undertaken using Stata/SE 10.1 for Windows (StataCorp, USA).

## Results

### Individual-level analysis

In 46 mainly low- and middle-income countries, individuals reporting a diabetes diagnosis were more likely to have symptoms of TB (coughing blood or blood in sputum, cough lasting 3 weeks or longer) (univariable OR: 2.39; 95% CI: 1.84–3.10; multivariable OR: 1.81; 95% CI: 1.37–2.39).

Table 2 describes the characteristics of included individuals from the 46 WHS countries. Of the study population, 1.4% reported coughing blood or blood in their sputum in the past 12 months as well as cough lasting ≥3 weeks and were classified as having TB, while 3.6% reported being diagnosed with diabetes. The study population included slightly more women than men, and the majority of individuals were <40 years of age. Most individuals were normal weight (BMI 20–25) though 5.5% had a BMI <17%, and 8.7% had a BMI >30. Of individuals, 59.3% reported ≥8 years of schooling. Nearly 20% were smokers, and 8.4% reported drinking at least 1 drink per day. Over two-thirds lived in homes with good floors and walls, but only half lived in homes with good toilet facilities. Approximately half lived in urban areas. There was substantial variation in household size (5th to 95th percentile: 1–11 people), crowding (5th to 95th percentile: 0.3–5.0 people per room) and monthly per-capita household expenditure (5th to 95th percentile: I$3.0–I$474 per person). Approximately 40% of individuals reported having health insurance.

Table 2

Characteristics of WHS respondents

Included in main analyses (N = 124 545)a All responding to a given question

Variable Value N Value
Symptoms of TB (%) 1.4 193 867 1.7
Self-reported T2DM (%) 3.6 206 174 3.5
Male (%) 47.7 227 480 47.8
Age <30 years (%) 30.0 227 116 31.5
Age 30–39 years (%) 21.2 227 116 21.1
Age 40–49 years (%) 18.6 227 116 17.9
Age 50–59 years (%) 13.0 227 116 13.0
Age ≥60 years (%) 17.2 227 116 16.5
BMI <17 (%) 5.5 161 353 6.1
BMI 17–20 (%) 20.7 161 353 44.4
BMI 20–25 (%) 43.9 161 353 20.9
BMI 25–30 (%) 21.2 161 353 20.4
BMI ≥30 (%) 8.7 161 353 8.2
School <1 year (%) 13.5 226 615 23.6
School 1–7 years (%) 27.2 226 615 28.5
School 8–12 years (%) 44.0 226 615 36.0
School  >12 years (%) 15.3 226 615 11.9
Current daily smoker (%) 19.8 223 402 22.7
Time daily smoker, years, [mean (SD)] 3.5 (9.3) 223 402 3.9 (10.0)
At least one drink per day (%) 8.4 216 281 7.3
House with good floor and walls (%) 72.8 216 487 62.1
House with good toilet (%) 48.9 223 217 44.7
Urban (%) 49.4 234 491 45.0
Household members (people), [mean (SD)] 5.1 (2.8) 235 157 5.4 (2.9)
People per room (people), [mean (SD)] 1.8 (1.6) 231 209 2.0 (1.7)
Per-capita household expenditure (I$2005 per month), [mean (SD)] 142.9 (357.1) 225 341 125.0 (344.8) Respondent insured (%) 39.8 199 664 29.8 Included in main analyses (N = 124 545)a All responding to a given question Variable Value N Value Symptoms of TB (%) 1.4 193 867 1.7 Self-reported T2DM (%) 3.6 206 174 3.5 Male (%) 47.7 227 480 47.8 Age <30 years (%) 30.0 227 116 31.5 Age 30–39 years (%) 21.2 227 116 21.1 Age 40–49 years (%) 18.6 227 116 17.9 Age 50–59 years (%) 13.0 227 116 13.0 Age ≥60 years (%) 17.2 227 116 16.5 BMI <17 (%) 5.5 161 353 6.1 BMI 17–20 (%) 20.7 161 353 44.4 BMI 20–25 (%) 43.9 161 353 20.9 BMI 25–30 (%) 21.2 161 353 20.4 BMI ≥30 (%) 8.7 161 353 8.2 School <1 year (%) 13.5 226 615 23.6 School 1–7 years (%) 27.2 226 615 28.5 School 8–12 years (%) 44.0 226 615 36.0 School >12 years (%) 15.3 226 615 11.9 Current daily smoker (%) 19.8 223 402 22.7 Time daily smoker, years, [mean (SD)] 3.5 (9.3) 223 402 3.9 (10.0) At least one drink per day (%) 8.4 216 281 7.3 House with good floor and walls (%) 72.8 216 487 62.1 House with good toilet (%) 48.9 223 217 44.7 Urban (%) 49.4 234 491 45.0 Household members (people), [mean (SD)] 5.1 (2.8) 235 157 5.4 (2.9) People per room (people), [mean (SD)] 1.8 (1.6) 231 209 2.0 (1.7) Per-capita household expenditure (I$2005 per month), [mean (SD)] 142.9 (357.1) 225 341 125.0 (344.8)
Respondent insured (%) 39.8 199 664 29.8

All means and standard deviations are weighted using sampling weights.

aThose included in the main analyses (univariable and multivariable) were required to have responded to questions for all covariates in the model.

In univariable analyses (Table 3), TB (coughing blood or blood in sputum and cough lasting ≥3 weeks) is positively associated with reporting a diabetes diagnosis (OR: 2.39; 95% CI: 1.84–3.10), increasing age (≥60 years; OR: 3.54; 95% CI: 2.53–4.94), lower BMI (BMI ≤ 17; OR: 4.01; 95% CI: 2.56–6.27), longer duration of daily smoking (each additional year, OR: 1.02; 95% CI: 1.01–1.04) and greater household crowding (additional person per room, OR: 1.01; 95% CI: 1.01–1.03). TB was negatively associated with more education (>12 years schooling, OR: 0.17; 95% CI: 0.09–0.31), a home with good floors and walls or with a good toilet (OR: 0.59; 95% CI: 0.43–0.83 and OR: 0.82; 95% CI: 0.68–0.98, respectively), an urban location (OR: 0.67; 95% CI: 0.53–0.85), the number of people in the household (additional person, OR: 0.95; 95% CI: 0.93–0.97); and having health insurance (OR: 0.72; 95% CI: 0.52–0.99).

Table 3

Univariable and multivariable relationship between symptoms of TB, self-reported T2DM and other covariates

Univariable analyses Multivariable analysis without T2DM Multivariable analysis with T2DM
Variable Value N TB (%) Unadjusted OR (95% CI) Adjusted OR (95% CI) Adjusted OR (95% CI)
Self-reported T2DM No 120 104 1.33
Yes 4441 3.16 2.39 (1.84–3.10)  1.81 (1.37–2.39)
Sex Female 68 001 1.38
Male 56 544 1.65 1.18 (0.90–1.54) 1.34 (0.92–1.95) 1.33 (0.92–1.93)
Age (years) 20–29 30 081 1.04
30–39 30 411 1.20 1.37 (0.97–1.95) 1.44 (1.01–2.04) 1.44 (1.01–2.05)
40–49 22–981 1.14 1.49 (1.24–1.78) 1.49 (1.20–1.86) 1.46 (1.17–1.83)
50–59 14 951 1.78 2.28 (1.38–3.75) 2.12 (1.32–3.39) 2.03 (1.25–3.29)
≥60 19 931 2.37 3.54 (2.53–4.94) 2.93 (2.10–4.09) 2.74 (1.93–3.89)
BMI (m/kg2<17 4038 4.56 4.01 (2.56–6.27) 3.89 (2.45–6.18) 3.91 (2.47–6.19)
17–20 18 417 1.53 1.48 (1.25–1.75) 1.53 (1.27–1.83) 1.54 (1.28–1.84)
20–25 57 911 1.07
25–30 31 510 1.26 1.24 (0.92–1.68) 1.14 (0.87–1.50) 1.12 (0.85–1.47)
≥30 12 669 1.12 1.12 (0.81–1.54) 0.99 (0.74–1.33) 0.96 (0.72–1.27)
School (years) None 16 874 3.06
1–7 41 342 1.86 0.79 (0.54–1.15) 0.91 (0.60–1.38) 0.90 (0.59–1.37)
8–12 53 655 0.99 0.42 (0.27–0.64) 0.64 (0.39–1.04) 0.64 (0.40–1.04)
>12 12 674 0.28 0.17 (0.09–0.31) 0.27 (0.13–0.53) 0.27 (0.13–0.53)
Current daily smoker No 105 000 1.32
Yes 19 545 1.71 1.26 (0.97–1.62) 0.87 (0.59–1.27) 0.87 (0.59–1.27)
Time daily smokera <8 109 514 1.35 1.02 (1.01–1.03) 1.01 (0.99–1.03) 1.01 (0.99–1.03)
(years) 8–15 4075 1.18
15–25 5579 1.11
>25 5377 2.63
At least one drink per day No 115 135 1.44
Yes 9410 0.97 0.91 (0.53–1.56) 0.84 (0.49–1.46) 0.84 (0.48–1.48)
House with good floor and walls No 32 055 2.40
Yes 92 490 1.02 0.59 (0.43–0.83) 0.75 (0.57–1.00) 0.74 (0.56–0.99)
House with good toilet No 61 049 1.66
Yes 63 496 1.13 0.82 (0.68–0.98) 1.24 (0.94–1.63) 1.23 (0.94–1.62)
Urban No 58 339 1.77
Yes 66 206 1.02 0.67 (0.53–0.85) 0.80 (0.67–0.97) 0.80 (0.66–0.96)
Household membersa  0.95 (0.93–0.97) 0.95 (0.93–0.98) 0.95 (0.93–0.98)
Q1 27 190 0.93
Q2 20 773 1.32
Q3 43 370 1.16
Q4 33 212 1.86
People per rooma  1.03 (0.99–1.07) 1.04 (0.99–1.09) 1.04 (0.98–1.09)
Q1 30 963 0.90
Q2 30 176 1.07
Q3 24 907 1.28
Q4 38 499 2.02
Per-capita household  1 (1.00–1.00) 1 (1.00–1.00) 1 (1.00–1.00)
expenditurea Q1 31 124 1.89
Q2 30 976 1.52
Q3 31 283 1.02
Q4 31 162 0.95
Respondent insured No 80 306 1.77
Yes 44 239 0.83 0.72 (0.52–0.99) 0.85 (0.65–1.12) 0.84 (0.63–1.10)
Univariable analyses Multivariable analysis without T2DM Multivariable analysis with T2DM
Variable Value N TB (%) Unadjusted OR (95% CI) Adjusted OR (95% CI) Adjusted OR (95% CI)
Self-reported T2DM No 120 104 1.33
Yes 4441 3.16 2.39 (1.84–3.10)  1.81 (1.37–2.39)
Sex Female 68 001 1.38
Male 56 544 1.65 1.18 (0.90–1.54) 1.34 (0.92–1.95) 1.33 (0.92–1.93)
Age (years) 20–29 30 081 1.04
30–39 30 411 1.20 1.37 (0.97–1.95) 1.44 (1.01–2.04) 1.44 (1.01–2.05)
40–49 22–981 1.14 1.49 (1.24–1.78) 1.49 (1.20–1.86) 1.46 (1.17–1.83)
50–59 14 951 1.78 2.28 (1.38–3.75) 2.12 (1.32–3.39) 2.03 (1.25–3.29)
≥60 19 931 2.37 3.54 (2.53–4.94) 2.93 (2.10–4.09) 2.74 (1.93–3.89)
BMI (m/kg2<17 4038 4.56 4.01 (2.56–6.27) 3.89 (2.45–6.18) 3.91 (2.47–6.19)
17–20 18 417 1.53 1.48 (1.25–1.75) 1.53 (1.27–1.83) 1.54 (1.28–1.84)
20–25 57 911 1.07
25–30 31 510 1.26 1.24 (0.92–1.68) 1.14 (0.87–1.50) 1.12 (0.85–1.47)
≥30 12 669 1.12 1.12 (0.81–1.54) 0.99 (0.74–1.33) 0.96 (0.72–1.27)
School (years) None 16 874 3.06
1–7 41 342 1.86 0.79 (0.54–1.15) 0.91 (0.60–1.38) 0.90 (0.59–1.37)
8–12 53 655 0.99 0.42 (0.27–0.64) 0.64 (0.39–1.04) 0.64 (0.40–1.04)
>12 12 674 0.28 0.17 (0.09–0.31) 0.27 (0.13–0.53) 0.27 (0.13–0.53)
Current daily smoker No 105 000 1.32
Yes 19 545 1.71 1.26 (0.97–1.62) 0.87 (0.59–1.27) 0.87 (0.59–1.27)
Time daily smokera <8 109 514 1.35 1.02 (1.01–1.03) 1.01 (0.99–1.03) 1.01 (0.99–1.03)
(years) 8–15 4075 1.18
15–25 5579 1.11
>25 5377 2.63
At least one drink per day No 115 135 1.44
Yes 9410 0.97 0.91 (0.53–1.56) 0.84 (0.49–1.46) 0.84 (0.48–1.48)
House with good floor and walls No 32 055 2.40
Yes 92 490 1.02 0.59 (0.43–0.83) 0.75 (0.57–1.00) 0.74 (0.56–0.99)
House with good toilet No 61 049 1.66
Yes 63 496 1.13 0.82 (0.68–0.98) 1.24 (0.94–1.63) 1.23 (0.94–1.62)
Urban No 58 339 1.77
Yes 66 206 1.02 0.67 (0.53–0.85) 0.80 (0.67–0.97) 0.80 (0.66–0.96)
Household membersa  0.95 (0.93–0.97) 0.95 (0.93–0.98) 0.95 (0.93–0.98)
Q1 27 190 0.93
Q2 20 773 1.32
Q3 43 370 1.16
Q4 33 212 1.86
People per rooma  1.03 (0.99–1.07) 1.04 (0.99–1.09) 1.04 (0.98–1.09)
Q1 30 963 0.90
Q2 30 176 1.07
Q3 24 907 1.28
Q4 38 499 2.02
Per-capita household  1 (1.00–1.00) 1 (1.00–1.00) 1 (1.00–1.00)
expenditurea Q1 31 124 1.89
Q2 30 976 1.52
Q3 31 283 1.02
Q4 31 162 0.95
Respondent insured No 80 306 1.77
Yes 44 239 0.83 0.72 (0.52–0.99) 0.85 (0.65–1.12) 0.84 (0.63–1.10)

aThese variables enter the regression models as continuous variables. We present quartiles in the Table to clearly illustrate the change in TB prevalence across the range of variable values.

When we assessed these relationships in the multivariable analysis including all covariates except diabetes (Table 3), the positive association between TB and male gender strengthened (OR: from 1.18 to 1.34). The multivariable adjustment attenuated the apparent effect of age (≥60 years, OR: from 3.54 to 2.93), more education (>12 years schooling, OR: from 0.17 to 0.27), good floor and walls (OR: from 0.59 to 0.75), good toilet (OR: from 0.82 to 1.24), urban location (OR: from 0.67 to 0.80) and health insurance (OR: from 0.72 to 0.85). Other significant ORs changed by <0.05 between univariable and multivariable analyses.

In the multivariable model that included diabetes, TB was positively associated with diabetes (OR: 1.81; 95% CI: 1.37–2.39)—reduced from the univariable analysis (OR: 2.21) (Table 3). The relationship between TB and age was further attenuated (≥60 years, OR: 2.74; 95% CI: 1.93–3.89). The remaining relationships remained virtually unchanged, with ORs differing by ≤0.03.

Repeating the multivariable model for individuals living in African, Asian, European and Latin American countries separately (Table 4), diabetes and TB maintained a remarkably consistent and generally significant relationship (OR: 1.96; 95% CI: 1.23–3.12; OR: 1.74; 95% CI: 0.82–3.72; OR: 2.38 95% CI: 1.08–5.24; OR: 1.99; 95% CI: 1.44–2.75, respectively for each region). TB risk was also positively associated with increasing age most strongly in Africa and Asia; and household crowding most strongly in Europe. TB risk was negatively associated with schooling and urban location.

Table 4

World region comparison of the relationship between symptoms of TB, self-reported T2DM and other covariates*

All countries (N = 124 545) Africa (N = 35 896) Asia (N = 37 153) Europe (N = 16 341) Latin America (N = 35 155)
OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI)
Self-reported T2DM 1.81 1.37–2.39 1.96 1.23–3.12 1.74 0.82–3.72 2.38 1.08–5.24 1.99 1.44–2.75
Male 1.33 0.92–1.93 1.17 0.68–2.04 1.69 1.11–2.56 0.84 0.28–2.49 0.95 0.62–1.44
Age (years)
30–39 1.44 1.01–2.05 1.42 0.90–2.24 1.39 0.88–2.20 4.60 1.92–11.0 1.01 0.87–1.18
40–49 1.46 1.17–1.83 1.36 0.70–2.63 1.63 1.30–2.05 2.00 0.62–6.50 1.12 0.90–1.38
50–59 2.03 1.25–3.29 2.24 1.10–4.56 2.71 1.09–6.79 1.22 0.22–6.74 1.02 0.50–2.07
>60 2.74 1.93–3.89 1.58 1.10–2.28 4.47 3.34–5.98 2.29 0.69–7.55 1.10 0.55–2.17
BMI (years)
<17 3.91 2.47–6.19 0.99 0.48–2.05 4.50 3.12–6.51 1.69 0.51–5.61 0.39 0.14–1.08
17–20 1.54 1.28–1.84 1.32 0.86–2.02 1.49 1.34–1.64 1.99 0.92–4.32 1.93 0.60–6.16
25–30 1.12 0.85–1.47 0.58 0.31–1.09 1.12 0.62–2.02 2.97 0.99–8.88 1.26 0.99–1.62
≥30 0.96 0.72–1.27 1.05 0.69–1.59 0.39 0.14–1.07 2.34 1.21–4.49 0.97 0.71–1.31
School (years)
1–7 0.90 0.59–1.37 0.79 0.60–1.05 1.04 0.53–2.04 0.39 0.23–0.66 0.64 0.41–1.01
8–12 0.64 0.40–1.04 0.48 0.31–0.72 0.65 0.28–1.54 0.38 0.15–0.97 0.53 0.35–0.78
>12 0.27 0.13–0.53 0.23 0.07–0.82 0.26 0.08–0.84 0.24 0.08–0.78 0.04 0.01–0.46
Current daily smoker 0.87 0.59–1.27 0.89 0.45–1.78 0.91 0.64–1.29 0.18 0.02–1.39 1.05 0.82–1.34
Time daily smoker 1.01 0.99–1.03 1.00 0.99–1.01 1.00 0.98–1.03 1.05 0.98–1.13 0.98 0.94–1.03
At least one drink per day 0.84 0.48–1.48 1.55 0.73–3.29 0.76 0.33–1.72 0.77 0.26–2.22 0.51 0.12–2.19
House with good floor and walls 0.74 0.56–0.99 0.71 0.51–1.00 0.79 0.55–1.15 0.56 0.06–5.63 0.82 0.77–0.87
House with good toilet 1.23 0.94–1.62 0.91 0.40–2.09 1.39 0.95–2.04 0.78 0.51–1.21 1.46 1.27–1.68
Urban 0.80 0.66–0.96 0.74 0.36–1.50 0.80 0.61–1.04 1.31 0.76–2.25 0.70 0.57–0.86
Household members (#) 0.95 0.93–0.98 0.99 0.96–1.03 0.92 0.88–0.96 0.97 0.77–1.21 0.94 0.91–0.97
Household members per room 1.04 0.98–1.09 1.03 0.92–1.15 1.03 0.95–1.11 1.58 1.08–2.30 1.07 0.98–1.16
Per-capita household expenditure 1.00 1.00–1.00 1.00 1.00–1.00 1.00 0.99–1.00 1.00 1.00–1.00 1.00 1.00–1.00
Respondent insured 0.84 0.63–1.10 0.62 0.37–1.03 0.96 0.47–1.96 1.09 0.34–3.50 0.93 0.87–0.99
All countries (N = 124 545) Africa (N = 35 896) Asia (N = 37 153) Europe (N = 16 341) Latin America (N = 35 155)
OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI)
Self-reported T2DM 1.81 1.37–2.39 1.96 1.23–3.12 1.74 0.82–3.72 2.38 1.08–5.24 1.99 1.44–2.75
Male 1.33 0.92–1.93 1.17 0.68–2.04 1.69 1.11–2.56 0.84 0.28–2.49 0.95 0.62–1.44
Age (years)
30–39 1.44 1.01–2.05 1.42 0.90–2.24 1.39 0.88–2.20 4.60 1.92–11.0 1.01 0.87–1.18
40–49 1.46 1.17–1.83 1.36 0.70–2.63 1.63 1.30–2.05 2.00 0.62–6.50 1.12 0.90–1.38
50–59 2.03 1.25–3.29 2.24 1.10–4.56 2.71 1.09–6.79 1.22 0.22–6.74 1.02 0.50–2.07
>60 2.74 1.93–3.89 1.58 1.10–2.28 4.47 3.34–5.98 2.29 0.69–7.55 1.10 0.55–2.17
BMI (years)
<17 3.91 2.47–6.19 0.99 0.48–2.05 4.50 3.12–6.51 1.69 0.51–5.61 0.39 0.14–1.08
17–20 1.54 1.28–1.84 1.32 0.86–2.02 1.49 1.34–1.64 1.99 0.92–4.32 1.93 0.60–6.16
25–30 1.12 0.85–1.47 0.58 0.31–1.09 1.12 0.62–2.02 2.97 0.99–8.88 1.26 0.99–1.62
≥30 0.96 0.72–1.27 1.05 0.69–1.59 0.39 0.14–1.07 2.34 1.21–4.49 0.97 0.71–1.31
School (years)
1–7 0.90 0.59–1.37 0.79 0.60–1.05 1.04 0.53–2.04 0.39 0.23–0.66 0.64 0.41–1.01
8–12 0.64 0.40–1.04 0.48 0.31–0.72 0.65 0.28–1.54 0.38 0.15–0.97 0.53 0.35–0.78
>12 0.27 0.13–0.53 0.23 0.07–0.82 0.26 0.08–0.84 0.24 0.08–0.78 0.04 0.01–0.46
Current daily smoker 0.87 0.59–1.27 0.89 0.45–1.78 0.91 0.64–1.29 0.18 0.02–1.39 1.05 0.82–1.34
Time daily smoker 1.01 0.99–1.03 1.00 0.99–1.01 1.00 0.98–1.03 1.05 0.98–1.13 0.98 0.94–1.03
At least one drink per day 0.84 0.48–1.48 1.55 0.73–3.29 0.76 0.33–1.72 0.77 0.26–2.22 0.51 0.12–2.19
House with good floor and walls 0.74 0.56–0.99 0.71 0.51–1.00 0.79 0.55–1.15 0.56 0.06–5.63 0.82 0.77–0.87
House with good toilet 1.23 0.94–1.62 0.91 0.40–2.09 1.39 0.95–2.04 0.78 0.51–1.21 1.46 1.27–1.68
Urban 0.80 0.66–0.96 0.74 0.36–1.50 0.80 0.61–1.04 1.31 0.76–2.25 0.70 0.57–0.86
Household members (#) 0.95 0.93–0.98 0.99 0.96–1.03 0.92 0.88–0.96 0.97 0.77–1.21 0.94 0.91–0.97
Household members per room 1.04 0.98–1.09 1.03 0.92–1.15 1.03 0.95–1.11 1.58 1.08–2.30 1.07 0.98–1.16
Per-capita household expenditure 1.00 1.00–1.00 1.00 1.00–1.00 1.00 0.99–1.00 1.00 1.00–1.00 1.00 1.00–1.00
Respondent insured 0.84 0.63–1.10 0.62 0.37–1.03 0.96 0.47–1.96 1.09 0.34–3.50 0.93 0.87–0.99

*Reference categories for variables like age, BMI, and schooling have ORs or 1.0 and are not shown in the table.

Including a measure of relative expenditure (per-capita household expenditure tertile), compared either with other individuals of the same country or with to individuals within the same GDP quintile, did not alter the direction or magnitude of our main findings (Appendix Table 1, available at IJE online). Similarly, when relative expenditure and its interaction with diabetes were included, the direction and magnitude of our main findings did not change. Importantly, no specific pattern of excess TB risk conditional on diabetes at particular relative expenditure levels emerged. Similarly, when we considered interactions between relative expenditure and gender, age, smoking and alcohol consumption, the direction and magnitude of the relationship between TB and diabetes did not change, nor did we observe strong gradients between relative expenditure and any of these other variables.

In multivariable analyses restricted to nine countries with high burdens of both TB and diabetes (Table 5), the relationship between TB and diabetes strengthened slightly (OR: from 1.81 to 2.00) (Appendix Table 2, available at IJE online).

Table 5

Countries with high prevalence of both T2DM and TB

Country T2DM prevalence (%) TB prevalence (%) Adult population Included in WHS analysis
India 5.9 0.31 603 677 000 Yes
Brazil 5.2 0.08 109 901 000 Yes
Russian Federation 9.2 0.16 105 244 000 Yes
Pakistan 8.5 0.33 72 760 000 Yes
Republic of Korea 6.4 0.13 34 147 000 No
Romania 9.3 0.19 16 392 000 No
Peru 5.1 0.22 15 397 000 No
Malaysia 9.4 0.13 13 280 000 Yes
Iraq 7.7 0.20 11 962 000 No
Afghanistan 8.2 0.66 11 130 000 No
Yemen 7.7 0.14 8 137 000 No
Ecuador 4.8 0.20 7 548 000 Yes
Guatemala 5.5 0.11 5 620 000 No
China, Hong Kong SAR 8.8 0.08 5 424 000 No
Dominican Republic 10.0 0.12 4 991 000 Yes
Bolivia 4.8 0.29 4 480 000 No
El Salvador 6.2 0.07 3 620 000 No
Croatia 5.8 0.07 3 412 000 Yes
Honduras 5.7 0.10 3 302 000 No
Nicaragua 6.1 0.08 2 567 000 No
Mauritius 10.7 0.14 786 000 Yes
Qatar 16.0 0.08 393 000 No
Djibouti 4.9 1.14 300 000 No
Suriname 8.6 0.10 251 000 No
Brunei Darussalam 10.7 0.06 209 000 No
Belize 5.7 0.06 124 000 No
Micronesia 6.7 0.06 82 000 No
Kiribati 6.2 0.06 60 000 No
Palau 8.7 0.09 12 000 No
Country T2DM prevalence (%) TB prevalence (%) Adult population Included in WHS analysis
India 5.9 0.31 603 677 000 Yes
Brazil 5.2 0.08 109 901 000 Yes
Russian Federation 9.2 0.16 105 244 000 Yes
Pakistan 8.5 0.33 72 760 000 Yes
Republic of Korea 6.4 0.13 34 147 000 No
Romania 9.3 0.19 16 392 000 No
Peru 5.1 0.22 15 397 000 No
Malaysia 9.4 0.13 13 280 000 Yes
Iraq 7.7 0.20 11 962 000 No
Afghanistan 8.2 0.66 11 130 000 No
Yemen 7.7 0.14 8 137 000 No
Ecuador 4.8 0.20 7 548 000 Yes
Guatemala 5.5 0.11 5 620 000 No
China, Hong Kong SAR 8.8 0.08 5 424 000 No
Dominican Republic 10.0 0.12 4 991 000 Yes
Bolivia 4.8 0.29 4 480 000 No
El Salvador 6.2 0.07 3 620 000 No
Croatia 5.8 0.07 3 412 000 Yes
Honduras 5.7 0.10 3 302 000 No
Nicaragua 6.1 0.08 2 567 000 No
Mauritius 10.7 0.14 786 000 Yes
Qatar 16.0 0.08 393 000 No
Djibouti 4.9 1.14 300 000 No
Suriname 8.6 0.10 251 000 No
Brunei Darussalam 10.7 0.06 209 000 No
Belize 5.7 0.06 124 000 No
Micronesia 6.7 0.06 82 000 No
Kiribati 6.2 0.06 60 000 No
Palau 8.7 0.09 12 000 No

Findings were generally robust to sensitivity analyses assessing selection bias, self-report of diabetes and the symptomatic definition of a TB case (Appendix Tables 3, 4, and 5, available at IJE online). Though some variables, namely those related to socioeconomic status, housing quality and crowding were collinear, findings from the multivariable analysis were largely unaffected by this collinearity.

### Country-level analysis

Over 10 years, TB was more likely to increase in countries where diabetes prevalence increased (for TB prevalence: OR: 4.7; 95% CI: 1.0–22.5; for TB incidence: OR: 8.7; 95% CI: 1.9–40.0) and in countries with lower per-capita, base year GDP (highest vs lowest GDP quintile for TB prevalence: OR: 0.09; 95% CI: 0.02–0.42; for TB incidence: OR: 0.03; 95% CI: 0.01–0.14)]. This is the case despite the fact that higher diabetes prevalence accompanied lower TB prevalence in both the 1990s and 2000s. Countries with higher diabetes prevalence tended to have lower TB prevalence and incidence in both periods (1990–95: Spearman’s rho: prevalence −0.53; P < 0.0001 and incidence −0.52; P < 0.0001; 2003–04: ρ: prevalence −0.62; P < 0.0001 and incidence −0.63; P < 0.0001). There is also a strong GDP gradient: TB prevalence and incidence had a negative association with per-capita GDP (1990–95: rho: prevalence −0.75; P < 0.0001 and incidence −0.72; P < 0.0001; 2003–04: rho: prevalence −0.77; P < 0.0001 and incidence −0.74; P < 0.0001), whereas diabetes prevalence was positively associated with per-capita GDP (1990–95: rho: 0.57; P < 0.0001; 2003–04: rho: 0.64; P < 0.0001).

In 2003–04, 29 countries had both higher burdens of diabetes (>4.6% prevalence) and TB (>0.06% prevalence) (Table 5). These high-TB/high-diabetes burden countries represent 28% of the adult population (approximately 1 billion people) of the 163 countries.

## Discussion

Individuals in lower income countries, where the majority of the world’s TB burden is located, were more likely to report symptoms of active TB disease if they also reported a prior diagnosis of T2DM. At the population level, between the 1990s and early 2000s, TB prevalence and incidence were more likely to increase in countries in which diabetes prevalence increased, conditioning on base year, per-capita GDP. Countries of particular concern given their size, substantial TB burdens and large projected increases in diabetes prevalence7,30 include Brazil, China, India, Peru and the Russian Federation.

While epidemiologic transitions classically involve shifts from infectious to non-communicable diseases, the lingering presence of both diseases heightens the risk of interaction. With nearly 1 billion individuals currently living in developing countries with substantial burdens of both TB and diabetes and rising trends of diabetes prevalence worldwide, our findings highlight the need for appropriate public health action.

This work contributes to a growing body of evidence on the importance of the TB/diabetes relationship. Notably, it considers population-representative data with large sample sizes from lower income countries, examining the relationship at both the country and individual level. The results confirm TB’s relationship with known risk factors, adding further confidence in the analysis. Using data from countries with higher TB burdens, its findings provide population-representative confirmation for prior systematic reviews of studies conducted largely in selected sub-populations from higher income countries.1,2

We interpret the association between a self-reported diabetes diagnosis and TB symptoms as evidence of an association between the biological presence of diabetes and active TB disease. Self-report can potentially bias this interpretation.32 In groups unlikely to know their diabetes status, true diabetes prevalence will likely be underestimated.33 In light of this under-reporting, our estimate may be biased toward the null, given that true diabetics who were misclassified as non-diabetics or were excluded from the analysis due to non-response are also likely to be at risk for TB. Self-reported TB symptoms also have less than perfect sensitivity and specificity compared with gold-standard diagnostics. Of particular concern is that TB symptoms may be non-specific since pneumonia and bronchitis can cause hemoptysis of substantial duration, as previously reported.34,35 Thus, the estimated relationship may also partly represent the relationship between diabetes and other respiratory ailments. If diabetes were also positively associated with these other respiratory ailments, the relationship we estimated would be biased away from the null. Unfortunately, we know of no population-representative data sets including laboratory tests for TB and diabetes. Of some reassurance, our estimates are strikingly consistent with non-population-representative studies employing gold standard diagnostics.1,2

It is possible that individuals who develop complications of diabetes severe enough to seek more frequent medical care are more likely to be diagnosed with TB—leading to an observed association between TB and diabetes, though tuberculosis may have preceded diabetes. In our analysis, this possibility is unlikely because symptoms of TB were self-reported (i.e. blood in cough and cough lasting ≥3 weeks) as occurring in the past year, while diabetes diagnosis could have occurred at any time in the person’s life. Furthermore, for many people, the timescale of TB infection to symptoms is shorter than the timescale from the development of diabetes to the experience of symptoms that they themselves notice.

To the extent that the risks of both TB and T2DM are increased by HIV/AIDS, increases in HIV/AIDS could account for some of the association we have estimated here. It is well accepted that TB resurgence in many countries has been increased by HIV/AIDS. However, a review of the literature36–42 on the relationship between diabetes and HIV/AIDS finds no studies reporting evidence of a causative increase—notably, Hepatitis C co-infection and the use of highly active anti-retroviral therapy (HAART) particularly when protease inhibitors are included increase the risk of diabetes. In our sample of countries, those with the highest HIV/AIDS prevalence generally had low access to HAART in 2003, and, therefore, we believe that the possibility is small that HIV/AIDS explains the TB/diabetes relationships that we have observed.

In our country-level analysis, we interpret the longitudinal increase in TB prevalence associated with diabetes prevalence as evidence suggestive of a causal relationship between diabetes and TB manifesting at the population health level. Differential changes in country health-care systems leading to increased diagnosis of both diabetes and TB could also produce this result; a possibility we cannot exclude. However, we note that most developing countries between the 1990s and early 2000s did not drastically expand their diabetes screening, diagnostic or management services. Also, in our country-level analyses, we adjusted for country-fixed effects and per-capita GDP levels as proxies for the types of care provided by health systems. Additionally, in the individual-level analysis, even after adjusting for health insurance status and per-capita income, individuals with diagnosed diabetes were more likely to report symptoms of TB.

While the sample size in the study is large, a substantial number of individuals are excluded in the multivariable analysis because of non-response to survey items. Those not providing complete responses tended to be poorer and have less education, though reassuringly, in univariable analyses, the association between TB and the covariates were similar for the restricted study sample and all individuals providing responses to a given item.

The mechanisms by which diabetes increases TB risk at the individual and population levels remain to be elucidated. Our individual-level analysis is cross-sectional and cannot distinguish the temporal ordering of diabetes onset, TB infection and TB symptom onset. Country-level analyses do not capture patterns of population mixing or other risk factors contributing to the observed associations. While the country- and individual-level relationships we estimate are broadly consistent, we do not observe even stronger associations of TB and diabetes in socioeconomic sub-populations where these two diseases are most likely to mix. However, if failure to be diagnosed with diabetes were associated with relative poverty, then differential under-diagnosis could explain our failure to detect an economic gradient.

Future research priorities include epidemiologic and modeling studies. Prospective cohort studies employing biochemical tests to definitively detect TB and diabetes at multiple time points can address the time ordering of diabetes onset and TB infection or reactivation. Such studies can also link individual and population risks including the interactions of risk factors and interplay of social determinants. Policy analyses using computer-based population models translate TB/diabetes associations and trends in prevalence into projections of future burdens. Such models can also assess the relative value of strategies including accelerating the availability of diabetes interventions and combining or targeting diabetes and TB initiatives.11,43 Interventions reducing diabetes could potentially reduce individual-level risks of TB and indirectly avert increases in TB population prevalence.

An increase in the individual risk of TB associated with T2DM has important public health implications. At a national level, the seriousness of this risk depends crucially on the current and future levels and distributions of TB, diabetes and their risk factors throughout the population. Proactive evidence-based prevention policies are crucial for countries facing large TB/diabetes burdens.

## Supplementary data

Supplementary data are available at IJE online.

## Acknowledgements

J.D.G.-F.: study conception and design, data collection, analysis and interpretation, drafting of manuscript and critical revision. C.Y.J.: study design, interpretation of results, critical manuscript revision. T.C.: study design, interpretation of results, critical manuscript revision. M.B.M.: study design, interpretation of results, critical manuscript revision.

Conflict of interest: The corresponding author confirms that he had full access to all the data in the study and had final responsibility for the decision to submit for publication. All authors affirm that they have no financial or personal relationships with other people or organizations that could inappropriately influence or bias their work. No funding source had any role in the study design; in the collection, analysis and interpretation of data; in the writing of the report; and in the decision to submit for publication.

KEY MESSAGES

• In lower income countries, symptoms of TB are more likely to occur in individuals reporting a past diabetes diagnosis.

• Countries with higher diabetes prevalence were more likely to see increases in TB between the mid-1990s and early 2000s.

• Of particular concern, populous countries including India, Peru and the Russia Federation, which have relatively large TB burdens, have seen rapid increases in their diabetes prevalence.

## References

1
Jeon
CY
Murray
MB
Diabetes mellitus increases the risk of active tuberculosis: a systematic review of 13 observational studies
PLoS Med
,
2008
, vol.
5
pg.
e152

2
Stevenson
CR
Critchley
JA
Forouhi
NG
, et al.  .
Diabetes and the risk of tuberculosis: a neglected threat to public health?
Chronic Illn
,
2007
, vol.
3
(pg.
228
-
45
)
3
Dye
C
Lonnroth
K
Jaramillo
E
Williams
BG
Raviglione
M
Trends in tuberculosis incidence and their determinants in 134 countries
Bull World Health Organ
,
2009
, vol.
87
(pg.
683
-
91
)
4
Lonnroth
K
Jaramillo
E
Williams
BG
Dye
C
Raviglione
M
Drivers of tuberculosis epidemics: the role of risk factors and social determinants
Soc Sci Med
,
2009
, vol.
68
(pg.
2240
-
46
)
5
Omran
AR
The epidemiologic transition. A theory of the epidemiology of population change
Milbank Mem Fund Q
,
1971
, vol.
49
(pg.
509
-
38
)
6
World Health Organization
The World health report: 1999: Making a difference
,
1999
Geneva
World Health Organization
7
Wild
S
Roglic
G
Green
A
Sicree
R
King
H
Global prevalence of diabetes: estimates for the year 2000 and projections for 2030
Diabetes Care
,
2004
, vol.
27
(pg.
1047
-
53
)
8
Raviglione
M
XDR-TB: entering the post-antibiotic era?
Int J Tuberc Lung Dis
,
2006
, vol.
10
(pg.
1185
-
87
)
9
Corbett
EL
Watt
CJ
Walker
N
, et al.  .
The growing burden of tuberculosis: global trends and interactions with the HIV epidemic
Arch Intern Med
,
2003
, vol.
163
(pg.
1009
-
21
)
10
Wells
CD
Cegielski
JP
Nelson
LJ
, et al.  .
HIV infection and multidrug-resistant tuberculosis: the perfect storm
J Infect Dis
,
2007
, vol.
196

Suppl
.
1
(pg.
S86
-
107
)
11
Stevenson
CR
Forouhi
NG
Roglic
G
, et al.  .
Diabetes and tuberculosis: the impact of the diabetes epidemic on tuberculosis incidence
BMC Public Health
,
2007
, vol.
7
pg.
234

12
Moussavi
S
Chatterji
S
Verdes
E
Tandon
A
Patel
V
Ustun
B
Depression, chronic diseases, and decrements in health: results from the World Health Surveys
Lancet
,
2007
, vol.
370
(pg.
851
-
58
)
13

World Health Organization: World Health Survey.http://www.who.int/healthinfo/survey/en/index.html (12 August 2009, date last accessed)
14
Cohen
R
Muzaffar
S
Capellan
J
Azar
H
Chinikamwala
M
The validity of classic symptoms and chest radiographic configuration in predicting pulmonary tuberculosis
Chest
,
1996
, vol.
109
(pg.
420
-
23
)
15
Rosen
MJ
Chronic cough due to tuberculosis and other infections: ACCP evidence-based clinical practice guidelines
Chest
,
2006
, vol.
129

Suppl. 1
(pg.
197S
-
201S
)
16
Ayles
H
Schaap
A
Nota
A
, et al.  .
Prevalence of tuberculosis, HIV and respiratory symptoms in two Zambian communities: implications for tuberculosis control in the era of HIV
PLoS One
,
2009
, vol.
4
pg.
e5602

17
Robinson
JR
Young
TK
Roos
LL
Gelskey
DE
Estimating the burden of disease. Comparing administrative data and self-reports
Med Care
,
1997
, vol.
35
(pg.
932
-
47
)
18
Wu
SC
Li
CY
Ke
DS
The agreement between self-reporting and clinical diagnosis for selected medical conditions among the elderly in Taiwan
Public Health
,
2000
, vol.
114
(pg.
137
-
42
)
19
Goldman
N
Lin
IF
Weinstein
M
Lin
YH
Evaluating the quality of self-reports of hypertension and diabetes
J Clin Epidemiol
,
2003
, vol.
56
(pg.
148
-
54
)
20
Okura
Y
Urban
LH
Mahoney
DW
Jacobsen
SJ
Rodeheffer
RJ
Agreement between self-report questionnaires and medical record data was substantial for diabetes, hypertension, myocardial infarction and stroke but not for heart failure
J Clin Epidemiol
,
2004
, vol.
57
(pg.
1096
-
103
)
21
El Fakiri
F
Bruijnzeels
MA
Hoes
AW
No evidence for marked ethnic differences in accuracy of self-reported diabetes, hypertension, and hypercholesterolemia
J Clin Epidemiol
,
2007
, vol.
60
(pg.
1271
-
79
)
22
Molenaar
EA
Van Ameijden
EJ
Grobbee
DE
Numans
ME
Comparison of routine care self-reported and biometrical data on hypertension and diabetes: results of the Utrecht Health Project
Eur J Public Health
,
2007
, vol.
17
(pg.
199
-
205
)
23
Kishore
B
Nagrath
SP
Mathur
KS
Hazra
DK
Agarwal
BD
Manifest, chemical and latent chemical diabetes in pulmonary tuberculosis
J Assoc Physicians India
,
1973
, vol.
21
(pg.
875
-
81
)
24
Oluboyo
PO
Erasmus
RT
The significance of glucose intolerance in pulmonary tuberculosis
Tubercle
,
1990
, vol.
71
(pg.
135
-
38
)
25
Goyal
BN
Dubey
AL
Joshi
LD
Saxena
HN
Study of diabetic status in pulmonary tuberculosis
J Diab Assoc India
,
1978
, vol.
18
(pg.
191
-
97
)
26
Singh
MM
Biswas
SK
AS. Impaired glucose tolerance in active pulmonary tuberculosis
Indian J Tuberculosis
,
1984
, vol.
31
(pg.
118
-
21
)
27
World Bank
World Development Indicators

http://www.worldbank.org/data (12 August 2009, date last accessed)
28
King
H
Aubert
RE
Herman
WH
Global burden of diabetes, 1995-2025: prevalence, numerical estimates, and projections
Diabetes Care
,
1998
, vol.
21
(pg.
1414
-
31
)
29
International Diabetes Federation
Diabetes Atlas

http://www.eatlas.idf.org/ (26 October 2008, date last accessed)
30
World Health Organization
The stop TB strategy, case reports, treatment outcomes and estimates of TB burden

31
Heston
A
Summers
R
Aten
B
University of Pennsylvania Center for International Comparisons of Production, Income and Prices: Penn World Table Version 6.2
,
2006

http://pwt.econ.upenn.edu (12 August 2009, date last accessed)
32
Spiegelman
D
Zhao
B
Kim
J
Correlated errors in biased surrogates: study designs and methods for measurement error correction
Stat Med
,
2005
, vol.
24
(pg.
1657
-
82
)
33
Danaei
G
Friedman
AB
Oza
S
Murray
CJ
Ezzati
M
Diabetes prevalence and diagnosis in US states: analysis of health surveys
Popul Health Metr
,
2009
, vol.
7
pg.
16

34
Joshi
N
Caputo
GM
Weitekamp
MR
Karchmer
AW
Infections in patients with diabetes mellitus
N Engl J Med
,
1999
, vol.
341
(pg.
1906
-
12
)
35
Muller
LM
Gorter
KJ
Hak
E
, et al.  .
Increased risk of common infections in patients with type 1 and type 2 diabetes mellitus
Clin Infect Dis
,
2005
, vol.
41
(pg.
281
-
8
)
36
Brar
I
Shuter
J
Thomas
A
Daniels
E
Absalon
J
A comparison of factors associated with prevalent diabetes mellitus among HIV-Infected antiretroviral-naive individuals versus individuals in the National Health and Nutritional Examination Survey cohort
J Acquir Immune Defic Syndr
,
2007
, vol.
45
(pg.
66
-
71
)
37
Butt
AA
McGinnis
K
MC
, et al.  .
HIV infection and the risk of diabetes mellitus
AIDS
,
2009
, vol.
23
(pg.
1227
-
34
)
38
Carr
A
Samaras
K
Thorisdottir
A
Kaufmann
GR
Chisholm
DJ
Cooper
DA
Diagnosis, prediction, and natural course of HIV-1 protease-inhibitor-associated lipodystrophy, hyperlipidaemia, and diabetes mellitus: a cohort study
Lancet
,
1999
, vol.
353
(pg.
2093
-
99
)
39
Chronister
CL
Gurwood
AS
Type 2 diabetes in association with HIV-1 protease inhibitors in HIV-infected patients
J Am Optom Assoc
,
1998
, vol.
69
(pg.
695
-
98
)
40
Tien
PC
Schneider
MF
Cole
SR
, et al.  .
Antiretroviral therapy exposure and incidence of diabetes mellitus in the Women's Interagency HIV Study
AIDS
,
2007
, vol.
21
(pg.
1739
-
45
)
41
Urso
R
Croce
GF
Tubili
C
, et al.  .
[Dysmetabolic syndrome related to HIV-1 protease inhibitors. Review of the literature and personal data]
Recenti Prog Med
,
2000
, vol.
91
(pg.
78
-
85
)
42
Visnegarwala
F
Chen
L
Raghavan
S
Tedaldi
E
Prevalence of diabetes mellitus and dyslipidemia among antiretroviral naive patients co-infected with hepatitis C virus (HCV) and HIV-1 compared to patients without co-infection
J Infect
,
2005
, vol.
50
(pg.
331
-
37
)
43
Harries
Billo
N
Kapur
A
Links between diabetes mellitus and tuberculosis: should we integrate screening and care?
Trans R Soc Trop Med Hyg
,
2009
, vol.
103
(pg.
1
-
2
)