American Journal of Epidemiology

The Author 2008. Published by the Johns Hopkins Bloomberg School of Public Health.
All rights reserved. For permissions, please e-mail: journals.permissions@oxfordjournals.org.

Vol. 167, No. 12

DOI: 10.1093/aje/kwn075
Advance Access publication April 8, 2008

Original Contribution
Diabetic Control and Risk of Tuberculosis: A Cohort Study

Chi C. Leung1, Tai H. Lam2, Wai M. Chan3, Wing W. Yew4, Kin S. Ho3, Gabriel M. Leung2, Wing S.
Law1, Cheuk M. Tam1, Chi K. Chan1, and Kwok C. Chang1
1

Tuberculosis and Chest Service, Department of Health, Hong Kong, Peoples Republic of China.
Department of Community Medicine, The University of Hong Kong, Hong Kong, Peoples Republic of China.
3
Elderly Health Service, Department of Health, Hong Kong, Peoples Republic of China.
4
Tuberculosis and Chest Unit, Grantham Hospital, Hong Kong, Peoples Republic of China.
2

Diabetes mellitus is associated with tuberculosis. A cohort of 42,116 clients aged 65 years or more, enrolled at 18
Elderly Health Service centers in Hong Kong in 2000, were followed up prospectively through the territory-wide
tuberculosis registry for development of tuberculosis from 3 months after enrollment to December 31, 2005, by use
of their identity card numbers as unique identier. The effects of diabetes mellitus and diabetic control on tuberculosis risk were assessed with adjustment for sociodemographic and other background variables. Diabetes
mellitus was associated with a modest increase in the risk of active, culture-conrmed, and pulmonary (with or
without extrapulmonary involvement) but not extrapulmonary (with or without pulmonary involvement) tuberculosis,
with adjusted hazard ratios of 1.77 (95% condence interval: 1.41, 2.24), 1.91 (95% condence interval: 1.45,
2.52), 1.89 (95% condence interval: 1.48, 2.42), and 1.00 (95% condence interval: 0.54, 1.86), respectively.
Diabetic subjects with hemoglobin A1c <7% at enrollment were not at increased risk. Among diabetic subjects,
higher risks of active, culture-conrmed, and pulmonary but not extrapulmonary tuberculosis were observed with
baseline hemoglobin A1c
7% (vs. <7%), with adjusted hazard ratios of 3.11 (95% condence interval: 1.63,
5.92), 3.08 (95% condence interval: 1.44, 6.57), 3.63 (95% condence interval: 1.79, 7.33), and 0.77 (95%
condence interval: 0.18, 3.35), respectively.
aged; diabetes mellitus; tuberculosis

Diabetes mellitus has been well reported to be associated

with increased risk of tuberculosis (14). However, few
studies examined specifically the effect of diabetes control.
The presence of a myriad of associated social factors, metabolic derangements, and comorbidities also poses major
difficulties in dissecting the effect of diabetes mellitus from
other potential confounders.
In Hong Kong, active tuberculosis disease is statutorily
notifiable to the Department of Health, and the notification
database has been computerized, with the identity card
number as the unique identifier. The annual tuberculosis
notification rate is about 90/100,000 population, and it is
substantially higher in males and those older than 65 years
(5). The Elderly Health Service provides a territory-wide

and community-based health maintenance program to the

elderly through 18 centers (6). All ambulatory elderly persons aged 65 years or more are eligible for voluntary enrollment. A nominal annual membership fee of about US
$13.75 is charged, with a waiver mechanism for those receiving public financial assistance. The availability of such
a community health program with minimum barrier access
provides an opportunity to study a representative sample of
the community-living elderly in Hong Kong. A standardized
health questionnaire was administered to each enrolled client, followed by medical examination, chest radiographic
examination, and regular screening for hypertension and
diabetes mellitus. Hypertension was diagnosed when blood
pressure was 140/90 mmHg or more after two repeated

Correspondence to Dr. Chi Chiu Leung, Wanchai Chest Clinic, 99 Kennedy Road, Wanchai, Hong Kong, Peoples Republic of China (e-mail:
cc_leung@dh.gov.hk).

1486

Am J Epidemiol 2008;167:14861494

Downloaded from http://aje.oxfordjournals.org/ at Georgetown University on May 21, 2015

Received for publication September 6, 2007; accepted for publication March 6, 2008.

Diabetic Control and Tuberculosis

measurements (7). Diabetes mellitus was diagnosed, mainly

by a fasting plasma glucose level of 7.0 mmol/liter or higher,
together with confirmatory symptoms and/or blood/plasma
glucose determinations (8). Patients with symptoms suspicious of active tuberculosis or radiologic abnormalities were
referred to the 18 chest clinics under a centralized Tuberculosis and Chest Service. Under such an appropriate service
infrastructure, a study was performed to examine prospectively the effect of diabetes mellitus and diabetic control on
the risk of development of active tuberculosis.

MATERIALS AND METHODS

Am J Epidemiol 2008;167:14861494

tuberculosis with extrapulmonary involvement, irrespective of

whether there was pulmonary involvement.
Disease incidence was calculated by assuming a Poisson
distribution in the rate of occurrence of the events. Univariate analysis was followed by Cox regression analysis of the
effect of diabetes mellitus on the respective development of
active tuberculosis, culture-confirmed tuberculosis, pulmonary tuberculosis, and extrapulmonary involvement. The
effect of diabetic control at baseline as reflected by the
hemoglobin A1c level was similarly assessed. The proportional hazards assumption was examined by use of graphical
methods (i.e., log minus log plots) and the time-dependent
covariate method. As a countercheck for the robustness of
the multivariate model, the analysis was repeated after exclusion of subjects with malignancies, body mass index
<18.5, recent weight loss of 5 percent or more in 6 months,
and hospital admission within 12 months of baseline. Twotailed p < 0.05 was taken as statistically significant.
The study was approved by the Ethics Committee of the
Department of Health of Hong Kong, Peoples Republic of
China.
RESULTS

A total of 42,659 clients aged 65 years or more were recruited into the health maintenance program of the Elderly
Health Service in 2000. Twenty-three duplicate entries, 304
subjects with a missing/invalid identity number, 127 with
missing/incomplete information on sex, age, weight, and/or
height, and 89 with known active tuberculosis on presentation or within 3 months of enrollment were excluded, leaving
42,116 subjects for analysis. The background characteristics
of the cohort are shown in table 1. The data were over 99.9
percent complete for the variables listed.
A total of 3,893 (9.2 percent) deaths were identified from
the death registry. The mean duration of follow-up from the
day of enrollment to death, development of active tuberculosis, or December 31, 2005 (whichever came first), was
1,827 (standard deviation: 325) days. A total of 510 new
tuberculosis notifications were identified from the tuberculosis notification registry during the period of follow-up.
Eight cases were duplicate entries due to early relapse.
Twenty-five cases (six cases of bronchogenic carcinoma,
13 cases of non-tuberculous mycobacterial infection, six
cases of old lung scars) were found to have an incorrect
diagnosis after review of the records, leaving for analysis
477 cases of active tuberculosis, 326 (68.3 percent) of which
were culture confirmed. The mean time interval of followup before notification was 881 (standard deviation: 583)
days. There were 395 new cases (82.8 percent) and 82 retreatment (with past tuberculosis history) cases (17.2 percent). There was no significant difference in the proportion
of retreatment tuberculosis cases between diabetic and nondiabetic subjects (13.8 percent vs. 18.0 percent: p 0.335).
Pulmonary involvement was found in 426 cases (89.3 percent), and extrapulmonary involvement was found in 87
cases (18.2 percent), including 36 cases (7.5 percent) with
both. The incidences of active, culture-confirmed, pulmonary, and extrapulmonary (all or alone) tuberculosis are
summarized in table 2. Of the 394 patients who underwent

Downloaded from http://aje.oxfordjournals.org/ at Georgetown University on May 21, 2015

A cohort of clients who enrolled at the 18 Elderly Health

Service centers from January 1, 2000, to December 31,
2000, was retrospectively assembled as previously reported
(9). The date of enrollment, name, sex, age, identity card
number, smoking status, alcohol use, language spoken, educational level, marital status, housing situation, working
status, public means-tested financial assistance status, coexisting medical conditions, recent weight loss, hospital admission during the past 12 months, Activities of Daily
Living score, body mass index, and hemoglobin A1c level
were retrieved from the baseline health assessment database
of the Elderly Health Service. The baseline database was
cross-matched prospectively with the death registry and the
tuberculosis notification registry by use of the identity card
number as the unique identifier, supplemented by name and
age, from 3 months (arbitrarily taken as 91 days) after enrollment to December 31, 2005. Diabetes mellitus and
hypertension were defined as the corresponding physiciandiagnosed conditions with or without treatment, and the
updated diagnoses according to World Health Organization
criteria (7, 8) after the screening at enrollment were used in
the analysis. The hemoglobin A1c level, which reflected
a time-weighted mean glycemic control over the previous
34 months, was checked at enrollment for those with
a known history of diabetes mellitus, and a hemoglobin
A1c level of less than 7 percent was taken as reflecting
satisfactory glycemic control at baseline (10). The duration
of follow-up in days was defined as the period from the start
of matching (91 days after enrollment) to the date of notification of tuberculosis, death, or December 31, 2005,
whichever occurred first. Information on the date of tuberculosis notification, bacteriologic status, form of tuberculosis, and previous tuberculosis history was retrieved from the
notification registry. The diagnosis of and clinical information on all identified tuberculosis cases were verified by
reviewing medical records retrieved from chest clinics and
other relevant sources, as well as the public health records
of the Tuberculosis and Chest Service. An active case of
tuberculosis was defined as disease proven by isolation of
Mycobacterium tuberculosis or, in the absence of bacteriologic
confirmation, disease diagnosed on clinical, radiologic, and/or
histologic grounds together with an appropriate response to
antituberculosis treatment. Pulmonary tuberculosis was defined as active tuberculosis with pulmonary involvement, irrespective of whether there was extrapulmonary involvement.
Similarly, extrapulmonary tuberculosis was defined as active

1487

1488 Leung et al.

TABLE 1. Background characteristics of the cohort, Hong Kong, Peoples Republic of

China, 20002005
Variable

All cohort
(n 42,116)

No diabetes
mellitus
(n 35,672)

With diabetes
mellitus
(n 6,444)

p value*

Male sex, %

34.7

34.7

34.5

0.741

Age, years (mean)

72.6

72.6

72.5

0.404

Cantonese speaking, %

98.3

98.4

98.3

0.548
<0.001

Marital, %
Never married

3.8

4.0

3.0

Married

60.6

60.6

60.7

Widowed/divorced/other

35.6

35.4

36.3

Education, %
Postsecondary

3.6

3.7

13.7

13.6

13.7

Primary

37.5

37.6

36.7

No formal

16.7

16.9

15.6

Illiterate

28.6

28.3

30.2

0.008

Housing, %
39.5

39.5

39.8

Privately rented

Public

4.8

4.8

4.8

Privately owned

49.1

49.2

48.9

6.6

6.6

6.5

Other
Working full-/part-time, %

0.954

4.5

4.6

3.9

0.014

18.0

17.9

18.9

0.057

Never

71.2

70.5

74.7

<0.001

Former drinker

10.1

9.7

11.9

Social

14.9

15.7

10.8

3.8

4.1

2.6

On public means-tested assistance, %

Alcohol, %

Regular
Smoking, %
Never

70.8

70.6

72.3

Former smoker

20.2

20.2

20.1

9.0

9.3

7.6

Current

<0.001

Body mass index, %

30
25<30

6.3

5.9

8.3

34.6

33.6

39.8

23<25

22.9

22.6

24.6

18.5<23

31.4

32.5

25.5

<0.001

4.8

5.4

1.8

Hypertension, %y

42.8

39.2

62.5

<0.001

Cardiovascular disease, %z

14.4

13.4

19.9

<0.001

COPD/asthma, %{

5.7

5.9

5.1

0.012

Malignant disease, %

0.7

0.7

0.6

0.308

Recent weight loss, %#

2.7

2.5

3.9

<0.001

13.9

13.3

17.0

<0.001

<18.5

Admission within 1 year, %

Activities of Daily Living, score (mean)

7.03

7.03

7.05

0.006

* Probability for the null hypothesis of no difference in the factor.

y Previously diagnosed by physician or newly diagnosed on enrollment.
z Previously diagnosed by physician, with or without drug treatment.
COPD, chronic obstructive pulmonary disease.
{ COPD and/or asthma as diagnosed by physician.
# Recent weight loss of
5% within 6 months.

Am J Epidemiol 2008;167:14861494

Downloaded from http://aje.oxfordjournals.org/ at Georgetown University on May 21, 2015

3.6

Secondary

Diabetic Control and Tuberculosis

1489

TABLE 2. Incidence of tuberculosis among diabetic and nondiabetic subjects, Hong Kong, Peoples Republic of China, 20002005
No diabetes mellitus
(n 35,672)

All cohort (n 42,116)

Tuberculosis
No. of
cases

Rate/
100,00
personyears

95%
condence
interval

No. of
cases

Rate/
100,00
personyears

With diabetes mellitus

(n 6,444)

95%
condence
interval

No. of
cases

Rate/
100,00
personyears

95%
condence
interval

Relative
risk*

95%
condence
interval

p value

Active

477

226

207, 248

383

214

193, 237

94

295

239, 362

1.38

1.09, 1.74

0.005

Culture conrmed

326

155

138, 172

258

144

127, 163

68

214

166, 271

1.48

1.12, 1.95

0.004

Pulmonary

426

202

183, 222

340

190

170, 211

86

270

216, 334

1.42

1.12, 1.80

0.003

All extrapulmonary

87

41

33, 51

75

42

33, 53

12

38

9, 66

0.90

0.49, 1.65

0.733

Extrapulmonary only

51

24

18, 32

43

24

17, 32

25

11, 50

1.05

0.49, 2.31

0.907

* Relative risk of tuberculosis for diabetic versus nondiabetic group.

found to have diabetes on enrollment. Within the diabetic

subjects, baseline hemoglobin A1c levels were available for
4,690 (72.8 percent) subjects with known diabetes mellitus
at enrollment, with a mean hemoglobin A1c of 7.62 (1.37

TABLE 3. Annual incidence of active tuberculosis and culture-conrmed tuberculosis among diabetic patients by baseline diabetic
(and control) status as reected by hemoglobin A1c level, Hong Kong, Peoples Republic of China, 20002005
Active tuberculosis
Diabetes mellitus

All cohort
No diabetes mellitus

No.

Rate/
No. of 100,000
cases personyears

95%
condence
interval

Culture-conrmed tuberculosis

Unadjusted
relative
risk*

95%
condence
interval

42,116 477

226

207, 248

35,672 383

214

193, 237

1.00

Referent

Rate/
No. of 100,000
cases personyears

95%
condence
interval

Unadjusted
relative
risk*

95%
condence
interval

326

155

138, 172

258

144

127, 163

1.00

Referent

Diabetes mellitus,
hemoglobin A1c <7%

1,620

11

136

68, 244

0.64

0.35, 1.16

99

43, 196

0.69

0.34, 1.39

Diabetes mellitus,
hemoglobin A1c
7%

3,070

64

422

325, 539

1.97

1.51, 2.57

45

297

216, 397

2.06

1.50, 2.82

Diabetes mellitus,
no hemoglobin A1cy

1,754

19

222

133, 346

1.03

0.65, 1.64

15

175

98, 288

1.21

0.72, 2.04

p < 0.001
Females
No diabetes mellitus

p < 0.001

27,498 199

143

124, 164

23,279 165

140

119, 163

1.00

Referent

125

90

75, 107

102

86

71, 105

1.00

Referent

Diabetes mellitus,
hemoglobin A1c <7%

1,019

78

21, 200

0.56

0.15, 1.45

58

12, 171

0.68

0.14, 2.03

Diabetes mellitus,
hemoglobin A1c
7%

2,014

24

237

152, 352

1.69

1.05, 2.61

45

158

90, 256

1.83

1.01, 3.11

Diabetes mellitus, no
hemoglobin A1cy

1,186

102

38, 223

0.73

0.26, 2.61

15

175

98, 288

0.79

0.21, 2.08

14,618 278

388

344, 437

201

281

243, 322

12,393 218

358

312, 409

1.00

Referent

156

256

218, 300

1.00

Referent

p < 0.001
Males
No diabetes mellitus

p < 0.001

Diabetes mellitus,
hemoglobin A1c <7%

601

239

96, 492

0.67

0.27, 1.40

171

55, 398

0.67

0.21, 1.59

Diabetes mellitus,
hemoglobin A1c
7%

1,056

40

793

567, 1,080

2.22

1.54, 3.12

29

575

385, 826

2.24

1.46, 3.35

568

13

480

256, 821

1.34

0.70, 2.34

11

406

203, 727

1.59

0.78, 2.92

Diabetes mellitus,
no hemoglobin A1cy

p < 0.001
* Relative to the group without diabetes mellitus.
y Hemoglobin A1c not checked at baseline, mainly because of diagnosis only at enrollment.

Am J Epidemiol 2008;167:14861494

p < 0.001

Downloaded from http://aje.oxfordjournals.org/ at Georgetown University on May 21, 2015

voluntary testing for human immunodeficiency virus, only

one (0.25 percent) was infected.
A total of 6,444 subjects (15.3 percent) of the entire cohort either had known history of diabetes mellitus or were

1490 Leung et al.

TABLE 4. Sex- and age-adjusted hazard ratios of active, culture-conrmed, pulmonary, and
extrapulmonary tuberculosis by baseline diabetic (and control) status in Cox regression analysis before
and after exclusion of those subjects with risk factors possibly related to subclinical tuberculosis at
baseline, Hong Kong, Peoples Republic of China, 20002005
Types/forms of tuberculosis
Active
Diabetic status

Culture conrmed

Pulmonary

Extrapulmonary*

Hazard
ratioy

95%
condence
interval

Hazard
ratioy

95%
condence
interval

Referent

1.00

Referent

1.00

Referent

0.68

0.33, 1.37

0.58

0.30, 1.12

0.87

0.28, 2.77

1.56, 2.65

2.13

1.55, 2.93

2.16

1.64, 2.84

0.80

0.32, 1.97

0.67, 1.69

1.25

0.74, 2.10

1.07

0.66, 1.75

1.13

0.41, 3.08

Hazard
ratioy

95%
condence
interval

Referent

1.00

0.63

0.34, 1.14

2.03
1.06

Hazard
ratioy

95%
condence
interval

No diabetes mellitus

1.00

Diabetes mellitus,
hemoglobin A1c <7%
Diabetes mellitus,
hemoglobin A1c
7%
Diabetes mellitus, no
hemoglobin A1c

Before exclusionz

p < 0.001

p < 0.001

p 0.950

After exclusion of other

risk factors
No diabetes mellitus

1.00

Referent

1.00

Referent

1.00

Referent

1.00

Referent

Diabetes mellitus,
hemoglobin A1c <7%

0.68

0.33, 1.36

0.62

0.26, 1.52

0.59

0.26, 1.32

1.22

0.38, 3.90

Diabetes mellitus,
hemoglobin A1c
7%

2.08

1.51, 2.85

2.07

1.40, 3.04

2.21

1.58, 3.08

1.08

0.43, 2.71

Diabetes mellitus, no
hemoglobin A1c

1.23

0.73, 2.08

1.47

0.82, 2.64

1.34

0.78, 2.30

1.16

0.36, 3.72

p < 0.001

p 0.001

p < 0.001

p 0.980

* Any extrapulmonary involvement.

y Sex- and age-adjusted hazard ratio.
z Adjusted for sex and age group without exclusion of any subjects.
Subjects with malignancies, body mass index of <18.5, recent weight loss of
5% in 6 months, and hospital
admission within 12 months at baseline.

percent). Diabetic subjects who developed active tuberculosis subsequently had a higher mean hemoglobin A1c level
than those who did not (8.30 percent vs. 7.61 percent: p <
0.001). Out of the 4,690 diabetic subjects, 1,620 (34.5 percent) had a hemoglobin A1c level below 7 percent, and
3,070 (65.5 percent) had a hemoglobin A1c level at or above
7 percent. Table 3 shows the incidence of active tuberculosis
and culture-confirmed tuberculosis by different baseline diabetic control status. Subjects with hemoglobin A1c
7
percent had about a twofold risk of active or cultureconfirmed tuberculosis (both p < 0.01) relative to elderly
subjects without diabetes. Similar hazard ratios were found
after adjustment for sex and age and after exclusion of subjects with some of the risk factors potentially associated
with subclinical tuberculosis disease at baseline (table 4).
The adjusted hazard ratios increased to 2.5-fold (2.80-fold
for pulmonary tuberculosis) after adjustment for all confounding or potentially confounding sociodemographic
and clinical variables (table 5). These adjusted hazard ratios
were further increased to 3-fold when diabetic subjects with
hemoglobin A1c <7 percent were used as the reference
group instead (table 5). Diabetic subjects diagnosed only
at enrollment and without known baseline hemoglobin
A1c had an intermediate risk between those with hemoglo-

bin A1c <7 percent and
7 percent. The potential interaction between diabetes mellitus and body mass index was
considered, but the interaction term of diabetes mellitus
categories 3 body mass index categories was removed
from the final model because it failed to reach statistical
significance. There was no evidence of violation of the proportional hazards on examination of the log minus log plots
and testing by the time-dependent covariate method. Figure
1 depicts the hazard function curves of active tuberculosis
for different diabetic control status categories in the overall
Cox model.

DISCUSSION

In this study, diabetes mellitus was associated with a modest increase in risk of active, culture-confirmed, and pulmonary tuberculosis, but not extrapulmonary tuberculosis
(tables 2 and 5). The increased risk was observed predominantly among diabetic subjects with baseline hemoglobin
A1c
7 percent, while those with baseline hemoglobin
A1c <7 percent were not at increased risk (tables 3, 4, and 5).
The association between diabetes mellitus and tuberculosis is well reported (14), but prospective cohort data are
Am J Epidemiol 2008;167:14861494

Downloaded from http://aje.oxfordjournals.org/ at Georgetown University on May 21, 2015

p < 0.001

Diabetic Control and Tuberculosis

1491

TABLE 5. Effects of baseline diabetes (and control) status on active tuberculosis, culture-conrmed
tuberculosis, pulmonary tuberculosis, and any extrapulmonary involvement after controlling for other
confounding variables in Cox proportional hazard analysis, Hong Kong, Peoples Republic of China,
20002005
Types/forms of tuberculosis
Active
Diabetic status
Hazard
ratioy

Culture conrmed

95%
condence
interval

Hazard
ratioy

Pulmonary

95%
condence
interval

Hazard
ratioy

Extrapulmonary*

95%
condence
interval

Hazard
ratioy

95%
condence
interval

Diabetes mellitus or not

No diabetes mellitus

1.00

Referent

1.00

Referent

1.00

Referent

1.00

Referent

Diabetes mellitus

1.77

1.41, 2.24

1.91

1.45, 2.52

1.89

1.48, 2.42

1.00

0.54, 1.86

p < 0.001

p < 0.001

p < 0.001

p 0.995

All categories
1.00

Referent

1.00

Referent

1.00

Referent

1.00

Referent

Diabetes mellitus,
hemoglobin A1c <7%

0.81

0.44, 1.48

0.86

0.42, 1.75

0.77

0.40, 1.50

0.99

0.31, 3.17

Diabetes mellitus,
hemoglobin A1c
7%

2.56

1.95, 3.35

2.69

1.94, 3.72

2.80

2.11, 3.70

0.88

0.35, 2.20

Diabetes mellitus, no
hemoglobin A1c

1.32

0.83, 2.10

1.54

0.91, 2.60

1.37

0.84, 2.23

1.24

0.45, 3.41

p < 0.001

p < 0.001

p < 0.001

p 0.967

With known hemoglobin A1c

Diabetes mellitus,
hemoglobin A1c <7%

1.00

Referent

1.00

Referent

1.00

Referent

1.00

Referent

Diabetes mellitus,
hemoglobin A1c
7%

3.11

1.63, 5.92

3.08

1.44, 6.57

3.63

1.79, 7.33

0.77

0.18, 3.35

p 0.001

p 004

p < 0.001

p 0.967

* Any extrapulmonary involvement.

y Hazard ratio adjusted for sex, age, smoking, alcohol use, language, marital status, education, housing, working
status, public means-tested nancial assistance status, body mass index, cardiovascular disease, hypertension,
chronic obstructive pulmonary disease/asthma, malignancy, recent weight loss of
5% within 6 months, hospital
admission within 12 months, and Activities of Daily Living scores, with categories as listed in table 1.

notably scarce, possibly because of the practical difficulty of

following up a large number of patients over a prolonged
period, and none of the available studies specifically examined diabetic control as reflected by hemoglobin A1c (14).
Indeed, notwithstanding the various acute and chronic infections frequently observed among diabetic subjects, relatively little is known about the underlying mechanism(s) or
the exact role of diabetic control on infection risks (11). In
vitro immune studies and skin testing among diabetic subjects have shown possible impairment of adaptive immune
responses among diabetic subjects (1215), but there have
been conflicting results with regard to the role of diabetic
control (11, 1416). The current study provided the first
unequivocal demonstration, in vivo, of the primary impact
of diabetic control on the development of tuberculosis.
The excess risk in this study is considerably lower than
that reported in another Asian population (1) or the Hispanics (3). Diabetic control was shown to be the predominant determinant of increased tuberculosis risk in this study.
Of some interest is the underlying reason why subjects with
well-controlled diabetes mellitus were not at increased risk
of pulmonary tuberculosis even after control of other confounding variables, including body mass index. Indeed, their
Am J Epidemiol 2008;167:14861494

risk was consistently on the low side, and this could have led
to a lower relative risk for diabetic subjects as a whole.
Major ethnic differences were reported in another study,
with relative risks varying from 2.95 among Hispanics,
1.31 among non-Hispanic Whites, and 0.93 among nonHispanic Blacks (17). The underlying mechanism(s) for
such major ethnic differences remains obscure, and differences in prevalence of latent infection and/or annual disease
incidence are not expected to affect internal comparisons
within population subgroups. Our results suggest that better
diabetic control would reduce tuberculosis risk, which might
therefore confound the ethnic differences as observed.
The increased risk of diabetes mellitus was relatively
specific for pulmonary tuberculosis but not extrapulmonary
tuberculosis. Although a type II error was possible with the
limited number of extrapulmonary tuberculosis cases, the
adjusted hazard ratios of diabetes mellitus for any extrapulmonary involvement were rather close to or below one,
irrespective of the diabetic control. Most of the previous
studies focused mainly on pulmonary disease (14), and
diabetes mellitus has been associated with more frequent
lower lung field lesions and increased cavity formation
(18, 19). In a case-control study in the United States,

Downloaded from http://aje.oxfordjournals.org/ at Georgetown University on May 21, 2015

No diabetes mellitus

1492 Leung et al.

pulmonary tuberculosis patients were found to have a higher

prevalence of diabetes mellitus than those with extrapulmonary tuberculosis (20). Differential effects of diabetes mellitus on pulmonary and extrapulmonary tuberculosis would
help to explain such an observation.
Extrapulmonary involvement is a hallmark of most immunocompromising conditions. Use of the tumor necrosis
factor inhibitors infliximab and etanercept has been associated with a very high tuberculosis risk (21, 22), but a high
proportion of extrapulmonary involvement was observed
(2123). Among individuals infected with human immunodeficiency virus, extrapulmonary tuberculosis involvement
is also a feature of advanced immunodeficiency (24). Besides diabetic control, a relatively specific effect on pulmonary tuberculosis has also been reported for smoking (25)
and body mass index (9, 26). Although the effect of smoking
on pulmonary tuberculosis is readily accounted for by its
site-specific action (27, 28) and the more frequent lower
lung field involvement in diabetics could also be compatible
with a perfusion-related factor, a close interplay between the
energy metabolism and immune system could well be implicated in the relatively specific pulmonary effect of both
diabetic control and body mass index. Further exploration of
the exact mechanism(s) mediating the link between diabetic
control and host defense might therefore be warranted.
Incomplete case ascertainment is always a major concern
in prospective cohort analyses. The cohort was followed up
through a territory-wide notification registry and a death
registry, which captured, respectively, tuberculosis cases
and deaths all over Hong Kong. The presence of a good

Am J Epidemiol 2008;167:14861494

Downloaded from http://aje.oxfordjournals.org/ at Georgetown University on May 21, 2015

FIGURE 1. Cumulative hazards for active tuberculosis with respect

to baseline diabetes mellitus (DM) status and hemoglobin A1c
(HbA1c) level after adjustment for potentially confounding variables
by Cox regression analysis, Hong Kong, Peoples Republic of China,
20002005.

health-care infrastructure, the statutory requirement for all

adult citizens to carry an identity card, the statutory tuberculosis notification system, and the widespread use of the
identity card number as a unique identifier were important
factors that facilitated the current study. Undernotification
remains a concern with regard to all notifiable diseases (29).
However, the cohort was already under the care of a service
in the Department of Health and had ready access to the
chest clinics in the same department for free tuberculosis
treatment. In a local audit of tuberculosis notifications, the
undernotification rate was only 3 percent in the chest clinics
even before the introduction of specific improvement measures (30). As we examined the differences between subgroups in this cohort, there should be good internal validity.
No information was available about the tuberculin status
or the newer interferon assays because such screening was
not regularly performed. The sensitivity of the tuberculin
skin test has been shown to be affected by advanced age
(31), while insufficient information is available for the new
interferon release assays (32). Although active tuberculosis
disease was the outcome actually measured in this study, it
would still be difficult to separate the effect of diabetes
mellitus on infection risk, disease development, or both in
the absence of baseline infection data. However, exposure is
a prerequisite for infection. If diabetes mellitus was indeed
associated with an increase in risk of tuberculosis infection,
such an association might be expected to be mediated
through increased exposure, possibly as a consequence of
various confounding sociodemographic and nosocomial
factors. In this regard, it is noteworthy from tables 4 and 5
that the association between diabetes mellitus and tuberculosis disease persisted even after control of these potential
confounders or hospital admissions. A study of molecular
clustering in Hong Kong also suggested that up to 80 percent of the tuberculosis cases in Hong Kong occurred as
a result of endogenous reactivation of remote infection
rather than recent transmission (33). With the significant
past burden of tuberculosis (5), a high proportion of latent
tuberculosis infection is expected among this elderly cohort.
A previous study has reported a tuberculin reactivity rate
(10 mm) of 68.6 percent, in the absence of recent contact
history, among inmates of old-age homes after two-stage
tuberculin skin testing (34). Furthermore, the effect of
diabetes mellitus was seen predominantly on pulmonary
tuberculosis and not extrapulmonary involvement. Such an
observation might be in favor of an effect on disease
manifestation, rather than solely as a result of increased
exposure.
This is a study on a cohort of elderly in community settings. Type II diabetes mellitus is increasingly prevalent in
both developed and developing areas (3538). In Hong
Kong, 97 percent of diabetes mellitus cases are type II diabetes mellitus (39). Although no specific attempt was made
to distinguish type I and type II diabetes mellitus in this
study, practically all diabetes mellitus cases included in
the cohort were expected to be suffering from type II diabetes mellitus, judging from the their advanced age (36).
Higher relative tuberculosis risks have been reported among
younger diabetic subjects (1, 16) or those with type I diabetes mellitus (3). Further studies will therefore be

Diabetic Control and Tuberculosis

ACKNOWLEDGMENTS

The authors wish to thank the staffs of the Elderly Health

Service and the Tuberculosis and Chest Service for their
assistance in collection of the raw data.
Conflict of interest: none declared.

REFERENCES
1. Kim SJ, Hong YP, Lew WJ, et al. Incidence of pulmonary tuberculosis among diabetics. Tuber Lung Dis 1995;76:52933.
2. Coker R, McKee M, Atun R, et al. Risk factors for pulmonary
tuberculosis in Russia: case-control study. BMJ 2006;332:
857.
3. Olmos P, Donoso J, Rojas N, et al. Tuberculosis and diabetes
mellitus: a longitudinal-retrospective study in a teaching
hospital. Rev Med Chil 1989;117:97983.
Am J Epidemiol 2008;167:14861494

4. Banyai AL. Diabetes and pulmonary tuberculosis. Am Rev

Tuberc 1931;24:65067.
5. Tuberculosis and Chest Service. Annual report of Tuberculosis
and Chest Service (2003). Hong Kong, Peoples Republic
of China: Department of Health, 2004.
6. Leung CC, Li T, Lam TH, et al. Smoking and tuberculosis
among the elderly in Hong Kong. Am J Respir Crit Care Med
2004;170:102733.
7. WHO Consultation: definition, diagnosis and classification of
diabetes mellitus and its complications. Part 1: diagnosis
and classification of diabetes mellitus. Geneva, Switzerland:
World Health Organization, 1999. (Publication no. WHO/
NCD/NCS/99.2).
8. Chalmers J, MacMahon S, Mancia G, et al. 1999 World Health
OrganizationInternational Society of Hypertension guidelines for the management of hypertension. Guidelines
subcommittee of the World Health Organization. Clin Exp
Hypertens 1999;21:100960.
9. Leung CC, Lam TH, Chan WM, et al. Lower risk of tuberculosis in obesity. Arch Intern Med 2007;167:1297304.
10. Saudek CD, Derr RL, Kalyani RR. Assessing glycemia in
diabetes using self-monitoring blood glucose and
hemoglobin A1c. JAMA 2006;295:168897.
11. Peleg AY, Weerarathna T, McCarthy JS, et al. Common
infections in diabetes: pathogenesis, management and
relationship to glycaemic control. Diabetes Metab Res Rev
2007;23:313.
12. MacCuish AC, Urbaniak SJ, Campbell CJ, et al. Phytohemagglutinin transformation and circulating lymphocyte
subpopulations in insulin-dependent diabetic patients.
Diabetes 1974;23:70812.
13. Casey JI, Heeter BJ, Klyshevich KA. Impaired response of
lymphocytes of diabetic subjects to antigen of Staphylococcus
aureus. J Infect Dis 1977;136:495501.
14. Eibl N, Spatz M, Fischer GF, et al. Impaired primary immune
response in type-1 diabetes: results from a controlled
vaccination study. Clin Immunol 2002;103:24959.
15. Pozzilli P, Pagani S, Arduini P, et al. In vivo determination of
cell mediated immune response in diabetic patients using a multiple intradermal antigen dispenser. Diabetes Res 1987;6:58.
16. Spatz M, Eibl N, Hink S, et al. Impaired primary immune
response in type-1 diabetes. Functional impairment at the
level of APCs and T-cells. Cell Immunol 2003;221:1526.
17. Pablos-Mendez A, Blustein J, Knirsch CA. The role of
diabetes mellitus in the higher prevalence of tuberculosis
among Hispanics. Am J Public Health 1997;87:5749.
18. Perez-Guzman C, Torres-Cruz A, Villarreal-Velarde H, et al.
Atypical radiological images of pulmonary tuberculosis in
192 diabetic patients: a comparative study. Int J Tuberc
Lung Dis 2001;5:45561.
19. Shaikh MA, Singla R, Khan NB, et al. Does diabetes alter the
radiological presentation of pulmonary tuberculosis? Saudi
Med J 2003;24:27881.
20. Antony SJ, Harrell V, Christie JD, et al. Clinical differences
between pulmonary and extrapulmonary tuberculosis: a 5-year
retrospective study. J Natl Med Assoc 1995;87:18792.
21. Keane J, Gershon S, Wise RP, et al. Tuberculosis associated
with infliximab, a tumor necrosis factor alpha-neutralizing
agent. N Engl J Med 2001;345:1098104.
22. Mohan AK, Cote TR, Block JA, et al. Tuberculosis following
the use of etanercept, a tumor necrosis factor inhibitor. Clin
Infect Dis 2004;39:2959.
23. Yang Z, Kong Y, Wilson F, et al. Identification of risk
factors for extrapulmonary tuberculosis. Clin Infect Dis 2004;
38:199205.

Downloaded from http://aje.oxfordjournals.org/ at Georgetown University on May 21, 2015

required to assess the effect of diabetic control in younger

subjects. The past history of tuberculosis was not specifically included for all enrolled subjects, but only 17.2 percent
of observed tuberculosis cases had a past tuberculosis history. There could well be concern that tuberculosis could
affect the diabetic control before it was clinically manifest.
However, only tuberculosis cases occurring 3 months after
recruitment were included in the analysis, and exclusion of
subjects with factors potentially associated with subclinical
tuberculosis disease did not alter the effect. The prospective
design of the study and the relatively uniform rate of tuberculosis in each subgroup during the 5 years of follow-up
(figure 1) are not in support of reverse causality. With the
urine and blood screening at enrollment, most diabetes mellitus patients should have been picked up. Some elderly
subjects could develop diabetes mellitus afterwards, but
the number of incident diabetes mellitus cases should be
relatively small and unlikely to affect our findings significantly. The hemoglobin A1c level was used to reflect longer
term diabetic control, rather than single blood sugar levels.
Only the baseline level of hemoglobin A1c for those with
known diabetes mellitus at enrollment was available, as
clinical follow-up of established conditions was regularly
carried out in other institutions. However, this would also
avoid the potential issue of reverse direction of causality for
interim hemoglobin A1c measurements.
With rapidly increasing prevalence of diabetes mellitus in
many parts of the world, intensified efforts on basic, clinical,
and epidemiologic research are called for to contain the
adverse health effects of this new epidemic. The impact of
diabetic control on tuberculosis risk also highlights the need
for good diabetic control among patients with known diabetes mellitus. Monitoring of hemoglobin A1c should be
carried out regularly among diabetic patients, especially in
a high tuberculosis prevalence country, where interaction
between the old scourge and the new enemy could further
aggravate human suffering. A heightened clinical awareness
of tuberculosis is also indicated among those with poor diabetic control and suggestive symptoms.

1493

1494 Leung et al.

32. Leung CC, Yam WC, Yew WW, et al. Comparison of

T-Spot.TB and tuberculin skin test among silicotic patients.
Eur Respir J 2008;31:26672.
33. Chan-Yeung M, Tam CM, Wong H, et al. Molecular and
conventional epidemiology of tuberculosis in Hong Kong:
a population-based prospective study. J Clin Microbiol 2003;
41:27068.
34. Chan-Yeung M, Chan FH, Cheung AH, et al. Prevalence
of tuberculous infection and active tuberculosis in old
age homes in Hong Kong. J Am Geriatr Soc 2006;54:
133440.
35. Leung CC, Yew WW, Chan CK, et al. Tuberculosis in older
people: a retrospective and comparative study from Hong
Kong. J Am Geriatr Soc 2002;50:121926.
36. Cheung BM, Wat NM, Man YB, et al. Development of
diabetes in Chinese with the metabolic syndrome: a 6-year
prospective study. Diabetes Care 2007;30:14306.
37. Meigs JB, Muller DC, Nathan DM, et al. The natural history of
progression from normal glucose tolerance to type 2 diabetes
in the Baltimore Longitudinal Study of Aging. Diabetes
2003;52:147584.
38. Mathers CD, Loncar D. Projections of global mortality and
burden of disease from 2002 to 2030. (Electronic article).
PLoS Med 2006;3:e442.
39. Chan BS, Tsang MW, Lee VW, et al. Cost of type 2 diabetes
mellitus in Hong Kong Chinese. Int J Clin Pharmacol Ther
2007;45:45568.

Am J Epidemiol 2008;167:14861494

Downloaded from http://aje.oxfordjournals.org/ at Georgetown University on May 21, 2015

24. Jones BE, Young SM, Antoniskis D, et al. Relationship of

the manifestations of tuberculosis to CD4 cell counts in
patients with human immunodeficiency virus infection. Am
Rev Respir Dis 1993;148:12927.
25. Leung CC, Yew WW, Chan CK, et al. Smoking and tuberculosis in Hong Kong. Int J Tuberc Lung Dis 2003;7:9806.
26. Tverdal A. Body mass index and incidence of tuberculosis.
Eur J Respir Dis 1986;69:35562.
27. Reynolds PR, Cosio MG, Hoidal JR. Cigarette smokeinduced Egr-1 upregulates proinflammatory cytokines in
pulmonary epithelial cells. Am J Respir Cell Mol Biol 2006;
35:31419.
28. Szulakowski P, Crowther AJ, Jimenez LA, et al. The effect of
smoking on the transcriptional regulation of lung inflammation
in patients with chronic obstructive pulmonary disease. Am
J Respir Crit Care Med 2006;174:4150.
29. Sheldon CD, King K, Cock H, et al. Notification of tuberculosis: how many cases are never reported? Thorax 1992;47:
101518.
30. Health Services Research Group, Department of Community
Medicine, The University of Hong Kong. Services for the
treatment of patients with tuberculosis in Hong Kong, final
report to the project steering group. Hong Kong, Peoples
Republic of China: The University of Hong Kong, 2000.
31. Nisar M, Williams CS, Ashby D, et al. Tuberculin testing
in residential homes for the elderly. Thorax 1993;48:
125760.