Supplementary appendixThis appendix formed part of the original submission and has been peer reviewed. We post it as supplied by the authors.
Supplement to: Dodd PJ, Gardiner E, Coghlan R, Seddon JA. Burden of childhood tuberculosis in 22 high-burden countries: a mathematical modelling study. Lancet Glob Health 2014; published online July 9. http://dx.doi.org/10.1016/S2214-109X(14)70245-1.
Modelling the 22 HBC paediatric TB burden
1
Supplementary Information for ‘Estimating the burden of childhood tuberculosis in the
twenty-two high TB burden countries: a mathematical modelling study’
P.J. Dodd, E. Gardiner, R. Coghlan and J.A. Seddon
Supplementary methods .............................................................................................................................................. 2
Description of data used ........................................................................................................................................... 2
Figure 1: The demography of the 22 HBC populations in 2010. ......................................................................... 2
Figure 2: WHO estimates for total per-capita TB prevalence. ............................................................................. 3
Figure 3: TB notifications by type of TB in 2010 for the 22 HBCs. .................................................................... 4
Figure 4: Types of TB notification in 2010 by age and sex. ................................................................................ 4
Table 1: Those HBC countries with comparable DHS data on household structure. ........................................... 5
Figure 5: DHS household demography - the example of India. ........................................................................... 6
Figure 6: Coverage of BCG vaccination in the 22 HBCs, data from WHO. ........................................................ 7
Figure 7: HIV prevalence in individuals under 15 ............................................................................................... 8
Infection rates ........................................................................................................................................................... 8
Table 2: Literature available used to inform community transmission parameters. ............................................. 9
Figure 8: Log-normal fit to literature estimates of the transmission parameter. ................................................ 10
Table 3: Choice of transmission parameters describing the rates of infection. .................................................. 11
Cohabitation analysis ............................................................................................................................................. 11
Figure 9: TB cohabitation probability (age-dependent model). ......................................................................... 12
Figure 10: TB cohabitation probability (age-uniform model). ........................................................................... 13
Figure 11: Cohabitation probability gradients. ................................................................................................... 14
Disease progression ................................................................................................................................................ 14
Table 4: Choice of parameters for TB disease progression risks. ...................................................................... 15
Protection from BCG .............................................................................................................................................. 15
Table 5: Choice of parameters describing the protective effect of BCG vaccination. ....................................... 16
The effect of HIV ..................................................................................................................................................... 16
Table 6: Choice of parameter describing the IRR for developing TB disease given HIV. ................................ 16
Numerical methods ................................................................................................................................................. 16
Limitations .............................................................................................................................................................. 16
Supplementary results ............................................................................................................................................... 18
Figure 12: Model estimates for all TB incidence in those aged under 15. ......................................................... 19
Table 7: Model estimates for all TB incidence in those aged under 15 years. ................................................... 19
Figure 13: A mosaic plot of the mean proportion of the estimated burden. ....................................................... 20
Table 8: Model estimates of total TB incidence in children 0-4 years and 5-14 years. ..................................... 21
Figure 14: The proportion of the incidence of paediatric TB in each country that is HIV associated. .............. 22
References .................................................................................................................................................................. 23
Modelling the 22 HBC paediatric TB burden
2
Supplementary methods
Description of data used
All of the 22 high TB burden countries (HBCs) had data available on their demography in 2010 (from UN ESA,
Population Division (1) see Figure 1). Where necessary, 5-year age categories were disaggregated under the
assumption of uniformly distributed ages. These data were used to generate the number of children at risk in each
country by age, and as denominators in the age-weighting of adult prevalence for the age-structured cohabitation
analysis (see below).
Figure 1: The demography of the 22 HBC populations in 2010, data from UN ESA.
Estimates of adult TB prevalence were available for all 22 HBCs in 2010, together with 95% uncertainty bounds
(from (2), see Figure 2). Uncertainty in per-capita prevalence was represented by gamma distributions,
parameterised by taking the quoted ranges defined by the upper and lower bounds as 1.96 x the standard deviation,
and the quoted point estimate as the mean. Similarly, all 22 HBCs reported breakdowns of TB notifications by type,
i.e. smear positive, smear negative, smear unknown, extrapulmonary, other (from WHO (2), see Figure 3). These
data were used to estimate the proportion of incident TB that is smear positive for the community ARI estimate. The
same estimate was used for all countries to avoid bias resulting from different case detection infrastructures etc.
Afghanistan Bangladesh Brazil Cambodia China
Democratic Republic of Congo Ethiopia India Indonesia Kenya
Mozambique Myanmar Nigeria Pakistan Philippines
Russian Federation South Africa Thailand Uganda United Republic of Tanzania
Viet Nam Zimbabwe
0
1000
2000
3000
4000
5000
0
5000
10000
15000
0
5000
10000
15000
0
500
1000
1500
0e+00
5e+04
1e+05
0
3000
6000
9000
0
5000
10000
0
25000
50000
75000
100000
125000
0
5000
10000
15000
20000
25000
0
2000
4000
6000
0
1000
2000
3000
4000
0
1000
2000
3000
4000
5000
0
10000
20000
0
5000
10000
15000
20000
0
3000
6000
9000
0
4000
8000
12000
0
2000
4000
0
2000
4000
0
2000
4000
6000
0
2000
4000
6000
8000
0
2500
5000
7500
0
500
1000
1500
2000
0−
45
−9
10
−1
41
5−
19
20
−2
42
5−
29
30
−3
43
5−
39
40
−4
44
5−
49
50
−5
45
5−
59
60
−6
46
5−
69
70
−7
47
5−
79
80
−
0−
45
−9
10
−1
41
5−
19
20
−2
42
5−
29
30
−3
43
5−
39
40
−4
44
5−
49
50
−5
45
5−
59
60
−6
46
5−
69
70
−7
47
5−
79
80
−
Age
Num
ber
(th
ousa
nds)
2010
Modelling the 22 HBC paediatric TB burden
3
Figure 2: WHO estimates for total per-capita TB prevalence and 95% uncertainty bounds for the 22 HBCs in
2010.
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Afghanistan
Bangladesh
Brazil
Cambodia
China
Democratic Republic of Congo
Ethiopia
India
Indonesia
Kenya
Mozambique
Myanmar
Nigeria
Pakistan
Philippines
Russian Federation
South Africa
Thailand
Uganda
United Republic of Tanzania
Viet Nam
Zimbabwe
0 500 1000
TB prevalence per 100,000
cou
ntr
y
2010
Modelling the 22 HBC paediatric TB burden
4
Figure 3: TB notifications by type of TB in 2010 for the 22 HBCs, data from WHO.
For 21 of the HBCs, breakdowns of TB notifications in 2010 were also available by age and sex ((2), see Figure 4).
These data were used as the numerator in the age-weighted assignment of prevalence TB cases in the age-structured
cohabitation analysis. Data were not available for Mozambique.
Figure 4: Types of TB notification in 2010 by age and sex for the 22 HBCs, data from WHO.
Afghanistan
Bangladesh
Brazil
Cambodia
China
Democratic Republic of Congo
Ethiopia
India
Indonesia
Kenya
Mozambique
Myanmar
Nigeria
Pakistan
Philippines
Russian Federation
South Africa
Thailand
Uganda
United Republic of Tanzania
Viet Nam
Zimbabwe
0 25 50 75 100
Proportion (%)
co
untr
y
new TB type:
smr+
smr−
smr?
EP
other
Types of new TB notifications in 2010
malemalemalefemalefemalefemale malemalemalefemalefemalefemale malemalemalefemalefemalefemale malemalemalefemalefemalefemale malemalemalefemalefemalefemale
malemalemalefemalefemalefemale malemalemalefemalefemalefemale malemalemalefemalefemalefemale malemalemalefemalefemalefemale malemalemalefemalefemalefemale
malemalemalefemalefemalefemale malemalemalefemalefemalefemale malemalemalefemalefemalefemale malemalemalefemalefemalefemale malemalemalefemalefemalefemale
malemalemalefemalefemalefemale malemalemalefemalefemalefemale malemalemalefemalefemalefemale malemalemalefemalefemalefemale malemalemalefemalefemalefemale
malemalemalefemalefemalefemale malemalemalefemalefemalefemale
Afghanistan Bangladesh Brazil Cambodia China
Democratic Republic of Congo Ethiopia India Indonesia Kenya
Mozambique Myanmar Nigeria Pakistan Philippines
Russian Federation South Africa Thailand Uganda United Republic of Tanzania
Viet Nam Zimbabwe
0−1415−2425−3435−4445−5455−64
65+
0−1415−2425−3435−4445−5455−64
65+
0−1415−2425−3435−4445−5455−64
65+
0−1415−2425−3435−4445−5455−64
65+
0−1415−2425−3435−4445−5455−64
65+
−0.2 −0.1 0.0 0.1 0.2 0.3 −0.2 −0.1 0.0 0.1 0.2 0.3
Proportion of 2010 notifications
Ag
e
type
smr+
smr−
EP
Modelling the 22 HBC paediatric TB burden
5
For 16 of the HBCs (see Table 1), comparable Measure DHS surveys were available (3) including full records of the
ages and sexs living in households, together with formal sample weightings for use in analysis. The DHS survey
data were used to generate synthetic populations with age-, sex- and household-structure representative of each of
these countries, for use in the cohabitation analyses described below. Figure 5 is a summary representation of the
household demography for India. Similar plots are available for all other countries from the authors on request.
Table 1: Those HBC countries with comparable DHS data on household structure, together with the year of
the survey.
Country Year
Bangladesh 2011
Brazil 1996
Cambodia 2010
Democratic Republic of Congo 2007
Ethiopia 2011
India 2006
Indonesia 2007
Kenya 2009
Mozambique 2011
Nigeria 2008
Philippines 2008
South Africa 2003
United Republic of Tanzania 2010
Uganda 2011
Viet Nam 2002
Zimbabwe 2011
Modelling the 22 HBC paediatric TB burden
6
Figure 5: DHS household demography - the example of India. Each panel represents the age and sex distribution
of those living in a given household size as a pyramid plot.
BCG vaccination coverage estimates were available for all 22 HBCs (from WHO (4), see Figure 6). BCG schedules
(5) all recommended vaccination at birth, with only Nigeria and The Russian Federation including repeat
vaccination of older children. Vaccine strains often differed between countries. The BCG vaccination coverages
were used to determine the fraction of children whose risks of progression from infection to disease were moderated
by BCG (in the manner described in the relevant section below).
1 2 3 4
5 6 7 8
9 10 11 12
13 14
0
20
40
60
80
0
20
40
60
80
0
20
40
60
80
0
20
40
60
80
density
Age
sex
female
male
India
Modelling the 22 HBC paediatric TB burden
7
Figure 6: Coverage of BCG vaccination in the 22 HBCs, data from WHO. N.B. the axis displaying coverage
does not begin at 0%.
HIV prevalence estimates in those aged under 15 were available for 9 of the HBCs (source (6), see Figure 7),
together with 95% uncertainty bounds. Countries for which there were not estimates reported from this source were
assumed to have negligible HIV prevalence in those under 15 years of age. Among the 22 HBCs taken to have zero
paediatric HIV prevalence for this reason, Thailand and the Russian Federation had the highest adult prevalences of
HIV, at 1.2% and 1.1% respectively, the rest being 0.6% or lower. Uncertainty in the prevalences was represented
by gamma distributions, parameterised by taking the quoted ranges defined by the upper and lower bounds were
taken as 1.96 x the standard deviation, and the quoted point estimate as the mean. This HIV prevalence was assumed
to be uniform by age in those under 15. Degree of immunosuppression or ART was not considered.
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Afghanistan
Bangladesh
Brazil
Cambodia
China
Democratic Republic of Congo
Ethiopia
India
Indonesia
Kenya
Mozambique
Myanmar
Nigeria
Pakistan
Philippines
Russian Federation
South Africa
Thailand
Uganda
United Republic of Tanzania
Viet Nam
Zimbabwe
70 80 90 100
% BCG coverage in 2010
cou
ntr
y
Modelling the 22 HBC paediatric TB burden
8
Figure 7: HIV prevalence in individuals under 15 in those countries reporting HIV prevalence in this age
range to UNAIDS. Other countries were assumed to have negligible HIV prevalence in this age group.
Infection rates
Karel Styblo first empirically estimated the coefficient between the prevalence of smear-positive adult TB and the
annual risk of infection, (7). Using European data on ARIs from TST surveys combined with prevalence surveys,
he inferred from a rate of 10 infections per year for a typical smear positive TB case. This quantity has been
revisited since (8–10), as lower average case severity from improved case-detection, the rise of HIV-related
tuberculosis, and social changes may all have led to departures from Styblo’s estimate; indeed, more recent
estimates tend to be lower. It should be noted that from a modelling perspective, use of this community relationship
between smear positive TB and the ARI does not imply that infections caused by smear negative TB have been
altogether neglected. Rather, it can be thought of as modelling the infections due to smear positive TB cases plus the
product of the number of smear negative cases per smear positive case and the relative infectiousness of a smear
negative case, i.e. as inflated to indirectly include the smear negative contribution to transmission.
We used data from the reviews of Bourdin Trunz and colleagues (10) and van Leth and colleagues (9) (see Table 2)
and fitted a log-normal distribution to these estimates of (see Figure 8 and Table 3), which was used to model the
distribution of values for this coefficient in the community ARI approach to estimating infection incidence.
●
●
●
●
●
●
●
●
●
Democratic Republic of Congo
Ethiopia
Kenya
Mozambique
Nigeria
South Africa
Uganda
United Republic of Tanzania
Zimbabwe
1 2 3 4
Percent of those aged <15 years with HIV (%)
cou
ntr
y
Modelling the 22 HBC paediatric TB burden
9
Table 2:Literature available used to inform community transmission parameters.
Country year ARI year smr+ TB
prevalence Ref
Cambodia 2002 2.06 20021 269 7.8 T21
China 1979 1.1 19791 187 5.7 T22
China 1990 1.0 19901 134 7.2 T22
China 2000 0.7 20001 110 6.5 T23
The Philippines 1981-83 2.5 1981-831 660 3.8 T25/5
The Philippines 1997 2.3 19971 310 7.4 T24/5
South Korea 1965 5.3 19651 668 7.9 T26
South Korea 1970 3.9 19701 559 7.0 T26
South Korea 1975 2.3 19751 480 4.8 T26
South Korea 1980 1.8 19801 309 5.8 T26
South Korea 1985 1.2 19851 239 5.0 T26
South Korea 1990 1.1 19901 143 7.7 T26
South Korea 1995 0.5 19951 93 5.4 T26
Kenya 1994-96 1.1 19952 132 8.3 T27
Tanzania 1993-98 0.9 19962 172 5.2 T28
Tanzania 2000 0.68 20002 214 3.1 T29
Malawi 1994 1.0 19952 248 4.0 T30
Madagascar 1991-94 1.2 19952 145 8.4 T31
Egypt 1995-97 0.3 19962 24 13.2 T32
Laos 1996 1.1 19962 165 6.7 T33
India 2000-03 1.5 20022 155 9.6 T34
South Korea 1975 1.78 19751 480 3.7 V7/8
South Korea 1980 1.24 19801 310 4.0 V7/8
South Korea 1985 1.12 19851 240 4.7 V7/8
South Korea 1990 0.6 19901 144 3.2 V7/8
China 1979 0.64 19791 187 3.4 V9
China 1985 0.59 19851 156 3.8 V9
China 1990 0.64 19901 134 4.8 V10
Modelling the 22 HBC paediatric TB burden
10
China 1996 0.7 19961 121 5.8 V10
The Philippines 1982 2.48 19821 950 2.6 V11
The Philippines 1994 2.30 19941 525 4.4 V11
1. Survey
2. WHO estimate
Tn =Bourdin Trunz and colleagues (10) - ref n; Vn =van Leth and colleagues (9) - ref n;
T21-T32 = (11), (12), (13), (14), (15), (16), (17), (18), (19), (20), (21), (22), (23);
V7-V11= (24), (16), (12), (13), (14)
Figure 8: Log-normal fit to literature estimates of the transmission parameter, beta.
Papers by Rutherford and colleagues (25) and by Lienhardt and colleagues (26) each described the infection rates
among children who were living with a smear-positive TB case by the time of notification. Rutherford and
colleagues in Indonesia (25) found 48% of child household contacts to be infected by tuberculin skin test (10mm
cut-off) and 51% positive by interferon gamma release assay. Lienhardt and colleagues (26) found 25.8% of child
household contacts positive by TST with 10mm cut-off and 33.6% positive by TST with 5mm cut-off. The
systematic review and meta-analysis of TB contact investigations by Fox and colleagues (27) reported rates of
infection of 45.4% [95% CI: 40.7% to 50.2%] among household contacts of all ages in low- and middle-income
countries, averaged over index smear-status. To interpret these numbers in terms of a transmission rate, one needs to
make an assumption about the duration of household exposure. The assumption of a typical duration before
0.00
0.05
0.10
0.15
0.20
0 5 10 15
Transmission parameter, b
den
sity
Modelling the 22 HBC paediatric TB burden
11
notification of ~1 year (as in (28); in the context of a ~3 year duration of TB disease without treatment (29))
translates these numbers into rates of 0.65y-1
and 0.71y-1
for (25), and 0.41y-1
and 0.30y-1
for (26). We therefore
modelled the uncertainty in this parameter using a log-normal distribution with an approximate correspondence
between its median and the midpoint of these estimates, and between the distribution’s inter-quartile range their
range, i.e. each model run sampled beta from this distribution (as were other parameters from their distributions).
See Table 3.
Table 3: Choice of transmission parameters describing the rates of infection for a given exposure, modelled as
log-normal distributions.
Quantity mu sigma median LQ UQ
community (y-1) 1.678 0.371 5.354 4.168 6.879
household (y-1) -0.693 0.500 0.500 0.357 0.701
Cohabitation analysis
The cohabitation analyses estimated the risk of infection in children via their risk of sharing a household with a TB
case. To calculate the probability of sharing a household with a TB case we took two approaches. In each approach,
the per-capita estimate of prevalence for the country was used in conjunction with the DHS household data, which
was taken to constitute a synthetic population. In the first, age-dependent cohabitation analysis (labelled ‘hhage’),
the probability of an individual being a TB case was dependent on age and sex, with the distribution being taken
from the smear positive age- and sex-stratified WHO notification data where available. Mozambique, for which
comparable DHS data were available, did not have age-structured notification data and so was not included in this
part of the analysis. Smear negative cases were assumed to follow the same distribution. To calculate the probability
that a child in a particular age group was sharing a household with a TB case, children of that age were repeatedly
sampled from the synthetic population, with weights corresponding to the survey design weights in the DHS sample-
frame. The mean probability of their sharing with a TB case was then calculated as a the mean of:
∏
for each pick, where is the probability of the i-th cohabiting individual being a TB case, and the product is taken
over the other individuals in the household. 10,000 children were sampled at each per-capita population TB
prevalence, and the mean over these used as an estimator for the probability of cohabiting with a TB case.
As the notification data are subject to various biases, and may not fairly reflect the age and sex distribution of
prevalent TB, we carried out a separate age-uniform cohabitation analysis (labelled ‘hhflat’) in which the
distribution of prevalent cases of TB was assumed to be flat across those aged 15 years or older. The procedure was
otherwise identical. Mozambique was included in this analysis.
The results for the probability of cohabiting with a TB case as per-capita population TB prevalence varies between 0
and 1000 per 100,000 are shown in Figure 9 (age-dependent analysis) and Figure 10 (age-uniform analysis). The
probabilities for the different age categories were very similar for a given TB prevalence and, to a good
approximation, linear in the per-capita population TB prevalence. This was used to simplify the onward analysis as
the gradients of the probability with respect to TB prevalence were extracted (compared for both methods in Figure
11), and used in the infection model. Where the gradient was not available for a given country and method, either
Modelling the 22 HBC paediatric TB burden
12
due to the absence of household data, or the absence of age-stratified notification data for the age-dependent analysis,
it was sampled for each run from those of other countries.
The probability that a cohabiting TB case was smear positive was then taken using the mean proportion of notified
cases that were smear positive from the data presented in Figure 3. Smear negative cases were assumed to be 23% as
infectious per unit time as smear positive cases, based on the molecular studies (30,31).
Figure 9: TB cohabitation probability (age-dependent model). The probability of sharing a household with a
smear positive TB case as TB prevalence increases, by age, and for each of the 15 HBCs for which both age-
structured notification data and a Measure DHS survey was available. This was calculated in a Monte Carlo fashion
using the age-distribution of TB from the smear positive notification data in Figure 4.
Bangladesh Brazil Cambodia Democratic Republic of Congo
Ethiopia India Indonesia Kenya
Nigeria Philippines South Africa United Republic of Tanzania
Uganda Viet Nam Zimbabwe
0
2
4
6
0
2
4
6
0
2
4
6
0
2
4
6
0 250 500 750 1000 0 250 500 750 1000 0 250 500 750 1000
TB prevalence per 100,000
Pro
babili
ty o
f co
hab
itin
g w
ith a
TB
case (
%)
Age
0
1
2−4
5−9
10−14
Modelling the 22 HBC paediatric TB burden
13
Figure 10: TB cohabitation probability (age-uniform model). The probability of sharing a household with a
smear positive TB case as TB prevalence increases, by age, and for each of the 16 HBCs for which a Measure DHS
survey was available. Those aged 15 year or older were assumed equally likely to be a prevalent case of TB.
Bangladesh Brazil Cambodia Democratic Republic of Congo
Ethiopia India Indonesia Kenya
Mozambique Nigeria Philippines South Africa
United Republic of Tanzania Uganda Viet Nam Zimbabwe
0
2
4
6
0
2
4
6
0
2
4
6
0
2
4
6
0 250 500 750 1000 0 250 500 750 1000 0 250 500 750 1000 0 250 500 750 1000
TB prevalence per 100,000
Pro
babili
ty o
f coh
abitin
g w
ith
TB
ca
se (
%)
Age
0
1
2−4
5−9
10−14
Modelling the 22 HBC paediatric TB burden
14
Figure 11: Cohabitation probability gradients. The rate of increase in the probability of a child sharing a
household with a case of TB as the TB prevalence in the population increases. Dots are the gradients according to
the age-neutral model; triangles are the gradients according to the age-structured model.
Disease progression
Disease progression risks were based on the review of the pre-chemotherapy literature by Marais and colleagues
(32). BCG vaccination had not been used in the populations reviewed in this study. We take these risks as
approximating the risk of progression within a year. The risks of developing TB disease in (32) were used to
parameterise the mean of beta distributions for each age category, with the variance taken as 0.125 x mean
(equivalent to taking a range of plausible variation as +/-25% of the mean, and the variance as a quarter of the
plausible range). The risk of extrapulmonary disease (miliary TB or TB meningitis) was modelled as a random
proportion of the risk of any disease, modelled by a beta distribution with mean equal to the range midpoints in (32)
and the variance taken as a quarter of these ranges. Where no ranges were given, +/-25% of the mean was used as
the range. Where the probability was expressed as <x, x/2 was used as the mean. Modelling the two types of disease
in this way ensured that the random total probability for any kind of disease never exceeded unity. These parameters
and their associated modes, lower- and upper-quartiles are presented in Table 4.
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Bangladesh
Brazil
Cambodia
Democratic Republic of Congo
Ethiopia
India
Indonesia
Kenya
Mozambique
Nigeria
Philippines
South Africa
United Republic of Tanzania
Uganda
Viet Nam
Zimbabwe
4 5 6 7
probability gradient with prevalence
cou
ntr
y
Modelling the 22 HBC paediatric TB burden
15
Table 4: Choice of parameters for TB disease progression risks, modelled as beta distributions.
quantity alpha beta median LQ UQ
probability TB disease, age [0,1) years 1.500 1.500 0.500 0.298 0.702
probability TB disease, age [1,2) years 1.250 3.750 0.250 0.108 0.360
probability TB disease, age [2,5) years 0.330 6.270 0.050 0.002 0.064
probability TB disease, age [5,10) years 0.137 6.703 0.020 0.000 0.013
probability TB disease, age [10,15) years 0.870 4.930 0.150 0.043 0.219
proportion TB EP, age [0,1) years 0.960 2.240 0.300 0.112 0.451
proportion TB EP, age [1,2) years 0.712 1.322 0.350 0.107 0.557
proportion TB EP, age [2,5) years 0.620 5.580 0.100 0.017 0.145
proportion TB EP, age [5,10) years 0.750 5.250 0.125 0.029 0.183
proportion TB EP, age [10,15) years 0.112 6.615 0.017 0.000 0.008
Protection from BCG
The reported efficacy of BCG in protecting against different forms of TB disease varies substantially. Rodrigues and
colleagues (33) conducted a meta-analysis of protection against TB meningitis and miliary TB, finding a protection
of 86% [65%,95%] from RCTs and 75% [61%,84%] from case-control studies. Colditz and colleagues (34) meta-
analysis of BCGs protection for infants found protections against all TB of 74% [62%,83%] from RCTs and 52%
[38%,64%] from case-control studies. They found a protection of against TB meningitis of 64% [12%,86%] and a
protection of 78% [58%,88%] against disseminated TB. Bourdin Trunz and colleagues (10) updated these meta-
analyses to arrive at protections of 73% for TB meningitis and 77% for miliary TB. We choose to model the
protection against extra-pulmonary TB by a hazard ratio (i.e. 1-protection) following a beta distribution, with
parameters presented in Table 5. We chose to model the protection against PTB as a certain fraction of the
protection against EPTB, with this fraction beta distributed with parameters given in Table 5. Taken together, this
approach gave a protection against PTB of median [LQ-UQ] = 54% [38%-69%] and a protection against EPTB of
median [LQ-UQ] = 70% [52%-84%].
There has been suggestion (e.g. (35,36)) that the observed protective effect of BCG varies with latitude, as a result
of differences in the prevalence of non-tuberculous mycobacteria in the environment. We carried out separate
analysis as a sensitivity analysis where we allowed 41% (36) of the protection from BCG to vary with latitude, so
that the protection at the equator was only 59% of that at the poles. The latitude of each country’s capital was used.
Modelling the 22 HBC paediatric TB burden
16
Table 5: Choice of parameters describing the protective effect of BCG vaccination, modelled as beta
distributions.
quantity alpha beta median LQ UQ
hazard ratio for EPTB, all ages 1.250 2.500 0.301 0.155 0.484
fractional protection for PTB, all ages 4.000 1.000 0.841 0.707 0.931
The effect of HIV
For each country, the proportion of children who were HIV-infected had their TB incidence multiplied by an
incidence rate ratio (IRR). There is limited evidence around the effect of HIV infection in children on the rate of
progression to TB disease following infection with Mycobacterium tuberculosis. Seddon and colleagues (37)
reported an HIV prevalence of 29% among paediatric TB cases diagnosed in Cape Town during 2007-2009. Taken
with the UNAIDS estimate of 3% HIV prevalence for those under 15 years of age in South Africa, this gives an IRR
of 13, suggesting that children with HIV are 13 times more likely to develop TB than those HIV negative. Madhi
and colleagues (38) estimated a higher incidence rate ratio (RR, 22.5; 95% CI, 13.4–37.6) among children
hospitalised with TB in Johannesberg, South Africa between 1996-1997 as did Hesseling and colleagues (39) (RR
24.2 95% CI, 17–34) among infants aged up to 12 months in the Western Cape of South Africa between 2004-2006.
We modelled the IRR with a log-normal distribution, described in Table 6.
Table 6: Choice of parameter describing the IRR for developing TB disease given HIV infection, modelled as
a log-normal distribution.
quantity mu sigma median LQ UQ
IRR for TB given HIV infection, all ages 2.996 1.000 20.000 10.118 39.261
Numerical methods
The model was run repeatedly to build up a distribution of outputs, every time sampling each the input parameter
from the probability distribution used to characterise its likely value. The reported analyses were based on sample
sizes of 50,000 runs, using a Latin hypercube sampling scheme to improve coverage of the parameter space.
Gaussian kernel density estimation was used to estimate distributions, unless otherwise stated. Paediatric proportions
were calculated using model estimates as the numerator and WHO all-TB estimates as the denominator. All analyses
were carried out using the R environment for statistical computing (40).
Limitations
This section details the limitations of the model, some of the assumptions that have been made and some of the
omissions.
1. The model only includes the 22 highest burden countries. Although these account for approximately 80%
of the burden, the model does not look at the remaining countries of the world. These other countries are
lower burden or smaller countries and it is likely that they have a very different population structure, health
care system, underlying health problems and TB epidemiology.
Modelling the 22 HBC paediatric TB burden
17
2. The model uses as a starting point the estimated TB prevalence data for adults with TB. These estimates are
themselves based on a simple model of the duration of TB disease; biases in this would reflect in our results.
3. The model uses TB prevalence estimates as an input of burden but uses notifications data as an input of
who in the population gets TB within countries in the age-dependent cohabitation analysis. There may be
systematic differences between the two. For this reason, we included the age-uniform cohabitation model
subvariant, where TB was assumed uniformly distributed across adults by age and sex.
4. It is assumed that the proportions of adults with smear positive and non-smear positive disease in the
notification data is representative of these proportions in the prevalence data.
5. We were not able to access household structure data for China and Russia, which are two of the largest
burden countries and may have very different TB epidemiology and household structure to other countries
due to their political and social history. We were not able to find notifications disaggregated by age for
Mozambique, and so were not able to include this country in the age-structured cohabitation analysis. We
were able to use the ARI approach for these countries, however.
6. We have neglected geographical (and other) heterogeneities within countries; in fact, TB is frequently very
heterogeneous at this scale. Similarly, we have looked at TB only in the general population and not in
special populations such as prisons, mines etc. Although these populations do not usually include children,
adults from these populations are very high risk and when returning home are likely to contribute to
childhood infections.
7. We have assumed that children are only infected by adults and not by other children. Although it is likely
that the majority of infections do result from adults, children, and especially older children, can certainly be
infectious.
8. We have not included any modelling of the specific issues around HIV and household exposure.
Households that include HIV-infected adults are more likely to both have TB but also have children that are
HIV-infected. The impact of HIV on the infectiousness of the adult has also not been included.
9. We have not included elements of the degree of smear positivity- certainly the more positive the sputum the
more likely children are to become infected. Smear status has been considered binary - smear positive or
smear negative.
10. We have not included any measure of the effects of drug resistance on whether children are more or less
likely to be exposed, become infected or develop disease.
11. When assessing the proportion of household-exposed children who become infected we have used studies
that used infection rates in child contacts at the time of diagnosis of the adult. Converting this into a rate
needed by the model required an assumption about the typical timing of case detection, and the assumption
of a uniform infectiousness through time.
12. In common with almost all infectious disease modelling, we have assumed a consistent and linear
relationship between adult TB prevalence and ARI. In reality this relationship may be non-linear and
variable.
13. The model ignores anergic children. We have used a positive test of infection as a surrogate for TB
infection and assumed that it is a necessary stage to pass through towards disease. However, many children
do not mount response to TST or IGRA and in cohorts of children with TB disease a significant proportion
of children have negative tests of infection.
14. We have looked at HIV status in the child as a binary issue (HIV or no HIV). The impact of degree of
immunosuppression and the use of ART has not been included. HIV prevalence in children has been
assumed not to vary with age.
15. BCG has been challenging to model. The literature surrounding its effects is hugely variable, with a
protective effect described as ranging from 0 to 70%. The implications of geographical location are also
unclear. However, we have included a summary protective effect for both pulmonary and nonpulmonary
disease and also modelled the effect of the latitude, based on the best available data. We have not included
any element of BCG strain used, the mode of delivery, efficacy of cold chain etc. We also neglected the
possibility that BCG vaccination protects against infection, as distinct from disease.
Modelling the 22 HBC paediatric TB burden
18
16. When modelling the risk of disease progression following infection we have not included factors such as M.
tuberculosis strain type, host genetics, host nutritional status, exposure to sunlight (and vitamin D levels) or
any other factors other than age, HIV status and BCG vaccination history. It is clear that factors other than
these can play a part in risk of disease progression.
17. We have assumed that risk of infection and risk of disease progression once infected are unrelated.
However, in reality, it is likely that the degree of exposure (and likely bacillary load exposed to), will affect
both the risk of infection as well as the risk of disease progression once infected.
18. From an infectious disease modelling perspective we have assumed that all children at the beginning of
their exposure were non-immune and had not been infected before. We have assumed that the risk of
disease progression following infection is the same for a child who has recently been infected for the first
time as it is for a child who has recently been infected but who has also been previously infected.
19. Literature risks of progression refer to a life-long time-frame, whereas we have interpreted these risks as
operating on a timescale shorter than the width of our age categories. This is necessary with the underlying
model structure to allow for age-dependent risks of progression, since we do not represent time-since-
infection. However, where our age-categories are narrowest, the progression is also likely to be most rapid,
which ameliorates this approximation.
20. The model does not include IPT. While IPT significantly reduces the likelihood of disease progression
following infection, in the high burden countries child contacts are rarely identified and IPT is rarely
provided.
21. The most recent demographic data was typically from 2010. We therefore used TB, HIV and BCG
estimates for this year.
Supplementary results
In this section, we present model estimates of the absolute TB incidence aggregated over extra-pulmonary and
pulmonary TB disease, for younger and older age groups ([0,5) and [5-15), respectively). Results from the
community ARI approach (labelled ‘comm’) and the both age-dependent cohabitation (labelled ‘hhage’) and age-
uniform cohabitation (labelled ‘hhflat’) approaches to modelling infection rates are presented. Results are also
presented from the latitude-independent (labelled ‘nolat’) and latitude-dependent (labelled ‘lat’) approaches to
modelling the protection from BCG vaccination. In total then, there are 6 model variants presented: 3 infection
variants x 2 protection BCG variants = (comm, hhage, hhflat) x (lat, nolat).
Modelling the 22 HBC paediatric TB burden
19
Figure 12: Model estimates for all TB incidence in those aged under 15 years in the 22 HBCs; all model
variants shown. The lines represent a histogram with binwidth 50,000.
In Figure 12 and Table 7 we compare the results from the different model variants aggregated over all age groups
and all of the 22 HBCs. The cohabitation estimates were similar to each other and less than the community
estimates: medians of around 400,000 compared with 650,000 for the latitude-dependent analysis, 300,000
compared with 500,000 for the latitude-independent analysis. Modes were approximately 20% lower than
corresponding medians. The community ARI model variant consistently gave the highest estimates and the age-
dependent cohabitation analysis the lowest. The latitude-dependent model variants of BCG protection consistently
estimated higher numbers than the latitude-independent model variants. Across all model variants considered, the
lowest lower quartile predicted was approximately 0.2 million cases per year, and the highest upper quartile
predicted was approximately 1 million cases per year.
Table 7: Model estimates for all TB incidence in those aged under 15 years in the 22 HBCs; all model variants
shown. (LQ=lower quartile; UQ=upper quartile.) N.B. from a different sample to results in main paper.
LAT ARI median LQ UQ
lat comm 651,731 424,628 988,953
lat hhage 395,167 271,453 567,114
lat hhflat 437,528 300,365 628,478
nolat comm 473,975 299,379 743,123
nolat hhage 288,140 190,632 428,417
nolat hhflat 319,248 210,991 474,510
0e+00
1e−06
2e−06
0 250000 500000 750000 1000000
22 HBC total paediatr ic TB incidence (per year)
de
nsity
LAT
lat
nolat
method
comm
hhage
hhflat
Modelling the 22 HBC paediatric TB burden
20
In Figure 13 we present a visualisation of the mean proportions of the total due to ages 0-4 and 5-14, split by type of
disease. The mean estimated incidence in ages 0-4 is similar to that in ages 5-14 (evidenced by similar bar widths).
Extrapulmonary TB is more common in the younger age groups (approximately 25% of incident cases), and makes
up approximately 10% of the total incidence in both age groups. The estimated summaries of model incidence in
each age-group across the 22 HBCs are presented in Table 8.
Figure 13: A mosaic plot of the mean proportion of the estimated burden (for the lat x comm method) that is
pulmonary vs extrapulmonary (here denoted PTB and DTB, respectively), and in younger vs older age groups (0-4
and 5-14). This is a diagrammatic representation of a cross-tabulation: both the widths and heights are meaningful.
age
type
0−4 5−14
PT
BD
TB
Modelling the 22 HBC paediatric TB burden
21
Table 8: Model estimates of total TB incidence in children 0-4 years and 5-14 years, in the 22 HBCs for the
BCG model variant with no variation of efficacy by latitude, and for the 3 ARI model variants. (LQ=lower quartile;
UQ=upper quartile.) N.B. calculated from a different sample to results in main paper – differences indicate level of
Monte Carlo error.
LAT ARI age median LQ UQ
lat comm 0-4 325,832 206,128 495,726
lat hhage 0-4 199,869 132,741 285,868
lat hhflat 0-4 221,517 146,854 316,775
nolat comm 0-4 227,357 136,076 363,585
nolat hhage 0-4 139,462 86,731 211,336
nolat hhflat 0-4 154,282 95,943 234,306
lat comm 5-14 283,290 134,568 528,877
lat hhage 5-14 172,911 84,577 309,817
lat hhflat 5-14 191,196 93,730 342,941
nolat comm 5-14 205,659 94,449 399,575
nolat hhage 5-14 125,664 59,415 235,267
nolat hhflat 5-14 139,271 65,666 260,711
Modelling the 22 HBC paediatric TB burden
22
Overall, a small proportion of incident paediatric TB cases were infected with HIV (median [LQ-UQ] = 5.0%
[2.4%-10.1%]), which is lower than the corresponding statistic for adults (13%, (28)); however, this is likely to vary
substantially by country. Figure 14 shows the proportion of incident paediatric TB predicted to be in those with HIV
coinfection for 9 HBC countries with non-zero paediatric HIV prevalence. While in the majority of these countries
the proportion was predicted to be under 20%, for South Africa and Zimbabwe the proportion of coinfected children
was predicted to be over 30%.
Figure 14: The proportion of the incidence of paediatric TB in each country that is HIV associated. In the
other HBCs, HIV prevalence in those aged under 15 was assumed to be zero.
Democratic Republic of Congo
Ethiopia
Kenya
Mozambique
Nigeria
South Africa
Uganda
United Republic of Tanzania
Zimbabwe
0 20 40
% of paediatr ic TB incidence HIV−associated
cou
ntr
y
Modelling the 22 HBC paediatric TB burden
23
References
1. World Population Prospects, the 2012 Revision [Internet]. UN Dep. Econ. Soc. Aff. Popul. Div. [cited 2013
Sep 13]. Available from: http://esa.un.org/wpp/Excel-Data/population.htm
2. WHO | TB data [Internet]. WHO Stop TB Data. [cited 2013 Sep 13]. Available from:
http://www.who.int/tb/country/en/
3. MEASURE DHS [Internet]. Demogr. Heal. Surv. [cited 2013 Sep 13]. Available from:
http://www.measuredhs.com/
4. WHO World Health Organization: Immunization, Vaccines And Biologicals. Vaccine preventable diseases
Vaccines monitoring system 2013 Global Summary Reference Time Series: BCG [Internet]. [cited 2013 Sep
13]. Available from:
http://apps.who.int/immunization_monitoring/globalsummary/timeseries/tscoveragebcg.html
5. The BCG World Atlas: A Database of Global BCG Vaccination Policies and Practices [Internet]. [cited 2013
Sep 13]. Available from: http://www.bcgatlas.org/
6. AIDSinfo [Internet]. UNAIDS. [cited 2013 Sep 13]. Available from:
http://www.unaids.org/en/dataanalysis/datatools/aidsinfo/
7. Styblo, Karel. The relationship between the risk of tuberculous infection and the risk of developing infectious
tuberculosis. Bull Int Union Tuberc Lung Dis. 1985;60:117–9.
8. Dye C. Breaking a law: tuberculosis disobeys Styblo’s rule. Bull World Health Organ. 2008;86(1):4–4.
9. Van Leth F, van der Werf MJ, Borgdorff MW. Prevalence of tuberculous infection and incidence of
tuberculosis: a re-assessment of the Styblo rule. Bull World Health Organ. 2008 Jan;86(1):20–6.
10. Bourdin Trunz B, Fine P, Dye C. Effect of BCG vaccination on childhood tuberculous meningitis and miliary
tuberculosis worldwide: a meta-analysis and assessment of cost-effectiveness. Lancet. 2006 Apr
8;367(9517):1173–80.
11. Yamada, N, Onozaki, I, Eang, MT. National TB Survey 2002, Cambodia: preliminary report of tuberculin
survey. Tuberculosis Surveillance Research Unit Progress Report 2004. Hague KNCV Tuberc Found.
2005;40–8.
12. Ministry of Public Health of the People’s Republic of China. Nationwide random survey for the epidemiology
of tuberculosis in 1990. Beijing Minist Public Heal People’s Repub China. 1990;
13. China Tuberculosis Control Collaboration. The effect of tuberculosis control in China. Lancet. 2004 Aug
31;364(9432):417–22.
14. Tupasi TE, Radhakrishna S, Rivera AB, Pascual ML, Quelapio MI, Co VM, et al. The 1997 Nationwide
Tuberculosis Prevalence Survey in the Philippines. Int J Tuberc Lung Dis Off J Int Union Tuberc Lung Dis.
1999 Jun;3(6):471–7.
15. Tupasi TE, Radhakrishna S, Pascual ML, Quelapio MID, Villa ML, Sarol J, et al. BCG coverage and the
annual risk of tuberculosis infection over a 14-year period in the Philippines assessed from the nationwide
prevalence surveys. Int J Tuberc Lung Dis. 2000;4(3):216–22.
16. Hong YP, Kim SJ, Lew WJ, Lee EK, Han YC. The seventh nationwide tuberculosis prevalence survey in
Korea, 1995. Int J Tuberc Lung Dis. 1998;2(1):27–36.
Modelling the 22 HBC paediatric TB burden
24
17. Odhiambo JA, Borgdorff MW, Kiambih FM, Kibuga DK, Kwamanga DO, Ng’ang’a L, et al. Tuberculosis
and the HIV epidemic: increasing annual risk of tuberculous infection in Kenya, 1986-1996. Am J Public
Health. 1999;89(7):1078–82.
18. Tuberculosis control in the era of the HIV epidemic: risk of tuberculosis infection in Tanzania, 1983-1998. Int
J Tuberc Lung Dis Off J Int Union Tuberc Lung Dis. 2001 Feb;5(2):103–12.
19. Egwaga, SM, Cobelens, FGJ, Muwinga, H, Borgdorff, MW. Tanzanian National Tuberculin Survey, 4th
Round Results of Interim Analysis of 12 Regions. Hague KNCV Tubercuclosis Found. 2003;Tuberculosis
Surveillance Research Unit Progress Report 2003-Volume II:107–22.
20. Salaniponi FML, Kwanjana J, Veen J, Misljenovic O, Borgdorff MW. Risk of infection with Mycobacterium
tuberculosis in Malawi: national tuberculin survey 1994. Int J Tuberc Lung Dis. 2004;8(6):718–23.
21. Champetier de Ribes, G, Ranaivoson, G, Rakotoherisoa, E, Andriamahefazafy, B, Blanchy, S. Annual risk of
tuberculosis infection in Madagascar: study from 1991 to 1994. Bull Soc Pathol Exot. 1997;90:349–52.
22. El Ibiary S, de Coster EJ, Tolba FM, van Maaren P, Wasily L, van Cleeff M, et al. Trend in the annual risk of
tuberculous infection in Egypt, 1950-1996. Int J Tuberc Lung Dis Off J Int Union Tuberc Lung Dis. 1999
Apr;3(4):294–9.
23. Arnadottir TH, Soukaseum H, Vangvichit P, Bounmala S, Vos E. Prevalence and annual risk of tuberculosis
infection in Laos. Int J Tuberc Lung Dis Off J Int Union Tuberc Lung Dis. 2001 May;5(5):391–9.
24. Hong YP, Kim SJ, Kwon DW, Chang SC, Lew WJ, Han YC. The sixth nationwide tuberculosis prevalence
survey in Korea, 1990. Tuber Lung Dis. 1993;74(5):323–31.
25. Rutherford ME, Hill PC, Maharani W, Apriani L, Sampurno H, van Crevel R, et al. Risk factors for
Mycobacterium tuberculosis infection in Indonesian children living with a sputum smear-positive case. Int J
Tuberc Lung Dis. 2012;16(12):1594–9.
26. Lienhardt C, Sillah J, Fielding K, Donkor S, Manneh K, Warndorff D, et al. Risk factors for tuberculosis
infection in children in contact with infectious tuberculosis cases in the Gambia, West Africa. Pediatrics.
2003;111(5):e608–e614.
27. Fox GJ, Barry SE, Britton WJ, Marks GB. Contact investigation for tuberculosis: a systematic review and
meta-analysis. Eur Respir J. 2013 Jan;41(1):140–56.
28. WHO 2012. Global tuberculosis control 2012. Geneva: World Health Organization, 2012.
29. Tiemersma EW, van der Werf MJ, Borgdorff MW, Williams BG, Nagelkerke NJD. Natural history of
tuberculosis: duration and fatality of untreated pulmonary tuberculosis in HIV negative patients: a systematic
review. PloS One. 2011;6(4):e17601.
30. Behr MA, Warren SA, Salamon H, Hopewell PC, Ponce de Leon A, Daley CL, et al. Transmission of
Mycobacterium tuberculosis from patients smear-negative for acid-fast bacilli. Lancet. 1999 Feb
6;353(9151):444–9.
31. Tostmann A, Kik SV, Kalisvaart NA, Sebek MM, Verver S, Boeree MJ, et al. Tuberculosis transmission by
patients with smear-negative pulmonary tuberculosis in a large cohort in the Netherlands. Clin Infect Dis Off
Publ Infect Dis Soc Am. 2008 Nov 1;47(9):1135–42.
32. Marais BJ, Gie RP, Schaaf HS, Hesseling AC, Obihara CC, Starke JJ, et al. The natural history of childhood
intra-thoracic tuberculosis: a critical review of literature from the pre-chemotherapy era. Int J Tuberc Lung
Dis Off J Int Union Tuberc Lung Dis. 2004 Apr;8(4):392–402.
Modelling the 22 HBC paediatric TB burden
25
33. Rodrigues LC, Diwan VK, Wheeler JG. Protective effect of BCG against tuberculous meningitis and miliary
tuberculosis: a meta-analysis. Int J Epidemiol. 1993;22(6):1154–8.
34. Colditz GA, Berkey CS, Mosteller F, Brewer TF, Wilson ME, Burdick E, et al. The efficacy of bacillus
Calmette-Guerin vaccination of newborns and infants in the prevention of tuberculosis: meta-analyses of the
published literature. Pediatrics. 1995;96(1):29–35.
35. Palmer CE, Long MW. Effects of infection with atypical mycobacteria on BCG vaccination and tuberculosis.
Am Rev Respir Dis. 1966;94(4):553.
36. Fine PEM. Variation in protection by BCG: implications of and for heterologous immunity. The Lancet.
1995;346(8986):1339–45.
37. Seddon JA, Hesseling AC, Marais BJ, Jordaan A, Victor T, Schaaf HS. The evolving epidemic of drug-
resistant tuberculosis among children in Cape Town, South Africa. Int J Tuberc Lung Dis Off J Int Union
Tuberc Lung Dis. 2012 Jul;16(7):928–33.
38. Madhi SA, Huebner RE, Doedens L, Aduc T, Wesley D, Cooper PA. HIV-1 co-infection in children
hospitalised with tuberculosis in South Africa. Int J Tuberc Lung Dis. 2000;4(5):448–54.
39. Hesseling AC, Cotton MF, Jennings T, Whitelaw A, Johnson LF, Eley B, et al. High incidence of tuberculosis
among HIV-infected infants: evidence from a South African population-based study highlights the need for
improved tuberculosis control strategies. Clin Infect Dis. 2009;48(1):108–14.
40. R Core Team (2013). R: A language and environment for statistical computing. R Foundation for Statistical
Computing, Vienna, Austria. URL http://www.R-project.org/.