1
Appendix
2
3
Estimation of country-specific prevalence
4
A binomial model was used to estimate prevalence rates taking into account the number
5
of individuals screened, the number of positive individuals and the sensitivity coefficient
6
of the diagnostic test used. The model was formulated and implemented under a
7
Bayesian framework.
8
Let Yijk denote the number of positive samples out of N ijk individuals screened using a
9
diagnostic tool of sensitivity category k ( k  1, 2,3 ) in country i, i  1,..., C and study
10
j  1,..., ni , where ni is the total number of surveys in country i and C  236 is the total
11
number of countries analysed. The model assumes that Yijk is distributed binomially,
12
that is: Yijk ~ Binomial ( N ijk , pijk ) , where pijk is the observed prevalence of S. stercoralis.
13
If Di  denotes an individual who is truly infected by S. stercoralis in country i , the
14
observed prevalence pijk is related to the true prevalence through the equation
15
pij k  P(Tijk , Di )  P(Tijk  | Di ) P( Di )
(1)
16
where Tijk  denotes that an individual in country i , survey j has tested positive using a
17
diagnostic tool with sensitivity category k . Let  i indicate the true prevalence, i.e.
18
 i  P( Di ) , equation (1) can be written as pijk  Sijk  i
19
where Sijk  P(Tijk  | Di ) is the sensitivity value of the diagnostic tool used in study j .
20
Assuming that only one diagnostic tool was used in each study j and that its sensitivity
21
depends only on the category (low, moderate or high), equation (2) becomes pij  S k  i .
(2)
1
22
Since actual prior data for prevalence were unavailable, an uninformative prior was
23
elicited on  i ,i.e.  i ~ Beta(a, b) .Several choices of the hyperparameters a and b were
24
implemented, with the aim of conducting a probabilistic sensitivity analysis, leading to
25
very similar estimates. The hyperparameters a and b where chosen in order to define a
26
unimodal distribution with large variance.
27
Based on information found in the literature, the prior distributions on the sensitivity of
28
the diagnostic tools of category low, moderate and high were specified as follows:
29
S1 ~ Beta(1,1) I (0.129,0.689) , S2 ~ Beta(1,1) I (0.471,0.968) , S3 ~ Beta(1,1) I (0.68,0.982) .
30
Estimation of prevalence in specific risk groups
31
For the second analysis, the same prior information on the sensitivity of diagnostic tools
32
was used. The model was formulated in the odds ratio scale, considering each risk
33
group separately. Let Yjkr ~ Binomial ( N rjk , p rjk ) model the distribution of infected
34
individuals in risk group r and Yjkc denote the one in the control group, i.e.
35
Yjkc ~ Binomial ( N cjk , pcjk ) ,
36
37
38
39
a logistic regression equation can be written on the risks
logit( pcjk )   cjk ,
logit( prjk )   cjk   jkr where a prior distribution is given on the  jkr ~ N (c, d ) , where the
latter represents the risk effect in the scale relative to the control group. For each risk
group a separate regression model was fitted.
40
The results of the models were expressed in terms of posterior medians and Bayesian
41
Credible Intervals.
42
2
43
44
45
46
47
48
49
50
51
52
53
54
55
56
Table A1: Posterior diagnostic test sensitivity estimates: community-based
studies
Sensitivity
Prior
1. Low
2. Moderate
3. High
U(0.13-0.69)
U(0.47-0.97)
U(0.68-0.98)
Posterior
(median, 95%
BCI)
0.17, (0.15-0.18)
0.84, (0.77-0.90)
0.95, (0.88-0.98)
Prior
Posterior (median,
95% BCI)
Beta(3.7,5.69)
Beta(7.13,2.54)
Beta(10.76,2.09)
0.17, (0.15-0.18)
0.84, (0.78-0.90)
0.95, (0.90-0.98)
Table A2: Posterior diagnostic test sensitivity estimates: studies among
immigrants.
Sensitivity
Prior
1. Low
2. Moderate
3. High
U(0.13-0.69)
U(0.47-0.97)
U(0.68-0.98)
Posterior
(median, 95%
BCI)
0.13, (0.13-0.14)
0.70, (0.49-0.95)
0.97, (0.94-0.98)
Prior
Beta(3.7,5.69)
Beta(7.13,2.54)
Beta(10.76,2.09)
Posterior
(median, 95%
BCI)
0.15, (0.15-0.16)
0.73, (0.52-0.95)
0.97, (0.94-0.98)
Table A3: Posterior diagnostic test sensitivity estimates: hospital-based studies.
Sensitivity
Prior
1. Low
2. Moderate
3. High
U(0.13-0.69)
U(0.47-0.97)
U(0.68-0.98)
Posterior
(median, 95%
BCI)
0.19, (0.17-0.21)
0.47, (0.47-0.48)
0.98, (0.97-0.98)
Prior
Posterior
(median, 95%
BCI)
Beta(3.7,5.69)
0.20, (0.20-0.22)
Beta(7.13,2.54)
0.50, (0.48-0.51)
Beta(10.76,2.09) 0.98, (0.97-0.98)
57
58
3
59
Figure Captures
60
Figure A1a: Risk of S. stercoralis infection in HIV/AIDS patients (meta-analysis of case-
61
62
control studies, excluding sensitivity of diagnostic test)
Figure A1b: Risk of S. stercoralis infection in patients with HTLV-1 infection (meta-analysis
63
of case-control studies, excluding sensitivity of diagnostic test)
64
Figure A1c: Risk of S. stercoralis infection in alcoholics (meta-analysis of case-control
65
66
67
studies, excluding sensitivity of diagnostic test)
Figure A1d: Risk of S. stercoralis infection in patients with diarrhoea (meta-analysis of casecontrol studies, excluding sensitivity of diagnostic test)
68
4
69
70
Figure A1a
71
5
72
73
Figure A1b
74
6
75
76
Figure A1c
77
7
78
79
Figure A1d
80
81
8