Figure 1 shows the PRISMA flow chart of the study selection process. A total of 521 abstracts were identified from the searches. After removing duplicates, the titles and abstracts of 182 articles were screened for eligibility. Of these, 62 full text articles were accessed and screened out of which 43 studies met the inclusion criteria and were included in the meta-analysis [21–63].
Characteristics of included studies
Out of the 43 studies, two were carried out before 1995, five between 1995 and 2005, and thirty-five between 2006 and 2020. Only eleven of the forty-three studies had data derived from national representative samples. The internal and external validity of the included studies were determined using a 9 point score (Table 1). Most of the studies (37/43) had low risk of bias with the rest having intermediate risk (Table 2).
Table 2
Means and standard deviations (SD) derived from the medians of the included studies
First author, year
|
Country, period of data collection
|
Sample size
|
median
|
Derived mean
|
Derived SD
|
Chaouki, 1994
|
Algeria (1994)
|
982
|
17.9
|
17.9
|
0.1
|
Ngo, 1997
|
DRC (1991 -92)
|
306
|
39.3
|
43.7
|
21.5
|
Hess, 1999
|
Cote de'Ivore
|
72
|
351
|
351
|
455
|
Hess, 1999
|
Cote de'Ivore
|
66
|
136
|
184.2
|
192.3
|
Ojule 1998
|
Nigeria, 1998
|
90
|
213
|
213
|
9.9
|
Ojule 1998
|
Nigeria, 1998
|
105
|
149
|
149
|
14.7
|
Dillon, 2000
|
Senegal 1996–1997
|
462
|
60
|
60
|
39
|
Eltom, 2000
|
Sudan 1998–1999
|
47
|
38
|
33.6
|
29.1
|
Lwenje, 2000
|
Swaziland
|
165
|
295
|
299
|
30
|
Akdader-O, 2020
|
Algeria 2016–2017
|
173
|
233
|
234.8
|
37
|
Garnier, 2016
|
Burkina Faso
|
946
|
74
|
74
|
28.2
|
Kavishe, 2020
|
Burundi 2018
|
87
|
86.7
|
87
|
33.2
|
Habimana, 2013
|
DRC, 2009–2011
|
225
|
138
|
138.4
|
50
|
IGN, 2017
|
Djibouti, 2015 (NS)
|
230
|
265
|
294
|
205
|
MOHP, 2017
|
Egypt, 2014–2015
|
1498
|
135
|
135
|
50.5
|
Hamza, 2007
|
Egypt, 2006
|
113
|
102
|
102.9
|
31.1
|
Elsayed, 2016
|
Egypt,2016
|
400
|
170
|
173.5
|
56.5
|
Mohammed, 2019
|
Ethiopia, 2013–2014
|
562
|
120.6
|
136.1
|
109.6
|
Fereja, 2018
|
Ethiopia
|
354
|
85.7
|
89.3
|
67.2
|
Kedir, 2014
|
Ethiopia, 2012
|
435
|
58.1
|
63.8
|
66.7
|
Ersino, 2013
|
Ethiopia, 2009
|
172
|
15
|
16.9
|
22.8
|
Takele, 2018
|
Ethiopia, 2017
|
403
|
137
|
137
|
59.5
|
Keno, 2017
|
Ethiopia, 2014
|
40
|
88.6
|
96.8
|
51.2
|
Negeri, 2014
|
Ethiopia, 2011
|
423
|
48
|
48
|
17.9
|
NaNA, 2019
|
Gambia 2018
|
118
|
113.5
|
123.7
|
116.7
|
GHS, 2017
|
Ghana, 2015
|
102
|
183.5
|
183.5
|
69.5
|
Gyamfi, 2018
|
Ghana, 2016 (ss)
|
239
|
159
|
159
|
59.7
|
Adu-Afarwuah, 2018
|
Ghana, 2009–2011
|
295
|
137
|
145.8
|
106.5
|
Farebrother, 2018
|
Kenya
|
162
|
337
|
347.2
|
229.6
|
Randremanana, 2019
|
Madagascar, 2014
|
170
|
53
|
50.2
|
59.8
|
Stinca, 2017
|
Morocco 2013 − 14
|
245
|
32
|
35.9
|
30.6
|
Sadou 2013
|
Niger 2012
|
240
|
119
|
119
|
44.8
|
Hess, 2016
|
Niger, 2014–2015
|
662
|
69
|
73.9
|
56.4
|
Jibril, 2016
|
Nigeria, 2014
|
300
|
193
|
193
|
71.5
|
Kayode, 2019
|
Nigeria, 2012
|
133
|
135
|
138.5
|
58.5
|
Ujowundu, 2010
|
Nigeria, 2009
|
302
|
151.1
|
152.1
|
41.7
|
Rohner, 2016
|
Sierra Leone, 2013
|
154
|
175.8
|
176
|
65.9
|
MOH-FGS, 2020
|
Somalia, 2018–19
|
236
|
269
|
424
|
454
|
Mabasa, 2019
|
South Africa, 2012 -13
|
565
|
164
|
183.3
|
147.9
|
Stinca, 2017
|
South Africa
|
207
|
174
|
189.8
|
151
|
Mtumwa, 2017
|
Tanzania, 2009–2010
|
947
|
136.8
|
151.9
|
147.9
|
Ba, 2020
|
Tanzania, 2015–2016
|
266
|
156.1
|
160.6
|
146
|
Stinca, 2017
|
Tanzania, 2016
|
330
|
422
|
434.3
|
252.4
|
Chinyanga, 2006
|
Zimbabwe, 2006
|
94
|
115.5
|
128.5
|
137
|
The prevalence of insufficient iodine intake (UIC < 150 µg/L) among pregnant women on the various African countries before 1995, 1995–2005, and 2006–2020
Before 1995, available data from two studies revealed moderate countrywide iodine deficiency in pregnancy in the Democratic Republic of Congo at the time and severe iodine deficiency in pregnancy in a subnational sample from North-Eastern Algeria [21, 22]. The pooled median UIC across the two studies was 28.6 µg/L (95% CI 7.6–49.5), with considerable heterogeneity (I2 99.73 %, p < 0.001, Fig. 2).
Between 1995 and 2005 four subnational studies from Ivory coast, Nigeria Sudan and Senegal, and one national survey from Swaziland [23–27] yielded a pooled pregnancy UIC of 174.1 µg/L (95% CI 90.4–257.7, Fig. 2); with considerable heterogeneity (I2 99.96 %, p < 0.001).
Between 2005 and 2020, 35 studies from 18 countries had pregnancy median UIC data. Eleven of the studies were national surveys from 10 countries. These national surveys revealed more than adequate intake in Djibouti and Somalia [28, 29]; adequate iodine intake in Ghana, Sierra Leone, and Tanzania [30–32]; mild inadequate intake in Egypt, Gambia and Tanzania [33–35], and moderate insufficient iodine intake in Burkina Faso, Burundi and Madagascar [36–38]. The remaining 24 [39–63] were subnational studies. The pooled median pMUIC across the 35 studies conducted between 2005 and 2020 was 145 µg/L (95% CI 126–172), with substantial heterogeneity (I2 99.81 %, p < 0.001) (Fig. 2). There was a significant increase in pregnancy median UIC between 1995 and 2020 compared to the period before 1995 (Kendaull’s tau correlation co-efficient 0.270, p = 0.032).
Derived mean UIC by time-period
The pooled derived mean pregnancy UIC (Table 2, Fig. 3) was 27.96 µg/L (95% CI 11.6–67.04, tau 0.630) before 1995; 143.22 µg/L (95% CI 108.65–188.78, tau 0.362) between 1995 and 2005; and 127.99 µg/L (95% CI 108.59–150.85, tau 0.493); with significant difference across time-period (Q = 12.24, d.f. = 2, p = 0.002).
Assessment of publication bias
Publication bias was assessed using funnel plots. The funnel plot for the studies in the period 1995–2004 was not suggestive of potential publication bias (Fig. 4) (R metabias linear regression test t = − 0.36005, p-value = 0.7335). No additional studies were imputed after checking for funnel asymmetry using the Twedie and Duval’s trim and fill test. The funnel plot for the studies carried out between 2005 and 2020 was asymmetrical (Fig. 5). The trim and fill test imputed sixteen potential missing studies suggesting potential publication bias (Fig. 6). The funnel plot asymmetry was confirmed by the R metabias linear regression test (t = 3.872, p < 0.001).