Skip to main content

Table 6 Results of meta-regression analyses to identify associations with HIV prevalence, sources of between-study heterogeneity, and trend in HIV prevalence in FSWs in the Middle East and North Africa (MENA)

From: HIV epidemiology among female sex workers and their clients in the Middle East and North Africa: systematic review, meta-analyses, and meta-regressions

Variables

 

Studies

Samples

Univariable analyses

Multivariable analysis

Total N

Total N

OR (95% CI)

LR test p value

Variance explained R (%)

AOR (95% CI)

p value

LR test p value¥

Country/subregion*

         

 Eastern MENA

Afghanistan, Iran, Pakistan

122

47,533

1.00

< 0.001

39.80

1.00

 

< 0.001

 Fertile Crescent

Egypt, Iraq, Jordan, Lebanon, Syria

136

132,758

0.17 (0.10–0.27)

  

0.21 (0.12–0.36)

< 0.001

 

 Bahrain and Yemen

Bahrain and Yemen

11

2491

2.60 (0.78–8.67)

  

1.77 (0.52–6.01)

0.357

 

 Horn of Africa

Djibouti, Somalia, South Sudan

93

29,509

33.45 (19.77–56.58)

  

45.43 (24.66–83.68)

< 0.001

 

 North Africa

Algeria, Libya, Morocco, Sudan, Tunisia

312

75,428

3.14 (2.09–4.72)

  

2.90 (1.80–4.68)

< 0.001

 

Population type

Street-based, venue-based, and other FSWs

619

220,363

1.00

0.002

1.29

1.00

 

0.163

Bar girls

55

67,356

0.33 (0.17–0.67)

  

0.66 (0.37–1.18)

0.163

 

Total sample size of tested FSWs

< 100 participants

75

4008

1.00

0.001

1.54

1.00

 

< 0.001

≥ 100 participants

599

283,711

0.36 (0.20–0.65)

  

0.35 (0.21–0.56)

< 0.001

 

Median year of data collection**

< 1993

104

36,038

1.00

0.001

1.96

1.00

 

0.005

1993–2002

169

98,221

0.31 (0.17–0.56)

  

1.18 (0.71–1.95)

0.522

 

≥ 2003

401

153,460

0.57 (0.33–0.97)

  

2.03 (1.24–3.33)

0.005

 

Sampling methodology

Non-probability sampling

570

254,072

1.00

0.217

0.08

Probability-based sampling

104

33,647

0.72 (0.42–1.21)

  

Response rate

≥ 60%

96

31,161

1.00

0.043

0.64

1.00

 

0.544

< 60%/unclear

62

14,102

2.76 (1.24–6.13)

  

1.17 (0.60–2.27)

0.645

 

Not applicable

516

242,456

1.37 (0.80–2.37)

  

1.33 (0.79–2.23)

0.279

 

Validity of sex work definition

Clear and valid definition

117

36,431

1.00

0.161

0.25

Poorly defined/unclear

41

8832

2.35 (0.96–5.73)

  

Not applicable

516

242,456

1.15 (0.70–1.90)

  

HIV ascertainment

Biological assays

157

44,894

1.00

0.786

0

Self-report, unclear, and not applicable

517

242,825

0.94 (0.60–1.47)

  

  1. Abbreviations: AOR adjusted odds ratio, CI confidence interval, FSWs female sex workers, LR likelihood ratio, OR odds ratio
  2. *Countries were grouped based on geography and similarity in HIV prevalence levels. Given the large fraction of studies with zero HIV prevalence, particularly in the Fertile Crescent, an increment of 0.1 was added to a number of events in all studies when generating log odds, and Eastern MENA was thus used also as a statistically better reference. While this choice of increment was arbitrary, other increments yielded the same findings, though some of the effect sizes changed in scale
  3. **Year grouping was driven by independent evidence identifying the emergence of HIV epidemics among both men who have sex with men [10] and people who inject drugs [11] in multiple MENA countries around 2003. Missing values for year of data collection (only six stratified measures) were imputed using data for year of publication adjusted by the median difference between year of publication and median year of data collection (for studies with complete information)
  4. A large fraction of studies did not separate the different forms of female sex workers, and thus it was not possible to analyze these as separate categories
  5. Measures extracted only from routine databases with no reports describing the study methodology were not included in the ROB assessment
  6. Predictors with p value ≤ 0.1 were considered as showing strong evidence for an association with (prevalence) odds and were hence included in the multivariable analysis
  7. £Adjusted R2 in the final multivariable model = 49.21%
  8. ¥Predictors with p value ≤ 0.1 in the multivariable model were considered as showing strong evidence for an association with (prevalence) odds