Skip to main content

Table 2 Secondary analyses for meta-analyses reporting both continuous and dichotomous outcomes (N = 18)

From: Dichotomisation of a continuous outcome and effect on meta-analyses: illustration of the distributional approach using the outcome birthweight

Meta-analysis Published data Distributional estimates for low birthweight  
  Number of studies (sample size) Mean difference (g) (95% CI) P-value Number of studies (sample size) RR (95% CI) P-value Number of studies (sample size) Distributional RR (95% CI) P-value Comments
Bupassiri 2011 [34] 21 (8,319) 65 (16, 114) 0.01 5 (13,638) 0.83 (0.63, 1.09) 0.18 21 (8,319) 0.72 (0.58, 0.89) < 0.01  
Crowther 2010 [35] 4 (417) 75 (−17, 167) 0.11 7 (1,452) 0.84 (0.68, 1.04) 0.12 4 (417) 0.99 (0.88, 1.06) 0.42  
Dodd 2010 [36] 2 (282) −75 (−210, 61) 0.28 1 (49) 0.41 (0.04, 4.20) 0.45 2 (282) 1.33 (0.78, 2.26) 0.29 2/3 primary studies of mean birthweight outcome accessed
Gouin 2011 [37] 18 (6,855) −441 (−532, −350) < 0.01 19 (38,796) 2.86 (2.36, 3.48) < 0.01 18 (6,855) 2.76 (2.12, 3.45) 0.01  
Gülmezoglu 2011 [38] 1 (208) −100 (−240, 40) 0.16 1 (604) 1.38 (0.92, 2.06) 0.12 1 (208) 1.40 (0.87, 2.24) 0.16  
Kawai 2011 [39] 13 (35,015) 45 (28, 62) < 0.01 13 (35,015) 0.92 (0.83, 1.02) 0.09 13 (35,015) 0.82 (0.75, 0.91) < 0.01 13/15 primary studies accessed
Kenyon 2010 [40] 13 (6,480) 49 (14, 85) 0.01 2 (4,876) 1.00 (0.96, 1.04) 0.94 13 (6,480) 0.99 (0.99, 1.00) 0.53  
Ladhani 2011 [41] 4 (880) −279 (−485, −74) 0.01 2 (26,026) 3.28 (2.25, 4.78) < 0.01 4 (880) 2.41 (1.42, 4.09) < 0.01  
Lamont 2011 [42] 1 (485) −12 (−128, 104) 0.89 2 (876) 0.96 (0.62, 1.47) 0.83 1 (485) 1.03 (0.77, 1.38) 0.84  
Mathanga 2011 [43] 2 (640) 121 (27, 214) 0.01 2 (624) 0.80 (0.54, 1.18) 0.25 2 (640) 0.75 (0.60, 0.94) 0.01  
McDonald 2010 [44] 9 (5,225) −120 (−248, 6.8) 0.06 9 (5,225) 0.92 (0.72, 1.16) 0.46 9 (5,225) 1.12 (0.99, 1.26) 0.07 9/10 primary studies accessed
Murphy 2011 [45] 8 (179,589) −121 (−199, −43) < 0.01 12 (1,110,176) 1.45 (1.21, 1.73) < 0.01 8 (179,589) 1.46 (1.10, 1.94) 0.01  
Reveiz 2011 [46] 3 (237) 15 (−111, 142) 0.81 1 (100) Not estimated NA 3 (237) 0.95 (0.59, 1.52) 0.83 Zero cases of LBW in both treatment arms of primary study
Salmasi 2010 [47] 44 (71,663) 13 (−105, 131) 0.83 18 (40,790) 1.09 (0.88, 1.35) 0.44 44 (71,663) 0.98 (0.77, 1.23) 0.85 18/19 primary studies of LBW outcome accessed
Whitworth 2010 [48] 5 (23,213) 11 (−20, 41) 0.49 8 (19,337) 1.04 (0.82, 1.33) 0.73 5 (23,213) 0.97 (0.87, 1.08) 0.56  
Wiysonge 2011 [49] 3 (1,809) 68 (19, 118) 0.01 4 (2,606) 0.83 (0.68, 1.01) 0.07 3 (1,809) 0.84 (0.74, 0.95) 0.01  
  Number of studies (sample size) Mean difference (g) (95% CI) P -value Number of studies (sample size) OR (95% CI) P- value Number of studies (sample size) Distributional OR (95% CI) P -value  
Pope 2010 [50] 5 (13,955) 100 (73, 128) < 0.01 8 (Unclear) 1.38 (1.25, 1.52) < 0.01 5 (13,955) 0.81 (0.76, 0.86) < 0.01 Pooled published pre-calculated estimates for LBW outcome (that is Log(OR) and SE)
Salvig 2010a[51]           
(Fixed effects model) 4 (1,187) 66 (1.6, 131) 0.04 3 (785) 0.98 (0.66, 1.46) 0.93 4 (1,187) 0.81 (0.65, 1.01) 0.07  
Salvig 2010a[51]           
(Random effects model) 4 (1,187) 68 (−75, 212) 0.35 3 (785) 0.95 (0.49, 1.85) 0.88 4 (1,187) 0.79 (0.48, 1.29) 0.34  
  1. CI: confidence interval; LBW: low birthweight; NA: not applicable; OR: odds ratio; RR: risk ratio; SE: standard error; aFixed effects model used in published meta-analysis but there was significant heterogeneity between studies (P = 0.003) so secondary analysis was repeated here using the random effects model.