Skip to main content

Table 4 Reviews on missing data and ITT in the BMJ, JAMA, NEJM and the Lancet 1

From: Handling missing data in RCTs; a review of the top medical journals

Study Timing Study inclusion criteria Number of trials included Number (%) of papers with missing data2 Number (%) of papers with more than 10% missing data2 Missing data approaches in primary analysis Number (%)3 Number (%) of papers reporting sensitivity analysis3 Number (%) of papers reporting ITT4
Hollis et al., 1999 [2] 1997 All RCTs 249 89/119 (75) 29/119 (24) Complete case 44 (49) 1 (1) 119/249 (48)
Simple imputation 15 (17)
Multiple imputation 0
Model based 29 (33)
Unclear 1 (1)
Wood et al., 2004 [22] July-Dec, 2001 All RCTs with non-survival outcomes 71 63/71 (89) 36/71 (51) Complete case 41 (65) 13 (21) 26/63 (41)
Simple imputation 14 (22)
Multiple imputation 1 (2)
Model based 5 (8)
Unclear 2 (3)
Gravel et al., 2007 [1] 2002 Sample of RCTs1 403 152/249 (61) 52/249 (21) Complete case 89 (59) Not reported 249/403 (62)
Simple imputation 32 (21) 201/283 (71)5
Multiple imputation 1 (1)
Model based6 0
Unclear6 30 (20)
Fielding et al., 2008 [21] 2005-2006 Random sample of RCTs with Quality of life outcomes 61 55/61 (90) 22/61 (36) Complete case 30 (55) 6 (11) Not reported
Simple imputation 11 (20)
Multiple imputation 1 (2)
Model based 9 (16)
Unclear 4 (7)
Bell et al., (current study) July-Dec, 2013 All RCTS with non-survival outcomes 77 73/77 (95) 36/77 (47) Complete case 33 (45) 27 (37) 62/73 (85)
Simple imputation 20 (27)
Multiple imputation 6 (8)
Model based 14 (19)
  1. 1Gravel et al. reported on 10 journals, including the BMJ, JAMA, NEJM and the Lancet.
  2. 2Denominator is the number of trials included except for Hollis et al. and Gravel et al., where denominators are the number of papers reporting ITT.
  3. 3Denominator is the number of papers with missing data.
  4. 4Denominator is the number of papers with missing data except for Hollis et al. and Gravel et al. where denominators are the number of trials included in the review.
  5. 5Sub-analysis of RCTs from the four journals (BMJ, JAMA, NEJM and the Lancet) out of the 10 journals included in Gravel’s review.
  6. 6Three reported as “other” might be model based (added to 27 marked “unclear”).