Skip to main content

Table 2 Key characteristics of included RCTs and methods for handling missing data

From: A review of the use of controlled multiple imputation in randomised controlled trials with missing outcome data

  The Lancet
(n = 51)
NEJM
(n = 67)
Total
(n = 118)
Type of Intervention
 Behaviour (diet, exercise, cognitive) 5 10% 4 6% 9 8%
 Medical device 6 12% 1 1% 7 6%
 Diagnostic 2 4% 0 0% 2 2%
 Drug 19 37% 47 70% 66 56%
 Health service strategies 5 10% 2 3% 7 6%
 Psychological 5 10% 0 0% 5 4%
 Surgical 9 18% 13 19% 22 19%
Number of Participants
  < 100 2 4% 4 6% 6 5%
 100 to 499 24 47% 20 30% 44 37%
 500 to 999 16 31% 21 31% 37 31%
 1000 to 4999 8 16% 14 21% 22 19%
  ≥ 5000 1 2% 8 12% 9 8%
Proportion of Missing Primary Outcome Data
  < 10% 21 41% 33 49% 54 46%
 10 to 19% 17 33% 23 34% 40 34%
 20 to 29% 6 12% 4 6% 10 8%
  ≥ 30% 3 6% 4 6% 7 6%
 not clear 4 8% 3 4% 7 6%
Type of Primary Outcome
 Binary 20 39% 34 51% 54 46%
 Continuous 27 53% 30 45% 57 48%
 Count 4 8% 0 0% 4 3%
 Time-to-event 0 0% 3 4% 3 3%
Assessed Differences Between Participants with Complete and Incomplete Data
 Yes 5 10% 6 9% 11 9%
 No 46 90% 61 91% 107 91%
Method of Handling Missing Data in Primary Analysis
 MAR MIa 22 44% 21 31% 43 36%
 Controlled MI 1 2% 1 1% 2 2%
 Complete case 19 38% 39 57% 58 49%
 Single imputation 5 10% 5 7% 10 8%
 Last observation carried forward 3 6% 2 3% 5 4%
Conducted Missing Data Sensitivity Analysis
 Yes 36 71% 59 88% 95 81%
 No 15 29% 8 12% 23 19%
Method of Handling Missing Data in Sensitivity Analysis Where Sensitivity Was Conductedb
 MAR MI 28 68% 42 54% 70 59%
 Controlled MI 3 7% 11 14% 14 12%
 Complete case 6 15% 10 13% 16 13%
 Single imputation 1 2% 8 10% 9 8%
 Last observation carried forward 1 2% 6 8% 7 6%
 Others 2 5% 1 1% 3 3%
Method of Controlled MI (primary or sensitivity analysis)
 Reference-based 2 40% 5 45% 7 44%
 Delta-based 3 60% 6 55% 9 56%
  1. Data presented as n and %. aOne trial using MI under MAR used a hybrid of MI under MAR and worst observation carried forward (single imputation)
  2. b23 trials using two or more different statistical method for sensitivity analysis
  3. Percentages are rounded to 0 decimal places so may not sum exactly to 100%