Article type
Year
Abstract
Background:
We recently proposed a number of assumptions for handling missing participant data (MPD), and assessing the associated risk of bias.
Objectives:
To assess the impact of applying the proposed assumptions about the outcomes of participants with missing data on the statistically significant effect estimates in five Cochrane Reviews.
Methods:
We conducted this study during the process of updating a series of five Cochrane Reviews on anti-coagulation in patients with cancer. We considered patients described as “withdrew consent”, “lost to follow-up” or “outcome not assessable” as having MPD. We focused on outcomes for which the primary analysis, a complete case analysis, found statistically significant results. We applied nine assumptions about the outcomes of participants with missing data on the effect estimates. Four of those assumptions are commonly used (e.g. best and worst case scenario). The remaining five and increasingly stringent assumptions are considered more plausible as they are based on incidences observed among followed-up participants. We assessed how many outcomes lost statistical significance or changed direction with each of the assumptions.
Results:
We included 12 outcomes that had statistically significant results in the complete case analysis. The table lists for each of these outcomes the effect of each assumption on its statistical significance. The impact of the common assumptions varied significantly with no change for two of them (best case scenario and ‘none had the event’) and major change for the other two (‘all had the event’ and, particularly, worst case scenario). Under the plausible assumptions (based on the review and lost to follow up/follow up: RILTFU/FU), an increasing number of outcomes, up to five, lost statistical significance, with one changing direction
Conclusions:
The proposed assumptions may help in assessing risk of bias associated with MPD. The increasingly stringent plausible assumptions may be more helpful with assessing risk of bias associated with MPD.
We recently proposed a number of assumptions for handling missing participant data (MPD), and assessing the associated risk of bias.
Objectives:
To assess the impact of applying the proposed assumptions about the outcomes of participants with missing data on the statistically significant effect estimates in five Cochrane Reviews.
Methods:
We conducted this study during the process of updating a series of five Cochrane Reviews on anti-coagulation in patients with cancer. We considered patients described as “withdrew consent”, “lost to follow-up” or “outcome not assessable” as having MPD. We focused on outcomes for which the primary analysis, a complete case analysis, found statistically significant results. We applied nine assumptions about the outcomes of participants with missing data on the effect estimates. Four of those assumptions are commonly used (e.g. best and worst case scenario). The remaining five and increasingly stringent assumptions are considered more plausible as they are based on incidences observed among followed-up participants. We assessed how many outcomes lost statistical significance or changed direction with each of the assumptions.
Results:
We included 12 outcomes that had statistically significant results in the complete case analysis. The table lists for each of these outcomes the effect of each assumption on its statistical significance. The impact of the common assumptions varied significantly with no change for two of them (best case scenario and ‘none had the event’) and major change for the other two (‘all had the event’ and, particularly, worst case scenario). Under the plausible assumptions (based on the review and lost to follow up/follow up: RILTFU/FU), an increasing number of outcomes, up to five, lost statistical significance, with one changing direction
Conclusions:
The proposed assumptions may help in assessing risk of bias associated with MPD. The increasingly stringent plausible assumptions may be more helpful with assessing risk of bias associated with MPD.