Long term oncological outcomes for laparoscopic versus open surgery for rectal cancer – A population‐based nationwide noninferiority study

Abstract

This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.

Impute‐then‐exclude versus exclude‐then‐impute: Lessons when imputing a variable used both in cohort creation and as an independent variable in the analysis model

“…In the cancer literature, some studies have used what we refer to as an "exclude-then-impute" strategy. [17][18][19] Using this strategy, subjects who have documented evidence of a stage IV cancer are excluded from the sample. The resultant sample consists of those subjects with documented evidence of stage I, II, or III cancer or who have missing data on cancer stage.…”

“…Despite the lack of methodological studies examining the relative merits of these two strategies, the "exclude-then-impute" strategy has been used in the cancer literature. [17][18][19] From a theoretical perspective, the "impute-then-exclude" strategy would appear to be the better option, as it allows imputed values of stage to take on any of the initially observed values (stage I to stage IV) (assuming those with stage 0 were initially excluded). However, the "impute-then-exclude" approach could lead to bias in the pooled variance estimation due to incompatibility, in that the sample used for imputation differs from the sample used for the substantive analysis (ie, the analysis sample differs from the imputation sample).…”

“…18 Similarly, in a recent study comparing laparoscopic versus open surgery for rectal cancer, the authors excluded subjects with stage IV cancer and then imputed stage for those with missing stage in the resultant sample. 19 Given the lack of methodological studies evaluating the performance of the impute-then-exclude and the exclude-then-impute strategies, it is not known what the consequences were of using the exclude-then-impute approach in these studies.…”

“…The best strategy has not been determined for imputing missing data when a variable that is subject to missingness is used both for inclusion/exclusion criteria and as an independent exposure variable in the subsequent analysis model. In the cancer literature, some studies have used what we refer to as an “exclude‐then‐impute” strategy 17‐19 . Using this strategy, subjects who have documented evidence of a stage IV cancer are excluded from the sample.…”

“…In each of the resultant restricted samples, the analysis model is fit in which the outcome is regressed on cancer stage and a set of covariates. Despite the lack of methodological studies examining the relative merits of these two strategies, the “exclude‐then‐impute” strategy has been used in the cancer literature 17‐19 . From a theoretical perspective, the “impute‐then‐exclude” strategy would appear to be the better option, as it allows imputed values of stage to take on any of the initially observed values (stage I to stage IV) (assuming those with stage 0 were initially excluded).…”