FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W4205197184> ?p ?o ?g. }

• W4205197184 endingPage "1494" @default.
• W4205197184 startingPage "1492" @default.
• W4205197184 abstract "We thank Drs. L. LaVange, M. Zhang, B. Zhang, and M. Proschan for their insightful commentaries on our manuscript. We respond to each in turn. Dr. LaVange Dr. LaVange highlighted the importance of covariate adjustment in large randomized trials. We could not agree more! Related to this, it is important to avoid the common mistake of selecting baseline variables to adjust for based on which ones have statistically significant imbalances across study arms (Pocock et al., 2002). The variables should either be selected before the trial starts (selecting those that are most prognostic for the outcome based on prior data), or selected using the trial data based on a completely prespecified algorithm that aims to select the most prognostic variables. Furthermore, Dr. LaVange emphasizes the importance of effect estimates that are valid “even in the presence of model misspecification.” We entirely agree with this point which was the impetus for the model-robust, covariate adjusted estimators in our paper. Related to this point, we agree with Uno et al. (2015); Pak et al. (2017), who advocate for using estimands (i.e., targets of inference) such as the restricted mean survival time that have a model-free interpretation. This is in contrast to commonly used estimands such as the hazard ratio (for time-to-event outcomes under a proportional hazards model) or odds ratio (for ordinal outcomes under a proportional odds model), which are not model-free estimands, and therefore may be difficult or impossible to interpret under model misspecification (i.e., if the proportional hazards model or proportional odds model is misspecified). Drs. M. Zhang and B. Zhang Drs. M. Zhang and B. Zhang describe a general approach for constructing covariate adjusted estimators that applies to many estimands. In particular, one of their estimators has improved precision compared to the methods that we proposed, when applied to our simulation distributions. This is impressive and makes a good argument for using the corresponding estimator in practice. One difference between their general approach and the approach in our paper is that the latter produces substitution estimators. Substitution estimators have the potential advantage of always being in the parameter space, for example, being between 0 and 1 when estimating a probability. Though in many cases this may not matter, it may be important when the true parameter values are close to the boundary. It is an area of future research to compare the general approach of Drs. M. Zhang and B. Zhang versus our general approach (using substitution estimators) across a variety of simulation studies that mimic features of completed trial data sets. Drs. M. Zhang and B. Zhang state that “there is a dilemma in that covariate adjustment is more useful in improving efficiency of inferences when sample size is large, in which case efficiency is of less a concern.” We respectfully disagree, and think that efficiency can be a major concern in large trials (as well as small trials). For example, a relative efficiency of 0.83 at sample size 1000 (as in the last row of Table 6) is approximately equivalent to a 17% reduction in the required sample size to achieve a desired power; this is approximately equivalent to a sample size reduction of 170 participants, which we consider to be important. More generally, at a fixed relative efficiency that is less than 1, the sample size reduction due to covariate adjustment is approximately proportional to the trial's sample size; this means that the impact can be substantial at large sample sizes. The importance of improving precision by covariate adjustment in large trials was highlighted by Dr. LaVange in her commentary. Dr. Proschan Dr. Proschan correctly pointed out that our methods were presented in the context of simple randomization. Stratified randomization is often used in phase 2 and 3 clinical trials (Lin et al., 2015). The methods that we presented for binary and ordinal outcomes can be directly applied to this case, except that the variance estimator should be modified to account for the stratified randomization procedure (thereby potentially increasing power in a corresponding hypothesis test). This modification can be done using the method in Wang et al. (2020), which gives a general formula for the asymptotic variance of M-estimators in randomized trials that use stratified randomization. For time-to-event outcomes, it is currently an open problem to determine the asymptotic variance for the covariate adjusted estimator that we used from Díaz et al. (2019), under stratified randomization. We conjecture that variance formulas (4-5) in Wang et al. (2020) can be used to consistently estimate the asymptotic variance in this case. An alternative estimator to consider for time-to-event outcomes is the augmented, inverse probability weighted estimator from Díaz et al. (2019), for which the variance estimation method in Wang et al. (2020) can be directly applied. Dr. Proschan clarified the contrast between conditional and marginal treatment effects. We much appreciate this, since it has been a common source of confusion in the context of covariate adjustment. Dr. Proschan posed the important question of which is more efficient in the context of logistic regression for binary outcomes: conditional or marginal tests (where both are covariate adjusted, and where the marginal test is a Wald test based on the estimator from Section 3.1 of our paper)? It was shown by Rosenblum and Steingrimsson (2016) that these are equally efficient, asymptotically, when the logistic regression model includes an intercept and main terms for treatment and baseline variables; this result holds under arbitrary model misspecification. Since the marginal treatment effect is interpretable without requiring model assumptions (unlike the conditional effect), we recommend to estimate and test marginal treatment effects in practice (as we did in our manuscript). Dr. Proschan suggested the use of randomization inference, which can also easily incorporate covariate adjustment. It has the advantage of not requiring distributional assumptions, and may be especially useful at smaller sample sizes where asymptotic arguments may be less applicable. A potential downside is that the confidence intervals produced by randomization inference are typically based on inverting tests of the null hypothesis that the treatment effect is identical to a fixed value for all participants. If treatment effects differ across individuals, then it may be difficult to interpret these confidence intervals. Lastly, we thank Dr. Proschan for pointing out our mistaken reference to the primary outcome in Beigel et al. (2020) as time to death; it was time to recovery." @default.
• W4205197184 created "2022-01-25" @default.
• W4205197184 creator A5005567441 @default.
• W4205197184 creator A5019357485 @default.
• W4205197184 creator A5022039512 @default.
• W4205197184 creator A5023302717 @default.
• W4205197184 creator A5028193922 @default.
• W4205197184 creator A5084451114 @default.
• W4205197184 date "2021-05-29" @default.
• W4205197184 modified "2023-09-30" @default.
• W4205197184 title "Rejoinder: Improving precision and power in randomized trials for COVID‐19 treatments using covariate adjustment, for binary, ordinal, and time‐to‐event outcomes" @default.
• W4205197184 cites W1594438765 @default.
• W4205197184 cites W2035655846 @default.
• W4205197184 cites W2760699482 @default.
• W4205197184 cites W2963870422 @default.
• W4205197184 cites W3027630905 @default.
• W4205197184 cites W970855755 @default.
• W4205197184 doi "https://doi.org/10.1111/biom.13495" @default.
• W4205197184 hasPubMedId "https://pubmed.ncbi.nlm.nih.gov/34050931" @default.
• W4205197184 hasPublicationYear "2021" @default.
• W4205197184 type Work @default.
• W4205197184 citedByCount "1" @default.
• W4205197184 countsByYear W42051971842021 @default.
• W4205197184 crossrefType "journal-article" @default.
• W4205197184 hasAuthorship W4205197184A5005567441 @default.
• W4205197184 hasAuthorship W4205197184A5019357485 @default.
• W4205197184 hasAuthorship W4205197184A5022039512 @default.
• W4205197184 hasAuthorship W4205197184A5023302717 @default.
• W4205197184 hasAuthorship W4205197184A5028193922 @default.
• W4205197184 hasAuthorship W4205197184A5084451114 @default.
• W4205197184 hasBestOaLocation W42051971841 @default.
• W4205197184 hasConcept C105795698 @default.
• W4205197184 hasConcept C116675565 @default.
• W4205197184 hasConcept C119043178 @default.
• W4205197184 hasConcept C121332964 @default.
• W4205197184 hasConcept C126322002 @default.
• W4205197184 hasConcept C149782125 @default.
• W4205197184 hasConcept C159047783 @default.
• W4205197184 hasConcept C168563851 @default.
• W4205197184 hasConcept C2779134260 @default.
• W4205197184 hasConcept C2779662365 @default.
• W4205197184 hasConcept C3006700255 @default.
• W4205197184 hasConcept C3008058167 @default.
• W4205197184 hasConcept C33923547 @default.
• W4205197184 hasConcept C41008148 @default.
• W4205197184 hasConcept C48372109 @default.
• W4205197184 hasConcept C524204448 @default.
• W4205197184 hasConcept C62520636 @default.
• W4205197184 hasConcept C71924100 @default.
• W4205197184 hasConcept C85461838 @default.
• W4205197184 hasConcept C94375191 @default.
• W4205197184 hasConceptScore W4205197184C105795698 @default.
• W4205197184 hasConceptScore W4205197184C116675565 @default.
• W4205197184 hasConceptScore W4205197184C119043178 @default.
• W4205197184 hasConceptScore W4205197184C121332964 @default.
• W4205197184 hasConceptScore W4205197184C126322002 @default.
• W4205197184 hasConceptScore W4205197184C149782125 @default.
• W4205197184 hasConceptScore W4205197184C159047783 @default.
• W4205197184 hasConceptScore W4205197184C168563851 @default.
• W4205197184 hasConceptScore W4205197184C2779134260 @default.
• W4205197184 hasConceptScore W4205197184C2779662365 @default.
• W4205197184 hasConceptScore W4205197184C3006700255 @default.
• W4205197184 hasConceptScore W4205197184C3008058167 @default.
• W4205197184 hasConceptScore W4205197184C33923547 @default.
• W4205197184 hasConceptScore W4205197184C41008148 @default.
• W4205197184 hasConceptScore W4205197184C48372109 @default.
• W4205197184 hasConceptScore W4205197184C524204448 @default.
• W4205197184 hasConceptScore W4205197184C62520636 @default.
• W4205197184 hasConceptScore W4205197184C71924100 @default.
• W4205197184 hasConceptScore W4205197184C85461838 @default.
• W4205197184 hasConceptScore W4205197184C94375191 @default.
• W4205197184 hasFunder F4320332161 @default.
• W4205197184 hasFunder F4320332163 @default.
• W4205197184 hasIssue "4" @default.
• W4205197184 hasLocation W42051971841 @default.
• W4205197184 hasLocation W42051971842 @default.
• W4205197184 hasLocation W42051971843 @default.
• W4205197184 hasOpenAccess W4205197184 @default.
• W4205197184 hasPrimaryLocation W42051971841 @default.
• W4205197184 hasRelatedWork W1975991677 @default.
• W4205197184 hasRelatedWork W2024551269 @default.
• W4205197184 hasRelatedWork W2042833573 @default.
• W4205197184 hasRelatedWork W2610273986 @default.
• W4205197184 hasRelatedWork W2748952813 @default.
• W4205197184 hasRelatedWork W2899084033 @default.
• W4205197184 hasRelatedWork W3036314732 @default.
• W4205197184 hasRelatedWork W4256514411 @default.
• W4205197184 hasRelatedWork W4382894326 @default.
• W4205197184 hasRelatedWork W2478782556 @default.
• W4205197184 hasVolume "77" @default.
• W4205197184 isParatext "false" @default.
• W4205197184 isRetracted "false" @default.
• W4205197184 workType "article" @default.