Skip to main content
. Author manuscript; available in PMC: 2022 Apr 29.
Published in final edited form as: Biometrics. 2020 Oct 5;77(4):1342–1354. doi: 10.1111/biom.13372

Figure 3:

Figure 3:

HPV log-hazard ratio from Cox modeling of overall survival and time to recurrence using imputed head and neck cancer data*

* Five imputation strategies are considered: (1) MICE based only on X, (2) MICE based on X and the log of the event/censoring time, (3) MICE based on X, the event indicator Delta, and the Nelson-Aalen estimate of the cumulative hazard H(t), (4) method of Bartlett et al. (2014), and (5) proposed strategy, where covariates are imputed ignoring the outcome and analysis involves stacking the imputations and weighting by f(Y|X). Imputation for strategies (2)-(4) use the overall survival outcome. For all methods, we present log-hazards ratios associated with HPV status from a Cox regression model for (A) overall survival and (B) time to recurrence based on the imputed data. This figure appears in color in the electronic version of this article, and any mention of color refers to that version.