Jackknife-based diagnostics for non-monotonic hazard survival model with interval-censored data
Jackknife-based diagnostics for non-monotonic hazard survival model with interval-censored data

Summary/Abstract: This study focuses on jackknife-based model diagnostics for a non-monotonic twoparameter hazard survival regression model (TBPR) when data is interval and rightcensored. This distribution is very flexible, because it accommodates both monotonic and bathtub-shaped hazard rates. This research proposes a bias-corrected jackknife harmonic mean and a ran-dom imputation technique to obtain the altered Cox-Snell (r*i), adjusted Martingale (r*i) and Schoenfeld (r*Si ) residuals. Two simulation studies were conducted to assess the perfor-mances of the altered residuals and their ability to detect extreme observations and outliers at various censoring proportions (cp) and sample sizes (n) for this model. The results indicated that the altered residuals based on jackknife outperformed other residuals at cp and n levels. The proposed methods are then illustrated using a real dataset on Hodgkin’s Disease with the prior treatment group as the covariate. The results showed that the altered residuals work well to address model adequacy and identify potential outliers in the dataset.

• Details
• Contents