K-means for shared frailty models.

BACKGROUND: The purpose of this research was to see how the k-means algorithm can be applied to survival analysis with single events per subject for defining groups, which can then be modeled in a shared frailty model to further allow the capturing the unmeasured confounding not already explained by the covariates in the model. METHODS: For this purpose we developed our own k-means survival grouping algorithm to handle this approach. We compared a regular shared frailty model with a regular grouping variable and a shared frailty model with a k-means grouping variable in simulations as well as analysis on a real dataset. RESULTS: We found that in both simulations as well as real data showed that our k-means clustering is no different than the typical frailty clustering even under different situations of varied case rates and censoring. It appeared our k-means algorithm could be a trustworthy mechanism of creating groups from data when no grouping term exists for including in a frailty term in a survival model or comparing to an existing grouping variable available in the current data to use in a frailty model.

Survival analysis of hierarchical learning curves in assessment of cardiac device and procedural safety.

Many Americans rely on cardiac surgical procedures and devices such as pacemakers and thrombolytic catheters to treat or manage their cardiovascular diseases. However, the failure of these cardiac devices and procedures could have grave consequences. One reason cardiac devices tended to fail was due to physician error; there is a learning effect for the physician or operator to come up to speed in skillfully implanting devices and conducting procedures. In order to better understand these learning effects, we had previously modeled the resulting learning curve effects in simulations a hierarchical setting with physicians clustered within institutions using our unique methodology (see the work of Govindarajulu et al 2017). Previously, we had employed these in hierarchical linear modeling and also in generalized estimating equations. In this setting, we have demonstrated how to apply similar methodology but revised in a survival analytic framework or time-to-event analyses. Through simulations and real dataset applications, we found that, out of the three shapes modeled to fit the learning curve, the logarithmic shape tended to have the best fit, similar to previous work (see the work of Govindarajulu et al 2017). However, as seen before, modeling the learning rate can be dataset specific and one shape may be better than another. We learned that modeling the learning rate could also be applied in the survival analysis setting through this new methodology. The goal of this paper is to model cardiac device and procedure learning curve effects in a time-to-event setting so that this knowledge may allow for the improvement of both short and long-term patient survival.