AAIC 2023

Systolic blood pressure and probable dementia: a secondary analysis of a randomized clinical trial using joint longitudinal and survival models to analyze how change in systolic blood pressure over time associates with risk for dementia.

R/Medicine Demo day - aorsf

A live demonstration of the aorsf R package, which provides fast routines to fit and interpret oblique random survival forests (RSFs). The talk also provides a light introduction to supervised learning, random forests, and oblique trees.

Accelerated oblique random survival forests

The oblique random survival forest (RSF) is an ensemble supervised learning method for right-censored outcomes. Oblique RSF ensembles often have higher prediction accuracy than standard RSF ensembles. However, computational overhead and lack of interpretability are pervasive limitations of the oblique RSF, In this talk, I introduce a method to increase computational efficiency of the oblique RSF and a method to estimate importance of individual predictor variables with the oblique RSF. Both methods are available in the aorsf R package.

Making Machine Learning Fast and Interpretable

Risk prediction can help direct treatment and interventions to patients who are most likely to benefit. The oblique random survival forest, a machine learning algorithm for risk prediction, has been used to develop a risk prediction algorithm for heart failure and to identify specific factors for adults who are black or white that drive predicted heart failure risk, with adverse social determinants of health being a major driver for adults who are black. This talk covers these topics and also introduces methods to increase the computational efficiency and interpretability of the oblique random survival forest.

Disseminating Prediction Methods

Statisticians are trained to develop novel statistical techniques that can be used to engage with complex problems. However, we are less likely to receive training in software development. Without efficiently coded algorithms, intuitive documentation, and friendly APIs, the methods we ‘share’ in our R packages may cause frustration and turn potential users away (possibly to a less valid method!). In this talk, I focus on efficiently writing the core algorithms in R packages using Rcpp. I share my experience developing R packages with statistical methods and present four ideas that have made a positive impact on my work.