Seminars @ Biostatistics, UCPH
Upcoming seminars
Friday, September 11, 14:00
Constraint-based methods for causal discovery use conditional independence tests to estimate causal graphs. The performance of these methods depends crucially on the choice of test, as both poor calibration and low power can lead to errors in the estimated graph. In this talk, I will present the Generalised Kernel Covariance Measure (GKCM), a kernel-based conditional independence test built on the Generalised Hilbertian Covariance Measure framework (Lundborg et al., 2022), and discuss its potential for causal discovery. Through kernel embeddings, GKCM accommodates mixed continuous and categorical data and is sensitive to a broad range of conditional dependencies. Additionally, GKCM accommodates a broad class of regression estimators. We use this flexibility to replace the kernel ridge regression used in existing kernel-based tests with tree-based methods, addressing the computational cost of tuning kernel ridge regression and the poor calibration that can result when it is left untuned. In simulations, GKCM with tree-based regression methods often achieves better type I error control and competitive or superior power compared with state-of-the-art tests. I will also discuss its current limitations: GKCM scales poorly with sample size. This is particularly challenging for causal discovery algorithms, since these typically require running a large number of conditional independence tests.
Monday, September 14, 13:30
Clinical trials with survival endpoints lose information when participants are censored before death is observed. We develop target-preserving estimators that use posttreatment disease history, such as recurrence or progression, to recover information lost to censoring for marginal survival and restricted mean survival effects. The difficulty is that the intermediate event is downstream of treatment: naive adjustment can change the causal estimand, and the useful information enters only through the observed coarsening. We derive observed-data influence functions with and without recurrence history and obtain an exact gain identity. The identity shows that efficiency improvement is driven by the censoring hazard, the split of the alive risk set into recurrence states, and the residual-survival separation between those states. In the no-covariate illness–death model, the Aalen–Johansen estimator realizes the recurrence-augmented efficient score after standardization to the marginal target. With covariates, correctly specified Cox–Breslow transition hazards provide a root-n plug-in benchmark, while a hazard-induced one-step estimator gives rate robustness and, under transition-law stability, canonical inference with flexible learners. A semi-synthetic metastatic breast cancer study calibrated from digitized progression-free survival and overall survival curves illustrates the gain identity. The framework applies broadly to censored time-to-event studies with informative intermediate histories.
Map of CSS
You can find CSS next to the Botanical Garden, 5 minutes from Nørreport station.
Meeting room 5.2.46 is the library of the Biostatistics section, located in building 5, 2nd floor, room 46. See the map below for directions inside CSS.
