Project Summary: Developing disease-modifying therapies for neurodegenerative diseases has been challenging, in part because accurate statistical models to identify the optimal time for intervention do not exist. Models of how symptoms worsen over time (i.e., the symptom trajectory) before and after a clinical diagnosis can help identify that optimal time. These models can help pinpoint when a therapy could prevent a clinical diagnosis or slow the disease after a clinical diagnosis. Yet modeling the symptom trajectory is not easy even for Huntington disease, where researchers can track asymptomatic patients guaranteed to develop the disease and its symptoms. Like other neurodegenerative diseases, Huntington disease progresses slowly over decades, so studies that track symptoms often end before clinical diagnosis. This makes time to clinical diagnosis right-censored (i.e., a patient's motor abnormalities will merit a clinical diagnosis sometime after the last study visit, but exactly when is unknown), leaving researchers with the challenge of trying to model the symptom trajectory before and after clinical diagnosis without full information about when clinical diagnosis occurs. The challenge creates a unique statistical problem of modeling the symptom trajectory as a function of a right-censored covariate, time to clinical diagnosis (hereafter, simply “time to diagnosis”). Tackling this problem by modeling the distribution for time to diagnosis has long been thought to be the best strategy. For years, we and others worked to develop reliable distribution models, but we found that if the model was even slightly wrong, we would get biased estimates of how the symptom trajectory changes as a function of time to diagnosis. This bias causes problems for clinical trials because they are incorrectly powered to determine if a therapy modifies the disease course. We began seeking a strategy that does not require us needing to accurately model the distribution for time to diagnosis. Our team developed such a strategy for a related problem: estimating a regression model that has a covariate measured with error. Like a right-censored covariate, when a covariate is measured with error, the covariate's true value and distribution are unknown. Rather than finding the correct distribution, our model-free strategy accurately estimates the regression model even when the distribution for the covariate is mismodeled. Our overarching objective is to develop a similarly robust, model-free strategy when we have a right-censored covariate, which requires tackling challenges in three new areas: when the study ends before clinical diagnosis occurs (noninformative censoring; Aim 1), when worsening symptoms lead to study dropout (informative censoring; Aim 2), and when the data are longitudinal (Aim 3). This Landis Award Supplement will fund the training and development of a postdoctoral researcher, allowing them to integrate all project aims into a new software package for ...