|
Abstract:
|
I describe innovative statistical methods to address challenging problems in the diagnosis of complex diseases and characterization of their underlying mechanism, in two different settings: mental health and nuclear medicine. Mental disorders are complex and often, lack reliable biomarkers for diagnosis. Severity of a mental disorder is often assessed by instruments that attempt to quantify the same trait but in different scales (e.g., continuous and ordinal), and evaluating the replaceability of differently scaled instruments remains as a challenging problem. I present methods that address this problem to support accurate characterization of complex mental disorders by assessing inter-rater reliability between continuous and ordinal scales, and present its extensions to adjust for covariates and accommodate other outcome types, including time-to-event data. In a different setting, to characterize complex biological mechanisms in kidney obstruction, physicians analyze complex modalities of clinical data, including renal images and curves. Clinical judgement of kidney obstruction depends on the experiences of radiologists, typically has poor interrater agreement. I discuss statistical models that can effectively integrate information from different modalities to produce accurate interpretations and predictions of kidney obstruction. I conclude with some remarks.
|
