Back to search

Boston University

Causality, uncertainty and falsification in clinical research

Abstract

dc:description.abstract

We address two foundational issues of inductive reasoning and related applications. We first consider the problem of inferring the causal effect of active versus control treatment in randomized clinical trials. We regard the pairwise differences in outcome between active and control subjects as fixed quantities which may or may not be observed depending on treatment allocation. Causal inference consists of observing a sample of pairwise differences in order to estimate the mean of all possible pairwise differences, which constitutes the complete causal effect. However, because this complete effect is unobservable, there is an unavoidable observational uncertainty, which is a fundamental feature of the physical world. We follow with a focus on the problem of falsification of scientific claims based on experimental data. A scientific claim resulting from an experiment is a function of the observed data, which induces a stochastic model on the space of possible claims, and a probability of falsification by a follow-up experiment. The map from the data into the claim space, the structure of the claim space, and the claim falsification probability, provide a mathematical structure for exploring the statistical approach to inductive reasoning. We then turn from foundational issues and causal inference to the weaker question of association in the high-dimensional setting of microarray studies. We propose a permutation test based on the matrix of pairwise distances between individual genomic profiles to test for an association between a binary classifier and gene expression. If the two groups defined by the binary classifier differ in gene expression then the mean between group distance between profiles should exceed the mean within group distance. Finally, we extend the study of association in genomic studies to consider the accuracy of genomic-based prediction of uncomplicated recovery from severe trauma, for which we seek upper bounds. In this case the dimension of the model space precludes an exhaustive model search, so that we bracket this space in the two dimensions of smoothness and complexity. A sequence of models can be thought of as a visualization of the structure of a cross-section of the gene-outcome space.

Degree

thesis:*
Name thesis:degree_name
Doctor of Philosophy
Level thesis:degree_level
doctoral
Discipline thesis:degree_discipline
Mathematics
Grantor dc:publisher
Boston University
Year dc:date.issued
2012

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Hayden, Douglas

Rights

Language dc:language.iso
en_US

Identifiers

dc:identifier.*
Dc Identifier Barcode
11719032088512
Dc Identifier Bibid
99199459540001161

Chain of custody

source
Harvested from
Boston University
Base URL
open.bu.edu/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Hayden, Douglas. Causality, uncertainty and falsification in clinical research. doctoral thesis, Boston University, 2012. https://hdl.handle.net/2144/31566