COURSE AIM

The course provides a comprehensive introduction to trial design features used to mitigate bias, important aspects of trial design, conduct, analysis and reporting, and challenges and solutions for conducting RCTs with some focus on behavioural interventions. This will include some coverage of methods for elucidation of treatment mechanisms (e.g. mediation). Throughout the course the emphasis will be on practical issues faced by researchers in the conduct and analysis of RCTs through the lens of the mental health setting, and participants will be provided with skills to design, conduct and analyse rigorous RCTs in this research area.

REQUIREMENTS

This course requires an application to be made, please see more information tab on how to apply. Some practical sessions will involve analyses and interpretation using practice data sets using the Stata and/or R programs.  Stata and R syntax will be provided and explained, but some familiarity with software, especially Stata, is necessary, and individuals should have some knowledge of fitting regression models. Successful participants will need to bring their own laptop computer to the course with working installations of: 1) Stata version 14 or higher, 2) the R program.

Content Overview

Clinical prediction research develops models that try to predict the chances of a clinical outcome (such as death, diagnosis, treatment success or other future outcomes) based on characteristics related to the patient. Such models can be used to help clinician communicate the chances of clinical outcomes to their patients and to improve their management. It is therefore of crucial importance that such models are developed and tested appropriately. This 10-day course is aimed at PhD students and researchers in mental health and will provide an introduction to key components of prognosis and stratified medicine research using cutting edge statistical and machine learning modelling techniques.

The course covers all major steps of developing and accessing a clinical prediction model, including study design and data preparation, the problem of over-fitting in regression models, how to overcome over-fitting using penalized regression and cross-validation methods, how to deal with missing data, feature variable selection using random forests, and performance assessment and clinical usefulness of a model. Each day a short presentation of an application in prediction modelling will be presented. Teaching will be through lecturers and practical computer lab session interspersed with short presentations of prediction modelling researchers on current work. Practical sessions will involve the analyses and interpretation of practice datasets using the software R. Syntax of all procedures will be provided and explained but some familiarity with a syntax-based software (R, Stata, SAS) is advised. A short 2 h introduction to R will be provided at the beginning of the course.

Course Aim:

This module is an introduction into path analysis and structural equation modelling using the Stata software. The module features an introduction to the logic of SEM, including assumptions, model

specification, identification and estimation. Models for continuous and discrete response variables and continuous and discrete latent variables will be covered. Growth, autoregressive, MIMIC, and

instrumental variable models will be included.

Requirements:

This workshop will assume that participants have a good knowledge of:

• Basic regression and statistical modelling
• Writing code to run statistical models/syntax based statistical software, and familiarity with the Stata software specifically (such as can be obtained from the BHI Introduction to Stata course). Participants will need to provide their own laptop computer with Stata installed.

Course Aim:

The 10-day course provides an introduction to the nature of medical text, and the technical and organisational challenges encountered when processing.

The course will be based around practical examples and widely used NLP tools.

The course aims to provide an introduction to the major techniques of natural language processing, and to provide participants with the skills to create their own NLP applications, using both rules based and statistical approaches. Methods will be introduced for extracting structured information from text, and for automatically classifying text, together with the selection of data for training and for evaluation.

The course will discuss aspects of NLP that are specific to medical and biomedical texts, and develop a critical awareness of the issues associated with these texts. Participants will be equipped with the skills needed to analyse their own language processing problems, to design and test solutions, and to understand the limitations of those solutions.

The course will provide a practical instruction in the use of some widely used tools in NLP, including nltk (a Python toolkit).

COURSE DESCRIPTION

Starting from the scale construction and gradually moving to the most recent statistical methods employed in measurement, the course provides a complete methodological framework for applied researchers. While the classical test theory ideas on reliability and validity, often used in the literature, are presented, the more methodologically sound item response theory counterparts are explained (difficulty and discrimination, item and total characteristic and information curves).

For the measurement of the latent variable(s) the course presents exploratory and confirmatory factor analysis (EFA and CFA) for numerical data. For categorical data, both the item response theory approach (IRT; 2-parameter logistic, grated response and partial credit models) as well as the item factor analysis model (IFA; both with regard to EFA and CFA). The course also presents the methods to explore measurement differences (measurement invariance) between groups (MG CFA), measurement differences due to covariates (MIMIC models), and measurement differences between raters or time points (longitudinal IFA), for all types of data.

Lectures and lab sessions will be based on the use of the Stata software.

Potential students should already be familiar with and have a good understanding of regression and factor analysis. While no prior knowledge of Stata will be assumed, this will not be a formal Stata course. The course will be of interest to statisticians, researchers and PhD students of Psychology and Psychiatry, and mental health professional who wish to acquire and/or consolidate the knowledge and skills necessary for mental health research.