Statistical Modelling and Inference for Health

IIDS67642 Statistical Modelling and Inference for Health 15 Semester 1 Postgraduate Taught Level 7 Lesley-Anne Carter Unit coordinator Matthew Sperrin Unit coordinator Division of Informatics, Imaging and Data Sciences 1 FHEQ level (Framework for Higher Education Qualifications) ' Masters/Integrated Masters P4 ' European Credit Transfer & Accumulation System Rating : 7.5 The health sector is rich with data that currently remains under-utilised and often uses data to look at past healthcare delivery rather than using the data in order to provide insight to enhance healthcare delivery. A key component of the skill set of a data scientist is to be able to understand and implement a suite of modelling and statistical inference methods in order to utilise such data. This unit will build on central concepts and methods introduced in the pre-requisite unit Statistics for Health Data Science’ to provide the modelling and inference techniques required to handle complex data structures (such as nested data) and address causal hypotheses. The unit will be application-driven, with case-studies and examples drawn from health research across the University of Manchester. The unit will cover the following indicative topics:Section 1: Longitudinal Data Analysis<ul><li>Longitudinal data analysis using two-level and three-level random effects models</li><li>Experience Sampling Methodology - Study design elements for intensive longitudinal data</li><li>Methodological challenges in intensive longitudinal designs, including power and missing data</li></ul> Section 2: Causal Inference<ul><li>Confounding, selection bias and measurement error.</li><li>Directed acyclic graphs, d-separation</li><li>Study design principles</li><li>Adjustment methods: matching, stratification and inverse matching, propensity scores</li><li>Role of machine learning and AI in causal inference</li></ul> The health sector is rich with data that currently remains under-utilised and often uses data to look at past healthcare delivery rather than using the data in order to provide insight to enhance healthcare delivery. A key component of the skill set of a data scientist is to be able to understand and implement a suite of modelling and statistical inference methods in order to utilise such data. This unit will build on central concepts and methods introduced in the pre-requisite unit Statistics for Health Data Science’ to provide the modelling and inference techniques required to handle complex data structures (such as nested data) and address causal hypotheses. The unit will be application-driven, with case-studies and examples drawn from health research across the University of Manchester. The unit will cover the following indicative topics:Section 1: Longitudinal Data Analysis<ul><li>Longitudinal data analysis using two-level and three-level random effects models</li><li>Experience Sampling Methodology - Study design elements for intensive longitudinal data</li><li>Methodological challenges in intensive longitudinal designs, including power and missing data</li></ul> Section 2: Causal Inference<ul><li>Confounding, selection bias and measurement error.</li><li>Directed acyclic graphs, d-separation</li><li>Study design principles</li><li>Adjustment methods: matching, stratification and inverse matching, propensity scores</li><li>Role of machine learning and AI in causal inference</li></ul> This unit aims to build on the skills introduced in the unit ‘Statistics for Health Data Science’ to provide students with an understanding of techniques in statistical modelling and inference that allow a deeper insight into bio-health data. The unit will develop students’ literacy in the strengths, characteristics and correct application of modelling techniques, and how to interpret results. On completion students will also be able to implement analyses in an appropriate scripting language. In addition, the unit will develop students ability to critically appraise literature that describe previously implemented methods to address healthcare problems. Across all categories, students should be able to: <ul><li>LO1: Explain and discuss modelling and causal inference techniques and appraise their application in healthcare </li><li>LO2: Appraise the strengths and weakness of longitudinal modelling methods </li><li>LO3: Discuss the challenge of causal inference and the strong assumptions required.  </li></ul> <ul><li>LO4: Critically appraise literature that uses mathematical/statistical methods for health data </li><li>LO5: Assess the effectiveness and fitness for purpose of a modelling tool or technique</li><li>LO6: Apply modelling techniques and methods to healthcare data </li><li>LO7: Interpret analytical results</li></ul> <ul><li>LO8: Design and write scripts to implement statistical /mathematical methods to analyse health data</li></ul> <ul><li>LO9: Communicate and write analytical methods based on completed work and available literature in this area </li><li>LO10: Develop problem solving skills </li><li>LO11: Demonstrate a critical understanding of technical descriptions of statistical/mathematical analysis methods</li></ul> Analytical skills Understand technical descriptions of statistical/mathematical analysis methods Problem solving Solve problems The unit will be taught in a blended-learning format: basic knowledge and directed reading will be provided via eLearning so as to introduce students with key knowledge. The face-to-face time will consist of a series of lectures and discussions in which core concepts (introduced through pre-reading) will be re-capped and any further development discussed, as well as supervised computer time during which practical software and programming problems will be explored. Lectures will be recorded and distributed online. Alongside this, designated tutorials will be made available for students with academic staff as well as on-going support both online and F2F. <figure class="table"><table><thead><tr><th>Assessment</th><th>Length</th><th>How and when feedback is provided</th><th>Weighting in unit</th></tr></thead><tbody><tr><td>Summative: 1 x Short Report The report will include statistical scripts demonstrating the analysis of data and a written report (in paper style) to justify methods and explanation of work. There is a focus on selecting appropriate methods to solve applied problems.</td><td>1000 words</td><td>On Canvas / Turnitin, within 15 working days</td><td>40%</td></tr><tr><td>Summative: 1 x Written Examination  The written exam will test understanding of core concepts, interpretation of results, and critical thinking.</td><td>90 mins</td><td>After each assessment feedback is returned, a drop-in session will also be held for students to get further feedback as required.</td><td>60%</td></tr><tr><td>Formative: Formative assessment and feedback to students is a key feature of the on-line learning materials for this unit. Students will be required to engage in a wide range of interactive exercises to enhance their learning and test their developing knowledge and skills. In addition, there will be a series of supervised practical hands-on exercises that will allow for verbal feedback.</td><td> </td><td>Real-time feedback, and available through regular office hours.</td><td> </td></tr></tbody></table></figure> Feedback for summative assessments will be on Canvas / Turnitin, within 15 working days. After each assessment feedback is returned, a drop-in session will also be held for students to get further feedback as required. IIDS67631 Statistics for Health Data Science Pre-Requisite Compulsory IIDS67901 Machine Learning for Health Data Science Pre-Requisite Compulsory Health Data Science - CPD Health Data Science - CPD PGDT Taught Component Optional N Kirkwood, B. R., & Sterne, J. A. (2010). Essential medical statistics. John Wiley & Sons.Hernán MA, Robins JM (2024). Causal Inference: What If. Boca Raton: Chapman & Hall/CRC. (Parts I and II only)  eAssessment 30 Lectures 18 Practical classes & workshops 24 Tutorials 2 0 76