Latent class analysis in imaging studies as a method to estimate the true disease status based on multiple test results

Research question : Is latent class analysis (LCA) a viable alternative to panel diagnosis in imaging studies and clinical practice? Background:

Diagnosis is challenging for many medical conditions, requiring extensive testing. Clinicians must then combine every test result to determine the disease status and recommend the correct treatment. To help decision-making,we assemble a panel of clinicians to arrive at a diagnosis together.

It is the same situation in studies evaluating new tests - "panel diagnosis" acts as the reference to compare the new test against, to determine accuracy. However, organising and conducting panel diagnosis is laborious, resource-intensive, and subjective (Bertens et al., 2013; Keating et al., 2013). LCA is a statistical modelling method that combines multiple test results to estimate the disease status.

Latent class models (LCMs) are more rapid, cheap, and objective than panels (MacLean & Dendukuri, 2021). There are

also publications on designing and reporting studies specifically for LCA in diagnostic testing (Cheung et al., 2021; Kostoulas et al., 2017). However, uptake is slow due to a lack of validations against panel diagnosis. Translating LCA into diagnostic testing will make diagnosis more efficient, incurring substantial benefits for patients, the NHS, and funders.

Aims: 1)Evaluate if, how, and when LCA is more effective than panels in completed studies 2)Apply and evaluate LCA prospectively in ongoing studies 3)Implement and evaluate LCMs in clinical practice as an alternative to panel diagnosis Methods: The project is organised into four work packages:

Work Package 1 (from 04/23 to 03/25) will evaluate when LCMs have sufficient agreement with established panel diagnoses in six NIHR-funded diagnostic accuracy studies to enable researchers to trust the approach.

Work Package 2 (from 04/25 to 06/26) will determine in what situations LCA can reliably estimate disease prevalence and test accuracy by applying it in simulated studies representative of real-world scenarios. We will incorporate our findings into guidelines for designing diagnostic accuracy studies that apply LCA.

Work Package 3 (from 07/26 to 11/27) will design data collection to evaluate LCA against a panel in an ongoing prostate cancer study investigating whether MRI can replace biopsy. We will develop an LCM to construct a reference standard in future sub-studies.

Work Package 4 (from 07/26 to 03/28) will compare LCA to a panel in an ongoing Crohn's disease study investigating a novel MRI technique for detecting disease activity. We will incorporate routine clinical data into the LCMs to evaluate new MRI techniques without an expert panel and to explore the potential of LCMs as diagnostic tools in a clinical setting.

Impact:

Given the environment for this project, I am exceptionally well-placed to disseminate LCMs in both research and clinical settings immediately following completion, leading to rapid impact. I secured mentoring and support from three Professors of Radiology, a Professor of Diagnostic Statistics, and a world-leader in LCA methods. Importantly, I developed this project and will continue to collaborate with patients at IBDrelief to ensure the output is pertinent and patient-centric.