Development and validation of prediction models for secukinumab therapy in psoriasis: a machine learning approach enabling personalised medicine

Research question Can machine learning methods identify clinical markers predictive of secukinumab treatment response? Background Psoriasis is a chronic, immune-mediated disease with significant associated comorbidities.

The use of high-cost biologics have revolutionised treatment of psoriasis, however only around 50% of patients with psoriasis will remain on a biologic for 3-years due to loss of effectiveness. Little is known about what predicts variability in biologic treatment response.

Despite extensive research using genetic and molecular biomarkers, there are currently no predictors accurate enough for use in clinical practice.

Recent advances in data science have led to breakthroughs in machine learning with potential to discover insights into underlying structures of large and complex datasets.

Secukinumab (IL-17A inhibitor) is one of the current leading choice of biologics for psoriasis, which has been studied in the largest cohort of randomised controlled trials (RCTs).

The British Association of Dermatologists Biologics and Immunomodulators Register (BADBIR) is the largest psoriasis and pharmacovigilance registry in the world.

We propose a unique opportunity to apply state-of-the-art machine learning approaches to interrogate these high-quality resources.

The central hypothesis of this project is that baseline clinical characteristics can be used to predict treatment response to secukinumab in psoriasis.

Aims I aim to take an unsupervised, data-driven approach to identify baseline clinical predictors associated with responses to secukinumab treatment.

To accomplish these aims, I will receive further training from leading data science teams in Manchester, Newcastle and industry to develop expertise in machine learning methods. (Months 0-6) Methods. (Months 6-12) I will use secukinumab RCTs as a training dataset to develop predictive models. (Months 12-18) I will apply latent class longitudinal analysis to investigate whether patients can be divided into distinct groups according to the trajectory of change in psoriasis clearance following secukinumab therapy. (Months 18-24) I will apply different risk prediction methods (support vector machine and random forests) to classify baseline clinical characteristics, mapped to the response trajectories, to discriminate long-term treatment responder and non-responder. (Months 24-30) I will develop a web-based prognostic calculator and validate using data from real-world psoriasis registry (BADBIR).

Summary Using machine learning to inform clinical predictors could improve classification of treatment responders for better treatment stratification in the management of psoriasis patients. This may potentially shift the current biologic prescribing paradigm in psoriasis. Impact.

This project aligns well with the NIHR's Academy strategic theme for Health Data Science to deliver precision medicine through data.

It is ambitious, but will take maximal advantage from the research partnership between the Universities of Manchester-Newcastle-industry (Novartis); and equip me with high-level computational skills to become a future leader in this field.

Patient groups have been involved in the inception and development of the project, providing constructive feedback to translate the findings into a web-based decision aid. Dissemination.

I will ensure timely dissemination of results by publishing original research papers in high-impact journals, and presenting work at international meetings. I will make the codes and software developed open source; the methods would be of interest to a wide audience.