Background: Lupus is a prototype of a chronic complex autoimmune disease. Non-adherence rate to treatment is surprisingly high and impairs its management. Adherence to drug treatment is a complicated and multifactorial phenomenon, including characteristics of treatment.

Objectives: This study used unsupervised clustering analysis to identify profiles among lupus patients with regards to their treatment preferences (apart from efficiency).

Methods: An online survey among adult lupus patients from Carenity was conducted between August 2018 and April 2019. Multiple Correspondence Analysis (MCA) of a French lupus patient dataset was used with 3 unsupervised clustering methods (hierarchical, kmeans and partitioning around medoids).

Results: The 268 participating lupus patients were mostly female (96%), with a mean age of 44.3 years, and 83% fulfilled the ACR SLE diagnostic criteria. Overall, the preferred galenic form was oral (62%), and the most important characteristic was fewer side effects (32%).

A MCA was performed using 8 profile variables and the best unsupervised clustering method for this dataset (hierarchical clustering) identified 3 clusters (Table 1). Cluster 1, main one (59%), comprised patients with few comorbidities, less capacity to identify coming flares, and that wanted mostly (84%) oral treatments with limited side effects, most of them (83%) already on oral treatments. Cluster 2, the smallest one (13%), comprised younger patients, having already participated in clinical trials, favoring implants and the compatibility of treatments with pregnancy. Cluster 3 comprised patients (28%) that had longer lupus duration but less controlled disease with more comorbidities. A third had injectable treatments, wanting mainly implants and injections and with the main goal of reducing corticosteroids. The robustness of these results was confirmed by external validation and sensitivity analyses.

Conclusion: Most lupus patients preferred a drug galenic favoring the conservation of their autonomy. The identified clusters could help physicians to tailor their therapeutic proposition taking into account patients’ preferences and to maximize therapeutic adherence. Our study also highlights the potential of using unsupervised machine learning techniques in combination with direct-access patient community data to provide new a priori knowledge in the field of rare and complex chronic diseases.

References: [1]Meunier B, et al., Lupus. 2016 Apr;25(4):370-5.

[2]Qu H, et al., Arthritis Care Res (Hoboken). 2016 Dec;68(12):1787-1794.

[3]Fraenkel L, et al., Med Care. 2001 Nov;39(11):1203-16.

Disclosure of Interests: None declared