The symptoms of psoriatic arthritis are nonspecific, similar to those of other forms of chronic inflammatory rheumatism. However, early diagnosis of psoriatic arthritis can not only relieve pain but also prevent irreversible stiffness. Israeli researchers have developed PredictAI, an ML algorithm to shorten the time to diagnosis. They presented their study at the European Academy of Dermatology and Venereology (EADV) symposium held in Ljubljana last May.

Psoriatic arthritis is a chronic inflammatory disease that affects the joints and connective skin, particularly in patients with psoriasis. It progresses in flare-ups that may be mild and spaced several years apart, and may therefore go undiagnosed, especially since the most common symptoms, namely pain and swelling of the joints, are common to all rheumatic diseases.

In psoriatic arthritis, many patients also develop irreversible erosive joint disease and deformity.

Diagnosing Psoriatic Arthritis (PSA) with ML

PSA can be diagnosed by a dermatologist or rheumatologist, yet few patients make the connection between psoriasis and PSA. Dr. Jonathan Shapiro, dermatologist, medical advisor at Predicta Med analytics LTD, head of tele-dermatology at Maccabi Healthcare Services in Israel and author of the study states:

“Many psoriasis patients themselves may be unaware that they have PSA and will contact a general practitioner or orthopedic specialist about joint or back pain – without linking it to their skin condition, especially since the nonspecific nature of these symptoms makes it difficult for a clinician to diagnose on the first visit.”

The goal for the researchers was to shorten the time to diagnosis of PSA, which requires an average of 2.5 years after the first symptoms appear.

In this retrospective study, Dr. Shapiro and his team used the database of Maccabi Health Services, Israel’s second-largest health organization, and analyzed more than 2,000 records of patients with PSA, aged 21 to 85, who visited between 2008 and 2019. The PredictAI algorithm accurately identified 32-51% of them one to four years before a clinician’s diagnosis:

• 32% of patients in the study were identified 4 years prior to a clinician’s PSA diagnosis;
• 43% one year prior;
• In the analysis of medical records of psoriasis patients only, 51% of patients subsequently diagnosed with PSA were identified 1 year before the first diagnosis.

Dr. Jonathan Shapiro states:

“What PredictAI™ brings is the ability to scan large medical databases and use AI methods to look for clues such as complaints of joint pain, orthopedic specialist visits, lab results and many other parameters that can help identify an undiagnosed PSA patient up to 4 years before the first suspicion of PA and can detect more than 50% of these patients.”

Professor Dedee Murrell, Professor of Dermatology at the University of New South Wales, Sydney and Chair of the EADV Communications Committee considers these results “A step towards a better treatment pathway for patients with this painful condition” and states:

“Arthritis due to psoriasis can cause permanent joint damage and can present years before any skin psoriasis is apparent…Early diagnosis and therefore earlier treatment that could prevent pain and permanent joint destruction would be welcome. I would be interested to know if these patients already had joint destruction and a prospective randomized study could be done to determine if earlier diagnosis prevented joint destruction and the future development of other comorbidities associated with psoriasis.”

The team intends to continue research to increase the accuracy and sensitivity of PredictAI and assures:

“Our next step is to verify the performance of PredictAI on more databases around the world, which can help validate its results and improve it. We are starting a prospective study that aims to identify currently undiagnosed PSA patients so that we can consider referring them to a rheumatologist for further investigation.”

Article sources:

Shapiro J et al. Development of a machine learning tool for the early diagnosis of psoriatic arthritis in a primary care setting: a population-based study. Abstract #629. Presented at the EADV 2022 Spring Symposium.

Translated from Accélérer le diagnostic du rhumatisme psoriasique grâce à l’intelligence artificielle