Jonathan Shapiro 1, Tahel Ilan Ber 2, Or Ramni 2, Yonatan Jenudi 2, Amir Ben Tov 3,4,5, Sivan Gazit 3, Shlomit Steinberg-Koch 2, Benny Getz 2, Yehuda Shoenfeld 6, Ora Shovman 1,3,7

1 Maccabi Healthcare Services

2 Predicta Med, LTD

3 Maccabi Institute for Research & Innovation, Maccabi Healthcare Services, Tel Aviv, Israel.

4 Sackler Faculty of Medicine, Tel Aviv University, Tel Aviv, Israel.

5 Pediatric Gastroenterology Unit, Dana-Dwek Children’s Hospital, Tel-Aviv Sourasky Medical Center, Tel Aviv, Israel.

6 Zabludowicz Center for Autoimmune Diseases, Sheba Medical Center, affiliated to Tel-Aviv University, Israel

7 Department of Medicine B, Center for Autoimmune Diseases, Sheba Medical Center, Tel Hashomer, Israel



Psoriatic Arthritis (PsA) is an autoimmune inflammatory disease characterized by heterogeneous clinical manifestations. Timely diagnosis is challenging due to non-specific and overlapping symptoms, and delayed diagnosis may lead to irreversible joint damage and disability. Machine learning decision support tools can alert physicians to patients who would otherwise be misdiagnosed, thus improving patient outcomes. A proprietary machine learning algorithm, PredictAI™, was developed with the aim of identifying undiagnosed PsA in the primary care setting.


This retrospective study included electronic medical records (EMR) from approximately 2.5 million patients aged 21-85 belonging to Maccabi Healthcare Services, Israel’s 2nd largest Health Maintenance Organization, between 2008 and 2019. Inclusion criteria were: (i) at least 2 PsA diagnoses by a Rheumatologist, OR at least 1 PsA diagnosis by a Rheumatologist and 1 Psoriasis diagnosis by a Dermatologist, (ii) at least 6 years of data antedating first ever PsA diagnosis recorded.


2,020 patients the inclusion criteria. 86% were used for model training and the rest for validation. PredictAI™ identified 100 (45%) patients with discriminatory area under the curve (AUC) of 89%, 1 year prior to first ever recorded PsA diagnosis by any physician. It identified 88 (40%) patients 2 years prior and 76 (35%) patients 3 years prior to the diagnosis with AUC of 88% and 86% respectively (figure 1). Specificity was set to 99%.


PredictAI™ accurately identified 35-45% of PsA patients presenting to primary care, 1 or more years prior to first ever PsA diagnosis recorded by any physician. PredictAI™ can potentially substantially reduce time to diagnosis.


Figure 1- Identification of Psoriatic Arthritis Patients 1-3 Years Prior to First Ever Recorded Diagnosis, ROC Plot (TPR vs FPR)