POS1411 ARTIFICIAL INTELLIGENCE FOR IDENTIFYING NEW DISEASE CLUSTERS IN PATIENTS WITH PSORIATIC ARTHRITIS/PSORIASIS: A PROOF-OF-CONCEPT STUDY

Abstract

BackgroundPsoriatic arthritis (PsA)/Psoriasis (PsO) patients present with multiple long-term conditions (MLTCs), and more than half of PsA patients have ≥1 LTC, which have an impact on the quality of life. We used the Clinical Practice Research Datalink (CPRD) in the UK to determine multimorbidity in patients with PsA/PsO. CPRD is a database of routinely collected UK patient data that can be used to examine multimorbidity over the life course of the disease. A understanding of clusters and timing may allow the development of tailored programmes for surveillance and early interventions to reduce MLTcs in PSA/PO.ObjectivesThe overall objective of our study was to use CPRD to identify and interpret new and frequently-occurring disease clusters in the PsA/PsO population.MethodsWe identified PsA/PsO patients from the CPRD GOLD for the UK from 2009 to 2018 with at least one year of follow up but excluded patients in practices that had migrated to Aurum, a different electronic health record system held by CPRD. All patients were matched to controls at a 1: 4 ratio by age, sex, practice. We analysed 40 common MLTCs outlined by Barnett et al. to identify and interpret multimorbidity clusters. Multimorbidity clusters were identified using the network bi-clustering. Our methodology can be divided into three steps- (a) Separate the “case” and the “control” population, (b) Create a patient-condition matrix for each of “case” and “control” population and (c) From the patient-condition matrix in each case use a the Euclidean distance criterion to compute a similarity matrix of all the possible conditions among those patients.ResultsWe identified 67,827 incident or prevalent PsA/PsO patients aged 20 years and above who were matched to 271,308 controls by age, sex and practice (Table 1). The median number of long-term conditions (LTCs) was higher in the cases than in the controls.Table 1.CharacteristicCASE, N = 67,827CONTROL, N = 271,308Age Group, n (%)<206,491 (9.6%)27,897 (10%)20-5442,357 (62.4%)169,008(62.3%)55 and above19,148 (28.2%)74,403 (27.4%)Ethnicity, n (%)White28,200 (41.6%)93,485 (34.5%)Black, S.Asian and other956 (1.4%)3,735 (1.4%)Unknown38,671 (57%)174,088(64.2%)Sex, n (%)F35,431 (52%)141,724 (52%)BMI category, n (%)20 to <2511,183 (16.5%)43,240 (15.9%)25 to <3013,565 (20%)45,783 (16.9%)30 to <357,584 (11.2%)22,690 (8.4%)35 to 403,048 (4.5%)8,172 (3.0%)Alcohol status, n (%)Non-drinker/ Ex-drinker8,649/2,159 (15.9%)33,983/7,579 (15.3%)Drinker/Heavy drinker47,316/3,030 (74.2%)180,888/8,831 (69.9%)Smoking status, n (%)Ex-smoker/Non-smoker14,137/29,289 (64%)48,110/142,295 (70.2%)Smoker24,018 (35%)71,779 (26%)Long term condition, Median (IQR)1.00 (0.00, 3.00)1.00 (0.00, 2.00)We present the results obtained via spectral clustering on the similarity matrix obtained in each “case” and “control” population via Euclidean distance. There are 40 LTCs which are represented via two clusters in the “case” and the “control” population network (Figure 1). The LTCs in the same cluster are strongly connected within themselves compared to those in the other cluster. The noticeable part was clustering of diabetes, hypertension and chronic heart diseases in one group, especially in the “case” population over the “control” population. Interestingly depression, dementia and chronic heart diseases were in the same group in the clustering results for both the “case” and the “control” population. There are a central group of diseases (HYP, DEM, DIV, CHD etc.) in Figure 1, which seems to be connected with both sets of clusters. As mentioned earlier, diabetes has connections with these central groups of diseases, and it tends to be more pronounced in the “case” population.Figure 1.ConclusionMLTCs including diabetes, hypertension, chronic kidney disease and heart disease occurred together more commonly in patients with PsA/PsO than those without PsA/PsO.The future goal is to identify frequently occurring clusters of MLTCs in the immune-mediated inflammatory disease population.Disclosure of InterestsPrasad Nishtala: None declared, Neil McHugh: None declared, William Tillett Speakers bureau: Abbvie, Amgen, Celgene, Eli-Lilly, Janssen, MSD, Novartis, Pfizer and UCB, Consultant of: Abbvie, Amgen, Celgene, Eli-Lilly, Janssen, MSD, Novartis, Pfizer and UCB, Grant/research support from: Abbvie, Celgene, Eli-Lilly, Janssen, and UCB, Sarah Skeoch: None declared, Jenny Humphreys: None declared, John Pauling: None declared, Olga Isupova: None declared, Anita McGROGAN: None declared, Julia Snowball: None declared, Sandipan Roy: None declared