MO516: Clusters of Chronic Conditions Across the Spectrum of Kidney Function: A Nationwide Cohort Study

Abstract BACKGROUND AND AIMS Multimorbidity is the presence of two or more chronic conditions. Patients with multimorbidity suffer from a high treatment burden as they have to cope with numerous medications and attend multiple specialists. Some chronic conditions co-exist in predictable ways, such as ischaemic heart disease and heart failure, but others may not. Multimorbidity is common in people with chronic kidney disease (CKD) and is associated with adverse outcomes. We aimed to explore: (1) what different clusters of conditions are associated with CKD and how these change as kidney function declines and (2) associations between clusters of conditions and adverse outcomes. METHOD We studied the Secure Anonymised Information Linkage Databank (SAIL): an electronic health records repository for 79% of the population of Wales (UK). We estimated the date at which patients crossed estimated glomerular filtration rate (eGFR) thresholds: 90, 75, 60, 45, 30 and 15 mL/min/1.73 m2. We used eGFRs recorded between 2006 and 2021 and estimated the dates of eGFR decline using mixed effects models. We identified 27 chronic conditions using ICD-10 codes and prescription data. We applied a k-modes clustering algorithm within each eGFR category, at the dates patients crossed eGFR thresholds. We chose the number of clusters via the elbow method. To help describe each cluster, we identified `key conditions’ that were common in the cluster and more common than in the overall eGFR category (prevalence >20% and more than double the prevalence in the overall eGFR category). To help determine whether these clusters were clinically meaningful, we studied the association between clusters and adverse outcomes risk using Cox proportional hazards models (all-cause mortality and major adverse cardiovascular events: MACE), with clusters having no key conditions as the reference group. RESULTS Overall, 533 362 patients were included in the analysis. The median age was lowest in the eGFR 90 category (56 years, IQI: 47–64) and the highest in the eGFR 30 category (81 years, IQI: 73–87). Patients in the low eGFR categories had the highest number of conditions (Fig. 1). The most frequently recorded condition was hypertension, which ranged from 34.4% for eGFR 90%–86.1% for eGFR 15. Diabetes ranged from 17.5% for eGFR 90%–53.4% for eGFR 15. Chronic pain ranged from 21.5% for eGFR 90%–41.6% for eGFR 30. The models with six clusters/eGFR category were chosen as they provided the best fit. In most eGFR categories, the majority of patients were included in one cluster with no key conditions (Fig. 2). Various conditions were clustered together, with cardiovascular conditions often co-existing at eGFR 30. Chronic pain featured prominently, either as the sole key condition or in combination with mental or physical conditions. The relative rates of each outcome were highest in the clusters with cardiovascular key conditions; for example, at eGFR 15, the hazard ratio for MACE in the `Pulmonary disease, asthma & heart failure’ cluster was 2.3 [95% confidence interval (95% CI) 1.9–2.9; adjusted for age and sex). CONCLUSION We have described the prevalence of chronic conditions at different levels of kidney function, and we have identified clusters of conditions. Multimorbidity became extremely common at lower eGFR, clustering in unexpected ways: at lower eGFR, combinations of cardiovascular diseases and depression became prominent, as well as chronic pain. Patients in clusters with cardiovascular key conditions were at the greatest risk of adverse outcomes. Identification of disease clusters with refinement by eGFR may allow targeted approaches to therapy.