CONTEXT: The Alliance for Healthier Communities provides primary health care through Community Health Centres (CHCs) to people in Ontario who are at high-risk for health needs or experience barriers to receiving regular care. There is heterogeneity both within and across CHCs due to complex client profiles and specialized programs at CHCs. Better understanding this heterogeneity and especially reasons for care that frequently co-occur has implications for resource planning and allocation and for guiding future research.

OBJECTIVES: 1) Identify clusters of co-occurring reasons for care within CHCs, 2) find clusters that are common across CHCs, and 3) explore patterns of clusters with sociodemographic characteristics.

STUDY DESIGN AND ANALYSIS: Retrospective cohort study design. Hierarchical agglomerative clustering analyses.

POPULATION STUDIED: Adult clients who received primary care through a CHC from 2015 to 2017.

DATASET: CHCs have a central, structured electronic health record for all care providers. The present study uses sociodemographic characteristics and ICD-10 or ENCODE-FM codes from a de-identified extract of these records for adults who had at least one encounter with a Physician or Nurse Practitioner from 2015 to 2017.

OUTCOME MEASURES: Characteristics of clusters (e.g. summary of conditions) organized as per the objectives.

RESULTS: Across 73 CHCs there are 212,173 clients meeting the inclusion criteria, who collectively had 10,447,830 service events which resulted in 4,192 and 8,988 unique ICD-10 and ENCODE-FM codes, respectively. Exploratory work is underway to identify the most meaningful clustering method. Completed analyses include hierarchical agglomerative clustering with an average linkage approach and cosine similarity metric on ENCODE-FM codes for 10 randomly selected CHCs. Key challenges are the lack of inherent ‘distance’ between ENCODE-FM codes and rarity of some; treating codes as dummy variables gives each code equal importance and results in a large, sparse matrix. Final results with reasons for decisions will be presented at the Meeting.

EXPECTED OUTCOMES: The primary contribution of this work will be identification of commonly co-occurring conditions seen by CHCs, which clusters are present regardless of CHC, and which vary by sociodemographic characteristics. A secondary contribution will be methodological insights into clustering methods for primary care electronic health records.