Co-occurrence of medical conditions: Exposing patterns through probabilistic topic modeling of snomed codes

Abstract

Patients associated with multiple co-occurring health conditions often face aggravated complications and less favorable outcomes. Co-occurring conditions are especially prevalent among individuals suffering from kidney disease, an increasingly widespread condition affecting 13% of the general population in the US. This study aims to identify and characterize patterns of co-occurring medical conditions in patients employing a probabilistic framework. Specifically, we apply topic modeling in a non-traditional way to find associations across SNOMED-CT codes assigned and recorded in the EHRs of >13,000 patients diagnosed with kidney disease. Unlike most prior work on topic modeling, we apply the method to codes rather than to natural language. Moreover, we quantitatively evaluate the topics, assessing their tightness and distinctiveness, and also assess the medical validity of our results. Our experiments show that each topic is succinctly characterized by a few highly probable and unique disease codes, indicating that the topics are tight. Furthermore, inter-topic distance between each pair of topics is typically high, illustrating distinctiveness. Last, most coded conditions grouped together within a topic, are indeed reported to co-occur in the medical literature. Notably, our results uncover a few indirect associations among conditions that have hitherto not been reported as correlated in the medical literature.

Columbia Open Health Data, clinical concept prevalence and co-occurrence from electronic health records

“…Ornstein et al estimated prevalence and multi-morbidity of 24 chronic conditions from an EHR database covering primary care practices 15 . Bhattacharya et al reported patterns of co-occurring conditions in patients with kidney disease by applying topic modeling on SNOMED codes 16 . Researchers who have access to an Observational Medical Outcomes Partnership (OMOP) database can use the open web applications ACHILLES and ATLAS to access useful statistics and scientific analyses, including counts per concept, prevalence rates, and frequencies of records per person 17 .…”

Machine Learning Approaches for Electronic Health Records Phenotyping: A Methodical Review

“…The most common phenotypes are summarized in Figure 3 and include chronic conditions as well as adverse drug events. Social determinants of health such as marital status and homelessness were considered in 7 of the 106 articles [25][26][27][28][29][30][31], while more nuanced phenotypes such as disease severity (n = 5, 4.7%) [32][33][34][35][36] and disease subtypes (n = 19, 17.9%) [37][38][39][40][41][42][43][44][45][46][47][48][49][50][51][52][53][54] are emerging areas of focus. We also classified each phenotype as either binary (eg.…”

“…In contrast to the previously mentioned approaches, unsupervised learning is primarily used for phenotype discovery [151]. 20 articles developed unsupervised models, with a majority using latent Dirichlet allocation (LDA) [38,41,43,46,48], matrix or tensor factorization [40,44,52], and hierarchical clustering [45,47,[49][50][51] (Supplementary Material Table S12). A deep autoencoder was used to discover depression and suicidal ideation patterns [47].…”

Topic Identification from Spanish Unstructured Health Texts

“…Recently, topic models have been applied in the health domain. In [4] applied LDA model to find associations across SNOMEDCT codes diagnoses in patients with kidney disease. Also, [5] used LDA to identify 25 phenotypes topics from information of two healthcare systems.…”