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

Finding Long-COVID: temporal topic modeling of electronic health records from the N3C and RECOVER programs

Post-Acute Sequelae of SARS-CoV-2 infection (PASC), also known as Long-COVID, encompasses a variety of complex and varied outcomes following COVID-19 infection that are still poorly understood. We clustered over 600 million condition diagnoses from 14 million patients available through the National COVID Cohort Collaborative (N3C), generating hundreds of highly detailed clinical phenotypes. Assessing patient clinical trajectories using these clusters allowed us to identify individual conditions and phenotypes strongly increased after acute infection. We found many conditions increased in COVID-19 patients compared to controls, and using a novel method to associate patients with clusters over time, we additionally found phenotypes specific to patient sex, age, wave of infection, and PASC diagnosis status. While many of these results reflect known PASC symptoms, the resolution provided by this unprecedented data scale suggests avenues for improved diagnostics and mechanistic understanding of this multifaceted disease.

Identifying Comorbidity Patterns in People with and without Alzheimer's Disease Using Latent Dirichlet Allocation

Background

Multimorbidity is common in older adults and complicates diagnosing and care for this population.

Objective

We investigated co-occurrence patterns (clustering) of medical conditions in persons with Alzheimer's disease (AD) and their matched controls.

Methods

The register-based Medication use and Alzheimer's disease study (MEDALZ) includes 70,718 community-dwelling persons with incident AD diagnosed during 2005-2011 in Finland and a matched comparison cohort. Latent Dirichlet Allocation was used to cluster the comorbidities (ICD-10 diagnosis codes). Modeling was performed separately for AD and control cohorts. We experimented with different numbers of clusters (also known as topics in the field of Natural Language Processing) ranging from five to 20.

Results

In both cohorts, 17 of the 20 most frequent diagnoses were the same. Based on a qualitative assessment by medical experts, the cluster patterns were not affected by the number of clusters, but the best interpretability was observed in the 10-cluster model. Quantitative assessment of the optimal number of clusters by log-likelihood estimate did not imply a specific optimal number of clusters. Multidimensional scaling visualized the variability in cluster size and (dis)similarity between the clusters with more overlapping of clusters and variation in group size seen in the AD cohort.

Conclusions

Early signs and symptoms of AD were more commonly clustered together in the AD cohort than in the comparison cohort. This study experimented with using natural language processing techniques for clustering patterns from an epidemiological study. From the computed clusters, it was possible to qualitatively identify multimorbidity that differentiates AD cases and controls.

Towards a practical use of text mining approaches in electrodiagnostic data

Healthcare professionals produce abounding textual data in their daily clinical practice. Text mining can yield valuable insights from unstructured data. Extracting insights from multiple information sources is a major challenge in computational medicine. In this study, our objective was to illustrate how combining text mining techniques with statistical methodologies can yield new insights and contribute to the development of neurological and neuromuscular-related health information. We demonstrate how to utilize and derive knowledge from medical text, identify patient groups with similar diagnostic attributes, and examine differences between groups using demographical data and past medical history (PMH). We conducted a retrospective study for all patients who underwent electrodiagnostic (EDX) evaluation in Israel's Sheba Medical Center between May 2016 and February 2022. The data extracted for each patient included demographic data, test results, and unstructured summary reports. We conducted several analyses, including topic modeling that targeted clinical impressions and topic analysis to reveal age- and sex-related differences. The use of suspected clinical condition text enriched the data and generated additional attributes used to find associations between patients' PMH and the emerging diagnosis topics. We identified 6096 abnormal EMG results, of which 58% (n = 3512) were males. Based on the latent Dirichlet allocation algorithm we identified 25 topics that represent different diagnoses. Sex-related differences emerged in 7 topics, 3 male-associated and 4 female-associated. Brachial plexopathy, myasthenia gravis, and NMJ Disorders showed statistically significant age and sex differences. We extracted keywords related to past medical history (n = 37) and tested them for association with the different topics. Several topics revealed a close association with past medical history, for example, length-dependent symmetric axonal polyneuropathy with diabetes mellitus (DM), length-dependent sensory polyneuropathy with chemotherapy treatments and DM, brachial plexopathy with motor vehicle accidents, myasthenia gravis and NMJ disorders with botulin treatments, and amyotrophic lateral sclerosis with swallowing difficulty. Summarizing visualizations were created to easily grasp the results and facilitate focusing on the main insights. In this study, we demonstrate the efficacy of utilizing advanced computational methods in a corpus of textual data to accelerate clinical research. Additionally, using these methods allows for generating clinical insights, which may aid in the development of a decision-making process in real-life clinical practice.

Machine learning approaches for electronic health records phenotyping: a methodical review

OBJECTIVE

Accurate and rapid phenotyping is a prerequisite to leveraging electronic health records for biomedical research. While early phenotyping relied on rule-based algorithms curated by experts, machine learning (ML) approaches have emerged as an alternative to improve scalability across phenotypes and healthcare settings. This study evaluates ML-based phenotyping with respect to (1) the data sources used, (2) the phenotypes considered, (3) the methods applied, and (4) the reporting and evaluation methods used.

MATERIALS AND METHODS

We searched PubMed and Web of Science for articles published between 2018 and 2022. After screening 850 articles, we recorded 37 variables on 100 studies.

RESULTS

Most studies utilized data from a single institution and included information in clinical notes. Although chronic conditions were most commonly considered, ML also enabled the characterization of nuanced phenotypes such as social determinants of health. Supervised deep learning was the most popular ML paradigm, while semi-supervised and weakly supervised learning were applied to expedite algorithm development and unsupervised learning to facilitate phenotype discovery. ML approaches did not uniformly outperform rule-based algorithms, but deep learning offered a marginal improvement over traditional ML for many conditions.

DISCUSSION

Despite the progress in ML-based phenotyping, most articles focused on binary phenotypes and few articles evaluated external validity or used multi-institution data. Study settings were infrequently reported and analytic code was rarely released.

CONCLUSION

Continued research in ML-based phenotyping is warranted, with emphasis on characterizing nuanced phenotypes, establishing reporting and evaluation standards, and developing methods to accommodate misclassified phenotypes due to algorithm errors in downstream applications.

Optimized Identification of Advanced Chronic Kidney Disease and Absence of Kidney Disease by Combining Different Electronic Health Data Resources and by Applying Machine Learning Strategies

Automated identification of advanced chronic kidney disease (CKD ≥ III) and of no known kidney disease (NKD) can support both clinicians and researchers. We hypothesized that identification of CKD and NKD can be improved, by combining information from different electronic health record (EHR) resources, comprising laboratory values, discharge summaries and ICD-10 billing codes, compared to using each component alone. We included EHRs from 785 elderly multimorbid patients, hospitalized between 2010 and 2015, that were divided into a training and a test (n = 156) dataset. We used both the area under the receiver operating characteristic (AUROC) and under the precision-recall curve (AUCPR) with a 95% confidence interval for evaluation of different classification models. In the test dataset, the combination of EHR components as a simple classifier identified CKD ≥ III (AUROC 0.96[0.93-0.98]) and NKD (AUROC 0.94[0.91-0.97]) better than laboratory values (AUROC CKD 0.85[0.79-0.90], NKD 0.91[0.87-0.94]), discharge summaries (AUROC CKD 0.87[0.82-0.92], NKD 0.84[0.79-0.89]) or ICD-10 billing codes (AUROC CKD 0.85[0.80-0.91], NKD 0.77[0.72-0.83]) alone. Logistic regression and machine learning models improved recognition of CKD ≥ III compared to the simple classifier if only laboratory values were used (AUROC 0.96[0.92-0.99] vs. 0.86[0.81-0.91], p < 0.05) and improved recognition of NKD if information from previous hospital stays was used (AUROC 0.99[0.98-1.00] vs. 0.95[0.92-0.97]], p < 0.05). Depending on the availability of data, correct automated identification of CKD ≥ III and NKD from EHRs can be improved by generating classification models based on the combination of different EHR components.

Topic analysis of Road safety inspections using latent dirichlet allocation: A case study of roadside safety in Irish main roads

Under the Safe System framework, Road Authorities have a responsibility to deliver inherently safe roads and streets. Addressing this problem depends on knowledge of the road network safety conditions and the number of funds available for new road safety interventions. It also requires the prioritisation of the various interventions that may generate benefits, increasing safety, while ensuring that reasonable steps are taken to remedy the deficiencies detected within a reasonable timeframe. In this context, Road Safety Inspections (RSI) are a proactive tool for identifying safety issues, consisting of a regular, systematic, on-site inspection of existing roads, covering the whole road network, carried out by trained safety expert teams. This paper aims to describe how topic modelling can be effectively used to identify co-occurrence patterns of attributes related to the run-off-road crashes, as well as the corresponding patterns of road safety interventions, as described in the RSI reports. We apply latent Dirichlet allocation (LDA), a widespread method for fitting a topic model, to analyse the topics mentioned in RSI reports, divided into two groups: problems found; and proposed solutions. For this study, 54 RSI gathered over six years (2012-2017) were analysed, covering 4011 km of Irish roads. The results indicate that important keywords relating to the "forgiving roadside" and "clear zone" concepts, as well as the relevant European technical standards (CEN-EN1317 and EN 12,767), are absent from the extracted latent topics. We also found that the frequency of topics related to roadside safety is higher in the problems record set than in the solutions record set, meaning that problems are more easily identified and related to the roadside area than interventions may be. This paper presents methodological empirical evidence that the LDA is appropriate for identifying the co-occurrence patterns of attributes related to the ROR crashes in road safety inspections' reports, as well as the interventions' patterns associated with these crashes. Also, it provides valuable information aimed to determine the extent to which national road authorities in Europe and their contractors are currently capable of implementing and maintaining compliance with roadside standards and guidelines throughout the life cycle of roads.

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

Columbia Open Health Data (COHD) is a publicly accessible database of electronic health record (EHR) prevalence and co-occurrence frequencies between conditions, drugs, procedures, and demographics. COHD was derived from Columbia University Irving Medical Center's Observational Health Data Sciences and Informatics (OHDSI) database. The lifetime dataset, derived from all records, contains 36,578 single concepts (11,952 conditions, 12,334 drugs, and 10,816 procedures) and 32,788,901 concept pairs from 5,364,781 patients. The 5-year dataset, derived from records from 2013-2017, contains 29,964 single concepts (10,159 conditions, 10,264 drugs, and 8,270 procedures) and 15,927,195 concept pairs from 1,790,431 patients. Exclusion of rare concepts (count ≤ 10) and Poisson randomization enable data sharing by eliminating risks to patient privacy. EHR prevalences are informative of healthcare consumption rates. Analysis of co-occurrence frequencies via relative frequency analysis and observed-expected frequency ratio are informative of associations between clinical concepts, useful for biomedical research tasks such as drug repurposing and pharmacovigilance. COHD is publicly accessible through a web application-programming interface (API) and downloadable from the Figshare repository. The code is available on GitHub.