TL;DR
This study introduces a framework for characterizing comorbidities in genetic research cohorts using the Johns Hopkins Adjusted Clinical Groups (ACG) system. By analyzing electronic health records (EHR) from the eMERGE network, the researchers found significant differences in the burden of chronic conditions between cases and controls for certain complex diseases, suggesting this approach can improve cohort selection for genome-wide association studies (GWAS).
What Problem Does This Paper Address?
Genome-wide association studies (GWAS) aim to uncover genetic links to diseases. However, complex diseases often involve multiple co-occurring conditions (comorbidities) that can influence genetic findings. Current GWAS methods typically focus on a single primary phenotype, potentially overlooking how comorbidities might affect the observed genetic associations. This can lead to a less nuanced understanding of disease etiology and hinder the identification of true genetic determinants. The eMERGE network, which links electronic health records (EHR) with genetic data, is a valuable resource for such studies. However, the variability in how EHR data is used to define patient cohorts (e-phenotypes) and the lack of standardized methods for accounting for comorbidities present challenges. This paper addresses the need for a systematic way to characterize and compare comorbidities within GWAS cohorts, particularly between cases and controls, to enable more robust genetic analyses.
Methods
The researchers employed the Johns Hopkins Adjusted Clinical Groups® (ACG®) System, a well-established tool for classifying patient health conditions and predicting healthcare utilization. They utilized de-identified EHR data from twelve eMERGE institutions, focusing on International Classification of Disease (ICD-9-CM and ICD-10-CM) codes.
The core methodology involved:
- Phenotype Selection: Identifying cohorts of cases and controls for three specific e-phenotypes: ACE-inhibitor induced cough, peripheral arterial disease (PAD), and heart failure (HF).
- Comorbidity Characterization: Using the ACG system's Expanded Diagnosis Cluster (EDC) markers to capture all relevant chronic conditions for each individual. The ACG system assigns ICD codes to one of 282 EDCs and calculates a Chronic Condition Count (CCC) for each person.
- Frequency Analysis: Summarizing the top ten most frequent EDC chronic condition markers for both cases and controls within each e-phenotype.
- Comparison: Comparing the mean CCC between cases and controls for each e-phenotype using t-tests.
This approach allows for a comprehensive assessment of the comorbidity landscape within genetic study populations.
Datasets
The study utilized de-identified electronic health record (EHR) data from twelve institutions participating in the National Institutes of Health (NIH)-funded Electronic Medical Records and Genomics (eMERGE) Network. The data included International Classification of Disease, Ninth Revision, Clinical Modification (ICD-9-CM) and International Classification of Disease, Tenth Revision, Clinical Modification (ICD-10-CM) codes for service dates ranging from 1978 to 2017.
The specific e-phenotypes analyzed were:
- ACE-inhibitor induced cough
- Peripheral arterial disease (PAD)
- Heart failure (HF) (including both preserved and reduced ejection fraction subtypes)
These e-phenotypes were selected because they are commonly studied in GWAS, and some have already led to published GWAS studies.
Key Insights and Ideas
- Comorbidities Matter for GWAS Cohort Selection: The study highlights that simply defining cases and controls based on a primary phenotype might not be sufficient. Differences in comorbidity profiles between cases and controls can confound genetic analyses.
- ACG System as a Powerful Tool: The Johns Hopkins ACG system provides a standardized and comprehensive framework for quantifying the burden of chronic conditions, making it suitable for characterizing comorbidities in large EHR datasets.
- Disease-Specific Comorbidity Patterns: The research revealed that while some top comorbid conditions were shared between cases and controls, their rank order often differed, suggesting distinct comorbidity landscapes for cases versus controls.
- Potential for Improved Control Selection: Identifying significant differences in chronic condition counts between cases and controls for PAD and HF suggests that this characterization can inform more precise control selection strategies in future GWAS. This could lead to more accurate identification of genetic associations.
- Framework for Data Sharing: The authors propose that measures of comorbidity derived from the ACG system could be valuable additions to genomic datasets shared through platforms like dbGaP, enriching the data available for future research.
Key Results
- ACE-inhibitor induced cough: No statistically significant difference was found in the mean Chronic Condition Count (CCC) between cases and controls (p=0.142).
- Peripheral Arterial Disease (PAD): Statistically significant differences in CCC were observed between cases and controls (p<0.0001), with cases having a higher mean CCC.
- Heart Failure (HF): Similar to PAD, statistically significant differences in CCC were found between cases and controls (p<0.0001), with cases exhibiting a higher mean CCC.
- Top Comorbidities: For ACE-inhibitor induced cough, 9 out of 10 top comorbid conditions were the same between cases and controls. For PAD, 4 out of 10 were the same, and for HF, 5 out of 10 were the same. However, the rank order of these top comorbidities differed between cases and controls for all three phenotypes.
Limitations and Future Work
- Site Bias: The study acknowledges potential site bias due to the development and validation of e-phenotype definitions by individual institutions. Future work could explore more tailored approaches for case and control selection across institutions.
- Gender and Age Imbalances: The impact of gender and age imbalances on observed differences in CCC needs further exploration.
- Confounding by Indication: For some conditions, like PAD, comorbidities might be intrinsically linked to the definition of the phenotype or its risk factors (e.g., hypertension for PAD). Further investigation is needed to disentangle these effects.
- Data Modalities: The study primarily used ICD codes. Future work could integrate administrative claims data and laboratory data, which are captured by eMERGE cohorts but not used in this ACG analysis, to potentially improve the sensitivity and predictive power of comorbidity identification.
- Comparison with Other Definitions: The authors suggest further research is needed to compare the Johns Hopkins ACG software with other comorbidity definitions (e.g., Charlson comorbidity index) to understand how different definitions might influence analyses.
Summary updated