PheProb: probabilistic phenotyping using diagnosis codes to improve power for genetic association studies

Area of Science:

• Genetics
• Biostatistics
• Health Informatics

Background:

• Electronic Health Records (EHR) are valuable for large-scale phenotypic screens.
• Standard methods using diagnosis code thresholds can be limited by code accuracy variations.
• Improved phenotype definition is crucial for powerful genetic association studies.

Purpose of the Study:

• To develop and evaluate a novel approach, Phenotype Probability (PheProb), for defining phenotypes from EHR data.
• To assess if PheProb improves statistical power in genetic association studies compared to traditional thresholding.
• To leverage diagnosis code information more effectively for phenotype identification.

Main Methods:

• The PheProb approach uses unsupervised clustering on diagnosis codes to group patients.
• Phenotype probability is assigned based on the count of relevant diagnosis codes.
• The method was validated using simulated and real-world EHR data, including hyperlipidemia (ICD-9 272.x) and low-density lipoprotein cholesterol (LDL-C) genetic risk alleles.

Main Results:

• Traditional thresholding methods (≥1, ≥2, or ≥3 codes) showed non-significant associations (p-values 0.126, 0.123, 0.142) between genetic risk scores and hyperlipidemia.
• The PheProb approach identified a significant association (p=0.001) between the genetic risk score and hyperlipidemia.
• PheProb demonstrated superior performance in detecting the genetic association.

Conclusions:

• PheProb enhances statistical power for genetic association studies by utilizing the full spectrum of diagnosis code information.
• This method offers a more robust alternative to simple thresholding for phenotype definition in EHR data.
• PheProb has significant implications for large-scale genetic studies, including Phenome-Wide Association Studies (PheWAS).