Evaluating phecodes, clinical classification software, and ICD-9-CM codes for phenome-wide association studies in the electronic health record.

Wei WQ, Bastarache LA, Carroll RJ, Marlo JE, Osterman TJ, Gamazon ER, Cox NJ, Roden DM, Denny JC. · 2017

Read Evaluating phecodes, clinical classification software, and ICD-9-CM codes for phenome-wide association studies in the electronic health record. by Wei WQ, Bastarache LA, Carroll RJ, Marlo JE, Osterman TJ, Gamazon ER, Cox NJ, Roden DM, Denny JC. (2017) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc5501393.txt.

Introduction

Wei and co-authors report a methods comparison in PLoS ONE on disease coding for health record studies. Phenome-wide studies scan many diseases for gene links at once. They rely on billing codes to define each disease. Raw ICD-9 codes are often too fine or poorly grouped, though. The team compares three grouping systems head-to-head.

Hand-built phecodes matched 83 of 100 target diseases exactly. Raw ICD-9-CM codes matched 53, and CCS groups matched 32. Phecodes also replicated more known gene and disease links. Their signals tended to be stronger as well. All three systems can serve such studies, they conclude. Phecodes simply align best with real clinical diseases.

Catalog record

CREATORWei WQ, Bastarache LA, Carroll RJ, Marlo JE, Osterman TJ, Gamazon ER, Cox NJ, Roden DM, Denny JC.
YEAR2017
COLLECTIONPAPERS
FORMATTXT
FILE SIZE47 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC5501393

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Evaluating phecodes, clinical classification software, and ICD-9-CM codes for phenome-wide association studies in the electronic health record By Wei-Qi Wei, Lisa A. Bastarache, Robert J. Carroll, Joy E. Marlo, Travis J. Osterman, Eric R. Gamazon, Nancy J. Cox, Dan M. Roden, Joshua C. Denny, Andrey Rzhetsky Published in: PLoS ONE DOI: 10.1371/journal.pone.0175508 ABSTRACT Objective To compare three groupings of Electronic Health Record (EHR) billing codes for their ability to represent clinically meaningful phenotypes and to replicate known genetic associations. The three tested coding systems were the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes, the Agency for Healthcare Research and Quality Clinical Classification Software for ICD-9-CM (CCS), and manually curated “phecodes” designed to facilitate phenome-wide association studies (PheWAS) in EHRs. Methods and materials We selected 100 disease phenotypes and compared the ability of each coding system to accurately represent them without performing additional groupings. The 100 phenotypes included 25 randomly-chosen clinical phenotypes pursued in prior genome-wide association studies (GWAS) and another 75 common disease phenotypes mentioned across free-text problem lists from 189,289 individuals. We then evaluated the performance of each coding system to replicate known associations for 440 SNP-phenotype pairs. Results Out of the 100 tested clinical phenotypes, phecodes exactly matched 83, compared to 53 for ICD-9-CM and 32 for CCS. ICD-9-CM codes were typically too detailed (requiring custom groupings) while CCS codes were often not granular enough. Among 440 tested known SNP-phenotype associations, use of phecodes replicated 153 SNP-phenotype pairs compared to 143 for ICD-9-CM and 139 for CCS. Phecodes also generally produced stronger odds ratios and lower p-values for known associations than ICD-9-CM and CCS. Finally, evaluation of several SNPs via PheWAS identified novel potential signals, some seen in only using the phecode approach. Among them, rs7318369 in PEPD was associated with gastrointestinal hemorrhage. Conclusion Our results suggest that the phecode groupings better align with clinical diseases mentioned in clinical practice or for genomic studies. ICD-9-CM, CCS, and phecode groupings all worked for PheWAS-type studies, though the phecode groupings produced superior results. INTRODUCTION The near ubiquity of electronic health records (EHRs) represents an unprecedented opportunity to leverage large-scale healthcare data for discovery.[1, 2] Compared with randomized controlled trials or traditional observational cohorts, EHR-based studies offer several distinct advantages, including cost efficiency[3, 4], scale[5, 6], and the ability to conduct longitudinal analyses[7–9]. These advantages make EHRs a viable and efficient model for clinical and genomic research[10], including the potential to analyze hundreds of human diseases, drug responses, and many observable clinical traits. EHRs have proven particularly useful for phenome-wide association studies (PheWAS); however, currently, there is no EHR-derived “reference phenome” available for such research.[10] Most PheWAS, and indeed many other EHR studies, leverage International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) billing codes to define phenotypes. In this study, we evaluated three different ICD-based coding systems to enable clinical and genomic research. The World Health Organization (WHO) established and maintained the original version of International Classification of Diseases, Ninth Revision (ICD-9) to track morbidity and mortality statistics across the world. The United States Department of Health & Human Services and the Centers for Medicare and Medicaid Services (CMS) further created ICD-9-CM as an extension of ICD-9 primarily for billing purposes. ICD-9-CM codes have been widely used as a record of patient diagnoses in clinical practice and health management for decades. ICD-9-CM uses three (e.g. 250 “diabetes mellitus”) to five digits (e.g. 250.00 “type 2 diabetes mellitus without mention of complication”) to describe diseases and syndromes. The 2015 edition of ICD-9-CM has 22,401 distinct codes. Many of these codes are not billable codes (e.g., diabetes mellitus requires a 5-digit specification, so code “250” is not an allowable code). These codes are arranged hierarchically into nineteen large chapters (e.g., “Diseases of the digestive system”), 160 sections (e.g., “Noninfectious enteritis and colitis”), and 1,247 3-digit categories (e.g., 250.* “diabetes mellitus”). The first three digits describe the general condition of a patient and therefore have been commonly used to represent disease categories. Despite their convenience, using ICD-9-CM for phenotypic analyses remains challenging because not all codes are organized meaningfully for the purpose of high-throughput phenotypic analyses. ICD-9-CM separates active diagnosis of many diseases from a history of the same disease into different chapters, e.g. malignant neoplasm of breast (174.9) and personal history of malignant neoplasm of breast (V10.3). In addition, other diseases are separated for other reasons: sleep apnea has codes in the “Diseases of the nervous system” chapter (327.2) and in the “Symptoms, signs, and ill-defined conditions” (780.57). In other cases, logical numerical groupings combine diseases that are genetically or pathophysiologically distinct: 250.* aggregates both type 1 and type 2 diabetes mellitus. ICD-10 solves the last example but not the former ones. Thus, simply using the natural, numerical ICD groupings can introduce inaccuracy when phenotyping diseases. To facilitate clinical research using ICD codes, the Agency for Healthcare Research and Quality introduced the Clinical Classification Software for ICD-9-CM-CM (CCS) in 1999.[11] CCS reorganizes disparate ICD-9-CM codes into a smaller number of clinically meaningful categories. These categories are more useful for presenting descriptive statistics than individual ICD-9-CM codes. Similar to three-digit ICD-9-CM codes, CCS is an adequate solution for research that requires disease representations at a gross level. Nevertheless, with only 294 mutually exclusive categories (850 leaf diagnoses) in its 2015 version, CCS may lack sufficient granularity for many clinical studies. In addition, most clinical studies using the EHR require not only a phenotype definition for cases (those with the disease phenotype) but also one for controls (those without the disease phenotype and related conditions that might be the phenotype of interest).[12–15] For example, studies of type 2 diabetes often exclude subjects with any other type of diabetes.[16] Neither ICD-9-CM nor CCS provides a ready-made approach to automatically exclude populations of patients that are “possible” cases or have similar or potentially overlapping disease states (e.g., type 1 diabetes and secondary diabetes mellitus). In prior analyses, researchers often use 3-digit ICD-9-CM codes (parent codes) for exclusions, which could miss codes in other parts of diagnosis system (e.g., gestational diabetes, abnormal glucose, and secondary diabetes all have different 3-digit codes than type 1 and type 2 diabetes mellitus). The absence of such codes can result in a contamination of cases in the control population and decrease statistical power. Knowledgeable clinical researchers conducting focused studies on a particular disease often address these issues with custom approaches; however, our goal was to compare multiple schemas for phenotyping in high-throughput manner, such as PheWAS. In the process of developing the PheWAS approach within EHRs, we developed an aggregation schema for ICD-9-CM codes that attempted to represent distinct diseases and traits for primary clinical or genetic research while also pairing with groupings to identify reasonable control groups.[17, 18] These phenotype codes, or “phecodes”, aggregate one or more related ICD-9-CM codes into distinct diseases or traits. We manually reviewed all ICD-9-CM codes (including useful E and V codes) to reorganize them into phecodes, boosted by clinical co-occurrence data to identify like phenotypes in disparate sections of the ICD-9-CM code system (for example the ICD-9-CM code group 162 representing lung cancers and the code group V10.1 representing history of lung cancers) and existing phenotypes used in prior GWAS studies.

Read the complete text