In OHDSI/Legend: Large-Scale Evidence Generation and Evaluation in a Network of Databases

knitr::opts_chunk$set(echo = TRUE)
library(dplyr)

Definition process {-}

We originally identified outcomes for LEGEND from clinical trial endpoints from clinical guidelines and systematic reviews. We augmented these with adverse events from US structured product labels of hypertension drugs (https://dailymed.nlm.nih.gov/dailymed/). For each outcome, we developed an operational phenotype definition to determine if observational data could in fact support evaluation of the outcome. We used the same approach to design, implement, and evaluate all phenotypes. Specifically, we conducted a PubMed literature review to identify prior observational studies that used the phenotype as an outcome, looking especially for studies where source record verification or other approaches validated the outcome. In addition, we reviewed eMERGE PheKB phenotype entries (https://phekb.org/phenotypes). Clinical guidelines and systematic review of clinical trials of hypertension treatments informed our clinical definitions of cardiovascular outcomes (@Reboussin2018 @Whelton2018 @Williams2018). Where possible, conceptsets originated with published codelists (e.g. ICD-9-CM and ICD-10). We augmented these with lexical search and semantic exploration of the OHDSI standardized vocabularies. A clinical adjudicator then reviewed the cohort definitions and associated conceptsets. We developed concept definitions using ATLAS, the OHDSI open-source platform (https://github.com/OHDSI/atlas). We initially executed these definitions across 7 databases (CCAE, MDCR, MDCD, Optum, Panther, JMDC, IMS Germany) to identify qualifying patients. Because the databases used in this study do not all consistently contain laboratory values, diagnosis records alone identified outcomes involving electrolyte imbalance (hypokalemia, hypomagnesemia, hyponatremia). To assess consistency across data sources as well as general clinical reasonableness, we utilized these cohorts to characterize outcome incidence, stratifying by age decile, gender, and index year. We did not perform source record verification or other validation methods.

tab <- read.csv("CohortDefinitionTable.csv")
tab$name <- paste0("[", tab$name, "](", tab$url, ")")
tab <- tab %>% select(name, description, citations)
names(tab) <- c("Phenotype", "Logical description", "Supporting references")
knitr::kable(tab)

References

OHDSI/Legend documentation built on Dec. 29, 2020, 3:52 a.m.