In ohdsi-studies/LegendT2dm: Large-Scale Evidence Generation and Evaluation in a Network of Databases

knitr::opts_chunk$set(echo = FALSE)
library(dplyr)
library(DT)

Class-vs-class data experiment

Available data sources:

CCAE
MDCR
MDCD
OptumDOD
OptumEHR
Germany_DA
Australia_LPD
France_LPD
CUIMC
UK_IMRD
OpenClaims

Results are now available on legend_t2dm_class_diagnostics

RMarkdown file PsModelAssessment.Rmd (in LegendT2dm repo) generates this document using the OHDSI public DB server

Sys.setenv(DATABASECONNECTOR_JAR_FOLDER = keyring::key_get("legendT2dmDriverPath"))

legendT2dmConnectionDetails <- DatabaseConnector::createConnectionDetails(
  dbms = "postgresql",
  server = paste(keyring::key_get("legendt2dmServer"),
                 keyring::key_get("legendt2dmDatabase"),
                 sep = "/"),
  user = keyring::key_get("legendt2dmUser"),
  password = keyring::key_get("legendt2dmPassword"))

con <- DatabaseConnector::connect(legendT2dmConnectionDetails)
auc <- DatabaseConnector::querySql(con, sql = "SELECT * FROM legendt2dm_class_diagnostics.ps_auc_assessment;", snakeCaseToCamelCase = TRUE)
ps <- DatabaseConnector::querySql(con, sql = "SELECT * FROM legendt2dm_class_diagnostics.ps_covariate_assessment;", snakeCaseToCamelCase = TRUE)
DatabaseConnector::disconnect(con)

auc <- auc %>% mutate(auc = sprintf("%1.4f", auc),
                      equipoise = sprintf("%1.2f", equipoise)) %>%
  select(auc, equipoise, comparison, databaseId)

ps <- ps %>% mutate(sign = ifelse(coefficient >= 0, "+", "-"),
                    absCoef = sprintf("%1.2f", abs(coefficient))) %>%
  select(absCoef, covariateName, comparison, databaseId)

Propensity score AUC and equipoise

AUC close to (but not less than) 0.5 is best

Equipoise computed for [0.3 - 0.7] interval

datatable(auc, rownames = FALSE,
          options = list(order = list(2, 'desc')))

PS diangostics from VA-OMOP

Several highly-separable conceptIds emerge (vs GLP1RAs):

4203722 - Patient encounter procedure ($\infty$)
44924604 - Novofine needles
2514610 - Medication therapy management service(s) provided by a pharmacist
710054 - Qualified nonphysician health care professional online digital assessment and management

Most separable covariates

Note: index year: 2013, BD PEN NEEDLE and INSULIN AND ANALOGUES

datatable(ps, rownames = FALSE,
          options = list(order = list(2, 'desc'),
                         pageLength = 5))

Negative controls

LEGEND-HTN list: 76 - Some are questionable for T2DM (eg, wrist joint pain, vaginitis, amputated foot) - Some are gender biased (eg, endometriosis, irregular periods, bbnormal cervical smear)

nc <- read.csv("../inst/settings/NegativeControls.csv") %>% select(name)

datatable(nc, rownames = FALSE,
          options = list(pageLength = 5))

Negative controls

New list from Anna and Patrick:

From CEM (literature, labelingm spontaneous reports)
~200 (http://atlas-demo.ohdsi.org/#/conceptset/1868831/expression) then sorted by frequency
manual review, leaving 46

# nc <- read.csv("../inst/settings/NegativeControlsV2.csv") %>% select(name)
# 
# datatable(nc, rownames = FALSE,
#           options = list(pageLength = 5))