Concordance between primary and secondary electronic healthcare databases: A multi-database self-controlled case series study

Study type

• Study type:
  Non-interventional study
• Scope of the study:
  Assessment of risk minimisation measure implementation or effectiveness

Non-interventional study

• Non-interventional study design:
  Other
• Non-interventional study design, other:
  Self-controlled case series

Study drug and medical condition

• Study drug International non-proprietary name (INN) or common name:
  ACENOCOUMAROL

  APIXABAN

  DABIGATRAN

  EDOXABAN

  PHENINDIONE

  PHENPROCOUMON

  RIVAROXABAN

  WARFARIN
• Medical condition to be studied:
  Haemorrhage

Population studied

• Age groups:
  Adults (18 to < 46 years)

  Adults (46 to < 65 years)

  Adults (65 to < 75 years)

  Adults (75 to < 85 years)

  Adults (85 years and over)
• Estimated number of subjects:
  54000

Study design details

• Main study objective:
  Objective 1: Describe the concordance between primary and secondary care data in both the United Kingdom and the Netherlands, Objective 2: Compare the incidence of outcomes identified from primary and/or secondary care data in a self-controlled case series study (SCCS) design
• Outcomes:
  1) Percentage overlap of bleeding events occurring in the primary and secondary healthcare data domains, 2) incidence rates of major bleeding using primary and/or secondary care data and 3) Incidence rate ratios of major bleeding in the exposed time (first 30 days or including the remaining length of prescription) versus unexposed (baseline) time comparing primary and/or secondary care data.
• Data analysis plan:
  The baseline characteristics will be stratified by treatment group (DOAC or VKA) and by data source (CPRD Aurum or PHARMO). The baseline period is defined as the unexposed reference period 30 days prior to use of a one of the exposures and unexposed time begins 30 days after the last calculated exposure. Means, standard deviations (SD) and (percentage) totals will be calculated. Median follow-up will be calculated per treatment group in each data source. Incidence rates (IRs) for events occurring within exposed and unexposed intervals will be calculated, along with incidence rate ratios (IRRs) comparing these two periods. The IRR and corresponding 95% confidence interval (CI) will be calculated using conditional Poisson regression. Time-varying confounders which are associated with the exposure and the outcome, such as age, will be accounted for in the adjusted model. The analysis will be stratified by sex (effect modifier).