Skip to main content

Level 3 Analyses

Level 3 analyses:

  • Identify cohorts of interest
  • Perform complex adjustment for confounding repeatedly as part of prospective sequential analysis 

Below you will find descriptions of the different types of cohort identification strategies and analytic modules to perform a level 3 analysis.

What this program does: 

  • Identifies exposures, follow-up time, outcomes, and covariates
  • Estimates a propensity score (based on predefined covariates and/or via a high-dimensional propensity score approach)
  • Uses matching or stratification and follows the analytic cohort for outcome assessment in a survival analysis framework
  • Adds new users to the cohorts over time and repeats the above steps

Prospective surveillance:

  • Multiple executions of a level 2 analysis (Type 2 +: Exposures and Follow-up Time with Propensity Score Analysis) are required over the course of a surveillance activity. Typically, a program package is executed each time a Data Partner updates their database. Multiple surveillance options for propensity score analyses are available and differ in how patient data changes are handled across multiple executions during a surveillance activity.
  • Total Type 1 error is controlled over the course of the analysis at a user-specified threshold.

Output metrics include:

  • Tables of patient characteristics (unadjusted and adjusted cohorts)
  • Measures of covariate balance
  • Distribution of propensity score by exposure group
  • Odds ratios (with 95% confidence intervals)
  • Kaplan-Meier Curves
  • Attrition table

Continue reading about prospective propensity score matching on Sentinel's Git Repository. 

What this program does:

  • Identifies exposure of a medical product of  interest 
  • Defines risk and control windows relative to the exposure date
  • Examines the occurrence of health outcomes of interest during the risk and control windows 

Prospective surveillance:

  • Multiple executions of a Level 2 analysis (Type 3: Self-Controlled Risk Interval Design) are required over the course of a surveillance activity. In this fixed risk and control window design, only one option for prospective surveillance is currently available: the evaluation of mutually exclusive periods over time. Once data has been analyzed for a specific time interval, it is not updated or analyzed again. Therefore, careful consideration of database completion dates is important. 
  • Total Type 1 error is controlled over the course of the analysis at a user-specified threshold.

Output metrics include:

  • Number of exposure episodes
  • Exposed individuals
  • Individuals with a health outcome of interest in the risk and/or control windows 
  • Censored individuals overall
  • Estimates of relative risk (RR) and 95% confidence intervals are available 
  • Attrition table

Stratified by requester-defined:

  • Age group
  • Sex
  • Year
  • Year-month
  • Time-to-event (in days)

Continue reading about sequential self-controlled risk interval design on Sentinel's Git Repository. 

 

Want more details on the functional and technical documentation of each type of level 3 analysis? Visit Sentinel's Git Repository