Specification of Time Zero in Causal Inference Studies Using Real-World Data: Challenges and Solutions.

Background: Some major discrepancies between RWD studies and their randomized counterparts stemmed from a misalignment of eligibility, start of follow-up and the assignment of treatment strategies. Target trial emulation emphasizes alignment of assignment to treatment and start of follow-up at ‘time zero’. Specifying time zero can be simple (e.g., when comparing the initiation of two treatment strategies), but it can also be challenging (e.g., when time zero is non-unique or when treatment strategies are not identifiable at time zero)

Objectives: - To review the importance of aligning time zero, eligibility and treatment assignment.
- To provide solutions for 2 common challenges in the handling of time zero based on published case studies:
- Challenge 1: time zero is non-unique
- Solution 1: Sequential trial emulation with a baseline, time-fixed exposure
- Solution 2: Single trial emulation with a time-varying exposure
- Challenge 2: treatment strategies non-distinguishable at time zero
- Solution 1: Cloning
- Solution 2: Random assignment

This symposium is aimed to researchers involved in the design, analysis and interpretation of RWD studies. The key takeaway will be a deeper understanding of methodologic solutions to challenges of time zero specification.

Description: 1. Motivation (5’). Dr. Garcia de Albeniz (RTI Health Solutions, Spain) will present discrepancies between RWD analyses and clinical trials motivated by time zero misspecification.

2. Time zero is non-unique: sequential emulation with a baseline, time-fixed exposure (15’). Dr. Caniglia (University of Pennsylvania, US) will illustrate this approach via a case study on antibiotic initiation and preterm delivery.

3. Time zero is non-unique: single trial emulation with a time-varying exposure (15’). Prof. Sterne (University of Bristol, Bristol, UK) will contrast the method explained by Dr. Caniglia with a single trial emulation using a time-varying exposure, using COVID-19 vaccination as a case study.

4. Treatment strategies are not discernible at time zero because of the use of grace periods. The use of cloning (15’). Dr. Maringe (London School of Hygiene and Tropical Medicine, London, UK) will explain cloning using a case study on surgery for lung cancer.

5. Treatment strategies are not discernible at time zero because of the evaluation of different durations of treatment. The use of random assignment (15’). Dr. Aguado (RTI Health Solutions, Spain) will contrast the method explained by Dr. Maringe, cloning, with random assignment using a case study of antihypertensive therapy and dementia.

6. Discussion. Led by Prof. Hernandez-Diaz (Harvard University, Boston, US) and Dr. Garcia de Albeniz. (25’)