Target Trial Emulation Using Hospital-Based Observational Data: Demonstration and Application in COVID-19

Abstract

Methodological biases are common in observational studies evaluating treatment effectiveness. The objective of this study is to emulate a target trial in a competing risks setting using hospital-based observational data. We extend established methodology accounting for immortal time bias and time-fixed confounding biases to a setting where no survival information beyond hospital discharge is available: a condition common to coronavirus disease 2019 (COVID-19) research data. This exemplary study includes a cohort of 618 hospitalized patients with COVID-19. We describe methodological opportunities and challenges that cannot be overcome applying traditional statistical methods. We demonstrate the practical implementation of this trial emulation approach via clone-censor-weight techniques. We undertake a competing risk analysis, reporting the cause-specific cumulative hazards and cumulative incidence probabilities. Our analysis demonstrates that a target trial emulation framework can be extended to account for competing risks in COVID-19 hospital studies. In our analysis, we avoid immortal time bias, time-fixed confounding bias, and competing risks bias simultaneously. Choosing the length of the grace period is justified from a clinical perspective and has an important advantage in ensuring reliable results. This extended trial emulation with the competing risk analysis enables an unbiased estimation of treatment effects, along with the ability to interpret the effectiveness of treatment on all clinically important outcomes.

Keywords: COVID-19; competing events; methodology; observational data; target trial emulation.

Figure 1

A hypothetical example of target trial emulation applying cloning and censoring techniques. Notes: In observed data (A), patients with identification (ID) numbers 1 to 6 received treatment at some time after hospital admission. Patients with ID 7-10 were never treated during their hospital admission. Patients with ID 1, 3, 5, 7, and 9 died, and 2, 4, 6, 8, and 10 were discharged alive. Two clones of the observed data were created for each patient, with one clone assigned to each treatment strategy. (B) All patients were assigned to the X-treated arm in cloned sample B, and patients were censored when they deviated from the planned protocol. Patients with IDs 5 to 8 who were never treated were censored at the end of the grace period. (C) All patients were assigned to the non-X-treated arm in cloned sample C, and patients were censored when they deviated from the planned protocol. Patients with IDs 1 to 4 treated within the grace period were censored at the time they received treatment.

Figure 2

A competing risk model for trial emulation. Notes: Competing risk model with counterfactual (assuming all patients are treated or untreated from hospital admission) cause-specific hazards for in-hospital death and discharged alive, assuming that hazards are time-constant.

Figure 3

Weighted cumulative incidence function curves for in-hospital death (A) and discharged alive (B) outcomes. Notes: follow-up time of 60 days was shortened to 30 days for visualisation purposes.

Figure 4

Risk differences estimated via weighted cumulative incidence functions for in-hospital death and discharged alive outcomes. Notes: follow-up time of 60 days was shortened to 30 days for visualisation purposes