Abstract
BackgroundThe target trial framework (TTF) is used to estimate the causal effect of treatments from observational data. The aim of this thesis is twofold: (1) to use the TTF to evaluate the feasibility of estimating cardiac treatment effects from electronic health record (EHR) data of a single EHR database, specifically that of King’s College Hospital (KCH), and (2) to generate real-world evidence to improve cardiac patient care.
Methods
To achieve these objectives, I:
(a) developed an EHR data extraction pipeline using the CogStack platform;
(b) conducted a scoping review of target trial emulation studies published up to February 2021 to understand the best practices for implementing the TTF;
(c) emulated two landmark cardiology trials: RALES, a superiority trial demonstrating the benefits of initiating spironolactone in heart failure (HF) patients, and ROCKET AF, a non-inferiority trial comparing rivaroxaban to warfarin for stroke prevention in atrial fibrillation (AFib);
(d) applied the TTF to investigate new superiority and non-inferiority questions regarding the RALES and ROCKET AF interventions.
Results
The RALES trial emulations, both strict and pragmatic, yielded two-year mortality intention-to-treat (ITT) HRs of 0.74 (95% CI: 0.33, 1.62) and 0.40 (95% CI: 0.22, 0.71). These estimates aligned with the directionality of the RALES trial’s estimate, fell within its 95% CI, or matched its statistical significance, demonstrating the feasibility of estimating ITT cardiac treatment effects from KCH’s EHR data. Building on the success of the RALES trial emulations, I applied the TTF to investigate new superiority questions regarding spironolactone. The new study findings suggest that initiating spironolactone resulted in a survival benefit over two years, which was reduced but maintained over five years in HF patients with an LVEF range of 0 to 40% (two-year ITT cumulative risk ratio (CRR): 0.44; 95% CI: 0.25, 0.67; five-year CRR: 0.63; 95% CI: 0.44, 0.86), but resulted in no significant survival benefit over five years in HF patients with an LVEF range of 0 to 35% (ITT CRR: 0.77; 95% CI: 0.52, 1.05).
The ROCKET AF emulation yielded a PP HR of 1.08 (95% CI: 0.56, 1.99). The emulation estimate did not align with the directionality of the ROCKET AF trial’s estimate, fell outside its 95% CI, and did not meet the non-inferiority margin established by the ROCKET AF trial (1.46), suggesting that estimating the PP cardiac treatment effect from KCH’s EHR data is more challenging. Assuming the emulation failed due to difficulties in identifying strokes and systemic embolism cases using KCH’s EHR data, I investigated a new non-inferiority research question about the ROCKET AF interventions with an alternative outcome of interest: all-cause mortality, while keeping the same non-inferiority margin of 1.46. This new study shows no evidence that, among patients with nonvalvular AFib, rivaroxaban is non-inferior to warfarin in preventing all-cause mortality over 840 days (PP CRR: 1.06, 95% CI: 0.71, 1.55).
Conclusions
This thesis is the first to use the TTF to evaluate the feasibility of estimating cardiac treatment effects from KCH's EHRs. It also provides real-world evidence to help improve cardiac patient care.
| Date of Award | 1 Sept 2025 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Sabine Landau (Supervisor), Ho Chung Wu (Supervisor) & Nilesh Pareek (Supervisor) |
Cite this
- Standard