Leveraging real-world data to assess treatment sequences in health economic evaluations: a study protocol for emulating target trials using the English Cancer Registry and US Electronic Health Records-Derived Database

Background

Considering the sequence of treatments is vital for optimising healthcare resource allocation, especially in cancer care, where sequence changes can affect patients’ overall survival and associated costs. A key challenge in evaluating treatment sequences in health technology assessments (HTA) is the scarce evidence on effectiveness, leading to uncertainties in decision making. While randomised controlled trials (RCTs) and meta-analyses are viewed as the gold standards for evidence, applying them to determine the effectiveness of treatment sequences in economic models often necessitates making arbitrary assumptions due to insufficient information on patients' treatment histories and subsequent therapies. In contrast, real-world data (RWD) presents a promising alternative source of evidence, often encompassing details across treatment lines. However, due to its non-randomised nature, estimates of the treatment effectiveness based on RWD analyses can be susceptible to biases if not properly adjusted for confounding factors. To date, several international initiatives have been investigating methods to derive reliable treatment effects from RWD — by emulating Target Trials that replicate existing RCTs (i.e. benchmarks) and comparing the emulated results against the benchmarks. These studies primarily seek to determine the viability of obtaining trial-equivalent results through deploying specific analytical methodologies and study designs within the Target Trial emulation framework, using a given database. Adopting the Target Trial emulation framework facilitates the analyses to be operated under causal inference principles. Upon validation in a particular database, these techniques can be applied to address similar questions (e.g., same disease area, same outcome type), but in populations lacking clinical trial evidence, leveraging the same RWD source. Studies to date, however, have predominantly focused on the comparison of individual treatments rather than treatment sequences. Moreover, the majority of these investigations have been undertaken in non-English contexts. Consequently, the use of RWD in evaluating treatment sequences for HTA, especially in an English setting, remains largely unexplored.

Objectives

The goal of this project is to investigate the feasibility of leveraging RWD to produce reliable, trial-like effectiveness estimates for treatment sequences. We aim to assess the capability of two oncology databases: the US-based Flatiron electronic health record and the National Cancer Registration and Analysis Service (NCRAS) database of England. To achieve this, we plan to harness the Target Trial Emulation (TTE) framework for replicating two existing oncology RCTs that compared treatment sequences, with the intent of benchmarking our results against the original studies. Further, we aim to detail the practicalities involved with implementing TTE in diverse databases and outline the challenges encountered.

Methods

1. We aim to emulate existing RCTs that compare the effect of different treatment sequences by constructing the study design and analysis plan following the TTE framework. Specifically, the following case studies are planned: (1) Prostate cancer case study 1 (PC1) - US direct proof-of-concept study (method direct validation): replicating the GUTG-001 trial using Flatiron data (2) Prostate cancer case study 2 (PC2) - US-England bridging study (method extension): emulating Target Trials that compare treatment sequences that have been common in England using Flatiron data (3) Prostate cancer case study 3 (PC3) - English indirect proof-of-concept study (method indirect validation): emulating the same Target Trial in PC2 using English NCRAS data (4) Renal cell carcinoma case study (RCC) - method direct validation in a single-arm setting: emulating the sunitinib followed by everolimus arm in the RECORD-3 trial using English NCRAS data 2. We will compare results of the emulated Target Trials with those from the benchmark trials. 3. We plan to compare different advanced causal inference methods (e.g. marginal structural models using IPW and other g-methods) in estimating the effect of treatment sequences in RWD.

Expected results

This study will provide evidence on whether it is feasible to obtain reliable estimates of the (comparative) effectiveness of treatment sequences using Flatiron data and English NCRAS data. If applicable, we intend to develop a framework that provides a systematic way of obtaining the (comparative) effectiveness of treatment sequences using RWD. It is possible that the data quality is insufficient to emulate the planned Target Trials. In this case, we will report reasons for the implausibility of data analysis. If applicable, we will make suggestions to whether the national health data collection may be enhanced to make the analyses possible. The results of this study will be submitted to peer-reviewed journals and international conferences.