OBJECTIVES:

To introduce Target Trial Emulation (TTE) methods for the analysis of real-world data (RWD) to support HTA decision making and demonstrate their application to derive causal average treatment effect estimate for survival outcomes from RWD with risk of bias from immortal-time and time-varying confounding.

METHODS:

We analyse longitudinal patient level data from the EU Myelodysplastic Syndrome (EUMDS) registry to quantify the causal average treatment effects of alternative protocols for using erythropoiesis-stimulating agents (ESA) as first line therapy in intermediate-1 to low-risk Myelodysplastic Syndrome patients. We apply causal inference methods for survival outcomes with time-varying exposures for static interventions using a TTE framework. We constructed (target trial) protocols to assess alternative treatment modalities and used a ‘clone and censor’ approach to account for the risk of immortal time bias. Inverse probability of censoring weights were calculated to address the induced selection bias and time-varying confounding. We compared naïve and weighted Kaplan Meier (counterfactual) curves for the alternative treatment strategies.

RESULTS:

We found that ESA is only beneficial for the first year with survival difference of 1.1% (-4.2% – 6.4%). However, towards the end of follow up at 6 years we found a difference in survival of -3.1% (-16.4% - 10.2%). A sensitivity analysis based on an alternative study protocol for the use of ESA in this patient’s population revealed a large benefit for ESA in terms of survival throughout the entire follow up period with a survival difference of 3% (-2.8% - 8.9%) at year 1 to 13.6% (-2.5% - 29.9%) at year 6.

CONCLUSIONS:

The recently launched NICE RWE framework highlights the opportunities of using RWD to support HTA. Our case study shows how to use causal inference methods to emulate a target trial and produce relevant estimates of treatment effect which can inform a clinical and funding decisions.