Making the Most of Your Observational Data: Creating a Synthetic Control from Your Natural History Study

August 7, 2020

Recently a biotech approached Cytel for support with a Phase 2 Study in oncology. Regulators had requested a natural history of disease study, which tracks disease progression in the absence of any form intervention. These studies are used to build disease-models that can then inform a range of development opportunities within a drug development program.

A March 2019 FDA Guidance highlighted the importance of such studies for rare diseases, with former FDA Director Scott Gottlieb acknowledging that a lack of knowledge about the natural history of certain diseases is a significant obstacle in rare disease drug development.

Natural history studies can be prospective observational studies, but given the limited patients available for enrolment, sponsors often prefer to save them for the actual clinical trial. For oncology patients, designing such observational studies might also be unethical.

Cytel advised using Real World Data for the purposes of conducting a natural history study, and then building a synthetic control arm using the same datasets to support regulatory submission. This allowed all new patients to enroll into the treatment arm of the trial.

The Cytel Strategy then developed as follows:

1: Defining Estimands

When using RWD for an observational study, it is critical to develop estimands that maintain scientific rigour. Estimands typically require trial sponsors to specify the population, the measure of comparison, endpoint and intervention to capture the treatment effect with precision. This might seem at odds with a natural history study where there is no ‘treatment effect’ per se. Regardless, it is important to go through the process of identifying the right population and determining what the endpoint should be. Constructing appropriate estimands for natural history studies can help inform models that guide clinical development decisions at every phase of product development.

2: Evaluate Datasets for Quality

As always, statistical analysis follows a ‘garbage in garbage out’ principle. Quality of datasets must be evaluated for any RWD study. Such datasets must be evaluated to ensure that the correct population has been studied, that the endpoints are similar (or ideally the same) as the ones defined by the estimands, or that statisticians can make adjustments using familiar techniques like propensity scoring.

3: Conduct Natural History Study

Using a high quality data set Cytel conducted a natural history study that informed protocol submission and satisfied regulators.

4: Build Synthetic Comparator Arm

Cytel then used this data to support regulatory submission by building a synthetic control arm to use as the comparator. This also served to preserve patients for the treatment arm.