Use of RWE in a regulatory context: issues and examples Rob Hemmings, MHRA
Disclaimer I don t particularly like the terms real world data, real world evidence. To be more precise, I will discuss data generated in clinical practice (DGCP) as a data source.
Contents General principles Acceptance of DGCP Regulatory context and learnings from the Adaptive Pathways pilot. Estimands Projects and initiatives IMI GetReal EMA registries Examples Which data source and methodology answered which research question Next steps
General principles - Acceptance of DGCP Long history in PhV. Reactive PhV is different to prospective studies for specific questions on safety or efficacy For some purposes DGCP is already extensively used. RWE vs RCT: wrong question Data source vs methodological approach + data source Scientific guidance on post-authorisation efficacy studies http://www.ema.europa.eu/docs/en_gb/document_library/scientific_guideline/2016/12/wc500219040.pdf Research question study design data source statistical analysis
General principles - Regulatory context Exceptional Circumstances requires learning through lifecycle Cond MA introduces the notion that some answerable questions can be addressed in the post-authorisation phase Revision to CHMP GL promotes early dialogue to agree the question in advance and how it will be addressed RMPs PAES The choice of study design will be based on the particular medicinal product and the scientific uncertainty to be addressed. Studies involving randomisation may be the preferred design in the PAES setting.
The evidence hierarchy (which is anyway flawed) doesn t apply
General principles - Adaptive Pathways pilot. Adaptive Pathways March 2014 and August 2016 A scientific concept for medicine development and data generation which allows for early and progressive patient access to a. medicine 62 applications; 18 for face-to-face meetings; 6 of the applicants received parallel advice from EMA and HTA bodies. All of the 18 proposals accepted in Stage II of the pilot included plans for the use of real-world data to supplement randomised clinical trials, with a plan that went beyond the traditional use of a registry to investigate safety aspects.
General principles - Adaptive Pathways pilot. Use of existing disease registries to identify natural history of the disease, current standard of care, resource utilisation, adherence to treatment; Single arm studies for rare diseases compared with outcomes and time-points inferred from disease registries; Open label salvage studies in patients with no therapeutic options remaining, with the purpose of obtaining an expansion of the indication; Collection of efficacy and safety data from early access/compassionate use programs to supplement RCTs in small populations; Investigation of non-serological outcomes for vaccines;
General principles - Adaptive Pathways pilot. Post-authorisation drug registries for effectiveness, longterm outcomes, drug utilisation, Patient Reported Outcomes (PROs), time to treatment failure, diagnosis confirmation; Linking drug registries to risk-sharing schemes for reimbursement (pay-per-performance, annuity payments) Expansion of the indication based on a mixture of disease registries and compassionate use data (for rare, severe diseases, where RCT data were available for less severe forms of the disease); Post authorisation studies to investigate biomarkers (or other subpopulation selection criterion) status of an allcomer population.
General principles - Adaptive Pathways pilot. Post-authorisation data gathering plans The challenge remains to identify methodologically sound strategies of real-world evidence collection to support the assessment of both efficacy and effectiveness. These issues need to be discussed further in appropriate stakeholder fora and research programmes. Two stumbling blocks: Lack of detail for post-authorisation work based on DGCP Can DGCP generate evidence to extend an Indication, i.e. off-label
General principles - Randomisation Studies involving randomisation may be the preferred design in the PAES setting. Randomisation (and blinding) are most important for avoiding bias. Avoiding bias helps to establish causality. The performance of techniques that exist to make comparisons between populations to which treatments are not allocated at random is difficult to verify. Is the effect size adequate to overcome the potential bias?
General principles - Estimands A new word for you? ICH E9(R1) Precision in study objectives, research questions in light of intercurrent events which can impact on measuring outcome. Multiple different treatment effects exist. In which are you interested? Are we all interested in the same one?
General principles - Estimands
General principles - Estimands
General principles - Estimands Research question estimand study design data source statistical analysis Alison: Characterising the patient population, identifying and measuring exposure and outcomes with sufficient sensitivity and specificity is difficult What questions do we want to answer? What questions can we answer through this data source? Which data source do I need to answer my question?
General principles Which data source do I need to answer my question? Take pragmatic trial (best of both worlds?) as an example Would tend to measure the effect of a treatment-policy : the effect regardless of intercurrent event, i.e. outcome in respect of the entire treatment pathway not necessarily that somehow attributable to the initially randomised / assigned treatment. Require a precise research question before determining a data source
Projects and initiatives - IMI GetReal Moving RWE earlier in the development cycle RWE navigator Investigating and improving the external validity of clinical trials Identifying drivers of effectiveness in several case studies exploring different therapeutic areas (Hodgkin s Lymphoma, schizophrenia and diabetes). PragMagic gives insight into the possible consequences of more pragmatic trial design
Projects and initiatives - EMA registries Registries in European post-marketing surveillance: a retrospective analysis of centrally approved products, 2005 2013 Of 392 products that received a positive CHMP, 31 registries were requested for 30 products in total. Sixty-five percent were product registries whereas 35% were disease registries and 71% of the registries had a primary safety objective. Most commonly reported issues with registries were delayed time to start and low patient accrual rates. The delays found in getting new registries up and running support the need to improve the timeliness of data collection in the post-marketing setting.
Projects and initiatives - EMA registries to improve stakeholder collaboration and optimise the use of registries to support regulatory decisions: exploring mechanisms for regulators and applicants to systematically consider the need for registries and interact with registry holders; sharing and disseminating information on patient registries in specific disease areas; recommendations on governance principles, core data elements and quality standards; identifying needs for data collection, methodological and technical guidance; sustainability
Projects and initiatives - EMA registries Engagement in product-specific discussion under Adaptive Pathways, Scientific Advice and Marketing Authorisation Applications Dedicated workshops on cystic fibrosis and multiple sclerosis
Examples a typical one? Detailed presentation and discussion of RCTs to support initial Conditional MA. we will collect data post-authorisation to confirm efficacy and safety and support full MA. What research questions? to collect more data What data source? an existing registry, or our own / a healthcare database What outcomes and variable are collected, what data quality? haven t looked yet What trial design? cohort What methodology? will look for external control Where? What? Etc etc
Examples an easy one? LentiGlobin BB305 (a gene therapy medicinal product for the treatment of transfusion dependent betathalassemia) Once-only administration Initial conditional approval might be foreseen in the EU. Database would be made comprehensive with longterm follow-up for duration of the effect and long-term Prospective discussion takes place on the data elements and design of long-term evidence generation. All patients, all outcomes.
Examples Sialanar PUMA for treatment of sialorrhoea (chronic pathological drooling) in children aged 2 to <18 years with neurological disorders. Article 10a WEU. The lack of sufficient and reliable qualitative and quantitative data and subsequent resulting uncertainties do not allow to establish that glycopyrronium bromide has been used in the European Union for the symptomatic treatment of sialorrhoea (chronic pathological drooling) in children and adolescents aged 3 to <18 years with neurological disorders with an acceptable level of safety. Clinical Practice Research Datalink (CPRD) study showed 115 children with a specific diagnosis of drooling and 313 children with conditions indicative of drooling could be identified.
Next steps Alison: Multiple examples where observational studies on the same safety issue produce disparate results highlighting the importance of study design Need to develop a deep understanding of the data, to define the strengths and limitations so that the evidence arising from its analysis can be appropriately challenged. This is not trivial.
Next steps Companies need to break down silos, invest time and resource. Progress will only be made through discussion of specifics. WD-40 I have not failed, I ve just found 10,000 ways that won t work. Regulators need to be prepared: space for dialogue has been created.
Use of RWE in a regulatory context: issues and examples Rob Hemmings, MHRA