Bridging Data Gaps: Curating Real-World Data for Oncology Research

Increasingly, real-world data (RWD) from electronic health records (EHRs) and other sources is playing an important role in oncology research.

Due to the increasing openness and acceptance by regulators and payers, RWD has moved beyond historic use cases of HEOR and market access studies to constructing external control arms, informing post-marketing studies, and enhancing clinical trial design and enrollment.However, getting RWD into a research-ready state is not a simple matter.

By nature, RWD reflects care delivered in countless settings, with different clinicians, systems, and unique patient situations. With these uncontrolled conditions, the consistency and completeness of data capture can vary significantly. Critical details are often hidden in unstructured physician notes or documented inconsistently across sites.

Bringing structure to data from uncontrolled environments

Turning the variability of real-world documentation into useful research datasets requires a robust data access and curation process. Despite the rapid advances in artificial intelligence, reliable oncology RWD curation demands a hybrid methodology that pairs technology under the guidance of deep clinical expertise and a strong quality management system.

Complex oncology concepts such as lines of therapy, treatment exposure, and treatment response require expert clinical guidance and review not only during model development but during testing and validation as well.

A prudent integration of clinical expertise into a technology-enabled curation approach not only ensures clinical plausibility and validity but also brings needed context to instances of ambiguous documentation.

Addressing data gaps

A common concern about RWD is missingness. Unrecorded or absent values in a dataset can be categorized according to different mechanisms: missing completely at random, missing at random, or missing not at random.

It is vital to address missing data in order to avoid biased statistical estimates and erroneous conclusions.

Transparency regarding missingness is critical to building trust. Depending on the nature of the omission, imputation may be appropriate, i.e. filling in missing values leveraging methods like mean/median replacement, regression, or advanced multiple imputation techniques.

In cases of applying oncology RWD to research, advanced modeling is often necessary. This necessitates the development of algorithms to bridge the gap between data availability and the desired application, for example an external control arm or other regulatory use cases.

Clarifying lines of therapy data

In oncology research, algorithm development is often necessary to identify lines of therapy (LoT) associated with real-world treatment data.

In oncology, lines of therapy refers to the sequential order in which treatments are given and is typically governed by treatment failure and subsequent disease progression. In clinical trials, LoT is protocol-defined and prospectively captured. However, in the real-world setting, LoT is often not clearly documented by the treating clinician.

While the EHR may contain dates of treatment administration and prescribed agents, without explicit annotations, inferring line progression can be difficult.

To address this challenge requires developing sophisticated LoT algorithms that leverage both clinical and data expertise. Importantly, the outputs of these algorithms should be further validated by clinical experts.

This degree of methodological rigor ensures that curated RWD is scientifically credible and analytically reliable.

Applying technique to research

With a suitably flexible data model, customized algorithms can be developed to assist research sponsors and investigators in tailoring datasets to specific protocols or regulatory requirements.

To date, this approach to data curation and algorithm development has been successfully applied across multiple oncology indications, including acute myeloid leukemia, chronic lymphocytic leukemia, and diffuse large B-cell lymphoma.

As regulatory agencies, life sciences companies, and clinicians continue to embrace RWD, it is essential to address the variance of real-world documentation with precision and transparency. Real-world oncology data must be carefully curated with clinical intent to meet the necessary standards required to support life-saving cancer research.

C.K. Wang, Chief Medical Officer and GM of Oncology at Verana Health