The Hidden Value in Oncology Case Reports

If you work in oncology and have ever tried to conduct research using real-world data, I’m sure you’ve been in this scenario before: After your RWD vendor conducts a feasibility study on their vast pool of data, they return a mere handful of patients that ultimately fit your pre-specified inclusion/exclusion criteria.

After filtering down based on diagnosis, line of therapy, prior treatment history, biomarkers, etc, you’ve essentially created a needle-in-the-haystack routine. So, what is one to do when a larger N is needed to achieve any level of actionable insights or statistical significance? You could try to add in other data sources, but that would likely double the cost of the project. And who has budget for that?

What if there was a way to “top off” your useable patient numbers with a much more cost-effective and efficient approach?

I implore you to consider the oft maligned, but nevertheless peer-reviewed and published case report.

In all my years in value, evidence, and access, I have rarely, if ever, seen oncology case reports leveraged in a way that would mirror “big data” RWD. And yes, I understand the reticence. These cases are only published because they’re unique, right? So why should they be included in an analysis aimed to represent larger populations?

I propose to you several counterarguments to your objections for your consideration.

Objection #1: case reports are only published because of some unique attribute(s) of the patient, they therefore can't be used to represent the general population:

?What is unique about case reports is highly correlated to the time of publication. For example, an adverse event?(AE) observed for the first time in the case report literature 2017 may likely be a common secondary reported AE in 2022. Think myocarditis with the use of immunotherapies.

?Additionally, case reports are rich in longitudinality. If the unique aspect of the patient is what occurred towards the end of the presented patient trajectory (which it commonly is), all data on the earlier part of the trajectory is just as 'normal' as that of RWD from EMRs.

?In rare oncology sub-populations (i.e. precision medicine) the cohort sample size is often so small that any patient you can get adds substantially to the data set.

?In some cancers where treatment options are scarce (e.g., bladder cancer and pancreatic cancer) the unique aspect is often some minor variation to known treatments (surgery or old SoC) for which no impact can be documented. Cohorts of case report RWD for these populations are therefore much closer to the “general EMR RWD population”'.

??

Objection #2: Regulatory bodies require access to the full EMR, so case report RWD cannot be used.

?While true, the majority of RWD projects are non-regulatory. Moreover, if the indication is deemed to be of sufficiently high medical need, regulatory authorities are always open to discussing alternative approaches?to optimizing the use of available evidence. Worst case scenario is that case report RWD gets excluded from the primary analysis but only included in a sensitivity analysis.

??

Objection #3: Case reports are anecdotal evidence which is at the bottom of the evidence hierarchy.

Unless you have patient consent and have the ability to follow patients across the entire healthcare continuum (like we do here at xCures), EMRs can potentially be even more messy and incomplete. In most cases, case report data offer more complete longitudinal data, containing rich data on genetic mutations and biomarkers as well as clinical outcomes that are typically used in clinical trials but absent from EMR RWD, e.g., RECIST criteria-based tumor response, ECOG status, and progression-free survival.

If you use real-world data, you are used to having to statistically adjust for known confounders. The perceived selection bias in case reports generally does not represent anything worse or more challenging than the usual EMRs.

?While it’s certainly true that we cannot adjust for unknown confounders, typical EMR-based RWD also have a high number of unknown confounders, and incomplete data on key variables makes it even more challenging to make sufficient statistical adjustments, (unless, of course, you have the entire medical records, as is the case here at xCures). ?As my colleagues at BioSpark have observed, the data on patient characteristics in case reports tend to be more complete.

Here at xCures, we have partnered with BioSpark to help them structure and database all oncology case reports dating back to 2015. The RWD data set contains a diverse set of patient records, including:

? 15,000+ breast cancer cases

? 10,000+ colorectal cancer cases

? 7,500+ pancreatic cancer cases

? 5,500+ bladder cancer cases

? 5,000+ lung cancer cases

?

But this is all oncology peer-reviewed published case reports, both US and globally. If there is a tumor type you are looking for and your current data set needs a “top off,” let’s discuss what you’re looking for, and decide if it might make sense to dip into this vast, untapped body of individual patient oncology data available in case reports.

If you are interested in potentially discussing more, please contact me at [email protected]

?

This blog was developed in collaboration with Kristian Thorlund, Co-Founder and President of BioSpark.