OHDSI Home | Forums | Wiki | Github

Opportunity to help the hepatocellular carcinoma research community

Researchers
#1

Team:

I’m currently sitting at National Institute of Health at a NCI workshop on hepatocellular cancer. I’m learning quite a lot, though I admit its like drinking from a firehose: HCC is a quite a serious disease, most patients diagnosed have a grim prognosis and while there are many different treatment modalities available, many will not provide long-term efficacy to most patients. There is a wide range of treatment modalities, including various surgical, radiology, and medical options, each of which have variable efficacy and challenging safety profiles. The treatments are being innovated fast-and-furious, and the HCC community is trying hard to maintain Level 1 (RCT) evidence to inform current practice, but many are challenged by having the research keep up with the innovations. I was struck by when listening to various presentations during the workshop: in a space where there are so many different treatments (which are used in various sequences or combinations) trying to improve survival for a serious disease and an urgent need to evaluate their comparative effectiveness of these alternatives, it seems an observational data network, like what we’ve built in OHDSI, can meaningfully support clinical practice by complementing the evidence being gathered by RCTs. It makes rational sense that RCTs in this space can only have dozens or perhaps a few hundred patients on any treatment arm, and its clear that the data demands for this type of oncology research is substantial. But there’s seems to be a wide range of unanswered questions that EHRs, registries, and claims could answer right away: questions on clinical characterization to summarize disease natural history and treatment utilization in current clinical practice; questions on patient-level prediction to identify which patients are at greatest risk of treatment toxicities or to respond to treatment; questions on population-level effect estimation to examine the comparative safety and effectiveness of alternative regimens.

To satisfy my own curiosity, I created this cohort definition (http://www.ohdsi.org/web/atlas/#/cohortdefinition/1769552) to allow me to identify how many patients I have within data I have access to, who have a first diagnosis of hepatocellular carcinoma with at least 2 years of medical history prior to first diagnosis. HCC is fairly rare, so I was surprised to see that I have patient records for >50,000 patients who satisfy this cohort definition across the databases I examined. Recognizing wholly the inadequacy of claims and EHR data, I’m excited by what would be possible to learn from such a large sample, given that this is 2 orders of magnitude larger population than what most HCC research seems to focus on.

My question to the community: is anyone out there interested in supporting the HCC community to better understand the disease and effects of the medical interventions to support these patients? If you have data, how many HCC patients do you have to learn from?

Assuming there is community interest, I hope we can leverage the OHDSI Oncology workgroup’s efforts, as was showcased at AMIA yesterday by @rimma, @mgurley, @rchen, @Christian_Reich, and others. It seems the HCC community would be a great place to realize the potential a true learning healthcare system.

#2

Hi Patrick,

I ran the cohort on our IPCI data but we do not have patients in this cohort. The reason is that we do not have the Liver Cell Carcinoma code in our data (CDM nor source), we do have the exclusion codes. In the source coding system there are only higher-level terms.

Peter

#3

Hi Peter,

Could you please provide specific source codes?

@Dymshyts, we’ll need to check if we have mappings for these codes.

Thank you.

#4

Thanks for highlighting another important area of need Patrick!

I copied over your cohort definition to run on Columbia data and found that we have 218 patients who meet this definition. If we reduce it to 1 year prior observation that increases to 243 pts. However, I’m not sure how well the condition code for ‘primary malignant neoplasm of liver’ performs for identifying HCC. It does seem like liver cell carcinoma, which you also have in your definition, might do a bit better as all its children seem to be HCC. So if we want to explore this further, validation and comparison to some gold standard may be needed to create a robust and accurate phenotype/cohort definition

#5

@Patrick_Ryan This is really good area without enough evidence of good quality. Unfortunately, HCC is the second cause of death among cancer (next to the lung cancer) in Korea because of the high prevalence of HBV infection. When I included descendant terms of liver cell carcinoma, I got about 2000 patients in Ajou university EHR data and about 8000 patients in national sample cohort.

As you said ,the treatment of HCC is kind of an art. Though it is relatively clear to treat other cancer such as lung and colon, treating HCC is so difficult. Because it requires to understand the whole 3D anatomy of liver with vascular supplying system, the functional / structural relationship between cancer and the rest of liver, and underlying hepatic function in each patient to make a decision. And as you said, there is so bunch of options. In other words, the treatment can often be decided by the anatomy of cancer or condition of patients, rather than evidence or guideline.

The low prevalence in Western country and the complexity of treatment in HCC might be the main reason why there is no clear guidance in treatment of HCC. The study for HCC might be very complex. But definitely, this is the area requiring more attention.

Coincidentally, I’m planning to integrate the mutation information and radiology information of HCC patients into CDM. (It would take more than a month but less than a year… ^^;)

t