What Prognos Found and Where We’ll Take It
By Jason Bhan, MD, Fernando Schwartz, Brian Saindon, & Lauren Wiener
Over the past few months, Prognos has been performing an extensive investigation into pancreatic cancer. Using data from our nationwide network of partner labs, we have sought to find a better solution to pancreatic cancer. If you’ve been following along with us over the past four weeks, keep reading for a better understanding of our findings and where we’ll take them next. If not, you can find more of our pancreatic cancer blogs here.
Why Pancreatic Cancer Screening Matters
As we mentioned in our first blog, pancreatic cancer, or pancreatic adenocarcinoma, grows slowly in the ducts of the pancreas and can often take 10 years or more to metastasize. Because of this slow growth, patients typically do not experience symptoms until the late stages of this cancer, and as a result, 80% of pancreatic cancers are found late. The only known way to cure pancreatic cancer is through surgery, but tumors detected so late are quite often inoperable.
While high-risk individuals commonly undergo imaging studies in efforts to find pancreatic tumors while they are still operable, imaging is expensive and cannot be conducted on the general population. Therefore, the search continues for blood-based early screening tests, and we are left with a startling statistic – less than 10% of patients diagnosed with pancreatic cancer are still alive 5 years after diagnosis.
Source: NIH, National Cancer Institute
Our Findings and What They Indicate
Prognos has access to a large amount of retrospective data on individuals diagnosed with pancreatic cancer, and our longitudinal laboratory data repository helps us signal noise in large samples of patients diagnosed with several conditions like Pancreatic Cancer. Among 11,541 patients, described in Blog 2, first diagnosed with Pancreatic Cancer in 2016, we selected 8 tests (based on prior research) to determine whether we could detect any trends in these tests prior to pancreatic cancer diagnosis.
Among the key tests identified in our study, our descriptive analysis suggests that the following tests have a signal within 5 years prior to pancreatic cancer diagnosis: Alkaline Phosphatase, Cancer Ag 19-9, Hemoglobin A1c/Hemoglobin.total in blood, Hemoglobin [Mass/Volume] in blood, and Platelets. In particular, as seen in the below image which is presented in Blog 3, the percentage of individuals with abnormal high results to these 5 tests change over the 5 year time period prior to diagnosis.
Furthermore, as seen in Blog 4, we applied a Loess Smoothing Curve, which allowed us to showcase a detailed view of how the test result values of these 5 tests change over time prior to to pancreatic cancer diagnosis. For example, Blog 4 present the below image which suggests that Alkaline Phosphatase test result values increase during the previous 2 years to pancreatic cancer diagnosis.
In finding a tendency for abnormal results for these five blood tests among later diagnosed pancreatic cancer patients, we have discovered evidence that they could be indicative of pancreatic cancer. If these trends do indicate what they seem to, we just may be looking at a new way to screen for pancreatic cancer – and one that’s quicker and more cost effective to boot.
How We Can Move Forward
Currently, screening for pancreatic cancer is only recommended in high risk populations. Risk factors that define a high risk population include the presence of pancreatic cysts, first-degree or second-degree relatives who have had pancreatic cancer, new onset diabetes at ages 50 or older, and genetic syndromes associated with pancreatic cancer. Pancreatic screening methods include imaging studies, some of which are invasive and carry their own risks.
Application of machine learning algorithms may lead to calculations of probabilities for developing pancreatic cancer among patients that are not diagnosed with pancreatic cancer. These algorithms may incorporate information on laboratory testing patterns or laboratory test result values, along with other currently accepted pancreatic cancer risk factors. More reliable probabilities could aid in defining high risk populations who should move forward with the pancreatic cancer screening process. Further analysis could also lend insight into the current pancreatic cancer screening panel in order to identify whether a wider selection of tests should be included for pancreatic cancer screening.
Conclusion
Our findings certainly indicate that we could be onto something, but time and further investigation are needed to confirm what we strongly suspect – that the blood test results we analyzed could detect pancreatic cancer up to five years sooner than typical diagnosis. We will continue our investigation and hope to follow up in the coming months with more results. Stay tuned!