We are inspired by research that suggests there is a “window of opportunity” for early detection of pancreatic cancer, as noted by Kim and Ahuja in their 2015 paper (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4560741/ ). They note that there appears to be a 10-year period when pancreatic cancer is in a stage when it is potentially curable. However, the majority of patients are diagnosed after this stage due to the lack of symptoms, such as jaundice or back pain, until late in the development of the tumor. We hope to find a pattern in the results of common lab tests in our analysis, to offer information that may lead to a panel of lab tests or an equation using test results that can estimate the risk of pancreatic cancer in the general population – a screening tool that may one day indicate risk, open the window of opportunity, help find early tumors, and reduce mortality rates.
Following up on the results reported in Blog 3, we elected to dive deeper into each of the five individual tests out of the original eight tests we selected, because these five showed promise as potential screening tools for pancreatic cancer. The Loess Smoothing Curve is a useful tool to see if there is a smooth curve between the two variables, without assuming that the data fits any type of distribution. We used this tool to look at the variables of test result value versus time for results of the serum tests: alkaline phosphatase, CA 19-9, hemoglobin, platelets and HbA1c.
We start by looking at the two blood tests indicated by our previous research in 2015 as good candidates for a pancreatic cancer screening tool, alkaline phosphatase (ALP) and CA 19-9. We found that there is a steady rise in the values of both tests in the 2 years prior to the 2016 diagnosis of pancreatic cancer in our patient cohort (N = 11,541). We note that an ALP test was available for the majority of our cohort (65.6%), but fewer patients in the cohort were tested for CA19-9 in the 2 years prior to diagnosis (21.2%). Alkaline phosphatase values rose steadily until about 200 days prior to diagnosis, when they rose more steeply; in contrast, CA19-9 rose steadily from about day 400 prior to day 200, then after a short dip, rose steeply from day 100 to diagnosis. It is interesting that the standard error for ALP is narrow, as shown the graph, which reflects the larger sample size.
Test results for hemoglobin and platelets were available for 54.4% and 49.1% of the patient cohort, respectively, over the 5-year period of the study. We note that hemoglobin values show a negative slope, which becomes more steep about 600 days prior to diagnosis of pancreatic cancer. In contrast, platelets show a flat slope until about 250 days prior, when they show a steep upward slope. Only 2 years of data are presented for platelets.
Up to 66% of patients with pancreatic cancer also have diabetes, and studies suggest that between 40 – 75% of these diabetes cases are new onset (less than 2 years in duration) (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3932318/pdf/nihms552557.pdf ). We note that 35% of patients in the pancreatic cancer cohort in our study were tested for HbA1c in the 5 years prior to diagnosis, and starting at the 5 year prior mark, the average HbA1c was already about 6.5%, the threshold for diabetes recommended by some expert groups. It then consistently climbed across the 5 years included in our retrospective analysis, indicating that diabetes is common already in this patient group prior to diagnosis.
To the best of our knowledge, a retrospective analysis of lab test results in patients diagnosed with pancreatic cancer has not been attempted previously. We recognize that we are in the early stages of this work, looking for a biomarker panel that might function as an effective screening panel to assess pancreatic cancer risk in the general (normal risk) population. Finding pancreatic tumors earlier in their natural history, perhaps during the window of time when they are operable and curable, would impact survival rates and the lives of patients and their families. We are optimistic that our findings, whether positive or negative, will be a valuable contribution to this global research effort. The Prognos database offers a unique datamining opportunity, with billions of tests and hundreds of millions of patients from around the US included. We welcome your comments and feedback on this work – do you have a recommendation regarding a hypothesis that we should add to our pancreatic cancer research plan? Is there a patient population that we should focus on? What control group or groups do you recommend we use for the statistical analysis? Please send us your thoughts, we look forward to hearing from you.
Search for biomarkers for pancreatic cancer:
Loess Smoothing Curve: