Early Detection of Pancreatic Cancer

Sushil Kumar Garg; Suresh T. Chari


Curr Opin Gastroenterol. 2020;36(5):456-461. 

In This Article

Clinical Models

Clinical models have been developed using EHR for patients with NOD to enrich this population further. Sharma et al.[24] developed a clinical model using EHR called 'Enriching New-Onset Diabetes for PC' (i.e., ENDPAC), which has three parameters include age, change in glucose, and change in weight. This model stratifies patients into three groups: low (<0.1%), intermediate (~0.5%), and very high (~4%).[24] This model was able to predict 3-year risk of development of cancer with 78% sensitivity and 80% specificity [area under the curve value (AUC) 0.87] in the high-risk group. This model was validated in a diverse and integrated healthcare setting of Kaiser Permanente Southern California.[31] The key to success to both of these models was a rigorous definition of NOD, and they used lab parameters to define NOD instead physician diagnosis of NOD. There have been several attempts by other studies to replicate these models using claims data and national databases.[32,33] Still, they have not been able to replicate the success because they relied on coded and physician diagnoses without laboratory parameters.


Biomarkers are crucial for enrichment of high-risk populations, but unfortunately, there are no validated biomarkers that are currently clinically available for early detection of pancreatic cancer. There are some recent studies on biomarkers that appear promising. Melby et al.[34] have identified a 29 serum antibody microarray panel that can differentiate stage I and II pancreatic cancer from controls receiver operating characteristic AUC of 0.96. Another recent study was done by Cohen et al.[35] using combined cell-free DNA mutations, and circulating showed sensitivities of 69–98% and a specificity of more than 99% in early pancreatic cancer diagnosis.