The following is the summary of “Risk Prediction of Pancreatic Cancer in Patients With Recent-onset Hyperglycemia” published in the January 2023 issue of Clinical gastroenterology by Chen, et al.
It has been proposed that new-onset diabetes (NOD) can serve as an early warning sign of pancreatic cancer. However, according to the ADA’s definition of NOD, 2 high glycemic measurements must be present simultaneously or in quick succession. Researchers wanted to use electronic health records and a machine learning approach to predict risk in individuals with newly-diagnosed hyperglycemia. People enrolled in a health plan between the ages of 50 and 84 with a glycated hemoglobin (HbA1c) result of 6.5% or above between January 2010 and September 2018 were included in this retrospective cohort analysis of newly diagnosed cases of hyperglycemia.
About 102 possible predictors were found. To deal with missing data, ten imputation datasets were created. Risk models were created and tested using the random survival forests method. C-index, calibration plot, sensitivity, specificity, and positive predictive value were used as measures of effectiveness. The average age of the cohort members was 63.6 years, with 109,266 of them. Following a cohort of 1,000 individuals over three years, the incidence rate was 1.4% (95% CI: 1.3%-1.6%).
Most of the 50 training samples fell into one of 3 models consisting of age, weight change over a year, HbA1c, and one of 3 variables: HbA1c change over a year, HbA1c in the preceding 6 months, or HbA1c in the past 18 months. All of the c-indices fell somewhere between 0.81 and 0.82. Patients in the top 20% of estimated risks had a sensitivity of 56-61%, a specificity of 80%, and a positive predictive value of 2.5-2.6 %. Patients with pancreatic cancer may benefit from early detection by focusing on assessment at the time of recent hyperglycemia based on elevated HbA1c levels.
Source: journals.lww.com/jcge/Abstract/2023/01000/Risk_Prediction_of_Pancreatic_Cancer_in_Patients.13.aspx