Development and Validation of a Prediction Model to Estimate Individual Risk of Pancreatic Cancer

Machine Learning Models for Pancreatic Cancer Risk Prediction Using Electronic Health Record Data-A Systematic Review and Assessment

INTRODUCTION

Accurate risk prediction can facilitate screening and early detection of pancreatic cancer (PC). We conducted a systematic review to critically evaluate effectiveness of machine learning (ML) and artificial intelligence (AI) techniques applied to electronic health records (EHR) for PC risk prediction.

METHODS

Ovid MEDLINE(R), Ovid EMBASE, Ovid Cochrane Central Register of Controlled Trials, Ovid Cochrane Database of Systematic Reviews, Scopus, and Web of Science were searched for articles that utilized ML/AI techniques to predict PC, published between January 1, 2012, and February 1, 2024. Study selection and data extraction were conducted by 2 independent reviewers. Critical appraisal and data extraction were performed using the CHecklist for critical Appraisal and data extraction for systematic Reviews of prediction Modelling Studies checklist. Risk of bias and applicability were examined using prediction model risk of bias assessment tool.

RESULTS

Thirty studies including 169,149 PC cases were identified. Logistic regression was the most frequent modeling method. Twenty studies utilized a curated set of known PC risk predictors or those identified by clinical experts. ML model discrimination performance (C-index) ranged from 0.57 to 1.0. Missing data were underreported, and most studies did not implement explainable-AI techniques or report exclusion time intervals.

DISCUSSION

AI/ML models for PC risk prediction using known risk factors perform reasonably well and may have near-term applications in identifying cohorts for targeted PC screening if validated in real-world data sets. The combined use of structured and unstructured EHR data using emerging AI models while incorporating explainable-AI techniques has the potential to identify novel PC risk factors, and this approach merits further study.

An Integrative Pancreatic Cancer Risk Prediction Model in the UK Biobank

Pancreatic cancer (PaCa) is a lethal cancer with an increasing incidence, highlighting the need for early prevention strategies. There is a lack of a comprehensive PaCa predictive model derived from large prospective cohorts. Therefore, we have developed an integrated PaCa risk prediction model for PaCa using data from the UK Biobank, incorporating lifestyle-related, genetic-related, and medical history-related variables for application in healthcare settings. We used a machine learning-based random forest approach and a traditional multivariable logistic regression method to develop a PaCa predictive model for different purposes. Additionally, we employed dynamic nomograms to visualize the probability of PaCa risk in the prediction model. The top five influential features in the random forest model were age, PRS, pancreatitis, DM, and smoking. The significant risk variables in the logistic regression model included male gender (OR = 1.17), age (OR = 1.10), non-O blood type (OR = 1.29), higher polygenic score (PRS) (Q5 vs. Q1, OR = 2.03), smoking (OR = 1.82), alcohol consumption (OR = 1.27), pancreatitis (OR = 3.99), diabetes (DM) (OR = 2.57), and gallbladder-related disease (OR = 2.07). The area under the receiver operating curve (AUC) of the logistic regression model is 0.78. Internal validation and calibration performed well in both models. Our integrative PaCa risk prediction model with the PRS effectively stratifies individuals at future risk of PaCa, aiding targeted prevention efforts and supporting community-based cancer prevention initiatives.

Derivation and External Validation of Machine Learning-Based Model for Detection of Pancreatic Cancer

INTRODUCTION

There is currently no widely accepted approach to screening for pancreatic cancer (PC). We aimed to develop and validate a risk prediction model for pancreatic ductal adenocarcinoma (PDAC), the most common form of PC, across 2 health systems using electronic health records.

METHODS

This retrospective cohort study consisted of patients aged 50-84 years having at least 1 clinic-based visit over a 10-year study period at Kaiser Permanente Southern California (model training, internal validation) and the Veterans Affairs (VA, external testing). Random survival forests models were built to identify the most relevant predictors from >500 variables and to predict risk of PDAC within 18 months of cohort entry.

RESULTS

The Kaiser Permanente Southern California cohort consisted of 1.8 million patients (mean age 61.6) with 1,792 PDAC cases. The 18-month incidence rate of PDAC was 0.77 (95% confidence interval 0.73-0.80)/1,000 person-years. The final main model contained age, abdominal pain, weight change, HbA1c, and alanine transaminase change (c-index: mean = 0.77, SD = 0.02; calibration test: P value 0.4, SD 0.3). The final early detection model comprised the same features as those selected by the main model except for abdominal pain (c-index: 0.77 and SD 0.4; calibration test: P value 0.3 and SD 0.3). The VA testing cohort consisted of 2.7 million patients (mean age 66.1) with an 18-month incidence rate of 1.27 (1.23-1.30)/1,000 person-years. The recalibrated main and early detection models based on VA testing data sets achieved a mean c-index of 0.71 (SD 0.002) and 0.68 (SD 0.003), respectively.

DISCUSSION

Using widely available parameters in electronic health records, we developed and externally validated parsimonious machine learning-based models for detection of PC. These models may be suitable for real-time clinical application.

Clinical Prediction Models for Pancreatic Cancer in General and At-Risk Populations: A Systematic Review

INTRODUCTION

Identifying high-risk individuals using a risk prediction model could be a crucial first stage of screening pathways to improve the early detection of pancreatic cancer. A systematic review was conducted to critically evaluate the published primary literature on the development or validation of clinical risk prediction models for pancreatic cancer risk.

METHODS

MEDLINE, Embase, and Web of Science were searched for relevant articles from the inception of each database up to November 2021. Study selection and data extraction were conducted by 2 independent reviewers. The Prediction model Risk Of Bias Assessment Tool (PROBAST) was applied to assess risk of bias.

RESULTS

In total, 33 studies were included, describing 38 risk prediction models. Excluding studies with an overlapping population, this study consist of 15,848,100 participants, of which 58,313 were diagnosed with pancreatic cancer. Eight studies externally validated their model, and 13 performed internal validation. The studies described risk prediction models for pancreatic cancer in the general population (n = 14), patients with diabetes (n = 8), and individuals with gastrointestinal (and other) symptoms (symptoms included abdominal pain, unexplained weight loss, jaundice, and change in bowel habits and indigestion; n = 11). The commonly used clinical risk factors in the model were cigarette smoking (n = 27), age (n = 25), diabetes history (n = 22), chronic pancreatitis (n = 18), and body mass index (n = 14). In the 25 studies that assessed model performance, C-statistics ranged from 0.61 to 0.98. Of the 33 studies included, 6 were rated as being at a low risk of bias based on PROBAST.

DISCUSSION

Many clinical risk prediction models for pancreatic cancer had been developed for different target populations. Although low risk-of-bias studies were identified, these require external validation and implementation studies to ensure that these will benefit clinical decision making.

Phenotype risk scores (PheRS) for pancreatic cancer using time-stamped electronic health record data: Discovery and validation in two large biobanks

BACKGROUND

Traditional methods for disease risk prediction and assessment, such as diagnostic tests using serum, urine, blood, saliva or imaging biomarkers, have been important for identifying high-risk individuals for many diseases, leading to early detection and improved survival. For pancreatic cancer, traditional methods for screening have been largely unsuccessful in identifying high-risk individuals in advance of disease progression leading to high mortality and poor survival. Electronic health records (EHR) linked to genetic profiles provide an opportunity to integrate multiple sources of patient information for risk prediction and stratification. We leverage a constellation of temporally associated diagnoses available in the EHR to construct a summary risk score, called a phenotype risk score (PheRS), for identifying individuals at high-risk for having pancreatic cancer. The proposed PheRS approach incorporates the time with respect to disease onset into the prediction framework. We combine and contrast the PheRS with more well-known measures of inherited susceptibility, namely, the polygenic risk scores (PRS) for prediction of pancreatic cancer.

METHODOLOGY

We first calculated pairwise, unadjusted associations between pancreatic cancer diagnosis and all possible other diagnoses across the medical phenome. We call these pairwise associations co-occurrences. After accounting for cross-phenotype correlations, the multivariable association estimates from a subset of relatively independent diagnoses were used to create a weighted sum PheRS. We constructed time-restricted risk scores using data from 38,359 participants in the Michigan Genomics Initiative (MGI) based on the diagnoses contained in the EHR at 0, 1, 2, and 5 years prior to the target pancreatic cancer diagnosis. The PheRS was assessed for predictability in the UK Biobank (UKB). We tested the relative contribution of PheRS when added to a model containing a summary measure of inherited genetic susceptibility (PRS) plus other covariates like age, sex, smoking status, drinking status, and body mass index (BMI).

RESULTS

Our exploration of co-occurrence patterns identified expected associations while also revealing unexpected relationships that may warrant closer attention. Solely using the pancreatic cancer PheRS at 5 years before the target diagnoses yielded an AUC of 0.60 (95% CI = [0.58, 0.62]) in UKB. A larger predictive model including PheRS, PRS, and the covariates at the 5-year threshold achieved an AUC of 0.74 (95% CI = [0.72, 0.76]) in UKB. We note that PheRS does contribute independently in the joint model. Finally, scores at the top percentiles of the PheRS distribution demonstrated promise in terms of risk stratification. Scores in the top 2% were 10.20 (95% CI = [9.34, 12.99]) times more likely to identify cases than those in the bottom 98% in UKB at the 5-year threshold prior to pancreatic cancer diagnosis.

CONCLUSIONS

We developed a framework for creating a time-restricted PheRS from EHR data for pancreatic cancer using the rich information content of a medical phenome. In addition to identifying hypothesis-generating associations for future research, this PheRS demonstrates a potentially important contribution in identifying high-risk individuals, even after adjusting for PRS for pancreatic cancer and other traditional epidemiologic covariates. The methods are generalizable to other phenotypic traits.

Early detection of pancreatic cancer

Pancreatic ductal adenocarcinoma is most frequently detected at an advanced stage. Such late detection restricts treatment options and contributes to a dismal 5-year survival rate of 3-15%. Pancreatic ductal adenocarcinoma is relatively uncommon and screening of the asymptomatic adult population is not feasible or recommended with current modalities. However, screening of individuals in high-risk groups is recommended. Here, we review groups at high risk for pancreatic ductal adenocarcinoma, including individuals with inherited predisposition and patients with pancreatic cystic lesions. We discuss studies aimed at finding ways of identifying pancreatic ductal adenocarcinoma in high-risk groups, such as among individuals with new-onset diabetes mellitus and people attending primary and secondary care practices with symptoms that suggest this cancer. We review early detection biomarkers, explore the potential of using social media for detection, appraise prediction models developed using electronic health records and research data, and examine the application of artificial intelligence to medical imaging for the purposes of early detection.

Prediction of Pancreatic Cancer Based on Imaging Features in Patients With Duct Abnormalities

OBJECTIVES

Abnormalities of the main pancreatic duct may be an early indicator of pancreatic ductal adenocarcinoma (PDAC). We develop and validate algorithms that predict the risk of PDAC using features identified on cross-sectional imaging and other clinical characteristics collected through electronic medical records.

METHODS

Adult patients with abdominal computed tomography or magnetic resonance imaging in January 2006 to June 2016 demonstrating dilatation of main pancreatic duct were identified. Pancreas-related morphologic features were extracted from radiology reports using natural language processing. The cumulative incidence of PDAC with death as a competing risk was estimated using multistate models. Model discrimination was assessed using c-index. The models were internally validated using bootstrapping.

RESULTS

The cohort consisted of 7819 patients (mean age, 71 years; 65% female). A total of 781 patients (10%) developed PDAC within 3 years after the first eligible imaging study. The final models achieved reasonable discrimination (c-index, 0.825-0.833). The 3-year average risk of PDAC in the top 5% of the total eligible patients was 56.0%, more than 20 times of the average risk among the bottom 50% of patients.

CONCLUSIONS

Prediction models combining imaging features and clinical measures can be used to further stratify the risk of pancreatic cancer among patients with pancreas ductal dilatation.

Development of a stratification tool to identify pancreatic intraductal papillary mucinous neoplasms at lowest risk of progression

BACKGROUND

Because most pancreatic intraductal papillary mucinous neoplasms (IPMNs) will never become malignant, currently advocated long-term surveillance is low-yield for most individuals.

AIM

To develop a score chart identifying IPMNs at lowest risk of developing worrisome features or high-risk stigmata.

METHODS

We combined prospectively maintained pancreatic cyst surveillance databases of three academic institutions. Patients were included if they had a presumed side-branch IPMN, without worrisome features or high-risk stigmata at baseline (as defined by the 2012 international Fukuoka guidelines), and were followed ≥ 12 months. The endpoint was development of one or more worrisome features or high-risk stigmata during follow-up. We created a multivariable prediction model using Cox-proportional logistic regression analysis and performed an internal-external validation.

RESULTS

875 patients were included. After a mean follow-up of 50 months (range 12-157), 116 (13%) patients developed worrisome features or high-risk stigmata. The final model included cyst size (HR 1.12, 95% CI 1.09-1.15), cyst multifocality (HR 1.49, 95% CI 1.01-2.18), ever having smoked (HR 1.40, 95% CI 0.95-2.04), history of acute pancreatitis (HR 2.07, 95% CI 1.21-3.55), and history of extrapancreatic malignancy (HR 1.34, 95% CI 0.91-1.97). After validation, the model had good discriminative ability (C-statistic 0.72 in the Mayo cohort, 0.71 in the Columbia cohort, 0.64 in the Erasmus cohort).

CONCLUSION

In presumed side branch IPMNs without worrisome features or high-risk stigmata at baseline, the Dutch-American Risk stratification Tool (DART-1) successfully identifies pancreatic lesions at low risk of developing worrisome features or high-risk stigmata.

Environmental Risk Factors of Pancreatic Cancer

Despite the advancement in medical knowledge that has improved the survival rate of many cancers, the survival rate of pancreatic cancer has remained dismal with a five-year survival rate of only 9%. The poor survival of pancreatic cancer emphasizes the urgent need to identify the causes or the risk factors of pancreatic cancer in order to establish effective preventive strategies. This review summarizes the current evidence regarding the environmental (non-genetic, including lifestyle, and clinical factors) risk factors of pancreatic cancer. Based on the current evidence, the established risk factors of pancreatic cancer are cigarette smoking, chronic diabetes, and obesity. Other strong risk factors include low consumption of fruits and vegetables, excess consumption of alcohol, poor oral hygiene, and the lack of allergy history. In the future, more studies are needed to identify additional risk factors of pancreatic cancer, especially the modifiable risk factors that could be included in a public health campaign to educate the public in order to reduce the incidence of pancreatic cancer.

Do changes in health reveal the possibility of undiagnosed pancreatic cancer? Development of a risk-prediction model based on healthcare claims data

Background and objective

Early detection methods for pancreatic cancer are lacking. We aimed to develop a prediction model for pancreatic cancer based on changes in health captured by healthcare claims data.

Methods

We conducted a case-control study on 29,646 Medicare-enrolled patients aged 68 years and above with pancreatic ductal adenocarcinoma (PDAC) reported to the Surveillance Epidemiology an End Results (SEER) tumor registries program in 2004–2011 and 88,938 age and sex-matched controls. We developed a prediction model using multivariable logistic regression on Medicare claims for 16 risk factors and pre-diagnostic symptoms of PDAC present within 15 months prior to PDAC diagnosis. Claims within 3 months of PDAC diagnosis were excluded in sensitivity analyses. We evaluated the discriminatory power of the model with the area under the receiver operating curve (AUC) and performed cross-validation by bootstrapping.

Results

The prediction model on all cases and controls reached AUC of 0.68. Excluding the final 3 months of claims lowered the AUC to 0.58. Among new-onset diabetes patients, the prediction model reached AUC of 0.73, which decreased to 0.63 when claims from the final 3 months were excluded. Performance measures of the prediction models was confirmed by internal validation using the bootstrap method.

Conclusion

Models based on healthcare claims for clinical risk factors, symptoms and signs of pancreatic cancer are limited in classifying those who go on to diagnosis of pancreatic cancer and those who do not, especially when excluding claims that immediately precede the diagnosis of PDAC.

Pancreatic Cancer Prediction Through an Artificial Neural Network

Early detection of pancreatic cancer is challenging because cancer-specific symptoms occur only at an advanced stage, and a reliable screening tool to identify high-risk patients is lacking. To address this challenge, an artificial neural network (ANN) was developed, trained, and tested using the health data of 800,114 respondents captured in the National Health Interview Survey (NHIS) and Pancreatic, Lung, Colorectal, and Ovarian cancer (PLCO) datasets, together containing 898 patients diagnosed with pancreatic cancer. Prediction of pancreatic cancer risk was assessed at an individual level by incorporating 18 features into the neural network. The established ANN model achieved a sensitivity of 87.3 and 80.7%, a specificity of 80.8 and 80.7%, and an area under the receiver operating characteristic curve of 0.86 and 0.85 for the training and testing cohorts, respectively. These results indicate that our ANN can be used to predict pancreatic cancer risk with high discriminatory power and may provide a novel approach to identify patients at higher risk for pancreatic cancer who may benefit from more tailored screening and intervention.

A review of lifestyle, metabolic risk factors, and blood-based biomarkers for early diagnosis of pancreatic ductal adenocarcinoma

We aimed to review the epidemiologic literature examining lifestyle and metabolic risk factors, and blood-based biomarkers including multi-omics (genomics, proteomics, and metabolomics) and to discuss how these predictive markers can inform early diagnosis of pancreatic ductal adenocarcinoma (PDAC). A search of the PubMed database was conducted in June 2018 to review epidemiologic studies of (i) lifestyle and metabolic risk factors for PDAC, genome-wide association studies, and risk prediction models incorporating these factors and (ii) blood-based biomarkers for PDAC (conventional diagnostic markers, metabolomics, and proteomics). Prospective cohort studies have reported at least 20 possible risk factors for PDAC, including smoking, heavy alcohol drinking, adiposity, diabetes, and pancreatitis, but the relative risks and population attributable fractions of individual risk factors are small (mostly < 10%). High-throughput technologies have continued to yield promising genetic, metabolic, and protein biomarkers in addition to conventional biomarkers such as carbohydrate antigen 19-9. Nonetheless, most studies have utilized a hospital-based case-control design, and the diagnostic accuracy is low in studies that collected pre-diagnostic samples. Risk prediction models incorporating lifestyle and metabolic factors as well as other clinical parameters have shown good discrimination and calibration. Combination of traditional risk factors, genomics, and blood-based biomarkers can help identify high-risk populations and inform clinical decisions. Multi-omics investigations can provide valuable insights into disease etiology, but prospective cohort studies that collect pre-diagnostic samples and validation in independent studies are warranted.

Improving risk prediction for pancreatic cancer in symptomatic patients: a Saudi Arabian study

Background

Imaging tests used in our center are usually inadequate to confirm the high risk for pancreatic cancer. We aimed to use a combination of potential predictors including imaging tests to quantify the risk of pancreatic cancer and evaluate its utility.

Methods

This was a retrospective cohort study of patients who were suspected as having pancreatic cancer and underwent biopsy examination of pancreatic mass at King Abdulaziz Medical City, Riyadh, Saudi Arabia, between January 1, 2013, and December 31, 2016. We retrieved data on demographics, clinical history, imaging tests, and final pancreatic diagnosis from medical records.

Results

Of the 206 who underwent pancreatic biopsies, the mean age was 63.6 years; 54.4% were male. Of all the biopsies, 57.8% were malignant and 42.2% were benign masses. Nine factors contributed significantly to the risk of pancreatic cancer and were noted: older age (adjusted odds ratio [aOR] =1.048; P=0.010), male gender (aOR =4.670; P=0.008), weight loss (aOR =14.810; P=0.001), abdominal pain (aOR =7.053; P=0.0.001), blood clots (aOR =20.787; P=0.014), pancreatitis (aOR =4.473; P=0.021), jaundice (aOR =7.446; P=0.003), persistent fatigue (aOR =22.015; P=0.015), and abnormal imaging tests (aOR =67.124; P=0.001). The model yielded powerful calibration (P=0.953), excellent predictive utility (area under the receiver operating characteristic curve 96.3%; 95% CI =94.1, 98.6), with optimism-corrected area under the curve bootstrap resampling of 94.9%. An optimal cut-off risk probability of 0.513 yielded a sensitivity of 94% and specificity of 84.7% for risk classification.

Conclusion

The study developed and validated a risk model for quantifying the risk of pancreatic cancer. Nine characteristics were associated with increased risk of pancreatic cancer. This risk assessment model is feasible and highly sensitive and could be useful to improve screening performance and the decision-making process in clinical settings in Saudi Arabia.

Prediction model for pancreatic cancer risk in the general Japanese population

Genome-wide association studies (GWASs) have identified many single nucleotide polymorphisms (SNPs) that are significantly associated with pancreatic cancer susceptibility. We sought to replicate the associations of 61 GWAS-identified SNPs at 42 loci with pancreatic cancer in Japanese and to develop a risk model for the identification of individuals at high risk for pancreatic cancer development in the general Japanese population. The model was based on data including directly determined or imputed SNP genotypes for 664 pancreatic cancer case and 664 age- and sex-matched control subjects. Stepwise logistic regression uncovered five GWAS-identified SNPs at five loci that also showed significant associations in our case-control cohort. These five SNPs were included in the risk model and also applied to calculation of the polygenic risk score (PRS). The area under the curve determined with the leave-one-out cross-validation method was 0.63 (95% confidence interval, 0.60–0.66) or 0.61 (0.58–0.64) for versions of the model that did or did not include cigarette smoking and family history of pancreatic cancer in addition to the five SNPs, respectively. Individuals in the lowest and highest quintiles for the PRS had odds ratios of 0.62 (0.42–0.91) and 1.98 (1.42–2.76), respectively, for pancreatic cancer development compared with those in the middle quintile. We have thus developed a risk model for pancreatic cancer that showed moderately good discriminatory ability with regard to differentiation of pancreatic cancer patients from control individuals. Our findings suggest the potential utility of a risk model that incorporates replicated GWAS-identified SNPs and established demographic or environmental factors for the identification of individuals at increased risk for pancreatic cancer development.