Pancreatic Cancer Danger In Diabetic Women
Overview
The study aimed to address the challenge of distinguishing pancreatic cancer-related diabetes from type 2 diabetes, particularly in women with new-onset diabetes. Leveraging retrospective cohort data from Australian women, the researchers developed and validated a predictive model to identify individuals at risk of pancreatic cancer within three years of diabetes diagnosis.
Among over 99,000 women aged 50 years and above with new-onset diabetes, a small proportion (0.6%) were diagnosed with pancreatic cancer within the study period. The developed risk prediction model demonstrated a reasonable performance, with an area under the receiver operating curve of 0.73. Notably, age and the severity of diabetes, characterized by changes or additions to medication shortly after diagnosis, emerged as the most influential predictors. Additionally, the use of certain medications, such as beta-blockers, acid disorder drugs, and lipid-modifying agents, also contributed to the predictive accuracy of the model.
By applying a risk threshold of 50%, the model achieved a sensitivity of 69% and a positive predictive value (PPV) of 1.3%. This doubling of the PPV from baseline suggests the potential utility of the model in enhancing the early detection of pancreatic cancer among women with new-onset diabetes. Furthermore, the study highlighted that pancreatic cancer occurred more frequently in individuals without typical risk factors for type 2 diabetes, underscoring the need for targeted surveillance strategies.
The findings underscore the significance of age and rapid progression of diabetes as key risk factors for pancreatic cancer in this population. The developed model, with its improved PPV, holds promise as an initial screening tool, particularly in identifying individuals who may benefit from further diagnostic evaluation. As new biomarkers emerge, the model could be refined and integrated into clinical practice to facilitate early detection and intervention, ultimately improving patient outcomes.
Introduction
Pancreatic cancer ranks as the seventh most common cause of cancer-related deaths worldwide, with projections indicating a shift to the second most common by 2030 in more developed regions. Early diagnosis significantly improves survival rates, yet the low lifetime risk of pancreatic cancer (1.7%) precludes population-based screening. While surveillance benefits high-risk groups, the majority of cases arise sporadically, lacking established surveillance strategies. Hence, identifying high-risk populations is imperative, with interest in cohorts with a 5% or greater lifetime risk.
In Australia, pancreatic cancer ranks third in cancer-related mortality, with an increasing burden alongside rising diabetes incidence. Understanding the bidirectional relationship between diabetes and pancreatic cancer is critical, where longstanding diabetes elevates pancreatic cancer risk, and new-onset diabetes may signify underlying pancreatic malignancy. However, distinguishing between pancreatic cancer-induced diabetes and typical type 2 diabetes remains challenging.
Studies have shown a higher risk of pancreatic cancer within three years of diabetes diagnosis, albeit still too low for routine surveillance. Risk prediction models offer promise but are limited by small case numbers and complex variables. Leveraging medication data as a proxy for underlying risk factors or direct influences on pancreatic cancer risk presents a pragmatic approach for identifying surveillance candidates.
This study aims to estimate the absolute risk of pancreatic cancer post-diabetes diagnosis and develop a predictive model within a national cohort of Australian women. By leveraging real-world data and medication records, the model seeks to identify individuals warranting pancreatic cancer surveillance, thus advancing early detection and improving outcomes in this challenging malignancy-diabetes nexus.
Method
This study utilized administrative health databases, compiled by the Australian Institute of Health and Welfare for the IMPROVE Study, aimed at investigating ovarian cancer. The IMPROVE Study encompassed all Australian women aged 18 or above, holding citizenship or permanent residency. The linked databases comprised the Australian Medicare Enrolment File, Australian Cancer Database, Australian National Death Index, and Pharmaceutical Benefits Scheme (PBS) database.
The Australian Medicare Enrolment File provided demographic details for citizens or permanent residents aged 18 and above, registered with Medicare. The Australian Cancer Database supplied data on cancer diagnoses, including date and type, from 1982 to 2013. The Australian National Death Index furnished information on date and cause of death from 2002 to 2018.
The PBS database contained data on prescription medicines dispensed under Australia’s medication-subsidy scheme. Medications dispensed from July 2002 onwards were included. Women were classified as concessional beneficiaries if they received at least one concessional prescription during a calendar year.
The study’s use of the IMPROVE dataset was ethically approved by relevant Human Research Ethics Committees. Diabetes diagnosis was established based on dispensed prescriptions of anti-diabetic medications (ADMs), utilizing anatomic therapeutic chemical (ATC) classification code class A10.
To enter the cohort, women needed at least two ADM prescriptions within six months after January 1, 2004, with no prior ADM prescriptions. The analysis was restricted to women aged 50 or older at diabetes diagnosis, and those diagnosed up to December 31, 2010. Exclusion criteria included prior pancreatic cancer diagnosis, less than one year of PBS data pre-diagnosis, and death within three years post-diabetes diagnosis without prior pancreatic cancer.
The primary outcome was pancreatic cancer diagnosis within three years post-diabetes diagnosis. The follow-up began at the diabetes diagnosis, with censoring at three years if pancreatic cancer was not diagnosed within that period.
The dataset was split into training (80%) and test (20%) subsets, stratified by pancreatic cancer case status. Predictor variables included baseline medication use, severity of diabetes, and age at diagnosis. Area-level socioeconomic status and remoteness index were considered but omitted due to missing data.
This methodological approach ensured rigorous analysis and interpretation of the relationship between diabetes and subsequent pancreatic cancer risk, leveraging comprehensive administrative health databases and robust statistical techniques.
Statistical Analysis
The study aimed to estimate the absolute risk of developing pancreatic cancer within three years after the onset of diabetes. Baseline characteristics of women with and without pancreatic cancer were compared using statistical tests, revealing significant differences in certain variables.
Logistic regression was employed to develop a prediction model, with efforts made to balance the data due to the low prevalence of pancreatic cancer. Techniques such as up-sampling were utilized to address data imbalances. Model performance was evaluated using metrics such as the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and positive predictive value (PPV). Calibration was also assessed to ensure the model’s accuracy.
The relative contributions of predictors to the final model were determined and visualized, providing insights into the key factors influencing pancreatic cancer risk prediction. To mitigate overfitting, 10-fold cross-validation was employed for internal validation.
Sensitivity analyses were conducted to assess the robustness of the findings, including redefining medication usage criteria and exploring different age groups for risk prediction models. The datasets were prepared using SAS, and prediction models were developed and validated using the “caret” package in R software.
Overall, the study adhered to the TRIPOD checklist to ensure transparent reporting of the methods used, thereby enhancing the credibility and reproducibility of the findings.
Result
The study identified 99,687 women with new-onset diabetes, of which 602 were diagnosed with pancreatic cancer within three years of follow-up, indicating an absolute risk of 0.6%. Women diagnosed with pancreatic cancer had a higher mean age at diabetes diagnosis (75.2 years) compared to those who did not develop pancreatic cancer (68.4 years). Additionally, women with pancreatic cancer exhibited more severe diabetes (32% vs. 12%) and differences in medication usage, notably acid-disorder drugs, anti-hypertensive drugs, anti-thrombotic drugs, lipid-modifying agents, cardiac drugs, and osteoporosis drugs.
Adjusted odds ratios (ORs) highlighted increased likelihoods of pancreatic cancer with advancing age, severity of diabetes, and the use of acid-disorder drugs, while the use of lipid-modifying agents and beta-blockers showed associations with a reduced risk of pancreatic cancer.
The main model’s area under the curve (AUC) was 0.73, indicating reasonable predictive ability, though it underestimated the risk for those at higher risk. Age and severity of diabetes at diagnosis were the most influential predictors of pancreatic cancer, followed by the use of beta-blockers, acid-disorder drugs, and lipid-modifying agents. The model exhibited a sensitivity of 69%, specificity of 69%, and positive predictive value (PPV) of 1.3% at a risk threshold of 50%.
Sensitivity analyses showed consistent AUC values, with notable associations between insulin treatment and pancreatic cancer. Women initiating insulin as their first drug or within 60 days of commencing any oral anti-diabetic medication had significantly higher adjusted odds ratios for pancreatic cancer compared to those who switched to or added a different oral anti-diabetic medication within the same timeframe.
Models developed for different age groups showed slightly higher AUC values, emphasizing age’s influential role in pancreatic cancer risk prediction. Overall, the study provides valuable insights into the association between diabetes severity, medication usage, and the risk of pancreatic cancer, suggesting potential avenues for improved risk assessment and management strategies.
Conclusion
The study focused on women aged 50 years or older who were newly diagnosed with diabetes and their subsequent risk of pancreatic cancer over a 3-year period. It was found that 0.6% of these women were diagnosed with pancreatic cancer during this time frame. A predictive model was developed and validated using medication data from the time of diabetes treatment initiation, showing promising discrimination with an AUC of 0.73. By setting a risk threshold of 50%, the positive predictive value (PPV) was doubled to 1.3%, although this figure remains relatively low.
While new-onset diabetes might serve as an indicator of early pancreatic cancer, the incidence of diabetes is high while the lifetime risk of pancreatic cancer is low. The study emphasized the importance of stratifying the diabetic population based on pancreatic cancer risk. Several risk prediction models, such as the Enriching New-Onset Diabetes for Pancreatic Cancer (ENDPAC) model, have been proposed, showing varying levels of accuracy and applicability. However, many of these models require complex data inputs or suffer from low sensitivity and PPV.
In contrast, the developed model in this study utilized age and medication data, which are readily accessible, yet demonstrated performance comparable to more complex models. Advanced age at diabetes diagnosis emerged as the most significant predictor of pancreatic cancer. Continued hyperglycemia despite medication treatment and relief of symptoms by acid disorder medications were identified as potential indicators of underlying pancreatic cancer. The inverse association between lipid-modifying agents and beta-blockers suggested that diabetes might indicate pancreatic cancer in individuals without typical metabolic risk factors for type 2 diabetes.
Despite the model’s ability to improve PPV, it still falls short of supporting routine imaging investigations due to its low absolute PPV. However, it could serve as an initial screening tool, particularly as new biomarkers emerge. The study emphasized the need for better risk prediction models that consider the stage of pancreatic cancer at diagnosis. Meanwhile, clinicians were advised to remain vigilant for pancreatic cancer in older diabetic patients without typical metabolic risk factors, especially if diabetes is poorly controlled or accompanied by other symptoms like abdominal pain.
Oncology Related Tools
Other
Latest Research
