Exploring the complex and paradoxical relationships between Type 2 Diabetes, obesity and cancer, as a framework for risk prediction models

  • Ellena Badrick

Student thesis: Phd


Background: T2D is associated with increased cancer incidence, especially obesity-related cancers. The increased ‘risk’ has two parts – cancer occurrence associated with T2D diagnosis and a longer term risk with respective opportunities for: (i) early cancer detection; and (ii) long-term prediction risk modelling. During the scoping of my thesis, in order to inform the later prediction modelling, it was important to understand the relationships between T2D, obesity and mortality, and the effect modification by smoking status. Methods: I used two electronic health record datasets, Salford Integrated Records, and CPRD, both linked with the National Cancer Intelligence Service (1995-2010). I defined people with T2D and extracted data on BMI, weight change, smoking, medications, and comorbidities; and coded for obesity-related versus non-obesity-related cancers. I employed logistic regression models for early cancer detection and survival analysis using Cox models for longer term risk, quantifying performance using c-statistic and calculated 5- and 10-year absolute risks, %, a threshold above which enhanced cancer screening might be triggered. I performed internal cross-validation and sensitivity analysis using competing risk models. Results 1 (Chapter 3): A matched cohort study 10,464 people with incident T2D paired (1:3) with 31,020 individuals without T2D. HRs for associations of BMI with all-cause mortality using Cox models, evidence of the obesity paradox was present in ever smokers, with and without T2D. In a cohort of 241,004 people newly diagnosed with T2D Cox models of cancer risk were developed. As BMI increased so did cancer risk, with maximum BMI prior to T2D a better measure of obesity-related cancer risk. There was evidence of the obesity paradox for mortality risk but not cancer risk. Results 2 (Chapter 4): In 191,435 new onset T2D from CPRD, I quantified the co-diagnosis of cancers, CHD and hypothyroidism within first 6 months, and found similar non-specific co-occurrence patterns rather than causal associations between T2D and concurrently diagnosed cancers. I specifically focused on the clinical utility of new onset T2D and diagnosis of pancreatic adenocarcinoma but (against common perception) found disappointing Positive Predictive Values, even when accounting for age, weight loss, smoking and BMI. Results 3 (Chapter 5): I constructed gender-specific prediction models, including multiple imputations, for late occurrence of obesity-related cancers For men c-statistic was 0.7279 with 35.1% of men identified as having a greater than 1 in 10 risk, and disappointingly low proportions of women identified (17.1%) with poor model performance c-statistic 0.6637. Conclusion: Novel findings in the thesis were: an obesity paradox between BMI and mortality (but not cancer risk) in ever smoker T2D and non-T2D populations; non-specific associations between T2D and cancer co-diagnosis; and models with poor clinical utility for long-term cancer risks prediction. Notable methodological achievements included: methods to address ‘detection-time bias’; clinically useful classification into obesity-related versus non-obesity-related cancers; and multiple imputations in the prediction models.
Date of Award1 Aug 2018
Original languageEnglish
Awarding Institution
  • The University of Manchester
SupervisorAndrew Renehan (Supervisor) & Matthew Sperrin (Supervisor)


  • Epidemiology
  • General Practice
  • Risk Prediction Model
  • Early Detection
  • Cancer
  • Type 2 Diabetes

Cite this