Observational studies consistently show positive associations between type 2 diabetes mellitus (T2DM) and a range of cancers linked to excess adiposity, commonly termed obesity-related cancers (ORCs). However, interpretation is complicated by shared risk factors—most notably body mass index (BMI)—and biases including immortal time bias and detection-time bias. The authors aimed to use causal inference methods within UK Biobank to minimise these sources of confounding and bias and to estimate possible BMI-independent associations between new-onset T2DM and incident cancer.
The analysis used a matched cohort design nested within UK Biobank. Participants with new-onset T2DM (n = 23,771) were matched to unexposed participants (n = 71,170) at a 1:3 ratio on BMI, age, and sex. Matching and a sequential longitudinal approach were implemented to align exposure timing and reduce immortal time bias. The study team reported that matching specifically accounted for BMI to address the major shared risk factor for both T2DM and many cancers.
Primary outcomes were total incident cancer, categorised into obesity-related cancers (ORCs) and non-obesity-related cancers (NORCs). Secondary analyses examined site-specific cancer incidence. Median follow-up time across matched participants was 5 years. Across the cohort there were 7,694 incident cancers, of which 2,432 occurred in the T2DM group and 5,262 in the unexposed group.
The investigators developed Cox proportional hazards models using a time-split approach to estimate time-split hazard ratios (tsHRs) with 95% confidence intervals, and analyses were stratified by sex. The methodological strategy was chosen to reduce confounding by BMI and to mitigate immortal time bias; the report also assessed and described detection-time bias.
Overall, the study identified an association between new-onset T2DM and increased incidence of obesity-related cancers independent of matched BMI. There was no observed effect of new-onset T2DM on incidence of NORCs in the matched analyses.
In men, the matched, time-split analyses produced evidence for an effect of new-onset T2DM on obesity-related cancers with a tsHR of 1.39 (95% CI 1.21–1.59). Site-specific associations in men were particularly notable for hepatocellular carcinoma (tsHR 3.97, 95% CI 2.38–6.65), pancreatic cancer (tsHR 1.77, 95% CI 1.15–2.72), and kidney cancer (tsHR 1.62, 95% CI 1.13–2.32).
In women, there was evidence of an effect of new-onset T2DM on obesity-related cancers overall (tsHR 1.33, 95% CI 1.16–1.52). The authors report that the specific cancers driving this association in women were less clearly defined in their results. Importantly, the analysis did not find associations between new-onset T2DM and post-menopausal breast cancer or endometrial cancer, two cancers that are consistently associated with elevated BMI.
The study explicitly addressed common biases in observational cancer risk research. Matching on BMI and using a sequential longitudinal matching design were intended to reduce confounding and immortal time bias. The authors also report evidence of detection-time bias, particularly among men, which they considered in interpreting the temporal relationship between diabetes diagnosis and cancer detection.
The investigators conclude this is the first large-scale study to present evidence for BMI-independent associations between new-onset T2DM and incident obesity-related cancers. In men, the association appears largely driven by hepatocellular carcinoma, pancreatic cancer, and kidney cancer. In women, an overall association with ORCs was observed but the report did not clearly identify the specific cancer sites responsible. No effect of new-onset T2DM was observed for NORCs, and established BMI-linked cancers such as post-menopausal breast and endometrial cancer were not associated with new-onset T2DM in this matched analysis.
The findings suggest that mechanisms beyond shared adiposity may contribute to increased ORC incidence after a new diagnosis of T2DM, though the authors note evidence of detection-time bias and emphasise caution in causal interpretation.
The study was funded by Cancer Research UK and administered through the Manchester Cancer Research Centre MB-PhD scheme, with additional funder declarations for the UKRI AI programme and the Engineering and Physical Sciences Research Council for a related hub. The authors declared no competing interests. Ethics approvals and participant consent were confirmed. The underlying UK Biobank data used in this analysis are available to researchers who apply to use the resource via the UK Biobank access process described by the authors.