Tsilidis · BMJ (Clinical research ed.) 2015 · umbrella review of meta-analyses · n=27 meta-analyses

Type 2 diabetes and cancer: umbrella review of meta-analyses of observational studies.

Cited 877 times in the scientific literature.

Level 3 - non-randomized controlled study

Umbrella review / systematic review of meta-analyses of observational studies

PubMed 25555821 · doi:10.1136/bmj.g7607 · record verified 2026-08-26

What was done

This umbrella review searched PubMed, Embase, the Cochrane Database of Systematic Reviews, and reference lists to identify meta-analyses and systematic reviews of observational studies evaluating type 2 diabetes and the risk of developing or dying from cancer. The authors evaluated summary random- and fixed-effects estimates, between-study heterogeneity ($I^2$), small-study effects, excess significance bias, and 95% prediction intervals across cancer sites.

What was found

Across 27 eligible meta-analyses covering incidence across 20 cancer sites and mortality across 7 sites, 20 meta-analyses (74%) reported statistically significant increased risks at $P \le 0.05$ under random-effects models. However, only 7 meta-analyses (26%) included over 1,000 cases, achieved $P \le 0.001$ under both fixed- and random-effects calculations, and showed no evidence of small-study effects or excess significance. Of those 7, only 6 (22%) lacked substantial heterogeneity ($I^2 \le 75\%$)—namely breast, intrahepatic cholangiocarcinoma, extrahepatic cholangiocarcinoma, colorectal, endometrial, and gallbladder cancers. The 95% prediction intervals excluded the null value for only 4 of these cancer sites: breast, intrahepatic cholangiocarcinoma, colorectal, and endometrial cancer.

Why it matters

While literature widely links type 2 diabetes to elevated risk across numerous cancer types, rigorous validity assessment shows that robust observational evidence without apparent biases supports associations in only a very select subset of cancers.

Limits

Findings rely entirely on observational studies, which remain vulnerable to residual confounding and exposure misclassification. The umbrella design evaluated published meta-analyses rather than individual participant data, inheriting any methodology or search limitations of the included reviews. The abstract does not provide specific point estimates, relative risk magnitudes, or the total number of underlying primary studies and participants.

Cited by