Endometriosis and cancer: a systematic review and meta-analysis

Abstract

Background: Endometriosis is an often chronic, inflammatory gynaecologic condition affecting 190 million women worldwide. Studies have reported an elevated cancer risk among patients with endometriosis. However, prior research has included methodologic issues that impede valid and robust interpretation.

Objective and rationale: We conducted a meta-analysis of studies investigating the association between endometriosis and cancer risk and analysed the results by methodologic characteristics. We discuss the implications of cancer screening in patients and management challenges faced by clinicians.

Search methods: We searched PubMed and Embase databases for eligible studies from inception through 24 October 2019. We included cohort and case-control studies examining the association between endometriosis and cancer risk; cross-sectional studies and case reports were excluded. Publications had to present risk/rate/odds estimates with 95% CI. Random effects meta-analysis was used to estimate summary relative risks (SRR) and CIs. Heterogeneity across studies was assessed by the Q test and I2 statistics, and publication bias using Egger's and Begg's tests. Risk of bias and quality of the included studies were assessed using the risk of bias in non-randomized studies of interventions (ROBINS-I) tool.

Outcomes: Forty-nine population-based case-control and cohort studies were included. Twenty-six studies were scored as having a 'serious'/'critical' risk of bias, and the remaining 23 'low'/'moderate'. Cancer-specific analyses showed a positive association between endometriosis and ovarian cancer risk (SRR = 1.93, 95% CI = 1.68-2.22; n = 24 studies) that was strongest for clear cell (SRR = 3.44, 95% CI = 2.82-4.42; n = 5 studies) and endometrioid (SRR = 2.33, 95% CI = 1.82-2.98; n = 5 studies) histotypes (Pheterogeneity < 0.0001), although with significant evidence of both heterogeneity across studies and publication bias (Egger's and Begg's P-values < 0.01). A robust association was observed between endometriosis and thyroid cancer (SRR = 1.39, 95% CI =1.24-1.57; n = 5 studies), a very small association with breast cancer (SRR = 1.04, 95% CI =1.00-1.09; n = 20 studies) and no association with colorectal cancer (SRR = 1.00, 95% CI =0.87-1.16; n = 5 studies). The association with endometrial cancer was not statistically significant (SRR = 1.23, 95% CI =0.97-1.57; n = 17 studies) overall and wholly null when restricted to prospective cohort studies (SRR = 0.99, 95% CI =0.72-1.37; n = 5 studies). The association with cutaneous melanoma was also non-significant (SRR = 1.17, 95% CI =0.97-1.41; n = 7 studies) but increased in magnitude and was statistically significant when restricted to studies with low/moderate risk of bias (SRR = 1.71, 95% CI = 1.24-2.36, n = 2 studies). The most robust finding both in terms of statistical significance and magnitude of effect was an inverse association with cervical cancer (SRR = 0.68, 95% CI =0.56-0.82; n = 4 studies); however, this result has a high potential to reflect heightened access to detection of dysplasia for women who reached an endometriosis diagnosis and is thus likely not causal. Several additional cancer types were explored based on <4 studies.

Wider implications: Endometriosis was associated with a higher risk of ovarian and thyroid, and minimally (only 4% greater risk) with breast cancer, and with a lower risk of cervical cancer. However, this meta-analysis confirms that: a majority of studies had severe/critical risk of bias; there is impactful heterogeneity across studies-and for ovarian cancer, publication bias; and causal inference requires temporality, which in many studies was not considered. We discuss the implications of these potential associations from the perspectives of patients with endometriosis, clinicians involved in their care, and scientists investigating their long-term health risks.

Keywords: bias; cancer; case-control studies; cohort studies; endometrioma; endometriosis; epidemiology; methodology.