Endometrial cancer is the fourth most common cancer in women in the developed world, and its incidence is rising in parallel with global obesity rates. Approximately 75% of cases are caught before the cancer has spread outside the uterus, earning it the reputation of a relatively curable cancer - but this optimism depends on early detection, which in turn depends on identifying high-risk women proactively.
Multiple risk factors for endometrial cancer are well recognized: obesity, type 2 diabetes, polycystic ovary syndrome (PCOS), nulliparity (never having been pregnant), non-continuous hormone replacement therapy (HRT), and lack of contraceptive use. However, prior studies tended to examine these risk factors individually or in small subsets, making it impossible to compare their relative importance or combine them into a single risk model.
This study addressed that gap with two complementary approaches: a systematic meta-analysis to quantify and rank the pooled risk associated with each factor across all available published studies, followed by the development of an artificial neural network that integrates these quantified risks into a personalized cancer risk calculator.
The researchers systematically searched the literature and ultimately included 51 studies in their meta-analysis, covering six risk factor categories: obesity (BMI), type 2 diabetes, PCOS, parity (pregnancy history), hormone replacement therapy (HRT), and contraceptive use. Only studies with clear controls and data from 2003 onwards were included, ensuring the analysis reflects contemporary clinical populations.
Data were expressed as relative risk - a ratio comparing cancer rates in people with the risk factor to those without. Forest plots were used to visualize and statistically combine results across all included studies, accounting for variation between studies. From these pooled relative risks, the team calculated percentage risk contributions - translating abstract statistical ratios into intuitive numbers expressing how much each factor increases or decreases an individual's lifetime cancer risk.
The baseline lifetime endometrial cancer risk for a woman in the UK is approximately 2.8% - meaning about 1 in 36 women will develop this cancer in her lifetime. The risk percentages calculated in this study reflect how much each factor shifts that baseline, allowing direct comparison of factor magnitudes and enabling the neural network to translate factor combinations into an individual's estimated risk.
The meta-analysis confirmed that obesity is overwhelmingly the dominant risk factor for endometrial cancer, and that the risk escalates with increasing BMI. Having a BMI between 25 and 30 (overweight) adds approximately 2.01% to lifetime risk. A BMI of 30 or above (obese) adds 5.24%, and a BMI above 40 (morbidly obese) adds 6.9% - more than doubling the baseline lifetime risk of 2.8%.
PCOS (polycystic ovary syndrome) was the second largest risk factor, adding 4.2% to lifetime risk. PCOS is characterized by elevated androgen and estrogen levels with reduced progesterone - a hormonal environment that promotes endometrial proliferation without the protective counterbalancing effect of progesterone during the normal menstrual cycle.
Type 2 diabetes added 1.54%, never having been pregnant (nulliparity) added 1.2%, and non-continuous HRT added 0.56% to risk. On the protective side, use of an intrauterine device (IUD) reduced lifetime risk by 1.34%, oral contraceptive use reduced it by 0.9%, continuous (combined) HRT reduced it by 0.75%, and having given birth at least once reduced it by 0.9%.
The artificial neural network was built to translate a patient's specific combination of risk factor values into a single percentage risk estimate and a binary prediction of whether she is at high risk for endometrial cancer. The network was trained on data from the National Cancer Institute (NCI) database - 1,200 patient records maintaining a realistic 95%/5% ratio of negative to positive diagnoses.
The network architecture used two hidden layers (4 neurons in the first, 2 in the second) - a design tested to balance complexity against overfitting. The six risk factors (age category, BMI, diabetes, contraceptive use, HRT type, and parity) were used as inputs. The network was implemented as a web application, making it accessible to clinicians without technical expertise.
Training a neural network on imbalanced data - where only 5% of examples are positive cases - is challenging. The network can achieve high overall accuracy by learning to predict nearly everyone as negative. The researchers addressed this by separately evaluating sensitivity (ability to correctly identify cancer cases) and specificity (ability to correctly identify non-cancer cases), rather than relying solely on overall accuracy.
Using the NCI test dataset, the neural network achieved an overall accuracy of 98.6% - an impressive headline figure. Specificity (correctly identifying cancer-free women) was approximately 98.78%. However, sensitivity (correctly identifying women who do have cancer) was only about 75% - meaning the model missed approximately 1 in 4 cancer cases in the test dataset.
The high overall accuracy reflects the data imbalance: since only 5% of the test set had a positive cancer diagnosis, a model that predicted everyone as negative would be 95% accurate without detecting a single cancer. The true measure of the model's value is its sensitivity, which at 75% is meaningful but leaves room for improvement - particularly as more training data with positive diagnoses becomes available.
Validation on 40 blind patients from a UK hospital showed that patients who developed endometrial cancer had a median modeled risk percentage of 6.9%, compared to 4% for those without cancer. The scatter plot of BMI against modeled risk showed a strong positive correlation - confirming that the model's risk estimates align with known biology and independently measured clinical data.
The neural network produces two outputs: a percentage risk estimate and a binary high/low risk prediction. For women with a high predicted risk, the system suggests proceeding to more detailed clinical investigations - potentially including transvaginal ultrasound to assess endometrial thickness. For low-risk women, unnecessary investigations are avoided.
Uniquely, the model can also suggest personalized prevention strategies based on which risk factors are driving the elevated estimate. A woman with elevated risk primarily due to BMI would receive guidance about weight loss; a woman with elevated risk due to non-continuous HRT might be advised to discuss switching to a continuous combined formulation with her physician. This tailored feedback transforms a risk calculator into a prevention support tool.
The tool's focus on modifiable risk factors - obesity, contraception choice, HRT type - is particularly clinically meaningful. Unlike genetic risk factors, these can be changed. A model that clearly communicates the quantitative impact of lifestyle and medical choices on cancer risk could motivate behavior change in ways that general health advice cannot.
The study provides a detailed biological explanation for why obesity ranks above all other risk factors. Adipose (fat) tissue converts androgens into estrogen using the enzyme aromatase, and converts weaker estrogens into the potent estradiol using 17-beta-hydroxysteroid dehydrogenase. In obese women, this peripheral estrogen production is amplified, creating chronically elevated estrogen levels without the opposing protective effect of progesterone.
Obesity also causes insulin resistance, elevating blood insulin levels. Insulin and insulin-like growth factor 1 (IGF-1) directly stimulate endometrial cell proliferation and inhibit programmed cell death (apoptosis). High insulin also breaks down a protein (IGFBP-3) that normally sequesters IGF-1, further amplifying growth-promoting signals in the endometrium.
Finally, obesity triggers chronic low-grade inflammation, with fat tissue producing inflammatory proteins (cytokines, adipokines) that damage DNA, inhibit apoptosis, stimulate aromatase activity, and create conditions favorable to tumor formation. This triple mechanism - excess estrogen, excess insulin/IGF-1, and chronic inflammation - explains why obesity's impact on endometrial cancer risk so greatly exceeds that of other individual risk factors.