You are here

Development and validation of prediction models for endometrial cancer in postmenopausal bleeding

European Journal of Obstetrics & Gynecology and Reproductive Biology, August 2016, Pages 220 - 224



To develop and assess the accuracy of risk prediction models to diagnose endometrial cancer in women having postmenopausal bleeding (PMB).


A retrospective cohort study of 4383 women in a One-stop PMB clinic from a university teaching hospital in Hong Kong. Clinical risk factors, transvaginal ultrasonic measurement of endometrial thickness (ET) and endometrial histology were obtained from consecutive women between 2002 and 2013. Two models to predict risk of endometrial cancer were developed and assessed, one based on patient characteristics alone and a second incorporated ET with patient characteristics. Endometrial histology was used as the reference standard. The split-sample internal validation and bootstrapping technique were adopted. The optimal threshold for prediction of endometrial cancer by the final models was determined using a receiver–operating characteristics (ROC) curve and Youden Index. The diagnostic gain was compared to a reference strategy of measuring ET only by comparing the AUC using the Delong test.


Out of 4383 women with PMB, 168 (3.8%) were diagnosed with endometrial cancer. ET alone had an area under curve (AUC) of 0.92 (95% confidence intervals [CIs] 0.89–0.94). In the patient characteristics only model, independent predictors of cancer were age at presentation, age at menopause, body mass index, nulliparity and recurrent vaginal bleeding. The AUC and Youdens Index of the patient characteristic only model were respectively 0.73 (95% CI 0.67–0.80) and 0.72 (Sensitivity = 66.5%; Specificity = 68.9%; +ve LR = 2.14; −ve LR = 0.49). ET, age at presentation, nulliparity and recurrent vaginal bleeding were independent predictors in the patient characteristics plus ET model. The AUC and Youdens Index of the patient characteristic plus ET model where respectively 0.92 (95% CI 0.88–0.96) and 0.71 (Sensitivity = 82.7%; Specificity = 88.3%; +ve LR = 6.38; −ve LR = 0.2). Comparison of AUC indicated that a history alone model was inferior to a model using ET alone (difference = 0.19, 95% CI 0.15–0.24; p < 0.0001) and History plus ET (difference = 0.19, 95% CI 0.16–0.23, p < 0.0001) and history plus ET was similar to that of using ET alone (difference = 0.001 95% CI −0.015 to 0.0018, p = 0.84).


A risk model using only patient characteristics showed fair diagnostic accuracy. Addition of patient characteristics to ET did not improve the diagnostic accuracy as compared to ET alone in our cohort.

Keywords: Endometrial cancer, Prediction models, Postmenopausal bleeding.


Endometrial cancer is currently the most common malignancy of the female genital tract in Europe, USA, and Hong Kong. Postmenopausal bleeding (PMB) is the initial presenting symptom in majority of the women with endometrial cancer [1], [2], [3], [4], and [5]. Women with PMB therefore require assessment in order to detect the possibility of any underlying malignancy [2], [6], [7], [8], [9], and [10].

Transvaginal ultrasound scanning (TVS) or endometrial biopsy has been recommended as the first-line investigations [7], [11], [12], [13], [14], and [15]. In our previous study, we reported the sensitivities and false positive rates for detection of endometrial cancer as endometrial thickness (ET) increased [16]. Our analysis indicated that performing endometrial biopsy in women with an ET >3 mm would detect 97.0% (95% confidence interval [CI] 94.5–99.6%) of endometrial cancers but would require ≈55% of the women to still have to undergo an endometrial sampling [16].

Obesity, increasing age, hypertension, diabetes, recurrent episodes of PMB, early menarche, late menopause, nulliparity and unopposed use of exogenous estrogens have been reported to be associated with an increased risk of endometrial cancer [8], [9], [17], [18], [19], [20], and [21]. Prediction models incorporating these clinical characteristics may help to inform patients of their risk of malignancy as well as to assist clinicians in triaging patients for further investigations or urgent referral or further reduce the number of endometrial sampling that may be needed.

Probabilistic statistical classification models and risk scoring systems based either on clinical risk factors or a combination of clinical risk factors plus ET have been reported [17], [18], [19], [20], [21], and [22]. However, some of these prediction models were of relatively small sample size or when ET measurements were less than 4 mm, patients were either excluded or did not have histological verification. Women who subsequently developed endometrial cancer with ET below these levels might therefore have been under represented. Our previous study indicated that ≈6% of women presenting with PMB and with a confirmed diagnosis of endometrial cancer had an ET of 4 mm or less [16].

The aim of the present study was to develop and assess the relative performance of clinical probabilistic statistical classification models for screening of endometrial cancer using data from a large cohort of women with PMB and histological verification irrespective of the thickness of the endometrium.


We identified consecutive women who presented with PMB, vaginal bleeding at least 12 months after the last menstrual period, between 2002 and 2013 inclusive from our PMB clinic database. All patient data, including medical history, physical examination findings and ET results were prospectively collected in our computerised database at the time of consultation. Management of women was in accordance with our departmental protocol for the investigation and treatment of women presenting with PMB. This protocol has been described in detail previously [16]. In short; all women underwent transvaginal ultrasound assessment and endometrial sampling with a pipelle sampler (3 mm outer diameter; Pipet Curet, Cooper Surgical, Trumbull, CT, USA) irrespective of the ET. The ET was obtained at the mid-sagittal plane, measured from the outer borders of the anterior and posterior endometrium at the thickest part. Hysteroscopy was performed if the TVS showed an ET ≥5 mm or if the endometrium appeared to be abnormal or if the image of the ET was unsatisfactory. The final histological diagnosis was determined from the endometrial sampling, uterine curettage, hysteroscopic procedures, and hysterectomy histology results [16].

The following clinical characteristics of the women and ultrasound measurement data were extracted from the database: age at presentation (years), age at menopause (years), recurrent bleeding, parity, body mass index (BMI), use of hormonal replacement therapy, Chinese herbal medication or tamoxifen, presence of diabetes, chronic hypertension, hyperlipidaemia, history of cancer of the breast, use of anti-coagulant and ET measurement (millimeter). Use of medication and past medical history were self-reported by the women and recorded at the time of examination. Age at presentation rather than age at the first episode of PMB was used to avoid potential recall bias. Recurrent vaginal bleeding was defined as repeated episodes of bleeding separated by periods of no bleeding. Categorical variables were dichotomised with subdivisions (e.g. use of tamoxifen Yes/No) whilst continuous variables were retained as per their original scales and not categorised in order to avoid inflating the type-I error rate [23].

Three hundred and thirty five women (7.6%) had missing data for either age at menopause (n = 172, (3.9%)), parity (n = 42, (1.0%)) or BMI (n = 219, (5.0%)). Ten (3%) of the women with missing data had endometrial cancer. In order to avoid bias, complete-case analysis was not adopted for the missing cases. Women with missing BMI and age at menopause were allocated the median value of the women according to their endometrial cancer status. Cases with missing parity data were initially accommodated by including a ‘missing’ factor and then subsequently assumed to be ‘multiparous’ after a preliminary multivariate logistic regression indicated that parity ‘missing’ was not statistically significant.

A split-sample internal validation scheme was adopted to develop and assess the constructed prediction models due to the relative size of the database available [24]. The study cohort was randomly divided into the “Development” and “Validation” datasets, with two thirds allocated to “Development”. Two sets of models were created and assessed: one based on patient characteristics only and the second model which combined patient characteristics with ET.

The models were formed by firstly identifying univariate predictors of endometrial cancer from the available clinical factors in the overall study cohort. Multivariate logistic regression was then used to identify independent predictors of endometrial cancer risk factors in the ‘Development’ set. Univariate predictors were entered if their significance level was ≤0.2 whilst that significance level for remaining in the model was set at 0.05. The efficiency of the models was determined by the Nagelkerke R2 value.

The final models developed from the ‘Development’ set were applied to the “Validation” set. The patient-specific risk for endometrial cancer was calculated from the formula: risk = odds/(1 + odds) where odds =expY and ‘Y’ was determined from the final models derived from the multivariate logistic regression. The models ability to discriminate (concordance) was determined by calculating the area under curve (AUC) of the receiver operator characteristic curve (ROC) and Younden's Index was used to determine the point of optimal performance in the validation cohort [25]. Comparison of the AUC in the validation cohort was performed using Delong et al. [26].

All analyses were performed using the Statistics Package for Social Science Version 20 [SPSS, IBM, Armonk, NY, USA] and Stata version 12 SE [StataCorp, College Station, Texas, USA]. Normality of continuous variables was assessed using the Kolmogorov–Smirnoff test. A p-value <0.05 was considered to indicate statistical significance and bootstrapping techniques were used to calculate 95% CIs for prevalence, model coefficients and adjusted odds ratio based on 1000 repetitions were appropriate. The bootstrap samples were the same size as the ‘Development’ data set and were drawn with replacement. The study was approved by the Institutional Review Board, the Institutional Joint Chinese University of Hong Kong-New Territories East Cluster Clinical Research Ethics Committee (CRE-2013.450).


There were 4383 women who attended for investigation of PMB satisfied the criteria for analysis during the period between 2002 and 2013 [16]. Endometrial cancer was diagnosed in 168 (3.8%, 95% CI 3.2–4.4%) of these women. Table 1 reports the clinical characteristics in the study cohort prior to missing value imputation. Age at presentation, age at menopause, BMI, nulliparity, recurrent vaginal bleeding, hypertension, diabetes and ET were statistically significantly different between those with and without endometrial cancer.

Table 1 Univariate association between endometrial cancer and clinical characteristics at presentation.

Clinical variable/endometrial assessment Without endometrial cancer (n = 4215) With endometrial cancer (n = 168) p value
Age (years) 55 (52–62) 60 (55–69) 0.0001a
Age at menopause (years) 50 (48–52) 50 (49–53) 0.0001a
Body mass index (kg/m2) 23.8 (21.6–26.6) 25.2 (22.6–28.8) 0.0001a
 Nulliparous 188 (4.5%) 21 (12.5%) 0.0001b
 Parous 3985 (94.5%) 147 (87.5%)
Vaginal bleeding
 Single episode 2867 (68.0%) 79 (47.0%) 0.0001b
 Recurrent episode 1348 (32.0%) 89 (53.0%)
Hypertension 1205 (28.6%) 63 (37.5%) 0.012b
Diabetes 474 (11.2%) 37 (22.0%) 0.0001b
Breast cancer 244 (5.8%) 6 (3.6%) 0.306c
Hyperlipidaemia 127 (3.0%) 9 (5.4%) 0.105c
Tamoxifen use 205 (4.9%) 3 (1.8%) 0.06c
Hormone therapy use 137 (3.3%) 1 (0.6%) 0.07c
Anticoagulant use 72 (1.7%) 3 (1.8%) 0.764c
Chinese herbal medicine use 204 (4.8%) 7 (4.2%) 0.854c
Endometrial thickness (mm) 3.2 (2.3–5.2) 15.7 (9.5–23) 0.0001a

a Using Mann–Whitney analysis.

b Using Chi-Square analysis.

c Using Fisher's exact test.

Values expressed as n (%) or median and interquartile range.

There were no significant difference in covariates between those selected in the ‘Development’ (n = 2886) and ‘Validation’ (n = 1497) data sets (p > 0.05 for all). Table 2 shows the results from the univariate and multivariate analyses performed in the ‘Development’ set. Only parity (P), age at presentation (A), age at menopause (A), BMI (M) and recurrent vaginal bleeding (R) remained as significant independent predictors after correcting for other characteristics. The final model was named as “RAAMP” to reflect the significant predictors. Diabetes and hypertension status were not significant. After including scanning of endometrial thickness(S), age at menopause (p = 0.25) and BMI (p = 0.49) were no longer significant independent predictors of endometrial cancer and were therefore excluded. This final model was named as “SPAR”. The same sets of independent predictor of endometrial cancer were identified when analyses were restricted to only those cases with non-missing data for all variables.

Table 2 Results of the univariate and multivariate analyses in the ‘Development’ dataset used to create the final ‘Risk Score’ and ‘Logit’ models.

Clinical and ultrasound characteristics Univariate analysis Patient characteristics only Patient characteristics + endometrial thickness
OR 95% CI p aOR 95% CI p aOR 95% CI p
Recurrent bleeding 2.14 1.45–3.17 <0.0001 1.98 1.33–2.94 0.001 1.67 1.05–2.65 0.031
Diabetes 2.12 1.30–3.46 0.002
Nulliparous 3.51 1.98–6.22 <0.0001 3.98 2.15–7.33 <0.0001 2.68 1.33–5.40 0.006
Hypertension 1.4 0.93–2.11 0.11
Hyperlipidaemia 1.82 0.91–3.65 0.09
Hormone therapy use 0.28 0.11–0.72a 0.008
Tamoxifen use 0.34 0.12–1.03a 0.056
Age at assessment per incremental yearb 1.03 1.02–1.05 <0.0001 1.03 1.02–1.04 <0.0001 1.02 1.00–1.04 0.028
Age at menopause per incremental yearb 1.07 1.00–1.13 0.024 1.07 1.03–1.11 0.037
Body mass index (kg/mb)b 1.07 1.03–1.11 0.001 1.07 1.02–1.11 0.003
Endometrial thickness per incremental (mm)b 1.17 1.13–1.22 <0.0001 1.16 1.12–1.21 <0.0001

a 95% CI based on less than 1000 replicates as some bootstrap datasets had no valid cases preventing determination of the OR.

b Continuous predictor, modelled as linear term in logistic regression analysis.

CI, confidence interval; OD, odds ratio.

95% CI and significance of adjusted OR were determined using bootstrap techniques following 1000 repetitions.

The final ‘RAAMP and ‘SPAR’ models derived from the ‘Development’ cases and tested on the ‘Validation’ cases were

‘RAAMP’ Model  Y = −9.987474 (SE, 1.499)

  • +0.6826576 (SE, 0.2024744) if recurrent bleeding
  • +0.0306029 (SE, 0.0074131)* Age at presentation
  • +0.0576946 (SE, 0.0275953) * Age at menopause
  • +0.06522524 (SE, 0.0217002) * BMI
  • +1.379942 (SE, 0.3124162) if nulliparous

‘SPAR’ Model  Y = −6.02053 (SE, 0.5472794)

  • +0.151391 (SE, 0.0200035) * Scanning Endometrial Thickness ET
  • +0.9843951 (SE, 0.3584834) if nulliparous
  • +0.0306029 (SE, 0.0092393)* Age at presentation
  • +0.5098766 (SE, 0.2363772) if recurrent bleeding

The area under the ROC curves and Youden Index for ‘RAAMP’ and ‘SPAR’ in the ‘Development’ cases were 0.71 (SE, 0.024; 95% CI 0.66–0.75) and 0.93 (SE 0.015; 95% CI 0.90–0.95) respectively. The difference in AUC between both models was significant (difference = 0.20, 95% CI 0.15–0.24; p < 0.0001) The ROC curves for the prediction of endometrial cancer in the ‘Validation’ cohort using both models are shown in Fig. 1. The AUC were 0.73 (SE, 0.034; 95% CI 0.67–0.80) and 0.92 (SE, 0.019; 95% CI 0.88–0.96) respectively and difference in AUC between both models was statistically significant (difference = 0.19, 95% CI 0.14–0.24; p < 0.001). Youden Index for ‘RAAMP’ model was 0.72 (Sensitivity = 66.5%; Specificity = 68.9%; +ve LR = 2.14; −ve LR = 0.49) at a risk cut off criterion of 1:25 or higher. Youden Index for ‘SPAR’ model was 0.71 (Sensitivity = 82.7%; Specificity = 88.3%; +ve LR = 6.38; −ve LR = 0.2) at a risk cut off criterion of 1:24 or higher. The diagnostic effectiveness, reflected by the AUC, of the ‘SPAR’ was identical to 0.92 (95% CI 0.89–0.94) achieved using ET alone in our previous study and was not statistically significant (difference = 0.001 95% CI −0.015 to 0.0018, p = 0.84) [16]. ‘RAAMP’ was inferior to ET assessment based on ET alone (AUC difference = 0.19, 95% CI 0.15–0.24; p < 0.0001).


Fig. 1 Receiver operating characteristic curves of prediction of endometrial cancer using either patient characteristics only, patient characteristics in combination with endometrial thickness or by endometrial thickness alone.


In this study we have derived two probability-based models and evaluated their potential diagnostic performance in identifying endometrial cancer in our cohort of women who presented with PMB. Our analyses indicated that a risk model based on patient characteristics alone was only of fair accuracy. Integrating patient characteristics with ET did not improve diagnostic accuracy as compared to ET alone.

The strengths of our study are its large sample size and that endometrial histology was available for all patients irrespective of the ultrasound findings. Women diagnosed with endometrial cancer who had an ET of ≤4 mm were not excluded and no assumption was made as to whether they had a normal uterus. Our data indicated that 1 in 283 women with an ET ≤4 mm still had cancer.

The limitations of our study were that we did not ascertain whether nulliparity was by choice or as a result of infertility and that we relied on data imputation for some missing data. Data imputation was previously employed by Opmeer et al. in ≈15% of women in their assessment of PMB in women to assess different potential clinical management protocols [20]. Our analysis indicated that the same predictors still remained statistically significant even when analysis was restricted to only those women with complete data.

Our analyses both supported and differed from previous studies. Similar to non-Chinese populations we observed that the risk of endometrial cancer increases with age, age at menopause, BMI and whether the women had recurrent vaginal bleeding or were nulliparous when we excluded ET [17], [18], [19], [20], [21], and [22].

In contrast to previous studies we did not find a significant association between incidence of endometrial cancer and history of diabetes, hypertension, use of hormone replacement therapy or anticoagulants [17], [18], [19], [20], [21], [22], [27], and [28]. We postulate that this may be due to some of the earlier studies with smaller sample size which could have resulted in potential model over fitting and a resulting overestimate of the odds ratios [29]. The lack of an association between diabetes and incidence of endometrial cancer would agree with the recent follow-up study of postmenopausal women participating in the Women's Health Initiative (WHI) [30]. That study concluded that the association between diabetes and endometrial cancer incidence may be attributable to the higher body weight in women with diabetes and that the association between diabetes and endometrial cancer was not significant after adjusting for BMI [30].

The lack of association between factors previously identified, but which were found not to be significantly associated with having endometrial cancer in our cohort may also be due to underlying differences both in the prevalence of malignancy and patient characteristics. The incidence of endometrial cancer in our cohort was only 3.8%, similar to 4.9% reported in a cohort of ≈3000 women in the United Kingdom [17]. In contrast, the observed incidence in other series ranged from ≈10% to ≈24%, whilst the meta-analyses by Gupta et al. indicated an expected pretest prevalence of 14% [14], [20], and [21]. The relatively lower incidence of endometrial cancer in our cohort could be reflective of the lower overall BMI in our population. An alternative explanation would be the different referral pattern in our locality compared to previous studies. We accepted direct referral to our one-stop clinic by family or primary care physicians without the need for any prior investigation.

The AUC using our patient characteristics risk model was only 0.73 and increased to 0.92 after including ET as a predictor. The AUC however, was not significantly different from the 0.92 obtained by using ET alone in our previous study [16]. This is in contrast to the previous study by Opmeer et al. in which AUC using ET alone (AUC = 0.76) was identical to that using patient characteristics (AUC = 0.76) and increased to 0.90 when ET was integrated with patient characteristics [20]. There are wide variations in the AUC (0.68–0.97) when assessing the diagnostic performance of transvaginal ultrasound in different centres [15]. Whether adding patient characteristics to ET will improve the diagnostic performance is likely to be determined by the baseline performance of the ET within centres. Also, the ROC curve for patient characteristics did not indicate a particular point at which measurement of ET could be withheld, indicating the necessity of offering ET to all women with PMB bleeding.

Our previous study and finding of this present study would suggest that ET alone can be used to triage women for further investigation provided that individual centres have determined the baseline accuracy of their ET measurement after ensuring it is measured in a consistent manner with resultant high performance. Our approach to measurement of ET is analogous to performing fetal nuchal translucency assessment for Down's syndrome. All sonographers were required to go through supervised training and internal credentialing program prior to performing scans independently [16]. Inter-sonographer ET measurement variation was minimised by requirements for an optimal imaging plane, image magnification, measuring the thickest part of the endometrium and placement of calipers.

In this study, we assessed the diagnostic accuracy of predictive models for endometrial cancer in women with PMB. Our model of patient-characteristics alone (“RAAMP”) is of fair accuracy and cannot be used to triage women to endometrial sampling. These predictive factors may be considered when prioritising women during referral. Addition of patient characteristics to ET did not improve the diagnostic accuracy compared to using ET alone. At present, ET-alone seems the best strategy in triaging women when the scans are performed under a stringent protocol.

Ethical approval

Ethical approval was obtained from the Institutional Joint Chinese University of Hong Kong-New Territories East Cluster Clinical Research Ethics Committee. Date of approval: 27 September 2013. CREC reference number: CRE-2013.450.

Authors’ contributions

WAS, CCW, FLW, LTT and SDS were involved in the study design and data collection. WAS, CCW, FLW, LTT, MBW and SDS contributed to literature search, draft and review of the manuscript. WAS and SDS performed data analysis, data interpretation and preparation of the figures. SDS was involved in setting up the computer record proforma.

Conflict of Interests

The authors report no conflict of interest.


There is no source of any financial support or funding.


We thank Miss Joyce HY Chan and Miss Jennifer SF Tsang for her help in database management.


  • [1] K. Astrup, N.D.F. Olivarius. Frequency of spontaneously occurring postmenopausal bleeding in the general population. Acta Obstet Gynecol Scand. 2004;83:203-207
  • [2] The American College of Obstetricians and Gynecologists. The role of transvaginal ultrasonography in the evaluation of postmenopausal bleeding. ACOG Comm Opin No. 440. Obstet Gynecol. 2009;114:409-411 (reaffirmed 2013)
  • [3] U. Chandavarkar, J.M. Kuperman, L.I. Muderspach, N. Opper, J.C. Felix, L. Roman. Endometrial echo complex thickness in postmenopausal endometrial cancer. Gynecol Oncol. 2013;131:109-112 Crossref
  • [4] I. Jacobs, A. Gentry-Maharaj, M. Burnell, et al. Sensitivity of transvaginal ultrasound screening for endometrial cancer in postmenopausal women: a case-control study within the UKCTOCS cohort. Lancet Oncol. 2011;12:38-48 Crossref
  • [5] Hong Kong Cancer Registry, Hospital Authority. Leading Cancer Sites in Hong Kong in 2012. [accessed 05.06.15].
  • [6] M.C. Renaud, T. Le, J. Bentley, et al. Epidemiology and investigations for suspected endometrial cancer. J Obstet Gynecol Can. 2013;35:380-383
  • [7] Clinical practice guideline. Investigation of postmenopausal bleeding. (Institute of Obstetricians and Gynaecologists, Royal College of Physicians of Ireland and Directorate of Clinical Strategy and Programmes, Health Service Executive. Guideline No. 26, Version 1.0, 2013)
  • [8] A. Ewies, P. Musonda. Managing postmenopausal bleeding revisited: what is the best first line investigation and who should be seen within 2 weeks? A cross-sectional study of 326 women. Eur J Obstet Gynecol Reprod Biol. 2010;153:67-71 Crossref
  • [9] M.C. Salman, G. Bozdag, S. Dogan, K. Yuce. Role of postmenopausal bleeding pattern and women's age in the prediction of endometrial cancer. Aust N Z J Obstet Gynaecol. 2013;53:484-488
  • [10] R. Tarling, A. Gale, P. Martin-Hirsch, L. Holmes, K. Kanesalingam, P. Dey. Experiences of women referred for urgent assessment of postmenopausal bleeding (PMB). J Obstet Gynaecol. 2013;33:184-187 Crossref
  • [11] B. Karlsson, S. Gransberg, M. Wikland, et al. Transvaginal ultrasonography of the endometrium in women with postmenopausal bleeding – a Nordic multicenter study. Obstet Gynecol Surv. 1996;51:100-101 Crossref
  • [12] E. Ferrazzi, V. Torri, D. Trio, E. Zannoni, S. Filiberto, D. Dordoni. Sonographic endometrial thickness: a useful test to predict atrophy in patients with postmenopausal bleeding. An Italian multicenter study. Ultrasound Obs Gynecol. 1996;7:315-321
  • [13] R. Smith-Bindman, K. Kerlikowske, V.A. Feldstein, et al. Endovaginal ultrasound to exclude endometrial cancer and other endometrial abnormalities. JAMA. 1998;280:1510-1517
  • [14] J.K. Gupta, P.F.W. Chien, D. Voit, T.J. Clark, K.S. Khan. Ultrasonographic endometrial thickness for diagnosing endometrial pathology in women with postmenopausal bleeding: a meta-analysis. Acta Obstet Gynecol Scand. 2002;81:799-816 Crossref
  • [15] A. Timmermans, B.C. Opmeer, K.S. Khan, et al. Endometrial thickness measurement for detecting endometrial cancer in women with postmenopausal bleeding: a systematic review and meta-analysis. Obstet Gynecol. 2010;116:160-167 Crossref
  • [16] A.S.W. Wong, T.T.H. Lao, C.W. Cheung, et al. Reappraisal of endometrial thickness for the detection of endometrial cancer in postmenopausal bleeding: a retrospective cohort study. BJOG. 2015; 10.1111/1471-0528.13342
  • [17] N. Burbos, P. Musonda, I. Giarenis, et al. Predicting the risk of endometrial cancer in postmenopausal women presenting with vaginal bleeding: the Norwich DEFAB risk assessment tool. Br J Cancer. 2010;102:1201-1206 Crossref
  • [18] N. Burbos, P. Musonda, T.J. Duncan, S.G. Crocker, E.P. Morris, J.J. Nieto. Estimating the risk of endometrial cancer in symptomatic postmenopausal women – a novel clinical prediction model based on patients’ characteristics. Int J Gynecol Cancer. 2011;21:500-506 Crossref
  • [19] P. Musonda, N. Burbos, T.J. Duncan, S.G. Crocker, E.P. Morris, J.J. Nieto. Comparing the performance of two clinical models in estimating the risk of endometrial cancer in symptomatic postmenopausal women. Eur J Obstet Gynecol Reprod Biol. 2011;159:433-438 Crossref
  • [20] B.C. Opmeer, H.C. Van Doorn, A.P.M. Heintz, C.W. Burger, P.M.M. Bossuyt, B.W.J. Mol. Improving the existing diagnostic strategy by accounting for characteristics of the women in the diagnostic work up for postmenopausal bleeding. BJOG. 2007;114:51-58
  • [21] G. Opolskiene, P. Sladkevicius, L. Valentin. Prediction of endometrial malignancy in women with postmenopausal bleeding and sonographic endometrial thickness ≥ 4.5 mm. Ultrasound Obstet Gynecol. 2011;37:232-240 Crossref
  • [22] L. Giannella, K. Mfuta, T. Setti, L.B. Cerami, E. Bergamini, F. Boselli. A risk-scoring model for the prediction of endometrial cancer among symptomatic postmenopausal women with endometrial thickness > 4 mm. Biomed Res Int. 2014;2014:130569 10.1155/2014/130569
  • [23] P. Royston, D.G. Altman, W. Saurbrei. Dichotomizing continuous predictors in multiple regression: a bad idea. Stat Med. 2006;25:127-141 Crossref
  • [24] J. Labarère, R. Bertrand, M.J. Fine. How to derive and validate clinical prediction models for use in intensive care medicine. Intensive Care Med. 2014;40:513-527
  • [25] R. Fluss, D. Faraggi, B. Reiser. Estimation of the Youden Index and its associated cutoff point. Biom J. 2005;47:458-472 Crossref
  • [26] J.A. Hanley, B.J. McNeil. The meaning and use of the area under a receiver Operating characteristic (ROC) curve. Radiology. 1982;143:29-36 Crossref
  • [27] E. Friberg, N. Orsini, C.S. Mantzoros, A. Wolk. Diabetes mellitus and risk of endometrial cancer: a meta-analysis. Diabetologia. 2007;50:1365-1374 Crossref
  • [28] Z.H. Zhang, P.Y. Su, J.H. Hao, Y.H. Sun. The role of preexisting diabetes mellitus on incidence and mortality of endometrial cancer: a meta-analysis of prospective cohort studies. Int J Gynecol Cancer. 2013;23:294-303 Crossref
  • [29] S. Nemes, J.M. Jonasson, A. Genell, G. Steineck. Bias in odds ratios by logistic regression modelling and sample size. BMC Med Res Methodol. 2009;9:56 Crossref
  • [30] J. Luo, S. Beresford, C. Chen, et al. Association between diabetes, diabetes treatment and risk of developing endometrial cancer. Br J Cancer. 2014;111:1432-1439 Crossref


a Department of Obstetrics and Gynaecology, The Chinese University of Hong Kong, Prince of Wales Hospital, Hong Kong SAR

b Department of Obstetrics and Gynaecology, Women's & Children's Hospital, The University of Adelaide, Australia

Corresponding author at: Department of Obstetrics & Gynaecology, 1E, 1/F, Block EF, Prince of Wales Hospital, Shatin, Hong Kong SAR. Tel.: +852 26321531; fax: +852 26360008.