Endometriosis Predictive Models Based on Self-Assessment Questionnaire, Evidence from Clinical Examination or Imaging Findings: A Narrative Review

For this study, we conducted an extensive narrative literature search covering the years 2005 to 2022. We meticulously searched databases such as PubMed and EMBASE using strategic keywords such as “endometriosis AND predictive models”, “fertility AND endometriosis AND predictive models” and “bowel surgery for endometriosis AND predictive models” ( Table 1 ). Our selection criteria included studies with patient-completed questionnaires and symptom-based screening tools. In addition, we included studies that looked at models that considered the localisation of endometriosis using imaging both in conjunction with questionnaires and independently. However, to maintain accuracy and focus, we deliberately excluded models that focused exclusively on postoperative findings or studies that focused exclusively on adolescent cohorts ( Table 2 ). From the original pool of 180 articles, 52 were identified as duplicates, 128 were excluded based on predetermined criteria, and a curated selection of 23 studies were found to fit our research objectives. To increase rigour, the studies that were deemed appropriate were reviewed by two different and impartial researchers. The selected studies were characterised by a great diversity, differing in design, objectives and results. Due to this heterogeneity, it was not possible to provide a comprehensive systematic review of the prediction models presented. Instead, Table 3 provides essential insights into the key features of each included study, explaining the population studied, the geographical context, the type of analytical tool used, a brief overview of the methodology and the clinical significance of the results. It is important to note that this narrative review is an attempt to distil and consolidate the wealth of information available given the complex landscape of predictive models associated with endometriosis.

Given the noticeable delay in endometriosis diagnosis, which affects women’s well-being and has a potential impact on fertility [ 17 , 29 ], Verket et al. ventured to create a prediction model to facilitate early identification of women at high risk of endometriosis in primary care [ 8 ]. Their study relied on an anonymous questionnaire that revealed a robust association between a family history of endometriosis and future manifestation of the disease [ 8 ]. Parallel studies focused on pain as a predictive factor [ 14 , 17 ], but its subjective nature limited its prognostic utility. Of note, the risk of endometriosis was almost 50% higher in women with a family history of endometriosis [ 21 , 22 ], emphasizing the crucial role of this factor in early detection [ 30 ]. In other endeavours, Forman et al. [ 9 ] developed a questionnaire that focused on women’s pain and health history to differentiate those with endometriosis from those with a healthy pelvis. Unfortunately, this questionnaire did not effectively differentiate between the two groups. Fasciani et al. [ 11 ] proposed an endometriosis index composed of 38 parameters, taking into account patient-reported pain, consultation with physicians and diagnostic evidence, including pelvic examinations, imaging and laboratory tests. In contrast, Yeung et al. [ 12 ] developed a mathematical model using a preoperative questionnaire similar to the World Endometriosis Research Foundation-Women’s Health Symptom Survey (WERF-WHSS). Although the model had a sensitivity of 80.5% and a specificity of 57.7%, it was difficult for clinical application due to its complicated nature. Five studies that take a different perspective examine pre-surgical prediction models [ 8 , 14 ]. The approach of Eskenazi et al. [ 13 ] aimed to predict surgical diagnoses based on patient interviews, combining clinical symptoms, history, examinations and ultrasound findings. Although this model was excellent in predicting ovarian endometriosis, its efficacy diminished for non-ovarian manifestations. Calhaz-Jorge et al. [ 14 ] attempted to predict endometriosis in sub fertile women using a questionnaire that included age at laparoscopy, menarche, demographic data, social variables, obstetric history and cycle characteristics. This study, although informative, was not validated beyond its sub fertile study population and included women with previous pelvic surgery [ 14 ]. The study by Ballard et al. [ 15 ] aimed to determine whether endometriosis cases could be recognised prior to laparoscopy based on pain characteristics, using a comprehensive questionnaire to assess various aspects of pain. Dyschezia was found to be more common in women with endometriosis, particularly those with deep infiltrative endometriosis. While this study is informative, it has yet to be validated. In the meantime, Hackethal et al. [ 16 ] examined medical records to formulate a comprehensive questionnaire covering endometriosis history, surgical history, allergies, chronic diseases, family history, fertility and obstetric history. Although the questionnaire is rich in data, its complexity compromises the usefulness of self-assessment, and the differential diagnosis of women with and without endometriosis was not prioritized [ 16 ]. An innovative study in 2012 introduced a symptom-based model that predicted not only the presence of endometriosis but also its different stages in symptomatic women without prior surgical diagnosis [ 31 ]. While the prediction for each endometriosis stage remained modest, the accuracy improved when ultrasound findings of ovarian endometriotic cysts or nodules were taken into account. The prediction for stage III/IV endometriosis in particular achieved a high level of accuracy. With the Endometriosis Research Centre’s self-test, a questionnaire was introduced that allowed women to self-assess the likelihood of endometriosis based on their symptoms and medical history [ 18 ]. Respondents who answered “yes” to three or more questions showed an increased potential for endometriosis, even in non-symptomatic areas such as family history, subfertility or miscarriage. Abdulai et al. [ 7 ] took a digital approach and developed a web-based system for women with endometriosis that serves as both an educational tool and a platform for self-assessment. However, the user-friendliness of the system proved to be insufficient, so that there was no significant user participation.

In the field of predictive modelling for deep infiltrating endometriosis (DIE), Chapron et al. [ 21 ] have developed an innovative approach using a self-assessment questionnaire to predict the presence of DIE in women struggling with chronic pelvic pain. In this ground-breaking study, a range of prevalent symptoms such as dysmenorrhoea, dyspareunia and bowel and urinary symptoms during menstruation were carefully recorded. It is noteworthy that posterior DIE was closely associated with the leading symptom of dyschezia. In contrast to the conventional questionnaires filled out by patients, this study was driven by the development of a complex predictive model. Meanwhile, Pillet et al. [ 22 ] embarked on a complicated journey and developed a robust regression model using a comprehensively curated preoperative questionnaire. Their model relied on 57 comprehensive variables, culminating in the identification of four key predictors primarily related to DIE symptoms—pain duration, dysmenorrhoea, gastrointestinal discomfort and dyspareunia or subfertility. The commendable accuracy of the results is emphasised, although it should be recognised that the model is complicated and challenging for patients to use effectively. It is worth noting that this study was conducted in a specialist endometriosis centre, which may limit the generalisability of the results to non-specialist settings. Perello et al. [ 23 ] contributed to the prediction of DIE on this topic with a retrospective study based on a dataset of women with histologically confirmed ovarian endometriomas. A number of variables were included in the model, including BMI (body mass index), age at baseline, history of surgery for endometriosis, and pain scores for dysmenorrhoea, dyschezia, dyspareunia and pelvic discomfort. Although the complexity of the model may be challenging for patients, its strength lies in its ability to predict the presence of ovarian endometriosis, potentially enabling prioritised treatment for affected patients.

In the field of endometriosis localisation prediction, two key studies shed light on this complicated terrain [ 19 , 20 ]. Fedele et al. [ 20 ] utilised the American Urologic Association Symptom Index (AUASI) questionnaire to develop a system for the pre-surgical diagnosis of endometriosis with bladder involvement. The effectiveness of the model was particularly evident in patients with suspected bladder involvement. Griffiths et al. [ 19 ] analysed the symptoms of their patients to detect patterns indicating an increased risk of rectovaginal endometriosis. Evidently, dyspareunia proved to be a common leading symptom in individuals struggling with this variant. The spectrum of symptoms included dysmenorrhoea, infertility, dyschezia, rectal pain, cyclical and non-cyclical rectal bleeding, diarrhoea and tenesmus. While these informative models are promising, they are tailored to populations who are symptomatic or already have endometriosis, limiting their applicability to the general population. Barcelos et al. [ 24 ] introduced a model for predicting the location of endometriosis by combining medical history, physical examination, parity, symptoms and imaging evaluation. The analysis of pre- and postoperative results facilitated the identification of variables relevant to the prognosis of endometriosis localisation. While this approach is impressively consistent with intraoperative findings in DIE cases, it requires imaging and is therefore not suitable as a stand-alone patient screening tool. Stegmann et al. [ 10 ] have developed a model to both predict the location of endometriosis and aid its identification during surgery. This dual-purpose model deciphered features that signify an increased or decreased likelihood of biopsy-confirmed endometriosis. While it provides valuable guidance for biopsy target selection, its dependence on additional elements emphasises that it requires additional validation. Chattot et al. [ 28 ] utilised a preoperative scoring paradigm to predict rectosigmoidal involvement in endometriosis patients. The remarkable efficacy of this study was slightly attenuated by the integration of magnetic resonance imaging (MRI) in conjunction with ultrasonography, which resulted in additional costs.

In the field of pregnancy prediction after endometriosis surgery, significant progress has been made by Xin Li et al. [ 32 ], who validated the Endometriosis Fertility Index score (EFI) in women with historical endometriosis, up to 48 months after laparoscopy. This robust score combines medical history with surgical findings and paves the way for predicting the likelihood of spontaneous pregnancy. It is noteworthy that a direct correlation between increased EFI scores and an increased likelihood of spontaneous pregnancy was found. However, it should be borne in mind that the EFI does not take into account ovarian reserve and severe uterine anomalies or adenomyosis. In summary, if the EFI score is five or higher, natural conception should be considered after laparoscopic surgery to treat endometriosis, with in vitro fertilization (IVF) being a compelling alternative if natural conception is not possible.

Desplats et al. [ 25 ] investigated that rectosigmoidal endoscopic ultrasonography (RS-EUS) could serve as a predictive factor for patch or segmental resection in endometriosis when the rectosigmoidal nodule exceeds 5.20 mm. Although the results indicated a trend that wider nodes correlate with a higher likelihood of resection, statistical significance was lacking. Although RS-EUS has shown promise as a predictor, it remains secondary to other diagnostic tools in the diagnosis of endometriosis, warranting further investigation. Goncalves et al. [ 26 ] demonstrated the efficacy of transvaginal ultrasonography (TVUSS) with a sensitivity and specificity of 97% and 100%, respectively, in detecting rectosigmoid endometriosis nodules. However, when assessing intestinal mucosal infiltration, sensitivity and specificity fell to 62% and 83% respectively. Further studies investigated the diagnostic accuracy of 2D and 3D ultrasound [ 33 ], without finding a clear superiority between the two techniques. Bergamini et al. [ 27 ] investigated RS-EUS with transvaginal sonography with water contrast in the rectum (RWC-TVS) and found a higher sensitivity and specificity, although this was not statistically significant. Chattot et al. [ 28 ] introduced a preoperative scoring system integrating self-assessment questionnaires, speculum and digital examination, TVUSS and pelvic MRI to predict rectosigmoid involvement in endometriosis cases. Although the model was not validated and was performed at a single reference centre, it showed good results, albeit with a possible selection bias.

Endometriosis is a persistent, estrogen-related disease characterised by the presence of endometrial-like tissue outside the uterus. This abnormality triggers inflammatory reactions and scarring of the tissue [ 1 ]. The exact prevalence of endometriosis remains difficult to determine, although it is estimated that approximately 10% of reproductive-age females have endometriosis with 30–50% of them experiencing pelvic pain and/or infertility. Especially in regions with higher socio-demographic indices, the prevalence and incidence of the disease increased significantly between 1990 and 2017 [ 2 ]. The consequences of endometriosis include chronic pelvic pain, dyspareunia, infertility and a reduced quality of life (QoL) [ 1 ]. In women who struggle with chronic pelvic pain due to endometriosis, the impairment of quality of life is more pronounced than in women with pelvic pain due to other causes. This chronic condition primarily affects areas such as pain, stress, anxiety and social adjustment. In addition, research has looked at the impact of endometriosis on everyday activities such as work, social bonding, sexuality and psychological wellbeing. The symptoms experienced by people with endometriosis cover a broad spectrum. In particular, deep infiltrating endometriosis (DIE), in which endometrial-like tissue extends at least 5 mm beyond the lining of the uterus, accounts for almost 20% of endometriosis cases [ 3 ]. However, even today, diagnosing endometriosis and determining the factors that influence its progression and associated symptoms is still a huge challenge. The process of diagnosing endometriosis often takes years, which adds to the frustration and anxiety of patients. As a result, treatment and care for these women is delayed, sometimes up to 6–12 years after the first onset of symptoms [ 4 ]. The available evidence favours active patient involvement in the detection and diagnosis of the disease. However, relying on pain alone is not enough to diagnose endometriosis. Apart from the complexity of the diagnosis and the prognosis of symptoms, there is also the question of the proportion of women who experience improvement after surgery [ 5 ]. Another pressing concern is the prediction of fertility problems in women with endometriosis. It is clear that endometriosis profoundly affects various facets of women’s lives. Consequently, the development of predictive models is essential to identify risk factors for the development of endometriosis, the symptoms and the severity of the disease. The aim of this study is to present predictive models from the existing literature based on self-assessment questionnaires with or without imaging examinations. These questionnaires are focused on women’s symptoms, life style, family history and the effect of pain at their everyday life. The central questions of this study include the prediction of early-stage endometriosis or deep infiltrating endometriosis (DIE), the preoperative localisation of endometriosis, and particularly the identification of bowel involvement. The development of efficient predictive models will help primary care physicians to identify and diagnose women with endometriosis and improve the service they offer. As a result, these patients will be treated earlier and potentially better by specialists. Specialists will be able to recognise and offer the best treatment according to the models and improve women’s life. Although some predictive models have emerged in the literature, such as anti-Mullerian hormone or urinary peptide patterns, these efforts are preliminary and lack the extensive data required for clinical integration [ 6 ]. The diversity of studies further complicates statistical synthesis and analysis. Furthermore, the heterogeneity within the endometriosis population exacerbates the challenges faced in subsequent analyses.

Timely diagnosis of endometriosis is crucial to reduce patient frustration, fertility concerns and impaired quality of life [ 34 ]. This study aims to identify effective screening tools, with a focus on predictive models to assess endometriosis localisation and bowel involvement in deep infiltrating endometriosis (DIE). Focusing on these parameters is of central importance for early detection of the disease and crucial for planning optimal interventions [ 7 ]. Numerous question-based models have been explored, but their adoption remains hampered by inconclusive evidence [ 35 ]. Although the prediction of bladder or bowel disease shows positive trends [ 24 ], further evaluation is warranted. Geysenbergh et al. [ 36 ] addressed adolescents by adapting questionnaires for adults to identify potential endometriosis risk. However, the lack of similar studies for adults and the focus on urinary symptoms prevent wider application. Predictive models play an important role in the context of endometriosis. They provide a strategic approach to understanding, diagnosing and managing this complicated and often debilitating condition. Their importance stems from several key factors in the field of endometriosis treatment. Firstly, these models enable early detection and diagnosis of endometriosis, even before extensive clinical signs appear. Early detection is crucial for timely intervention, which can lead to improved patient quality of life and more effective treatment outcomes. Secondly, predictive models are ushering in a new era of personalised medicine. By taking into account a range of variables, including symptoms, medical history and imaging results, these models enable customised patient care. This individualised approach makes it possible to tailor treatment plans to the unique characteristics of each case and predict outcomes with greater precision. In addition, predictive modelling provides healthcare professionals with improved clinical decision support. This informed decision-making extends to treatment options, surgical planning and overall management strategies [ 16 ]. This in turn can optimise resource allocation and contain medical costs while improving patient outcomes. In addition, such models help optimise resource utilisation by identifying the patients who would benefit most from specialised procedures or interventions—a particularly important aspect given the limited resources available in healthcare. In addition, predictive models contribute to improved patient counselling by providing accurate information about the disease and possible outcomes. This enables patients to make informed decisions about their treatment and actively participate in the organisation of their healthcare. In addition, these models drive research and development efforts and shed light on the underlying mechanisms and risk factors of endometriosis. Another benefit is the reduction of diagnostic delays, as these models can identify patients at higher risk, leading to earlier intervention and treatment. The models also facilitate long-term monitoring of health status, help in the selection of participants in clinical trials and promote interdisciplinary collaboration between different medical disciplines [ 10 ]. While questionnaires offer valuable insights for patients [ 17 ], their implementation must take into account patients’ technological access. Predicting endometriosis using questionnaires remains a challenge when it comes to overcoming selection bias and population heterogeneity. To make accurate predictions, the biases in the prediction models need to be removed, especially when taking into account the different symptoms and localisations of the disease [ 37 , 38 , 39 , 40 ]. The diagnosis of endometriosis is associated with delays that affect patient well-being. Early detection allows patients to make informed decisions and facilitates physician engagement. The development of predictive models using machine learning methods, similar to cancer prediction, remains a goal. However, the complexity of endometriosis makes such modelling difficult given the heterogeneity of the disease and the variety of symptoms. In summary, symptom-based screening tools can help patients to recognise and understand their disease and at the same time help physicians to make decisions. However, the lack of a universally applicable and validated endometriosis screening tool compromises its clinical value. Future research should strive for a concise, accurate and widely applicable tool to improve the diagnosis and treatment of endometriosis. Limitations include heterogeneity of studies, different methodologies and retrospective designs. Further research must strive for consistency and validation to advance the field.

Endometriosis, a chronic disease that significantly affects women’s daily lives, benefits greatly from early diagnosis. Rapid identification enables optimal treatment, comprehensive risk awareness and fertility preservation. It is therefore essential to develop predictive models that enable doctors to recognize early symptoms with greater precision and thus provide better treatment. Unfortunately, the models proposed in the current literature often fall short of expectations and have unbalanced sensitivity and specificity. Moreover, they are usually complicated, lengthy and user-unfriendly. While the scientific community recognizes the need for prediction, the development of effective tools remains a challenge. It is important to develop a model in the near future that does not allow for subjectivity and complexity. It will be user friendly, will not take too long and will be validated on a large population. Artificial intelligence offers potential solutions, but large-scale, multicenter studies remain essential for substantial progress.