{"paper_id":"0cf21371-ed66-4c70-b7e1-c663853c7890","body_text":"Endometriosis is a complex full-body inflammation disease, affecting women of\nreproductive age, as well as cisgender, transgender, and non-binary people in a\ncurrently unknown number. 1 – 3  Simply stated,\nthe disease affects anyone with a body part classified as ‘female’. \n 1 \n  Approximately 200 million people worldwide, including 10–15% of women of\nreproductive age and 2.5% of postmenopausal women are affected by\nendometriosis. 4 , 5 \nThe most recognised symptoms of endometriosis are chronic pelvic pain, dyspareunia,\ndysmenorrhea, menorrhagia, bowel symptoms, and infertility. Currently, not only is\nthere no effective treatment for endometriosis, but the time between the development\nof the first lesions and diagnosis can be by about 7–10 years. 6 , 7  Thus, two major challenges must\nbe met: identification of the earliest symptoms, which could support investigations\non biological processes, and then biomarkers for potential therapeutic targets. The\nbiological complexity under endometriosis has been addressed using computational\nbiology approaches applied on endometriosis-related alterations (development and progression), \n 8 \n  and endometriosis-related core symptoms. \n 9\nNevertheless, the question of how to capture early symptoms of endometriosis, so that\npotential prevention strategies (as no relevant treatment currently exists for\nendometriosis) may have the highest impact on women still remains a matter of\nconcern. Indeed, the identification unknown symptoms or early signs of endometriosis\n(i.e. those not clearly or frequently reported in the scientific literature), can be\nuseful for healthcare professionals in their screening and diagnosis approach. The\nemergence of the Internet and the popularity of medical forums have provided\nadditional health information about the patient experience. \n 10 \n  For example, studies on the detection of COVID-19 symptoms on Twitter have\nemerged and supplemented clinically known symptoms. \n 11 \n  Sarker et al. \n 11 \n  reported the expression of anosmia and/or ageusia as symptoms not documented\nin the literature by matching the word of the corpus to a dictionary of medical\nconcepts. The advantages of web-based data are its volumetric nature and level of\ndetail regarding the patient's experience.\nIn view of the current dearth of data on the patient-reported experience on early\nsignals of endometriosis, the aim of this study was to identify some of them, by\ncollecting patient's experiences on online social networks platforms.\n\nThis was an observational and cross-sectional study consisting in the extraction\nand analysis of open and patient-centred discussions retrieved from online\nforums, which involved French-speaking communities. The selection of forums was\nmade for posts released between February 2005 and March 2021, with the query\n‘endométriose inurl: forum’ via the Google search engine. The collection and\nreviewing of URLs was made in March 2021 by two independent authors. Resources\nthat were not forums, which presented a newsletter, which did not give rise to\nan exchange, or which required access rights were excluded. The collection of\nURLs containing the keyword ‘endometriosis’ accounted for the various threads on\nthis topic. All pages on each thread were crawled, that is the URLs of the\nwebsites were scanned and retrieved.\nAs shown in  Figure 5 (Supplementary material) , the frequency of use of\nendometriosis forums was not constant over the selected study period\n(2005–2021). The number of posts was higher in the period ‘before 2015’ (wave 1)\nwhen compared to period ‘after 2015’ (wave 2). We then divided and compared data\ncollected according to these two waves. We found with a kappa test that there is\na huge agreement ( k  = 87%) in the occurrence of words between\nthe considered two waves, and that >80% of the evoked symptoms observed in\nthe period ‘after 2015’ were similar to those of the period ‘before 2015’.\nConsequently, data were pooled together in analyses.\nWe used the French edition of MedDRA (Medical Dictionary for Regulatory\nActivities) v.24. MedDRA is a dictionary of medical concepts to explore the\nsymptomatology related to endometriosis. \n 12 \n  The medical concepts ontology are hierarchically arranged according to\nfive levels of detail, from system organ class (SOC) to lowest level term (LLT).\nWe retained the LLT, which includes medical terms such as symptoms, risk\nfactors, and quality of life, for example ‘reproductive organs and breast\ndisorders’.\nThe workflow was carried out in five steps ( Figure 1 ): (i) the automated extraction\nof posts, and extraction of the vocabulary of medical concepts chosen for the\nanalysis; (ii) cleaning the data from the 10 French online forums and the\ndictionary of medical concepts at the preprocessing step; (iii) detection of\nsymptoms present in the corpus of posts from the dictionary; (iv)\ncontextualisation of the list of symptoms detected in the text; and (v)\nidentification of early symptoms with the building of a dictionary of temporal\nmarkers for the identification of early symptoms.\nStudy design: steps of data sources and data extraction, preprocessing,\nsymptoms detection, symptoms contextualisation step and early symptoms\nidentification.\nThe posts of each discussion thread were scraped using BeautifulSoup, Selenium,\nPandas and Urllib libraries to extract the data from the HTML and XPATH trees in\nPython language. They were then gathered into a common corpus with the same\nstructure. Finally, both posts and MedDRA LLTs followed the same preprocessing\nprocess with the Re and SnowballStemmer libraries in Python.\nThe following phases allowed retaining only a clean text:  the suppression step allowed removing duplicate/empty posts,\nautomatic replies, URLs and special symbols on the corpus. We also\nremoved medical concepts containing less than 3 letters (e.g. ‘PA’ a\nFrench acronym for blood pressure, which can also have a negative\nconnotation like ‘pas’); the homogenisation steps removed accents, punctuation marks and\nchanged texts to lowercase; the stemming process tokenised texts, in order to remove the stop\nwords and bring the words back to their root-words. Tokenisation\naims to split a message into a list of words. A unique ID in our corpus to guarantee anonymisation replaced the user's\nnickname. The final information was organised in a 9-column table: forum name;\ndiscussion ID; post date; post time; user ID; scraped message; cleaned posts;\nstem posts; post ID. The final corpus included new columns corresponding to the\nmessage without punctuation, without characters and without capitalisation\n(‘message_clean’) and to the stemming process (‘message_stem’), is available in\n Supplementary Table 3 .\nthe suppression step allowed removing duplicate/empty posts,\nautomatic replies, URLs and special symbols on the corpus. We also\nremoved medical concepts containing less than 3 letters (e.g. ‘PA’ a\nFrench acronym for blood pressure, which can also have a negative\nconnotation like ‘pas’);\nthe homogenisation steps removed accents, punctuation marks and\nchanged texts to lowercase;\nthe stemming process tokenised texts, in order to remove the stop\nwords and bring the words back to their root-words. Tokenisation\naims to split a message into a list of words.\nA clinical symptom is defined as a finding that is reported by patients or by\nsomeone close to them. \n 13 \n  For the purpose of the current study, a symptom was then deemed a signal\nperceived and reported by women, and which does not require medical expertise\n(e.g. cysts are signs and cannot be considered as a symptom since their\nidentification requires the intervention of a health professional). We\nidentified the presence of LLTs in the corpus by exact matching each concept of\nthe MedDRA dictionary on each post. We then calculated frequency of occurrence\nof each retained symptom. A manual annotation of each LLT with frequency greater\nthan 10 was carried out according to five categories: symptoms, (risk or\nprotective) factors, outcome, diagnosis and treatment. Two independent\nannotations were performed. A Kappa test was used for the analysis of\nannotators’ agreement. The difference in the annotations were discussed and\nsolved during a third annotation phase.\nContextualisation of symptoms helped providing more information on signals.\nClarifying the meaning of each early symptom is helpful in understanding their\nsemantics. The tokenisation of the sub-corpus of the message (including the\ntemporal precocity markers) allowed the words of the corpus to be crossed with\nthe dictionary of symptoms previously established. A co-occurrence matrix was\napplied with a four-word window to generate the occurrence of word pairs found\nin proximity. \n 14 \n  This number was set before the analysis, and can be adapted as\nappropriate. The matrix enabled to retrieve the words next to each symptom. The\nresult of the co-occurrence matrix gave, for each symptom, a list of words found\nnearby. A common annotation of each association of a symptom with a context word\nwas made to remove words that did not provide any additional information about\nthe symptom. For example, the symptom ‘spotting’ associated with the word ‘pill’\nleads to thinking that spotting is a consequence of taking the pill. An\nassociation of a symptom with a context word with similar meaning (e.g.\nsynonyms), was grouped under a representative medical term. The visualisation of\nthe results was carried out in the form of relational graphs with the\ngraph-oriented database Neo4j. Each symptom was linked to the medical term by a\nlink representing the contextualisation word.\nThe unique terms of the corpus were studied individually by two readers, for\nempirical construction of a lexical field of temporal markers. Inside each post,\nproximity of a 10-word window between each temporal marker and each detected\nsymptom was set. 15 , 16  This threshold was selected empirically, after a\ncomprehensive reading of all the posts. The production of the co-occurrence\nmatrix allowed the detection of symptoms close to a temporal marker evocative of\nprecocity ( Supplementary Table 1 ). The early symptoms selected were the\nmost frequently found in co-occurrence with a temporal marker. A temporal marker\nwas often introduced at the beginning of a sentence while a clinical sign can be\nfound later in the sentence ( Table 1 ). In a second step, the\ncontextualisation of the all symptoms was filtered to keep only the elements of\nthese early symptoms.\nExamples of original posts and preprocessed posts, showing the proximity\nbetween a temporal marker and an early symptom.\nAll of the data collected in this study came from public discussions. Information\nfrom public sources represents a public act, and is made available for passive\ndata collection analysis. Such a type of study does not need any approval from\nan ethical committee, as in France, these committees are mainly involved in the\nassessment of studies that are intended to collect data de novo, and which may\nrequire patients’ information and/or consent to be carried out. In the present\ncase, as far as people have accepted the general conditions of use of the\nselected forums before partaking in exchanges, so their posts are made publicly\navailable and can be retrieved and used for research, but not for commercial\npurposes. Nonetheless, to enable people to exert their full information right, a\nsummary of the project, its results, and the coordinates of the corresponding\nauthor are displayed on the website of the academic laboratory (ULR\n2694-METRICS, University of Lille). According to the privacy policy of the\nselected forums, which are in line with the General Data Protection Regulation\n(GDPR), the use of these publicly available posts does not require individual\nconsent from users. 17 , 18  We further completely de-identified posts, and\npseudonyms were replaced with unique identifiers and messages were not fully\nquoted.\n\nThe query ‘endométriose inurl: forum’ via the Google search engine resulted in 68\ndifferent websites. After applying the inclusion and exclusion criteria, 10\nforums were selected: ‘ Doctissimo’, ‘Journal des Femmes’, ‘Au Féminin’,\n‘Forum Psychologies’, ‘Madmoizelle’, ‘Vinted’, ‘Forum Parents’, ‘Être\nEnceinte’, ‘RockieMag Forum’,  and ‘ Maman pour la\nvie’  ( Figure 2 a and b) .  These forums were about general\nhealth, women's media and specialised maternity. From 2006 to 2016, Doctissimo\nwas the preferential forum used for endometriosis in these online communities\n( Figure 2 a).\n( a) Trend in the use of forums between 2005 and 2021. (b) Number\nof posts per forum.\nOverall, we identified 7148 URL discussion threads related to endometriosis\n( Supplementary Table 2 ) including 78,905 posts. The preprocessing\nstep removed 1585 posts (2%). The cleaned corpus was finally composed of 77,320\nunique posts. A total of 9390 users are distributed over the 10 forums. The top\n3 forums by number of posts were ‘doctissimo.fr’ (82.1%,\n N  = 64,812 posts), ‘journaldesfemmes.fr’ (7.5%,\n N  = 5906 posts) and ‘aufeminin.com’ (4.3%,\n N  = 3436 posts) ( Figure 2 b).\nRegarding the extraction of medical concepts, 83,217 unique LLTs from MedDRA were\nused in the symptom detection analysis. The grouping by LLTs of this result gave\nthe total number of occurrences of each LLTs. This yielded 2064 LLTs detected in\nthe corpus. The 630 LLTs with more than 10 occurrences in the corpus were\nmanually annotated according to symptoms, factors, outcomes of the disease,\ndiagnostic methods, and treatments classes. One annotator obtained 117 selected\nsymptoms, 365 other annotations (e.g. factor, outcome, diagnosis, and treatment)\nand 148 unreferenced (e.g. death, marriage, divorce, unemployment). A second\nannotator yielded 248 symptoms, 269 other categories, and 113 unreferenced. The\ndifference in annotation revealed a Kappa score of 43%. This important\ndiscrepancy led to a third annotation in common. The final annotation ( Supplementary Table 4 ) translated into 195 symptoms, 120\nfactors, 53 outcomes, 33 diagnoses and 35 treatment terms ( Figure 3 ). Symptoms that can only be\ndetected after a medical examination were removed (e.g. ovarian cysts).\nDistribution of the final annotation of LLTs in five categories: symptom,\nfactor, outcome, diagnostic, and treatment. Symptoms. A. Factor:\ncontaining identified risk and protective factors. B. Outcome:\ncontaining identified symptoms (clinical signs detected pre-diagnosis).\nC. Diagnostic: containing diagnosis methods. D. Treatment: containing\ntreatment methods.\nThe resulting dictionary of the most relevant unique terms of the corpus was used\nfor the co-occurrence matrix. After the filtering columns that contain symptoms,\nthe matrix was formed of the 8002 words in row and 167 symptoms ( Supplementary Table 6 ). The word occurrences were kept for each\nsymptom, and then checked manually. For the entire symptoms, 353 symptom/word\ncontext pairs were obtained. After the common annotation, 82 symptoms were\ncontextualised grouped into 41 representative medical terms ( Supplementary Table 7 ).\nThe corpus included 8126 unique terms that were studied to extract the time\nmarkers. We identified 26 temporal markers: ‘amont’, ‘ancient’, ‘adolescent’,\n‘anteced’, ‘antecedent’, ‘anterieur’, ‘apparu’, ‘auparav’, ‘avant’, ‘debut’,\n‘depui’, ‘enfanc’, ‘jeun’, ‘jeuness’, ‘lenfanc’, ‘apparaissent’, ‘premi’,\n‘premier’, ‘quauparav’, ‘reapparaiss’, ‘reapparaissent’, ‘reapparaitr’,\n‘reapparit’, ‘reapparu’, ‘vecu’, ‘vecus’. The messages located by temporal\nmarkers consisted in 15,032 unique posts. These posts were used for the rest of\nthe analysis. From the co-occurrence matrix, the symptom occurrences were kept\nfor each temporal marker, and then checked manually. A list of 53 unique early\nsymptoms was identified ( Supplementary Table 5 ).\nThis annotation was conducted to keep these 53 early symptoms contextualised.\nSeventy-four symptom/word context pairs were kept ( Supplementary Table 8 ) with 20 general symptoms: ‘Weakened\ngeneral condition’, ‘Hot flush’, ‘Headache’, ‘Vaginal itching’, ‘Abdominal\npain’, ‘Muscle pain’, ‘Ovarian pain’, ‘Dysuria’, ‘Dysmenorrhea’, ‘Dyspareunia’,\n‘Haematuria’, ‘Urinary tract infection’, ‘Inflammation’, ‘Metrorrhagia’,\n‘Menorrhagia’, ‘Migraine’, ‘Neuralgia’, ‘Limb oedema’, ‘Infertility’, ‘Digestive\ndisorders’ ( Figure 4 ).\nNetwork representing the detected early symptoms associated with the\nsymptoms annotated with the contextualisation words (visualisation with\nNeo4j). A. Node blue: early symptoms detected in the corpus. B. Edges:\ncontextualisation words associated with the symptom. C. Node orange:\nannotated symptom class.\n\nThis is a text-mining approach, based on the exploration of exchange platforms, with\nthe goal to investigate the ‘early’ symptomatology of endometriosis. We collected 41\ngroups of symptoms, including 20 groups of them considered as ‘early symptoms’\nassociated with endometriosis, which were recontextualised based on the content of\nforums. An exploratory patient-centred approach was used in order to take the best\nadvantage from free rich posts released by interested parties on endometriosis, just\nlike a ‘big’ focus group on endometriosis. As such, forums can also be viewed as an\ninteresting space for people to actually have sufficient time to update, correct or\neven contradict their own initial ideas on a given subtopic. By using the MedDRA,\neach symptom and its synonyms, were translated into a unified lexicon, which then\neased the retrieve of only terms referred to as symptoms from the built corpus. Our\nstudy allowed pinpointing symptoms freely reported by women on forums since social\nmedia now appear as the preferred space for women with endometriosis to express\nthemselves and share experiences/advices about their condition. \n 19 \n  To the best of our knowledge, this approach has so far never been adopted in\nclinical and/or epidemiological studies.\nSome of the symptoms identified in our analysis are in agreement with previous\nfindings in the literature. Their underlying biological mechanisms in the\npathophysiology of endometriosis have already been comprehensively discussed. The\ndysregulation of factors involved in the pathological process of endometriosis is\ndirectly associated with migraines, pelvic pain and more specifically with\ndysmenorrhea, dyspareunia, and painful bladder syndrome and irritable bowel syndrome. \n 20 \n  Pelvic and abdominal pains reach the nervous system, and lead through\nnociception, to neuropathic or neuroinflammatory pain, which may explain the\nmigraines and headaches experienced by patients. \n 20 \n  The inflammatory environment of the disease may explain the infertility\nexperienced by the patients, since these dysregulations can cause an alteration of\nthe embryonic implantation and the inability of women with endometriosis to get pregnant. \n 21 \n  Endometriotic lesions have been strongly associated with irritation causing\nabdominal pain, pain on defecation, urinary tract infection, dysuria, and digestive disorders. \n 22 \n  Finally, endometrial lesions can cause pelvic pain, menorrhagia and metrorrhagia. \n 23 \n  Other extracted symptoms, such as urinary pain, urinary tract infection, or\ngastrointestinal disorders are in agreement with the literature. Newly detected\nsymptoms herein, whose relationship with endometriosis may be unknown need to be\nmore extensively studied. Practitioners, who have not yet looked at, can now address\nthese potentially new symptoms. Nonetheless, oedema for example would be associated\nwith some pain and digestive disorders, which could guide the diagnosis towards\nendometriosis. Since peripheral nerves are directly associated with endometrial tissue, \n 20 \n  earlier authors have assumed that implanting endometrial tissue around the\nsciatic nerve would cause an inflammatory reaction, severe pain and neuralgia. \n 24 \n  Furthermore, it was argued that early diagnosis of endometriosis could avoid\npermanent nerve damage, which may be associated with bladder incontinence, muscle\nweakness and fatigue. \n 20\nMany patients complain of vaginal itching caused by endometrial lesions, with no\nclear physiological or biological explanations. The identified dizziness in the\nalteration of the patient's general state was found to be an adverse effect of\nendometriosis treatments (e.g. opioid, postoperative effects of laparoscopic\nsurgery). 25 , 26  Fatigue is associated with somatic pain syndrome and is\nsignificantly related to endometriosis 27 , 28  or can be associated with the\nside effects of some treatments (e.g. contraceptive pill). \n 26 \n  Patients with gastrointestinal disturbances report the presence of nausea. \n 29 \n  Acute pain can justify this nausea, for abdominal pain being possibly a\nsource of nausea, as well as dysmenorrhea. Furthermore, attempted treatments can\nalso yield side effects such as nausea and vomiting. \n 30 \n  The main information concerning a possible link between endometriosis and hot\nflashes suggested a post hormonal treatment effect, which may translate into\nartificial menopause. 31 , 32  Symptoms not biologically explained and/or explored in the\nliterature may be consequences of other symptoms and/or due to treatments.\nTherefore, it would be interesting to further check and investigate these signals\nduring consultations. Because some patients can self-report some of the symptoms\nfound in this study during the consultation interview, such potentially novel signs\ncan complete those already known and/or used by practitioners in their routine.\nIt is important to note that the text-mining approach has a number of limitations.\nManual data extraction is time consuming and limited to the selected sources. An\nadaptation of the data extraction method to each source within a learning of each\nHTML code is required. This step can be facilitated by using an API to automatically\nextract data from all sources using a list of keywords. Regarding the identification\nof early symptoms, a dictionary was created in an empirical way. A more accurate\nmethod could be developed and applied at the place of the built timeline in the\nspeech. The analysis of verb tenses can be used to classify and identify the\nsymptoms that occur first. The lack of information about the contextualisation of\nsome symptoms can add noise to the results. The context of the entire sentence in\nwhich the symptom is evoked can also cause confusion. Furthermore, without a formal\nor declared diagnosis by people using the selected forums for this study, we have\npurposely merged posts released by people with endometriosis and any third party\n(e.g. caregiver, relatives) interested by this condition. Nevertheless, the findings\nfrom this study would need to be compared with those of a future study involving\nonly people with endometriosis. Moreover, the temporal markers are not rigorously\nlinked to the earliness of symptoms as described in this study although they are in\nclose proximity. An analysis of the context, taking into account the tenses of the\nverbs, might help to identify more precisely the temporality/earliness of the evoked\nsymptoms. Moreover, since different symptoms may correspond to different stages of\nthe disease, due the heterogeneous manifestation of endometriosis, it is difficult,\nonce again, to ascertain that the novel symptoms obtained in this study are\nabsolutely in the earlier phases of the disease. Nonetheless, these symptoms should\nbe considered as additional signs that can inform clinical decision-makers and/or\npractitioners on endometriosis. In future studies, it is noteworthy that a sentiment\nanalysis can also be applied on such sentences to make sure that the symptom is not\nevoked as a negation (e.g. ‘ I don’t have a headache ’). Finally, at\na methodological standpoint, our approach can be tested on another corpus of posts\nin order to assess its validity and reproducibility.\n\nThis study showed the relevance of forum discussions in detecting endometriosis\nsymptoms from patient experiences. Symptoms completing previous findings (e.g.\nheadache, neuralgia) were pointed out by our analysis. However, some of the detected\nsymptoms in the current study have not yet been linked to any biological mechanism\nof the disease, and this invites further studies. Some of these potentially new\nsymptoms highlighted could for example be used to inform a protein network analysis\nanalogous to what was recently published, using the core symptoms of\nendometriosis.\n\nClick here for additional data file.\nSupplemental material, sj-docx-1-dhj-10.1177_20552076231176114 for Identification\nof early symptoms of endometriosis through the analysis of online social\nnetworks: A social media study by Mathilde Fruchart, Fatima El Idrissi, Antoine\nLamer, Karim Belarbi, Mohamed Lemdani, Djamel Zitouni and Benjamin C Guinhouya\nin DIGITAL HEALTH","source_license":"CC0","license_restricted":false}