{"paper_id":"3110cb1f-1f53-452c-8128-4f629dec27a8","body_text":"Proceedings of Machine Learning for Health Care 1–22, 2018 Machine Learning for Health Care 2018\nPhenotyping Endometriosis through Mixed Membership\nModels of Self-Tracking Data\nI˜ nigo Urteaga inigo.urteaga@columbia.edu\nDepartment of Applied Mathemathics\nColumbia University, New York, NY, USA\nMollie McKillop mm4234@cumc.columbia.edu\nDepartment of Biomedical Informatics\nColumbia University, New York, NY, USA\nSharon Lipsky-Gorman srg2128@cumc.columbia.edu\nDepartment of Biomedical Informatics\nColumbia University, New York, NY, USA\nNo´ emie Elhadad noemie.elhadad@columbia.edu\nDepartment of Biomedical Informatics\nColumbia University, New York, NY, USA\nAbstract\nWe investigate the use of self-tracking data and unsupervised mixed-membership models to\nphenotype endometriosis. Endometriosis is a systemic, chronic condition of women in repro-\nductive age and, at the same time, a highly enigmatic condition with no known biomarkers\nto monitor its progression and no established staging. We leverage data collected through a\nself-tracking app in an observational research study of over 2,800 women with endometriosis\ntracking their condition over a year and a half (456,900 observations overall). We extend a\nclassical mixed-membership model to accommodate the idiosyncrasies of the data at hand\n(i.e., the multimodality of the tracked variables). Our experiments show that our approach\nidentifies potential subtypes that are robust in terms of biases of self-tracked data (e.g.,\nwide variations in tracking frequency amongst participants), as well as to variations in\nhyperparameters of the model. Jointly modeling a wide range of observations about par-\nticipants (symptoms, quality of life, treatments) yields clinically meaningful subtypes that\nboth validate what is already known about endometriosis and suggest new findings.\n1. Introduction\nSmartphones and mobile applications are a powerful way to connect medical researchers\nto individuals. Recent software platforms like Researchkit and ResearchStack facilitate\nthe use of mobile technology to recruit and consent patients into studies. The first wave\nof app-based studies shows that with the right engagement techniques, patients provide\nvaluable data through their phone that can, shedding new insight into diseases Bot et al.\n(2016); Chan et al. (2017). This work contributes to the emerging area of research on\ndigital phenotyping from patient-generated data, specifically from data collected through\nsmartphone applications (Hartsell and Heller, 2017; Torous et al., 2018; Zhan et al., 2018).\nc⃝ 2018 I. Urteaga, M. McKillop, S. Lipsky-Gorman & N. Elhadad.\narXiv:1811.03431v1  [cs.CY]  6 Nov 2018\n\nPhenotyping Endometriosis\nWe focus on endometriosis, a condition for which there are no known biomarkers to help\nits diagnosis or monitor its progression, and for which subtypes proposed in the literature are\nnot well established. Using self-tracking data collected through an app designed specifically\nfor the sake of characterizing endometriosis at scale, we explore the use of unsupervised\nmethods to identify subtypes of the disease that cluster patients based on their signs and\nsymptoms, quality of life, and treatments.\nPhenotypes are important as a first step towards a better understanding of endometrio-\nsis, so that better treatment and management of patients can be achieved via phenotype-\nbased characteristics. We validate our approach through likelihood evaluations of the model\nin unseen data, clinical interpretability of identified subtypes by endometriosis experts,\npurity assessment on a subset of patient phenotype assignments against clinical experts\nclustering, and hypothesis testing against a clinically validated standard questionnaire for\npatients with endometriois.\nOur experiments show that(1) our approach identifies potential subtypes that are robust\nin terms of biases of self-tracked data (e.g., wide variations in tracking frequency amongst\nparticipants), as well as to variations in hyperparameters of the model; and (2) modeling\na wide range of observations about participants (symptoms, quality of life, treatments)\njointly yields clinically meaningful subtypes that both validate what is already known about\nendometriosis and suggest new findings.\nEndometriosis is a chronic condition estimated to affect 10% of women in reproductive\nage (Wheeler, 1989). It is traditionally described as when tissue similar to the endometrium\n(lining of the uterus) grows outside the uterine cavity, and forms lesions in the pelvic and\ngastro-intestinal areas primarily. It impacts women at a systemic level (Kvaskoff et al.,\n2015) and presents with a heavy burden of disease (Simoens et al., 2012).\nDespite its high prevalence, endometriosis continues to be a highly enigmatic condition:\nthere is lack of specificity in the range of signs and symptoms of the disease, and its clinical\ncharacterization is poor (Vercellini et al., 2007). It is understood to be heterogeneous in\nnature, and several stages of the disease have been proposed in the literature. Nevertheless,\nnone correlate with severity of symptoms experienced by patients, nor do they explain the\ndiversity of symptoms experienced. In fact, subtyping of endometriosis is currently an open\nand important research question for the endometriosis community (Johnson et al., 2017).\nThere is a need for more accurate phenotyping of endometriosis so that better, more\ntargeted treatments and management strategies can be developed. However, because en-\ndometriosis is not well understood from the cinical point of view, traditional phenotyping\napproaches that leverage electronic health record data are not appropriate. We instead turn\nto patient-generated data towards that goal.\nTechnical Significance: We contribute to the emerging area of research on unsuper-\nvised digital phenotyping from patient-generated data, specifically, by extending mixed-\nmembership models for multi-modal data collected through smartphone applications. We\nshow that by jointly modeling multiple types of self-tracking variables, we identify potential\ndisease subtypes that are robust in terms of biases of self-tracked data, and to variations in\nmodel hyperparameters.\nClinical Relevance: Endometriosis phenotypes are necessary for a better understanding\nof the disease, which will lead towards better treatment and management of patients. The\n2\n\nPhenotyping Endometriosis\nproposed unsupervised approach produces clinically relevant groupings of endometriosis\nsigns and symptoms. These phenotypes, grouped by the severity of the condition, suggest\nnovel findings about the disease, as clinically meaningful associations were identified.\n2. Data and Materials\nThe data and materials used in this work are described next. All procedures were reviewed\nand approved by our institutional review board under protocol number AAAQ9812.\n2.1. Phendo: A Smartphone App to Self-Track Endometriosis\nUsing participatory design, we designed and developed Phendo, a smartphone app for\nwomen with endometriosis to self-track their condition (McKillop et al., 2016, 2018). The\nresearch app is available for iOS 1 and Android 2 phones. Participants were recruited\nthrough patient advocacy groups and, once enrolled, can self-track a variety of variables.\nAt the moment level (i.e., as many times in the day as participants desire), they can track\ntheir pain across 39 specific body locations with 15 modifiers (e.g., “cramping” or “twisting”)\nand 3 severity levels; any of the 15 gastro-intestinal and genito-urinary issues identified\nduring design work with 3 severity levels; 21 signs and symptoms commonly identified\nby our participants (e.g., “blurry vision”, “hot flashes”, “fatigue”) and their severity; 10\npositive mood and affects and 14 negative ones, 3 bleeding patterns (“clots”, “breakthrough\nbleeding”, “spotting”), and customized medication intake (see for instance Figure 1, where\n2 Aleves were tracked at 7am, and other medications appear in the customized medication\ntracking screen).\nAt the day level, users can track a functional assessment of their day (see question “How\nwas your day?” in Figure 1) from “great” to “unbearable”, which activities of daily living\nwere hard to do (customized for endometriosis needs), menstruation patterns, customized\nanswers for diet items they want to keep track of, customized supplements, customized\nexercises, customized hormonal treatments, sexual activity and potential dyspareunia, and\na daily journal.\n2.2. Gold-standard data\nAs part of the profile tab in the Phendo research app, participants can take a standardized\nquestionnaire designed by the endometriosis research community, called WERF EPHect (Vi-\ntonis et al., 2014). The questionnaire represents the gold-standard for clinical characteriza-\ntion of endometriosis. It contains information about medical and surgical history, as well\nas quality of life related questions.\n2.3. Data preprocessing and cohort selection\nThe data collected for this study contains time-stamped responses from participants about\nall the variables tracked in the app. Because the data was collected for a wide range\nof variables at the participants’ discretion, it is heterogeneous both in type and quantity\n1. Available at https://itunes.apple.com/us/app/phendo/id1145512423\n2. Available at https://play.google.com/store/apps/details?id=com.appliedinformaticsinc.phendo\n3\n\nPhenotyping Endometriosis\nFigure 1: Example screenshots of Phendo, the self-tracking endometriosis app.\nacross participants, and possibly within a given participant’s timeline as well. As a first\nstep towards investigating phenotyping of endometriosis, we ignore the temporal aspect of\nthe condition, and rather aggregate all observations per variables tracked in the app for\neach participant.\nThe following variables from Phendo were included in this study (descriptive self-\ntracking statistics are provided in Table 1): (1) pain location, (2) pain description, (3)\npain severity, (4) gastrointestinal and genitourinary (GI/GU) symptoms, (5) their severity,\n(6) other symptoms, (7) their severity, (8) period flow, (9) bleeding patterns, (10) sexual ac-\ntivity, (11) difficult daily living activities, (12) medications including hormonal treatments,\nand (13) quality of life (“How was your day?”).\nQuestion Number of observations\n(median/mean/95%percentile/max)\nNumber of tracked days\n(median/mean/95%percentile/max)\nWhere is the pain 4/29/120/2196 1/6/31/204\nDescribe the pain 4/27/115/1657 1/6/31/204\nHow severe is the pain? 2/9/43/527 1/6/31/204\nWhat are you experiencing 1/7/33/544 1/4/17/175\nHow severe is the symptom 1/5/22/544 1/4/17/175\nDescribe the flow 0/4/16/176 0/4/16/176\nWhat kind of bleeding 0/3/15/173 0/2/12/77\nDescribe GI/GU system 1/6/28/317 1/5/20/205\nHow severe is it 1/6/27/317 1/5/20/205\nDescribe sex 0/1/3/159 0/1/3/158\nActivities 7/36/151/2148 2/7/29/266\nHow was your day? 3/12/57/458 3/12/57/458\nMedications/hormones taken 2/14/60/800 2/9/37/356\nTotal 35/159/718/6065 19/71/327/458\nTable 1: Summary statistics per-tracked question.\nWe selected a cohort of participants who had self-reported diagnosis of endometriosis,\nand had at least one entry in one of the above questions between December 2016 (launch\nof the app) and March 2018, resulting in 2,872 participants (corresponding to 456,900\nobservations total). Among them, 648 had responded to the WERF EPHect questionnaire.\n4\n\nPhenotyping Endometriosis\nThe app provides a fixed set of possible responses to most of the questions (details about\nall per-question vocabularies are provided in the appendix). Medications and hormones,\nwhich are inputted as free text in the app, were mapped to a fixed size vocabulary by\nidentifying their medication classes. The WERF EPHect questionnaire responses contained\nbinary (yes/no), categorical (e.g., number of laparoscopies) and real numbers (e.g., weight)\nanswer types. Not all questions were answered by all selected participants. We only use\nquestions with sufficient number of responses when correlating participants in obtained\nphenotypes with their responses.\n3. Methods\nThere are several challenges to address when using self-tracking data for phenotyping an\nenigmatic condition. First, because endometriosis is a heterogeneous condition and its\nclinical characterization is poor, no gold-standard phenotypes exist. As such, unsupervised\nmethods are well-suited to the task of learning phenotypes. Second, self-tracking data is\nheterogeneous (because of the many variables available for self-tracking), inherently noisy,\nirregularly spaced due to lack of participant engagement, and may reflect biases of self-\ntracking (e.g., participants may track more often when they experience symptoms of disease\nor participants may tend not to self-track when they are sick). To account for these issues,\nwe leverage unsupervised probabilistic methods, and specifically mixed-membership models.\nMixed-membership models are Bayesian generative models used to capture the latent\nstructure of collections of groups of data. Topic models are their primary example (Blei,\n2012), where one is interested in inferring the latent topics of a corpora of documents.\nTopic models analyze the statistics of observed words in each document, to capture what\nthe topics are, and what is each document’s proportion of topics (Blei et al., 2003).\nIn this work, we cast phenotyping endometriosis with self-tracking data as a probabilistic\ntopic modeling problem, by considering the set of responses per participant as “documents”,\nall coming from the “corpus” of endometriosis patients. As such, each set of observations is\nmodeled as a mixture model, where the mixture components (the phenotypes) are shared\nacross the population, but the mixture proportions (the phenotypic profile) vary per par-\nticipant. The mixed-membership model infers phenotypes based on the co-occurrence of\nobservations across the studied set of participants. That is, “topics” produced by this\ntechnique are groupings of responses to self-tracked variables that describe endometriosis\nphenotypes. These models are flexible enough to describe participants with more than one\nof these topics (i.e., mixture of phenotypes).\nThe available self-tracked data however is not a standard document, but a collection\nof responses to different questions. Therefore, we extend the mixed-membership model\nto accommodate for multi-modal data, in a similar way done by Pivovarov et al. (2015)\nfor EHR phenotyping. In our case, each modality is an specific question q = 1,··· ,Q ,\nwith its vocabulary size Vq (see section 2.3 for details). We note that subjects are free to\ntrack whatever questions they want over time and thus, data is highly unbalanced across\nparticipants and tracked variables.\nThe per-question mixed-membership generative process for each subject s = 1,··· ,S ,\nfollows:\n5\n\nPhenotyping Endometriosis\n1. Draw per-subject phenotypic proportions φs∼ DirichletK(φ|α) of dimension K with\nhyperparameter α.\n2. Draw per-phenotype and per-question response proportions θk,q∼ DirichletVq(θ|βk,q),\nfor all phenotypes k = 1,··· ,K , and questions q = 1,··· ,Q , with vocabulary size Vq\nand hyperparameters βk,q.\n3. Draw per-subject observation phenotype assignments zs,n∼ CategoricalK(z|φs), for\nn = 1,··· ,Ns.\n4. Draw per-subject questions responses xs,n|zs,n,qs,n∼ CategoricalVqs,n(x|θzs,n,qs,n), for\nn = 1,··· ,Ns, where qs,n indicates the response n to question q by subject s.\nAfter observing a dataset with Nsq responses per-subject and question, the goal is to\ninfer the phenotypic proportions φs for each participant, and the set of K phenotypes of\nthe disease, parameterized per-question by θk,q.\nTo that end, and due to the conjugacy assumptions in the generative process, we resort\nto a collapsed Gibbs sampler that utilizes the following distribution for observation n∗,\ngivenN previously “seen” data points\np(zs,n∗ =k|xs,n∗,qs,n∗,αN,βN)∝p(zs,n∗ =k|αs,N)p(xs,n∗|qs,n∗,zs,n∗,βk,q,N ),\nwith\n\n\n\np(zs,n∗ =k|αs,N) = αk,s,N∑K\nk=1αk,s,N\n,\np(xs,n∗|qs,n∗,zs,n∗,βk,q,N ) =\nβk,q,vq ,N\n∑Vq\nvq =1βk,q,vq ,N\n,\n(1)\nwhere the parameters for the updated posteriors are of the form\np(φs|ZN,α 0) = DirichletK(φs|αs,N), withαk,s,N =αk,0 +Ns,k ,\np(θk,q|XN,QN,ZN,βk,q,0) = Dirichlet(θk,q|βk,q,N ), withβk,q,vq,N =βk,q,vq,0 +Nk,q,vq .\n(2)\nNote that for the unsupervised learning of phenotypes, we only consider the self-tracked\ndata, and leave the WERF EPHect questionnaire data for evaluation purposes.\n4. Evaluation\n4.1. Experimental Setup\nWe evaluate and validate the proposed method from different perspectives.\nLikelihood of the learned model on unseen data. The quality of the proposed model\n(and chosen hyperparameters) is evaluated by held-out data log-likelihood comparisons. We\nsplit our dataset into 80/20 train/test splits, and evaluate the model learned in the training\nset with its log-likelihood in the test dataset. We note that computing the log-likelihood of\nmixed-membership models is nontrivial, as discussed in Wallach et al. (2009). The results\npresented here are based on extending the “left-to-right” method proposed by (Wallach\net al., 2009) to the per-question mixed membership model of section 3.\n6\n\nPhenotyping Endometriosis\nClinical interpretability. The enigmatic nature of endometriosis and its poor clinical\ncharacterization makes indispensable the interpretability of the model. We want to under-\nstand the specificity of signs and symptoms of the disease for each learned phenotype, as they\nare likely to be heterogeneous. For interpretability, we focus our analysis on per-phenotype\nvariable posteriors. These per-question posteriors reflect not only which responses are more\ncommonly tracked per phenotype, but also how they correlate with each other.\nTo allow for easy and visually appealing clinical evaluation, we provide both raw poste-\nrior heatmaps and most salient response wordclouds. Due to the different support size for\neach considered question, we plot wordclouds conditioned on the vocabulary items that cover\n80% of the posterior mass. The maximum font size is fixed equal for all phenotypes within a\nquestion, and the relative size of responses follows the proportions of the conditional prob-\nability ratios. This allows for a more clear identification of the most salient responses and\na principled way of comparing different sized-vocabularies. Both the heatmaps and word\nclouds of the posterior variables per phenotypes were provided to endometriosis experts for\nreview.\nAgreement between expert clustering and phenotyping. We randomly selected 30\nparticipants from the cohort, who had at least 30 days of tracked data and had high posterior\nprobability (above 95% percent) of being assigned to a unique phenotype (10 participants\nper phenotype) in the learned model. Note that participants to be evaluated by clinical\nexperts were selected randomly within each subtype, based on the model’s output only, and\nthat experts were not involved in this selection process at all. The responses collected from\nthese patients were reviewed by two clinical experts, who were asked to group them into\nthree clusters based on their clinical understanding of patient signs and symptoms. The\nassignments by the experts and the model are compared, via confusion matrices and cluster\npurity metrics.\nAssociations with standardized questionnaire. We selected a subset of questions\nfrom the WERF EPHect questionnaire for validation purposes, which included responses\nabout family history of endometriosis, pelvic pain, menstrual characteristics, surgical pro-\ncedures, comorbidities, and activities of daily living, as well as general health indicators.\nWe perform association tests between participants assigned to each phenotype (hard clus-\ntering based on learned per-participant phenotypic posteriors) and their responses to the\nquestions of interest, and report correlations that are significant. Specifically, categorical\nquestion answers were collapsed to ‘yes’ or ‘no’ outcomes for each cluster, producing 2 × 2\ncontingency tables. Fisher’s exact test (Fisher, 1922) was used for each contingency table\nto produce an odds ratio, which represents the likelihood of saying ‘yes’ to a particular\nquestion and being in a particular cluster, versus the odds of saying ‘no’ to that question\nand not being in a particular cluster. The average for answers with continuous outcome\nquestions was computed for each cluster, and compared to the mean of those not in that\nparticular cluster, using Welchs t-test for unequal variances (Ruxton, 2006). Significance\nfor Fisher’s exact tests and Welchs t-tests was determined at the 0 .05 level.\n4.2. Results\nModel Likelihood. We study the proposed model’s phenotyping accuracy via 5-fold\nlog-likelihood evaluation. As shown in Figure 2, there is a significant improvement of\n7\n\nPhenotyping Endometriosis\nour method when compared to vanilla LDA, where responses to all questions are modeled\ntogether as a bag-of-words. As we allow for per-question modalities, our method is capable\nof capturing discriminative signals in each of the variables, thus distinguishing between\nphenotypes (Figure 4). We emphasize the robustness of the inference mechanism with\nrespect to particular choices of hyperparameters. There is a slight performance improvement\nfor sparse phenotypes, which we further take advantage of for interpretability purposes.\nK = 2, = 0.1, = 0.1\nK = 2, = 0.1, = 0.01\nK = 2, = 0.1, = 0.001\nK = 2, = 0.01, = 0.1\nK = 2, = 0.01, = 0.01\nK = 2, = 0.01, = 0.001\nK = 2, = 0.001, = 0.1\nK = 2, = 0.001, = 0.01\nK = 2, = 0.001, = 0.001\nK = 3, = 0.1, = 0.1\nK = 3, = 0.1, = 0.01\nK = 3, = 0.1, = 0.001\nK = 3, = 0.01, = 0.1\nK = 3, = 0.01, = 0.01\nK = 3, = 0.01, = 0.001\nK = 3, = 0.001, = 0.1\nK = 3, = 0.001, = 0.01\nK = 3, = 0.001, = 0.001\nK = 4, = 0.1, = 0.1\nK = 4, = 0.1, = 0.01\nK = 4, = 0.1, = 0.001\nK = 4, = 0.01, = 0.1\nK = 4, = 0.01, = 0.01\nK = 4, = 0.01, = 0.001\nK = 4, = 0.001, = 0.1\nK = 4, = 0.001, = 0.01\nK = 4, = 0.001, = 0.001\nK = 5, = 0.1, = 0.1\nK = 5, = 0.1, = 0.01\nK = 5, = 0.1, = 0.001\nK = 5, = 0.01, = 0.1\nK = 5, = 0.01, = 0.01\nK = 5, = 0.01, = 0.001\nK = 5, = 0.001, = 0.1\nK = 5, = 0.001, = 0.01\nK = 5, = 0.001, = 0.001\n450000\n400000\n350000\n300000\n250000\n200000\nlogp()\nFigure 2: 5-fold test data log-likelihood of the proposed method (in red) Vs LDA (in blue).\nInterpretability. In this experiment and those to follow, we focus on a selected model\nwith 3 phenotypes (as models with more subtypes did not capture new discriminating in-\nsights) and sparse parameters (α =β = 0.001). The sparsity of the model allows for (i) few\nvocabulary items per-question being distinctive for each phenotype, and (ii) having discrim-\ninative per-participant phenotypic profiles (i.e., located on the vertices of the probability\nsimplex and thus strongly clustered).\nThe proposed model is clinically useful in that, for representing a participant, it allows\nfor the tracked symptoms to be explained by a mixture of the learned phenotypes. We\nshow the phenotypic assignments of participants to each learned endometriosis subtype in\nFigure 3, and note that they do not correlate with the number of days (or observations)\nparticipants tracked. Although participants in all phenotypes have tracked similar number\nof days (39, 43 and 46 on average), participants associated with phenotype 0 have tracked\nmore observations (on average, 126, 80 and 80, respectively).\n8\n\nPhenotyping Endometriosis\n0 1 2\nk\n0\n500\n1000\n1500\n2000\n2500s\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0\nFigure 3: Learned per-participant posterior phenotypic distribution.\nWe now elaborate on the specificity of signs and symptoms of the disease for each\nlearned phenotype, based on learned per-question posteriors (Figure 4), and their word-\ncloud visualizations (Figures 5 – 17).\nPain is shared across all phenotypes, and its location is most often tracked for pelvic\nand lower back areas (Figure 5). The commonality of these pain locations is expected, given\nthat lower back and pelvic pain are stereotypical for endometriosis (Chiantera et al., 2017).\nPhenotype 0, however, has a much wider range of locations in which pain is experienced,\ncharacterized by reports of deep vagina pain, rectal pain, and pain projecting along the legs\n(Figure 5). Similarly, certain descriptions of pain are common across phenotypes, such as\ndeep, aching, and cramping (Figure 6). However, phenotype 0 has a wider range of descrip-\ntions of how pain is experienced, besides pain being predominantly characterized as severe\n(Figure 7). These characterizations of pain across phenotypes indicate that the burden on\npain is much heavier and severe for phenotype 0, while moderate or mild descriptors are\nassociated with phenotypes 1 and 2 (this trend is consistent for the majority of questions).\nEach phenotype had strong involvement of gastrointestinal symptoms and was char-\nacterized by “endo belly” across phenotypes (Figure 8). This is a very common disease\nsymptom where the abdomen severely bloats (Ek et al., 2015; Luscombe et al., 2009). Phe-\nnotype 0 is distinctively associated with genitourinary symptoms, like painful and frequent\nurination or dysuria. While genitourinary symptoms in endometriosis are known, their\nassociation with a subgroup of patients is novel (Denny and Mann, 2007a).\nAs for menstrual characteristics (visualized in Figure 10), phenotype 0 reports heavier\nflow or menorrhagia. Menorrhagia is a common endometriosis symptom but has not been\nassociated with a particular subgroup of endometriosis patients (Vercellini et al., 1997).\nMoreover, although all phenotypes have tracked spotting or bleeding outside of the pe-\n9\n\nPhenotyping Endometriosis\nriod (Figure 11), phenotype 0 is also likely to track clots, consistent with descriptions of\nmenorrhagia (Warner et al., 2004).\nPainful sex or dyspareunia is a wieldy known symptom for endometriosis (Denny and\nMann, 2007b). As shown in Figure 12, our model learned that all phenotypes avoided sex.\nFurthermore, phenotype 0 is distinguished by the active avoidance of sex, and does not\nexperience satisfying sex.\nGeneral quality of life, as measured by the “How was your day?” question and tracked\ndaily activities visualized in Figures 16 and 17 respectively, show similar severe, moderate,\nand mild patterns for each phenotype. All phenotypes have manageable days, though bad\ndays are more present for phenotype 0. Even if there are common difficulties associated\nwith activities of daily living across phenotypes, phenotype 0 has a wider range of tracked\nproblems.\nMedications and hormones clearly discriminate the patients by severity of disease, as\nshown in Figure 13. Phenotype 2 does not take medications, or may take a combination of\nhormonal medications like birth control pills (BCPs). BCPs are often used as a first line\ntreatment for endometriosis symptoms (Schrager et al., 2013). Phenotype 1 takes BCPs\nas well, but is further characterized by its use of analgesics. In stark contrast, phenotype\n0 has heavy medication use, taking anti-depressants and strong pain medications,including\nnarcotics, neuropathic pain medications and opioids.\nOther symptoms of endometriosis, which were collected via the “What else are you\nexperiencing?” question shown in Figure 14, reflect the chronic nature of endometriosis.\nFatigue, mental fogginess, and headache occur across all phenotypes and are similar to\nother complex chronic conditions like chronic fatigue syndrome. These symptoms are also\ncharacteristic of low grade inflammation (Holgate et al., 2011; Louati and Berenbaum, 2015).\n10\n\nPhenotyping Endometriosis\n0 1 2\nk\nbones_paincervix_paindeep_vagina_paindiaphragm_painhead_paininner_thighs_painintestines_painjoints_painleft_arm_painleft_breast_painleft_leg_painleft_lower_back_painleft_outer_hip_painleft_ovary_painleft_pelvis_painleft_ribs_painleft_shoulder_painleft_side_abdomen_painlegs_painlower_back_painlower_chest_painneck_painpelvis_painrectum_painright_arm_painright_breast_painright_leg_painright_lower_back_painright_outer_hip_painright_ovary_painright_pelvis_painright_ribs_painright_shoulder_painright_side_abdomen_painupper_abdomen_painupper_chest_painuterus_painvagina_entrance_painwhole_abdomen_pain\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0\n(a) Where is the pain.\n0 1 2\nk\naching_pain\nburning_pain\ncramping_pain\ndeep_pain\ndull_pain\nnauseating_pain\npressure_pain\npulling_pain\npulsating_pain\nradiating_pain\nsharp_pain\nshooting_pain\nstabbing_pain\nthrobbing_pain\ntwisting_pain\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0 (b) Describe the pain.\n0 1 2\nk\nmild_pain\nmoderate_pain\nsevere_pain\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0 (c) How severe is the pain?\n0 1 2\nk\nblood_in_stool\ncant_urinate\nconstipation\ndiarrhea\nendo_belly\nfrequent_urination\ngas\nheartburn\nmouth_sores\nnausea\npainful_bowel_movement\npainful_urination\nstomach_upset\nuncomfortably_full\nvomiting\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0\n(d) Describe GI/GU system.\n0 1 2\nk\nmild_GI\nmoderate_GI\nsevere_GI\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0 (e) How severe is it.\n0 1 2\nk\nheavy_flow\nlight_flow\nmedium_flow\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0 (f) Describe the flow.\n0 1 2\nk\nbreakthrough_bleeding\nclots\nno_bleeding\nspotting\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0\n(g) What kind of bleeding.\n0 1 2\nk\navoided_sex\nbleeding_from_sex\nno_sex\npainful_after_sex\npainful_during_sex\nsex_felt_good\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0 (h) Describe sex.\n0 1 2\nk\nno_med_hormones\nestrogen/progestin\nnsaids\nnoclass\nprogestin\nanalgesic\nantidepressant\nanalgesic/narcotic\nnarcotic\nneuropathic_pain_medication\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0 (i) Medications/hormones taken.\n0 1 2\nk\nallergies\nasthma\nblurry_vision\nchest_pressure\ndizziness\neczema\nfatigue\nfever\nheadache\nhives\nhot_flash\nitchy\nmentally_foggy\nnoise_sensitivity\nnumbness\nrash\nringing_in_ears\nsinus_congestion\nsweaty\nswelling\ntouch_sensitivity\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0\n(j) What are you experiencing.\n0 1 2\nk\nmild_symptoms\nmoderate_symptoms\nsevere_symptoms\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0 (k) How severe is the symptom.\n0 1 2\nk\nbad_day\ngood_day\ngreat_day\nmanageable_day\nunbearable_day\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0\n(l) How was your day?\n0 1 2\nk\nclimb_stairs\neat\nget_dressed\nget_out_of_bed\nhave_sex\nhousework\njump\nkneel\nlie_down\nlift\nno_trouble\nprepare_food\nrun\nshop\nshower\nsit_down\nsleep\nsocialize\nstand\nstretch\nuse_toilet\nwalk\nwork\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0 (m) Activities\nFigure 4: Visualization of per-question posteriors for learned endometriosis phenotypes.\n11\n\nPhenotyping Endometriosis\n(a) Phenotype k = 0\n (b) Phenotype k = 1\n (c) Phenotype k = 2\nFigure 5: Where is the pain.\n(a) Phenotype k = 0\n (b) Phenotype k = 1\n (c) Phenotype k = 2\nFigure 6: Describe the pain.\n(a) Phenotype k = 0\n (b) Phenotype k = 1\n (c) Phenotype k = 2\nFigure 7: How severe is the pain.\n12\n\nPhenotyping Endometriosis\n(a) Phenotype k = 0\n (b) Phenotype k = 1\n (c) Phenotype k = 2\nFigure 8: Describe GI/GU system.\n(a) Phenotype k = 0\n (b) Phenotype k = 1\n (c) Phenotype k = 2\nFigure 9: How severe is it.\n(a) Phenotype k = 0\n (b) Phenotype k = 1\n (c) Phenotype k = 2\nFigure 10: Describe the flow.\n(a) Phenotype k = 0\n (b) Phenotype k = 1\n (c) Phenotype k = 2\nFigure 11: What kind of bleeding.\n(a) Phenotype k = 0\n (b) Phenotype k = 1\n (c) Phenotype k = 2\nFigure 12: Describe sex.\n13\n\nPhenotyping Endometriosis\n(a) Phenotype k = 0\n (b) Phenotype k = 1\n (c) Phenotype k = 2\nFigure 13: Medications/hormones taken.\n(a) Phenotype k = 0\n (b) Phenotype k = 1\n (c) Phenotype k = 2\nFigure 14: What are you experiencing.\n(a) Phenotype k = 0\n (b) Phenotype k = 1\n (c) Phenotype k = 2\nFigure 15: How severe is the symptom.\n14\n\nPhenotyping Endometriosis\n(a) Phenotype k = 0\n (b) Phenotype k = 1\n (c) Phenotype k = 2\nFigure 16: How was your day?\n(a) Phenotype k = 0\n (b) Phenotype k = 1\n (c) Phenotype k = 2\nFigure 17: Activities\nClustering agreement. Table 2 shows the agreement between the learned model’s pos-\nterior assignments and those determined by two different experts, with purity values of 0 .6\nand 0.73 for the first and second experts, respectively. The expert assignments to model’s\nphenotype 0 (the severe case) were the most accurate, while deciding between the other\ntwo phenotypes was deemed to be harder. As a matter of fact, both experts raised several\nconcerns about needing more data to make more informed decisions, as they had difficulties\ndistinguishing between non-severe endometriosis cases.\nPhenotype Expert 1\n0 1 2\n0 8 2 0\n1 3 5 2\n2 3 2 5\nModel\nPhenotype Expert 2\n0 1 2\n0 7 1 2\n1 1 8 1\n2 0 7 3\nModel\nTable 2: Phenotype confusion matrices.\nConsequently, both experts were asked to discriminate between severe and non-severe\ncases, for which results are shown in Table 3, with purities of 0 .73 and 0.87 respectively.\nAssociation with gold-standard questionnaire. We now summarize our results on\nhow the learned phenotypes associate with responses to the WERF EPHect questionnaire.\nIn general, severity and quality of life indicators of endometriosis as specified by WERF\nstandards align well with how our model discriminates patients.\n15\n\nPhenotyping Endometriosis\nSevere case Expert 1\nYes No\nYes 8 2\nNo 6 14Model\nSevere case Expert 1\nYes No\nYes 7 3\nNo 1 19Model\nTable 3: Severe case confusion matrices.\nOverall, phenotype 0 (severe phenotype) is associated with a heavier burden of dis-\nease. It shows correlation with a number of comorbidities such as anxiety (odds ratio,\nOR=1.62), depression or mood disorders (OR=1.57), migraine (OR=2.27), high blood pres-\nsure (OR=1.69), and polycystic ovary syndrome PCOS (OR=1.81). It is also associated\nwith painful bladder problems like interstitial cystitis (OR=2.28), which did not occur in\npatients in other phenotypes. Symptoms of these comorbidities are salient in the obtained\nposteriors (see for example, how the severe phenotype is characterized by painful urina-\ntion). Furthermore, women with endometriosis are known to exhibit these comorbidities:\nanxiety, depression, and other mood disorders (Pope et al., 2015), migraines (Yang et al.,\n2012), high blood pressure (Mu et al., 2017), PCOS (Holoch et al., 2014), and painful\nurination/interstitial cystitis (Chung et al., 2005).\nParticipants assigned to phenotype 0 also show an important reduction in quality of life,\nas compared to those in other phenotypes. Patients within this subtype were 3.25 more likely\nto rate their health as poor. Patients within the severe phenotype are more likely (OR=1.70)\nto report problems regarding going to work or carrying out activities of daily living as\nwell. These participants had also higher odds of experiencing limitations with vigorous\nactivities such as running (OR=2.06), moderate activities like housework (OR=3.45), lifting\n(OR=3.86), climbing even one flight of stairs (OR=4.42), bending (OR=3.52), walking just\none block (OR=5.47), dressing, as well as bathing (OR=10.67). Problems with productivity\nand reduced quality of life among endometriosis patients has been previously reported\n(Shabanov et al., 2017).\nThe surgical burden of the disease is most evident for those subjects assigned to pheno-\ntype 0. Participants in this group had significantly higher number of laparoscopies (mean\n1.67) as compared to those in phenotypes 1 and 2 (mean 1.36). Consistent with the ex-\nisting literature (Ballard et al., 2006), patients also reported seeing several doctors before\ndiagnosis, with phenotype 0 being associated with the highest number of doctors seen\n(mean=6.16), versus those in other phenotypes (mean=4.59).\nSeveral other associations were consistent with current disease knowledge. The severe\nphenotype had 10.86 times higher odds of having mild endometriosis (Stage 2) as compared\nto phenotypes 1 and 2, and the mild phenotype had 0.48 times lower odds of having a surgery\nwhere no endometriosis was found as compared to phenotypes 0 (severe) and 1 (moderate).\nWhile these results might seem counterintuitive, they confirm the lack of correlation between\npatient experience and existing staging of disease as discussed in the literature (Vercellini\net al., 2007). In addition, the odds of having an extremely regular period were 0.47 times\nlower in phenotype 0 (severe phenotype) compared to others. Such menstrual irregularity\nhas been shown to be associated with endometriosis before (Signorello et al., 1997).\n16\n\nPhenotyping Endometriosis\nFinally, associations were also found that may indicate differences in the etiology of the\ndisease among the different phenotypes. Those in phenotype 2 had 2.50 times higher odds\nof having a sister with endometriosis than others. Many different causes of endometriosis\nhave been proposed including heritable tendencies, but the exact etiology remains unclear\n(Cramer and Missmer, 2004). The set of proposed underlying causes of the diseases high-\nlights the need for reducing the heterogeneity of clinical presentation of symptoms, and our\napproach seems promising for reducing this heterogeneity. Finally, we found no significant\ncorrelations between phenotypes and age, race, or time-to-diagnosis.\n5. Conclusion\nThis paper contributes to research in digital phenotyping from self-tracking data. Our joint\nmodeling of multiple types of self-tracking variables through mixed-membership models\nshow that we can produce robust, clinically meaningful groupings of self-tracked variables.\nThese phenotypes, along with participant clustering, suggest novel findings about disease.\nIn the case of endometriosis, a particularly enigmatic condition with a dire need for phe-\nnotyping and subtyping, our methods identified three clusters of patients roughly grouped\nby the severity of their condition. Further, clinically meaningful novel associations beyond\nwhat is currently known about the disease were identified. Endometriosis phenotypes are\nnecessary as a first step towards a better understanding of the pathophysiologic mechanisms\nof the disease, which will lead towards better treatment and management of patients based\non learned phenotypic characteristics.\nFuture work should include modeling the temporality of signs and symptoms of en-\ndometriosis, particularly since it is estrogen dependent and thus linked to the menstrual\ncycle. Nevertheless, the analysis in this study already shed novel insight and demonstrates\nthe value of patient-generated data in medical research.\nAcknowledgments\nThe authors thank the study participants and acknowledge the many endometriosis advo-\ncacy groups that helped in recruiting them. This work was supported by the Endometriosis\nFoundation of America, the National Science Foundation award SCH 1344668, and the Na-\ntional Library of Medicine award T15 LM007079. We also thank Dr. Shadi Safar Gholi and\nDr. Arnold Advincula for their time and expertise.\nReferences\nKaren Ballard, Karen Lowton, and Jeremy Wright. What’s the delay? A qualitative study\nof women’s experiences of reaching a diagnosis of endometriosis. Fertility and sterility ,\n86(5):1296–1301, 2006.\nDavid M. Blei. Probabilistic Topic Models. Communications of the ACM , 55(4):77–84,\nApril 2012. ISSN 0001-0782.\nDavid M Blei, Andrew Y Ng, and Michael I Jordan. Latent Dirichlet allocation. Journal\nof machine Learning research, 3(Jan):993–1022, 2003.\n17\n\nPhenotyping Endometriosis\nBrian M Bot, Christine Suver, Elias Chaibub Neto, Michael Kellen, Arno Klein, Christopher\nBare, Megan Doerr, Abhishek Pratap, John Wilbanks, E Ray Dorsey, et al. The mPower\nstudy, Parkinson disease mobile data collected using ResearchKit. Scientific data , 3:\n160011, 2016.\nYu-Feng Yvonne Chan, Pei Wang, Linda Rogers, Nicole Tignor, Micol Zweig, Steven G\nHershman, Nicholas Genes, Erick R Scott, Eric Krock, Marcus Badgeley, et al. The\nAsthma Mobile Health Study, a large-scale clinical observational study using ResearchKit.\nNature biotechnology, 35(4):354, 2017.\nVito Chiantera, Elene Abesadze, and Sylvia Mechsner. How to understand the complexity\nof endometriosis-related pain. Journal of Endometriosis and Pelvic Pain Disorders , 9(1):\n30–38, 2017.\nMaurice K Chung, Rosemary P Chung, and David Gordon. Interstitial cystitis and en-\ndometriosis in patients with chronic pelvic pain: The “Evil Twins” Syndrome. JSLS:\nJournal of the Society of Laparoendoscopic Surgeons , 9(1):25, 2005.\nDaniel W Cramer and Stacey A Missmer. Epidemiology of endometriosis. In Endometriosis\nin Clinical Practice, pages 79–94. CRC Press, 2004.\nElaine Denny and Christopher H. Mann. A clinical overview of endometriosis: a misunder-\nstood disease. British journal of nursing , 16(18):1112–1116, 2007a.\nElaine Denny and Christopher H. Mann. Endometriosis-associated dyspareunia: the impact\non women’s lives. BMJ Sexual & Reproductive Health , 33(3):189–193, 2007b.\nMalin Ek, Bodil Roth, Per Ekstr¨ om, Lil Valentin, Mariette Bengtsson, and Bodil Ohlsson.\nGastrointestinal symptoms among endometriosis patient’s: A case-cohort study. BMC\nWomen’s Health, 15(1):59, 2015.\nRonald A Fisher. On the interpretation of χ2 from contingency tables, and the calculation\nof P. Journal of the Royal Statistical Society , 85(1):87–94, 1922.\nL Hartsell and K Heller. Preliminary fatigue subtype discovery from the MS mosaic study.\nIn Multiple Sclerosis Journal , volume 23, pages 734–735, 2017.\nStephen T Holgate, Anthony L Komaroff, Dennis Mangan, and Simon Wessely. Chronic\nfatigue syndrome: understanding a complex illness. Nature Reviews Neuroscience, 12(9):\n539, 2011.\nKristin J Holoch, Ricardo F Savaris, David A Forstein, Paul B Miller, H Lee Higdon III,\nCreighton E Likes, and Bruce A Lessey. Coexistence of polycystic ovary syndrome and\nendometriosis in women with infertility. Journal of Endometriosis and Pelvic Pain Dis-\norders, 6(2):79–83, 2014.\nNeil P Johnson, Lone Hummelshoj, G David Adamson, J¨ org Keckstein, Hugh S Taylor,\nMauricio S Abrao, Deborah Bush, Ludwig Kiesel, Rulla Tamimi, Kathy L Sharpe-Timms,\net al. World Endometriosis Society consensus on the classification of endometriosis. Hu-\nman Reproduction, 32(2):315–324, 2017.\n18\n\nPhenotyping Endometriosis\nMarina Kvaskoff, Fan Mu, Kathryn L Terry, Holly R Harris, Elizabeth M Poole, Leslie\nFarland, and Stacey A Missmer. Endometriosis: a high-risk population for major chronic\ndiseases? Human reproduction update, 21(4):500–516, 2015.\nKarine Louati and Francis Berenbaum. Fatigue in chronic inflammation-a link to pain\npathways. Arthritis research & therapy, 17(1):254, 2015.\nGeorgina M Luscombe, Robert Markham, Mirari Judio, Ariadna Grigoriu, and Ian S Fraser.\nAbdominal bloating: an under-recognized endometriosis symptom. Journal of Obstetrics\nand Gynaecology Canada, 31(12):1159–1171, 2009.\nMollie McKillop, Natalie Voigt, Rebecca Schnall, and No´ emie Elhadad. Exploring self-\ntracking as a participatory research activity among women with endometriosis. Journal\nof Participatory Medicine, 2016.\nMollie McKillop, Lena Mamykina, and No´ emie Elhadad. Designing in the Dark: Eliciting\nSelf-tracking Dimensions for Understanding Enigmatic Disease. In Proceedings of the\n2018 CHI Conference on Human Factors in Computing Systems , page 565. ACM, 2018.\nFan Mu, Janet Rich-Edwards, Eric B Rimm, Donna Spiegelman, John P Forman, and\nStacey A Missmer. Association Between Endometriosis and Hypercholesterolemia or\nHypertensionNovelty and Significance. Hypertension, 70(1):59–65, 2017.\nRimma Pivovarov, Adler J. Perotte, Edouard Grave, John Angiolillo, Chris H. Wiggins,\nand Nomie Elhadad. Learning probabilistic phenotypes from heterogeneous EHR data.\nJournal of Biomedical Informatics , 58:156 – 165, 2015. ISSN 1532-0464.\nCarley J Pope, Verinder Sharma, Sapna Sharma, and Dwight Mazmanian. A systematic\nreview of the association between psychiatric disturbances and endometriosis. Journal of\nObstetrics and Gynaecology Canada , 37(11):1006–1015, 2015.\nGraeme D Ruxton. The unequal variance t-test is an underused alternative to Student’s\nt-test and the Mann–Whitney U test. Behavioral Ecology, 17(4):688–690, 2006.\nSarina Schrager, Julianne Falleroni, and Jennifer Edgoose. Evaluation and treatment of\nendometriosis. Am Fam Physician, 87(2):107–113, 2013.\nS Shabanov, JM Wenger, S Seidler, M Bolmont, F Bianchi-Demicheli, and N Pluchino.\nWhen sex hurts the couple: the case of endometriosis. Revue medicale suisse , 13(554):\n612–616, 2017.\nLisa B Signorello, Bernard L Harlow, Daniel W Cramer, Donna Spiegelman, and Joseph A\nHill. Epidemiologic determinants of endometriosis: a hospital-based case-control study.\nAnnals of epidemiology, 7(4):267–274, 1997.\nSteven Simoens, Gerard Dunselman, Carmen Dirksen, Lone Hummelshoj, Attila Bokor, Iris\nBrandes, Valentin Brodszky, Michel Canis, Giorgio Lorenzo Colombo, Thomas DeLeire,\net al. The burden of endometriosis: costs and quality of life of women with endometriosis\nand treated in referral centres. Human Reproduction, 27(5):1292–1299, 2012.\n19\n\nPhenotyping Endometriosis\nJohn Torous, Patrick Staples, Ian Barnett, Luis R Sandoval, Matcheri Keshavan, and Jukka-\nPekka Onnela. Characterizing the clinical relevance of digital phenotyping data quality\nwith applications to a cohort with schizophrenia. npj Digital Medicine , 1(1):15, 2018.\nP. Vercellini, L. Fedele, G. Aimi, G. Pietropaolo, D. Consonni, and P.G. Crosignani. As-\nsociation between endometriosis stage, lesion type, patient characteristics and severity of\npelvic pain symptoms: a multivariate analysis of over 1000 patients. Human Reproduc-\ntion, 22(1):266–271, 2007.\nPaolo Vercellini, Olga De Giorgi, Giorgio Aimi, Stefania Panazza, Anna Uglietti, and\nPier Giorgio Crosignani. Menstrual characteristics in women with and without en-\ndometriosis. Obstetrics & Gynecology, 90(2):264–268, 1997.\nAllison F Vitonis, Katy Vincent, Nilufer Rahmioglu, Amelie Fassbender, Germaine M Buck\nLouis, Lone Hummelshoj, Linda C Giudice, Pamela Stratton, G David Adamson, Chris-\ntian M Becker, et al. World Endometriosis Research Foundation Endometriosis Phenome\nand biobanking harmonization project: II. Clinical and covariate phenotype data collec-\ntion in endometriosis research. Fertility and sterility, 102(5):1223–1232, 2014.\nHanna M Wallach, Iain Murray, Ruslan Salakhutdinov, and David Mimno. Evaluation\nMethods for Topic Models. In Proceedings of the 26th Annual International Conference\non Machine Learning , ICML ’09, pages 1105–1112, New York, NY, USA, 2009. ACM.\nISBN 978-1-60558-516-1.\nPamela E Warner, Hilary OD Critchley, Mary Ann Lumsden, Mary Campbell-Brown, Anne\nDouglas, and Gordon D Murray. Menorrhagia I: measured blood loss, clinical features,\nand outcome in women with heavy periods: a survey with follow-up data. American\nJournal of Obstetrics & Gynecology , 190(5):1216–1223, 2004.\nJM Wheeler. Epidemiology of endometriosis-associated infertility. The Journal of repro-\nductive medicine, 34(1):41–46, 1989.\nMeng-Han Yang, Peng-Hui Wang, Shuu-Jiun Wang, Wei-Zen Sun, Yen-Jen Oyang, and\nJong-Ling Fuh. Women with endometriosis are more likely to suffer from migraines: a\npopulation-based study. PLoS One, 7(3):e33941, 2012.\nAndong Zhan, Srihari Mohan, Christopher Tarolli, Ruth B Schneider, Jamie L Adams,\nSaloni Sharma, Molly J Elson, Kelsey L Spear, Alistair M Glidden, Max A Little, et al.\nUsing Smartphones and Machine Learning to Quantify Parkinson Disease Severity: The\nMobile Parkinson Disease Score. JAMA neurology, 2018.\n20\n\nPhenotyping Endometriosis\nAppendix A.\nVocabulary for each tracked question:\n• Where is the pain : bones pain, cervix pain, deep vagina pain, diaphragm pain,\nhead pain, inner thighs pain, intestines pain, joints pain, left arm pain, left breast pain,\nleft leg pain, left lower back pain, left outer hip pain, left ovary pain, left pelvis pain,\nleft ribs pain, left shoulder pain, left side abdomen pain, legs pain, lower back pain,\nlower chest pain, neck pain, pelvis pain, rectum pain, right arm pain, right breast pain,\nright leg pain, right lower back pain, right outer hip pain, right ovary pain, right pelvis pain,\nright ribs pain, right shoulder pain, right side abdomen pain, upper abdomen pain,\nupper chest pain, uterus pain, vagina entrance pain, whole abdomen pain\n• Describe the pain: aching pain, burning pain, cramping pain, deep pain, dull pain,\nnauseating pain, pressure pain, pulling pain, pulsating pain, radiating pain, sharp pain,\nshooting pain, stabbing pain, throbbing pain, twisting pain\n• How severe is the pain? : mild pain, moderate pain, severe pain\n• What are you experiencing: allergies, asthma, blurry vision, chest pressure, dizzi-\nness, eczema, fatigue, fever, headache, hives, hotflash, itchy, mentally foggy, noise sensitivity,\nnumbness, rash, ringing in ears, sinus congestion, sweaty, swelling, touch sensitivity\n• How severe is the symptom: mild symptoms, moderate symptoms, severe symptoms\n• Describe the flow : heavy flow, light flow, medium flow\n• What kind of bleeding : breakthrough bleeding, clots, no bleeding, spotting\n• Describe GI/GU system : blood in stool, cant urinate, constipation, diarrhea,\nendo belly, frequent urination, gas, heartburn, mouth sores, nausea, painful bowel movement,\npainful urination, stomach upset, uncomfortably full, vomiting\n• How severe is it: mild GI, moderate GI, severe GI\n• Describe sex: avoided sex, bleeding from sex, no sex, painful after sex, painful during sex,\nsex felt good\n• Activities: climb stairs, eat, get dressed, get out of bed, have sex, housework, jump,\nkneel, lie down, lift, no trouble, prepare food, run, shop, shower, sit down, sleep,\nsocialize, stand, stretch, use toilet, walk, work\n• How was your day?: bad day, good day, great day, manageable day, unbearable day\n• Medications/hormones taken: adrenergic agonists, amphetamine, analgesic, anal-\ngesic/narcotic, analgesic/nsaids, analgesic/opioids, anesthetic, anorectic, anti-inflammatory,\nantiacid, antiacid/nsaids, antibiotics, anticholinergic, anticoagulant, anticonvulsant,\nantidepressant, antidiabetic medication, antidiarrheal, antiemetic, antihistamine, an-\ntihypertensive, antipsychotic, antispasmodic, antispasmodic/sedative, anxiolytic, anx-\niolytic/anesthetic/muscle relaxant, barbituate, barbituate/analgesic, beta blocker, bron-\nchodilator, calcium channel blocker, cough medicine, decongestant, diuretic, dopamine agonist,\n21\n\nPhenotyping Endometriosis\nestrogen, estrogen/progestin, gonadotropin-releasing hormone agonist, gonadotropin-\nreleasing hormone antagonist, hormone based chemotherapy, hormone replacement therapy,\nhuman chorionic gonadotropin, human follicle stimulating hormone, laxative, muscle relaxant,\nnarcotic, narcotic/nsaids, neuropathic pain medication, no med hormones, noclass,\nnonbenzodiazepine hypnotic, nsaids, opioids, progestin, sedative, statin, steroid, stim-\nulant, thyroid hormones, topical anti-tumor medication, triptan, vasoconstrictor, vi-\ntamin a derivative\n22","source_license":"CC0","license_restricted":false}