Subgroups of pelvic pain are differentially associated with endometriosis and inflammatory comorbidities: a latent class analysis

Cross-sectional data used in the present study were collected after enrollment in The Women’s Health Study: From Adolescence to Adulthood (A2A) (N=1255), a prospective cohort study that enrolled females aged 7–55 years between November 2012 to June 2018 (A2A consists of N=1549 females, of which 81% met the inclusion criteria for this study per Supplemental Figure 1 ). As described in previous studies [ 47 – 50 , 52 ], participants were recruited from two tertiary care centers and surrounding communities in Boston, Massachusetts, USA using in-clinic eligible patient identification and hospital-catchment community advertising and word of mouth [ 15 ]. The population was oversampled for adolescents and individuals with surgically-diagnosed endometriosis. Study participants completed a questionnaire upon enrollment on socio-demographic (e.g., age, race and ethnicity) and clinical characteristics (e.g., age at menarche, weight, medication use) that was followed up with yearly questionnaires. Cyclic and acyclic pelvic pain were assessed using questions from an expanded form of the validated World Endometriosis Research Foundation (WERF) Endometriosis Phenome and Biobanking Harmonization Project (EPHect) Endometriosis Patient Questionnaire (EPQ) [ 62 ]. The initial version of the A2A baseline questionnaire assessed demographics, body mass index (BMI), physical activity, diet, smoking, alcohol consumption, reproductive factors, and other medical conditions as well as details on pain symptoms, treatment regimen, and medication use. In January 2014, an expanded version of the WERF EPHect Patient Questionnaire [ 43 ] was adopted for use at baseline; there was very little change in the questionnaire with the vast majority of the questions being the same. For those <18 years of age at enrollment, parental consent and participant assent was obtained, and for those ≥18 years of age written informed consent was obtained. This study was approved by the Michigan State University Institutional Review Board and Boston Children’s Hospital (BCH) for both BCH and Brigham and Women’s Hospital. Six pelvic pain variables from the questionnaire were used as indicators, defined as observed variables that are used to derive latent subgroups. Indicators for severity, frequency, and life impact were collected for both acyclic and cyclic pelvic pain. Details of pelvic pain indicators are provided in Supplemental Table 1 . Acyclic pelvic pain was defined as worst pelvic/lower abdominal pain not caused by menstrual cramps, intercourse, surgery, pregnancy, or other injury and infections. Its severity and frequency were measured over the last 3 months (survey from January 2014 onward) or last 12 months (survey prior to January 2014). Life impact was measured over the last 3 months. Cyclic pelvic pain was defined as dysmenorrhea or cramping, shooting, or stabbing pain that occurs during menses in the past 12 months. Continuous numerical rating scale (NRS) variables associated with severity for both acyclic and cyclic pelvic pain were converted to 3-level ordered categorical variables: 0–3 “none or minimal pain,” 4–6 “moderate pain,” 7–10 “severe pain” [ 8 ]. Frequency for acyclic pelvic pain consisted of 3 categories: “no pain or <1 day/month or monthly but not weekly”, “weekly” and “everyday”. Frequency for cyclic pelvic pain consisted also of 3 categories: “never or occasionally”, “often or usually” and “always”. Measures of life impact, defined as the impact on carrying out daily activities such as work and school, were collapsed into a binary yes/no indicator for both acyclic and cyclic pelvic pain. Questionnaire items included age at completion of baseline survey (continuous), recruitment site (clinic-based, population-based), age at menarche (continuous), date of last menstrual period, self-reported weight and height was used to calculate body mass index categories (defined by WHO BMI categories for those ≥20 years, and CDC age and sex specific Z-score for <20 or less: underweight: BMI−2 to 2) and self-identified US-census socially-defined groups: race (Black, White and other racial categories consisting of Asian, American Indian/Alaska Native, Native Hawaiian or Pacific Islander, multiracial, other race and unknown/not reported) and ethnicity (Hispanic/Non-Hispanic). Participants reported a range of pre-existing comorbidities diagnosed by a physician and self-reported pain conditions [ 52 ]. Eighteen comorbid conditions related to immune dysregulation and/or systemic inflammation or to chronic pain conditions (eight with 10–100 cases and ten with >100 cases in the total cohort) were included in the analyses ( Supplemental Methods 1 ). Briefly, these included surgically-diagnosed endometriosis (over-sampled at enrollment by cohort design with prevalence in the analytic sample of 48%), other gynecologic or genitourinary conditions: fibrocystic or benign breast disease, painful bladder or interstitial cystitis, uterine fibroids, ovarian cysts, and polycystic ovarian syndrome; respiratory immune conditions: allergies (i.e., grasses, pollens, mold, food, latex, drugs, animals and other) and asthma; rheumatologic and neurological conditions: fibromyalgia, chronic fatigue syndrome, rheumatoid arthritis, migraine, lower back pain, muscle or joint pain unrelated to infections or sports injuries and leg pain; gastrointestinal or abdominal conditions: inflammatory bowel disease (Crohn’s disease or ulcerative colitis), irritable bowel syndrome and non-pelvic abdominal pain. Of 18 conditions, 10 comorbid conditions had a prevalence of 10% or more in the population and were included in the final association analysis to minimize empty cells and unstable estimates and imprecise confidence intervals. Data cleaning and descriptive analysis, consisting of proportions for categorical variables and means and 95% confidence intervals or medians and 25% and 75% percentiles for continuous variables, was conducted in R (version 4.0.2) [ 57 ]. Participants who had menstrual periods in the past year but did not provide answers to questions regarding acyclic or cyclic pelvic pain were considered missing at random (MAR), a portion due to the unavailability of the question in the first version of the survey ( Supplemental Figure 2 ). However, a portion of the cohort was missing values for cyclic measures specifically because they had not menstruated in the past 12 months – primarily due to hormonal ovarian suppression. This group was considered missing not at random (MNAR) as their missingness may have been conditioned on the presence and life impact of dysmenorrhea (N=170, 14% of total sample). For this group in particular, missing cyclic pelvic pain variables were forward filled from current and prior age-range data. Participants were asked to recall the severity, frequency, and life impact of both cyclic and acyclic pain during the following age ranges: up to age 15, 16–20, 21–30, 31–40 and >40 years. A sensitivity analysis was conducted to evaluate the impact of this forward filling. Questions regarding history of pelvic pain across age ranges were only available in the WERF EPHect compliant version, which consisted of 73.3% of surveys and was consequently completed by 58.1% of those with endometriosis and 87.1% of those without the condition. Remaining missing indicators were handled using the full information maximum likelihood (FIML) approach. FIML uses the information available for each individual to maximize the sample log-likelihood function for estimating parameters and standard errors, under the assumption that the indicators are missing at random [ 58 ]. The FIML procedure in LCA does not address missingness in non-indicator variables external to the latent class model. Therefore, in the association analyses between latent classes and other variables, participants who were missing demographic and clinical or comorbidity variables were excluded from 3-step approach LCA. The number of participants missing information for each characteristic variable and for each comorbid condition was small: age at menarche (N=1), body mass index (N=1), allergies (N=67), asthma (N=8), chronic fatigue syndrome (N=8), Crohn’s or ulcerative colitis (N=5), fibrocystic or benign breast disease (N=67), fibromyalgia (N=67), inflammatory bowel syndrome (N=5), leg pain (N=94), lower back pain (N=80), muscle/joint pain (N=90 total), non-pelvic abdominal pain (N=81), ovarian cysts (N=9), painful bladder/interstitial cystitis (N=337), polycystic ovarian syndrome (N=8), rheumatoid arthritis (N=67), uterine fibroids (N=5) Latent class analysis (LCA) was used to identify subgroups of women with similar pelvic pain characteristics based on six indicators of pelvic pain. LCA uses the EM (expectation-maximization) algorithm to produce maximum likelihood estimates of model parameters, including identifying the typologies (i.e., subgroups of people who are alike) within a population [ 13 , 34 , 44 ]. We selected this method, rather than other clustering techniques (e.g., k-means clustering), as probability-based techniques provide robust parameter estimates under the latent variable framework and take into consideration measurement errors of the indicators [ 36 , 38 ]. Latent class analysis was conducted using Mplus (version 8.6) [ 42 ]. We determined the optimal number of latent classes based on five criteria in step-wise order ( Supplemental Methods 2 ) [ 3 , 10 , 13 , 37 , 43 , 44 ]. We tabulated proportions of the study population in each latent class-defined subgroup based on estimated posterior probabilities and estimated pain item response probabilities, which are proportions endorsing each category of a variable conditional on class membership. Conditional response probabilities were examined to assign labels or pain type subgroups to facilitate interpretation. We conducted descriptive statistics of demographics and clinical covariates and comorbidities across classes. The estimates were based on most-likely latent class membership of each individual in each class, which does not consider uncertainty around the estimated posterior probabilities of latent class assignment for each individual [ 13 ]. To examine the relationships between comorbidities and the subgroups of pelvic pain, a bias-adjusted three step-approach was used for LCA with covariates ( Supplemental Figure 4 ). This approach estimates the measurement model (i.e., LCA without covariates), assigns latent class membership to participants, and associates the class to an external variable accounting for classification uncertainty [ 2 , 6 , 61 ]. Demographic, clinical variables, surgically-confirmed endometriosis status, and other comorbidities were tested as predictors of class membership using multinomial logistic regression via the R3STEP procedure which incorporates the most likely latent class indicator variable and uncertainties associated with each (Mplus 8.6) [ 2 ]. We estimated unadjusted and age-adjusted odds ratios. We adjusted for age assuming linearity based on substantive knowledge as age is a risk factor for endometriosis and a number of comorbidities included in the analysis, as well, in the univariable multinomial analysis had significant associations with assignment to some pelvic pain classes [ 33 ].

A total of 1255 participants were included in the analytic population. Overall, participants predominantly identified as White (81%) and ranged from 12 to 55 years of age (median=23 years, 37% younger than 21 years) ( Table 1 ). About half (48%) had surgically-diagnosed endometriosis (N=597). Those with surgically-diagnosed endometriosis were younger (median age=19 years compared to 24 years among those who did not have an endometriosis diagnosis). Those with endometriosis also were more likely to have been recruited from clinics (97% versus 13.7%) and fewer menstruated in the last 12 months (78% versus 94%) compared to those not diagnosed with endometriosis. The most prevalent comorbidities were lower back pain (69%), migraine (49%), non-pelvic abdominal pain (46%), muscle or joint pain unrelated to infections or sports injuries (39%), and allergies (33%). Of the 18 comorbidities evaluated, 17 had a higher prevalence among those with surgically-diagnosed endometriosis; the one exception was polycystic ovarian syndrome (PCOS) with 2.5% prevalence among those with endometriosis compared to 5.2% in those without. The median number of comorbidities for participants was 3 (range 0–14). According to the five criteria applied to 2-class to 9-class LCA, the five-class model was considered optimal ( Supplemental Table 2 ). We prioritized aBIC, choosing the class with the lowest aBIC, and that which was most interpretable. Our entropy was 0.76, suggesting moderate to strong separation of classes. The class distributions (class 1 to 5) consisted of subgroups subjectively labeled based on prevalent pain characteristics ( Table 2 and Supplemental Figure 3 ). Class 1 (‘none or minimal pelvic pain’) was a subgroup with no acyclic pelvic pain experienced by 97% of class members and no cyclic pelvic pain experienced by 83% of class members. Class 2 (‘moderate cyclic pelvic pain only’) was a subgroup where 56% individuals experienced moderate cyclic pelvic pain associated with periods and 99% experienced no acyclic pelvic pain and 92% experienced no life impact because of this pain. Class 3 (‘severe cyclic pelvic pain only’) was a subgroup where, like class 2, members experienced only cyclic pelvic pain with 88% experiencing no acyclic pelvic pain. However, unlike class 2, 93% of class 3 experienced severe cyclic pelvic pain. Class 4 (‘moderate or severe acyclic, moderate cyclic pelvic pain’) was a subgroup that, similar to class 2, experienced cyclic pelvic pain with low life impact. However, in class 4, 53% were characterized by severe acyclic pelvic pain (53%), and life impacted by acyclic pelvic pain (65%). Finally, class 5 (‘severe acyclic, severe cyclic pelvic pain’) consisted of individuals experiencing the most severe categories for five of six pelvic pain indicators. Sociodemographic characteristics (age and race) of individuals assigned to the five classes were comparable. Some clinical characteristics were also comparable for age at menarche and body mass index. However, clinical characteristics like the date of last menstruation and medication use varied widely between groups ( Supplemental Table 3 ). A sensitivity analysis excluding those who had not menstruated in the past 12 months (remaining sample size N=1085) confirmed a 5-class model, largely similar to the groups described above, to be the best fitted model ( Supplemental Table 4 ). Unadjusted associations of pain variable-defined subgroups with covariates are described in Supplemental Table 5 . Higher age at completion of the survey was associated with lower odds of being in the ‘severe acyclic, severe cyclic pelvic pain’ subgroup (class 5) (OR=0.88, 95% CI=0.83,0.94) compared to the reference none or minimal pelvic pain subgroup (class 1). Higher age at completion of survey also was associated with lower odds of being in the severe cyclic pain only subgroup (class 3) (OR=0.89, 95% CI=0.84,0.93) compared to the none or minimal pelvic pain subgroup (class 1). Similarly, higher age at menarche was associated with lower odds of being in class 5 (OR=0.67, 95% CI=0.58,0.77) and class 3 (OR=0.67, 95% CI=0.58,0.78) compared to class 1. Neither BMI nor US-census socially-defined groups, race and Hispanic ethnicity, were significantly associated with being in any of the classes. The distribution of surgically-diagnosed endometriosis ranged from 4.0% in class 1 to 94.1% in class 5 ( Supplemental Table 4 ). Having surgically-diagnosed endometriosis was quantitively the strongest predictor of being in the three subgroups marked by severe pain ( Table 3 ). Migraine was the only condition associated with probability of being in all four pain subgroups compared to the none or minimal pain group, ranging from the lowest magnitude of association with being in the ‘moderate cyclic pain, severe acyclic pain’ (OR=2.62, 95% CI=1.38,5.00) to highest in the ‘severe acyclic, severe cyclic pelvic pain’ class (OR=7.78, 95% CI=4.82,12.56). Higher number of comorbidities was also significantly associated with being in all the pelvic pain subgroups. All ten comorbidities with >100 cases were associated with the ‘severe cyclic pain only’ subgroup (class 3) compared to the ‘none or minimal pelvic pain’ subgroup (class 1) ( Table 3 and Supplemental Table 6 ). No comorbidity was exclusively associated with this group. Allergies and asthma were only associated with the ‘severe cyclic pain only’ (class 3) and the ‘severe acyclic, severe cyclic pelvic pain subgroup’ (class 5) subgroups. Eight of ten comorbidities with 100 cases in the population were associated with ‘moderate or severe acyclic, moderate cyclic pain’ subgroup (class 4) relative to the ‘none or minimal pelvic pain’ subgroup (class 1). Finally, all ten comorbidities with >100 cases and two with 10–100 cases were associated with the ‘severe acyclic, severe cyclic pelvic pain’ subgroup (class 5) relative to the ‘none or minimal pelvic pain’ subgroup (class 1); the only morbidity that was solely associated with the ‘severe acyclic, severe cyclic pelvic pain’ subgroup (class 5) relative to the ‘none or minimal pelvic pain’ subgroup (class 1), was fibromyalgia (OR=62.75, 95% CI=1.98,1993.3). Fibromyalgia had a very low prevalence in this population, only affecting 1.35% of the total population and thus, although statistically and empirically significant (11 of the 16 affected (67%) were in class 5), yielded wide confidence intervals.

Strengths of this study included a large participant sample that used a validated tool to provide information. This enabled us to use the best available indicators and effectively address missing data. Additionally, the relatively young age of the cohort, compared to most other studies, allowed us to examine earlier windows of pelvic pain across the life course. The overrepresentation of clinic-recruited individuals, including those with a surgical diagnosis of endometriosis, provided deep phenotyping that does not exist in standard clinical care. This composition allowed us to detect more extreme subgroups while reflecting a spectrum of pelvic pain presentations. In clinical populations, ovulatory suppression is a common treatment modality [ 23 , 55 ], making it challenging to quantify severity or frequency of cyclic pelvic pain among those receiving this treatment. In contrast, acyclic pelvic pain status can still be assessed in those receiving ovulatory suppression treatment. For this reason, we quantified cyclic pain information using age-range data for the 13.5% of our population who did not menstruate in the past year. We also conducted a sensitivity analysis excluding this group and found that our classification remained robust. Limitations include potential recall bias pelvic pain history, as those currently experiencing pain may recall past pain differently from those with resolved pain. However, it should be noted this self-reported medical history at enrollment aligns with typical clinical settings where symptoms are self-reported and rarely functionally tested. Most data used in this study was recent, in the last 3 or the most recent 12 months, without considering more distant recalled time periods. Medication use beyond its impact on menstrual cyclicity was not examined. Individuals using pain-modifying analgesics or hormonal medication, or those with inadequate pain relief from such interventions may be misclassified into less severe pain groups, potentially driving the associations toward the null. The cohort’s intentional over-representation of specific groups may affect generalizability to a broader chronic pelvic pain population. Assessing comorbidities may introduce detection bias, especially for those who have had greater medical system contact. Particularly, those within the cohort who had not been surgically diagnosed with endometriosis were largely enrolled from the communities within the hospitals’ catchment area, without the same pattern of potential medicalization around their pelvic pain experience compared to those surgically diagnosed with endometriosis. This detection bias would drive associations between pain and comorbidities away from the null. However, while there were strong associations with some of the subgroups and endometriosis, endometriosis was not associated with all four pain subgroups. Also, an important reality is that pelvic pain is often experienced without healthcare intervention. It is also possible that there are women in the non-pain group who have comorbidities not yet diagnosed. In the case of endometriosis, in the 5 to 10 years of follow-up since the baseline data was collected for these cohort participants, only 3 without an endometriosis diagnosis at baseline have subsequently been diagnosed de novo with endometriosis. The study tackled non-random missingness by propagating last observed pain values forward under the assumption that they more often than not would be representative of the current counterfactual. This could introduce misclassification among those with the most dynamic pain experience. Additionally, the study yielded wide confidence intervals (CIs), which we attribute to two drivers. First, we incorporated uncertainty intervals in class assignment using a deliberately conservative three-step approach that would over-, not underestimate variability. Second, there were small sample sizes of some specific comorbidities experienced by this study population. Robustness of these findings can be determined by replication in larger populations yielding greater estimate precision. Lastly, the study’s hospital-based population may have been influenced by systemic biases impacting global pelvic pain treatment access, leading to higher proportion of White participants compared to other race and ethnicities. Future research should prioritize enrolling historically underrepresented populations [ 7 ].

This study identified five distinct subgroups of pelvic pain in a cohort of women, demonstrating a spectrum of patterns of pelvic pain severity, frequency and life impact. The association between these subgroups and comorbidities related with immune dysregulation or systemic inflammation and other comorbid pain conditions were examined. Pelvic pain results from complex interactions between multiple systems and pathologies. This research identifies distinct pelvic pain patterns, raising questions about the unique inflammatory and immunologic mechanisms linked to subgroups with elevated pain levels given their association with chronic overlapping pain conditions, asthma and allergies. Methods described here can be used in future analyses, incorporating further indicators such as observed measures and biomarkers of pain. This may allow for classes to be further refined from descriptive to ones that reflect intrinsic subtypes reflecting ever more precise pathophysiologic pathways. Consequently, we may better understand the etiology, diagnosis, and prognosis of chronic pelvic pain and associated disorders.

In a cohort of 1255 women, we identified five subgroups of chronic pelvic pain based on pelvic pain severity, frequency, and life impact. We observed that 13 of 18 comorbidities associated with inflammation and pain, including endometriosis, were heterogeneously distributed across these subgroups, which may point to unique pathways and also to informative shared underlying pathways [ 49 ]. Among these subgroups, the severe acyclic, severe cyclic pelvic pain subgroup was associated with the highest number of comorbidities. Migraine was associated with all four subgroups experiencing pelvic pain. Systematic subgrouping of chronic pelvic pain in women has been limited, with highly varied approaches. Chen, et al. (2018) applied LCA to identify three subgroups based on only dysmenorrhea symptoms plus non-pain late luteal / early follicular phase symptoms such as bloating, irritable bowel, and reduced appetite [ 11 ]. Alternatively, Obbarius et al. applied LCA to identify four classes using pain intensity, depression, anxiety, and general quality of life [ 45 ]. Distinct from these, the aim of the current study was to identify pain subgroups characterized solely by their pelvic pain experiences, without introducing other variables that may or may not be directly related to pelvic pain. This cohort intentionally over-sampled for individuals with surgically-diagnosed endometriosis (48% at baseline). Endometriosis was significantly associated with three of four pain groups, except the moderate cyclic pain only subgroup. That those with and without endometriosis can experience similar pain patterns suggests commonalities in underlying pain pathologies. Compared to the none or minimal pelvic pain class, our analysis revealed two subgroups that included those experiencing only cyclic pelvic pain – one marked by moderate cyclic pain and the other severe cyclic pain. The two other groups were both marked by the presence of severe acyclic pelvic pain – one group accompanied by moderate cyclic pain, and one group accompanied by severe cyclic pain. The subgroup characterized by severe cyclic and severe acyclic pelvic pain (class 5) exhibited a high comorbidity burden and a strong association with endometriosis. It is possible that class 5 was affected by central pain sensitization or hypothalamic-pituitary-adrenal (HPA) dysregulation, both of which have been associated with widespread pain and increased comorbidity [ 18 ]. Central pain sensitization, characterized by pain hypersensitivity can manifest with or without inflammation and neural lesions [ 64 ]. Class 5 was associated with fibromyalgia, a prototypical central pain sensitization syndrome characterized by structural and function changes in the central nervous system [ 9 ]. Neuro-immune interaction plays a crucial role in pathologic pain, in both peripheral and central sensitization [ 24 ]. HPA axis dysregulation, linked to stress response and elevated levels of corticotropin-releasing factor (CRF) can increase cortisol production and mast cell activation, and subsequent activation of pain nociceptors [ 18 ]. Elevated basal levels of cortisol have been observed in fibromyalgia [ 56 ]. In a recent study using data from the A2A cohort, participants with any autoimmune or inflammatory conditions had an increased odds of also having endometriosis; suggesting potential shared immune profiles or other biological mechanisms [ 52 ]. Interestingly, while most comorbidities, including endometriosis, were unevenly distributed among the four pelvic pain symptom-defined groups, migraine was strongly associated with all four pelvic pain subgroups. Although women with endometriosis are more likely to experience migraines [65], lifetime prevalence of migraine headaches is approximately 67% among women with chronic pelvic pain, regardless of endometriosis status [ 28 ]. Growing evidence suggests that chronic migraine is associated with central sensitization and neuroinflammation, with promising therapeutic prospects for chronic migraine subtypes in comparison to acute subtypes, and in headache subtypes suggestive of altered processing such as post-traumatic headache [ 30 , 41 ]. Chronic migraines have recently been categorized as part of broader group of ‘nociplastic’ pain syndromes, characterized by altered nociceptive function within a complex biopsychosocial framework, and linked to multiple overlapping pain syndromes and comorbidities [ 21 , 29 , 63 ]. This research has clinical implications for categorizing patient heterogeneity in chronic pelvic pain. Identifying subgroups based on patterns of pain presentation and association with comorbidities may provide a new avenue for greater personalized therapeutic approaches. For example, individuals classified in the severe acyclic, cyclic pelvic pain subgroup with a high comorbidity burden may benefit from more aggressive pain management strategies or novel pain management strategies targeting pain processing and neuromodulation, non-pharmacological interventions including physical therapy and psychological therapies, and early comorbidity management.

Continuous or episodic chronic pain in the pelvis and lower abdomen lasting more than 6 months is associated with significant morbidity among women. Worldwide, 17% to 81% of reproductive aged women report pelvic pain associated with menstruation (cyclic), while 2% to 24% report pelvic pain that is not associated with menstruation, intercourse, or pregnancy (acyclic) [ 32 ]. Among adolescents, the prevalence of cyclic pelvic pain ranges from 30 to 90% [ 25 , 54 ]. Chronic pelvic pain is multifactorial, and approaches to understanding and managing the condition have centered on known underlying disease pathologies, including reproductive system conditions like endometriosis and non-gynecologic conditions such as inflammatory bowel disease [ 12 ]. However, as in the case of endometriosis, histopathologic staging does not correlate well with pelvic pain symptoms or comorbidities such as infertility, treatment response, or prognosis [ 26 , 27 , 60 ]. There is increasing evidence that regardless of the range of underlying causes, chronic pelvic pain patients have common mechanisms driving their symptomology. Examples of this include but are not limited to alterations seen in nerve fibers and central gray matter volume [ 1 , 4 , 46 ]. In recent years, there has been growing interest in using multi-dimensional data to identify subgroups of disease [ 40 ], with more than 400 clinical categorizing or classifying systems in active use in clinical care. For example, in breast cancer and ovarian cancer, subgroup identification based on clinical characteristics, histologic and molecular differences have yielded valuable insight into subgroup-specific risk factors, prognosis and treatments [ 5 , 17 , 53 , 59 ]. However, systematic approaches to chronic pelvic pain subgrouping have been limited and varied. These approaches have included expert-based subtyping [ 35 ], classification based on pain scores [ 51 ], exploratory factor analysis to identify pelvic pain subtypes based on pain ratings and quantitative pressure thresholds [ 19 , 20 ], k-means clustering of interstitial cystitis using symptoms and psychosocial profile [ 31 ], latent class analysis for dysmenorrhea subtypes [ 11 ] and for assessing chronic pelvic pain burden subgroups [ 45 ]. Pathologies such as endometriosis and irritable bowel syndrome can lead to pelvic pain but may also interact in other ways. For instance, recent research has revealed higher prevalence of irritable bowel syndrome in patients with endometriosis and bladder-associated pelvic pain [ 14 , 16 ]. As well, patients with multiple chronic diseases can experience increased burden of chronic pain, with variable manifestations and response to non-personalized standard of care treatments [ 22 , 39 ]. Previous research on pelvic pain subgroups has been limited by small sample sizes and integration of non-pain factors. A focused approach uniquely concentrating on patterns of pelvic pain addresses current research gaps and offers a new avenue to relate pain symptomology to pathophysiology across multiple morbidities. The purpose of the current study was to identify symptom-based subgroups of chronic pelvic pain and uncover associations of the subgroups with eighteen comorbidities related to inflammation and chronic pain. To achieve this goal, we used baseline data from a longitudinal cohort of adolescents and adults with deeply-phenotyped pain symptomology. We identified groups of similar symptomatology with latent class analysis (LCA), a cross-sectional latent variable mixture modelling approach that clusters people with similar characteristics to uncover hidden clinical phenotypes [ 40 ].