Quality of life identification by unsupervised cluster analysis: A new approach to modelling the burden of endometriosis

Endometriosis is identified as a chronic disorder where cells similar to those lining the interior of the uterus, known for their secretory function, are discovered in locations outside the uterus. These cells respond to the hormonal shifts of the menstrual cycle, leading to persistent inflammation. Symptoms of endometriosis, which overlap with those of various other conditions, prominently include intense pain during menstrual cycles and sexual activity, abdominal discomfort (at times extending to the sacral area), and pain while urinating and undergoing gynecological evaluations [ 1 – 3 ]. Endometriosis is thought to impact 7% to 15% of reproductive-age women, encompassing 30% to 50% of women facing infertility challenges and nearly half of those experiencing chronic pelvic pain syndrome. These figures are speculative due to the potential for endometriosis to be symptom-free, rendering precise prevalence rates elusive [ 4 , 5 ]. Recent investigations have emphasized the extensive influence of endometriosis on various life factors [ 6 ]. Considering the symptomatic and complicating nature of endometriosis, it is crucial to address not only the physical symptoms but also the social and psychological ramifications associated with the diagnosis, including the patient’s quality of life (QoL) [ 7 ]. Evaluating QoL is crucial to identify the most suitable management and treatment strategies, taking into account the patient’s overall health and their physical, psychological, and social wellness [ 8 ]. The WHO conceptualizes QoL as the individual’s perception of their life situation within their cultural and value system context, reflecting their aspirations, standards, and concerns, influenced by their environment. QoL metrics cover the capacity to maintain social roles, adaptability, psychological health, and social interactions. The significant prevalence of endometriosis, coupled with its social and economic consequences, garners considerable research interest in the QoL domain [ 6 , 9 ]. As QoL is inherently subjective, varying greatly based on numerous personal and external factors, it remains essential to better understand the factors associated with QoL in endometriosis. Moreover, in the context of new challenges in personalized, predictive, and preventive medicine, it is essential to understand the harmful factors which could influence QoL. Thus, creating a phenotype of women with risk of low QoL could be of interest in the personalized medicine which is currently implemented. Cluster analysis is a multivariate methodology that can be performed to identify groups of participants with similar characteristics in the context of complex mechanisms it is a methodology for performing groups in which the data are not scattered evenly by n‐dimensional space but instead form clusters. This approach has been recently performed for several models, including recently for endometriosis [ 10 , 11 ]. Cluster analysis based on clinical variables has been observed to be mainly effective in the exploration of the characterization of phenotypes in diseases. Several findings have suggested that cluster analysis could improve the characterization of a disease phenotype. This novel approach has not yet been applied to QoL in women suffering from endometriosis. Thus, our aim was to investigate the hypothesis that cluster analysis could be used to identify homogeneous phenotyping subgroups of women according to the burden of the endometriosis for their QoL, and then to investigate the phenotype differences observed between these subgroups.

1,586 women responded to the questionnaire. The characteristics of the women are shown in Table 1 . The 25 th percentile of EHP-5 was used to define good or low EHP-5, the cutoff was 600 in the dataset. *Comparison between all clusters. **Comparison between clusters 5 and 6. K‐means clustering was implemented with 8 clusters proving the best fit with the optimal CCC value of 17.2162 according to the different symptoms of endometriosis ( Table 2 ). The characteristics of each cluster are shown in Table 1 . A cluster was highlighted (cluster 5) showing a 100% rate of good QoL ( Table 3 ). This cluster presented only rates of pain during sexual intercourse, pain during bowel movements during periods, other digestive issues and pain, particularly excessive menstrual cramps, that is felt by more than 70%. Another cluster (cluster 6) showed a 100% rate of worse QoL ( Table 3 ). It presented rates of pain during sexual intercourse, abnormal or heavy menstruation, pain during bowel movements during periods, other digestive issues, worsening pain over time and pain, particularly excessive menstrual cramps, that is felt by more than 70%. Parallel coordinate plots for the display of the structure of the observations in each cluster show how the clusters differ. This figure presents the different cluster hierarchies ( Fig 1 ). Parallel coordinate plots for the display of the structure of the observations in each cluster showing how the clusters differ, and biplot 3B of the clusters. Clusters 6 and 8 with 100% of classified rates represent 644 women for 41% of the population ( Table 3 ). Cluster 5 and 6 were significantly different for all the parameters, except for menopause status (p = 0.274), being in a couple (p = 0.093), and having surgery for endometriosis (p = 0.076) ( Table 1 ). After applying a multiple regression logistic, the independent factors determining participants from cluster 8 to the cluster 6 according to QoL were age (over 45 years compared to below 25 years, OR = 0.17 [0.07–0.46], p<0.001), BMI (high vs low, OR = 0.47 [0.28–0.80], p = 0.005), having children (OR = 0.30 [0.18–0.48], p<0.001), having surgery for endometriosis (OR = 0.55 [0.32–0.94], p = 0.029), and education (high vs low, OR = 2.75 [1.75–4.31], p<0.001) ( Table 4 ).

Eight clusters were identified with two specific clusters based on the burden of sympotms of endometriosis for the classification of QoL. Independent determinants differencing these two specific clusters of women were age, education, BMI, having children and surgery of endometriosis. Cluster modeling permits the identification of homogeneous women phenotypes for QoL with endometriosis. The implementation of new methodological approaches could be essential to improve the QoL of women living with endometriosis and lead to implementation of appropriate preventive strategies.

We designed and conducted a cross-sectional survey using survey software developed by our hospital. The survey was completed anonymously to encourage honest and unbiased responses. The study link was disseminated via social media (Instagram) where participants were asked to forward this link to others they know. All registrants were free to accept or decline the invitation, with no monetary reward received in return. Participants were also informed that they could withdraw at any time. Following internationally accepted ethical codes, respondents were duly informed of the purpose of the survey and were reminded of their participation rights before proceeding to take the survey. A research protocol was conducted to obtain approval from an ethical committee. The distribution of the questionnaire occurred between November 2023 and January 2024 in France on social media (Instagram). We closed the survey link after the workshop ended. The questionnaire was developed and adapted based on a review of literature [ 12 – 14 ]. It was pretested among six health and social care professionals and modified according to their feedback. The survey was conducted in French and required approximately five minutes to complete. The questionnaire was divided into the following sections ( S1 File ): Sociodemographic questions (marital status, age, educational level, children, BMI level calculated as weight (in kg) divided by height squared (in meters) and categorized as high (BMI > 30 kg/m 2 ), moderate (BMI between 25 and 30 kg/m 2 ), and low (less than 25 kg/m 2 ). Questions related to the disorder (diagnosis, symptoms, treatment, age at diagnosis etc.) Symptoms of endometriosis were defined as: pain during sexual intercourse; abnormal or heavy menstruation; infertility; pain during urination during periods; pain during bowel movements during periods; other digestive issues (diarrhea, constipation, nausea); worsening pain over time; pain, particularly excessive menstrual cramps and other symptoms. EHP-5 questionnaire: The EHP-5 (Endometriosis Health Profile) is a tool for measuring health-related quality of life in endometriosis [ 15 ]. It is a two-part questionnaire reffering to the last 4 weeks. The first part is a 5-item core questionnaire including questions about pain, control and powerlessness, emotions, social support and self-image. The second part is a 6-item modular questionnaire that consists of questions that may not be applicable to every woman with endometriosis. These 6 items refer to work life, relationship with children, sexual intercourse, medical, treatment and infertility. Each of the 11 items is scored on a Likert-type scale with the range from 0 = never to 4 = always. The second part also has an option ‘not applicable.’ Scores are then transformed on a scale 0–100, with 0 = best possible health status, 100 = worst possible health status. Ethics statement. The study was approved by the Foch IRB: IRB00012437 (approval number: 23-07-05) on 27 July 2023. Written consent was obtained from all participants. Characteristics of the study population were described as the mean standard deviation (SD) for continuous variables. Categorical variables were described as numbers and proportions. Comparisons between groups were performed using the Mann–Whitney test or t Student test for continuous variables. Pearson’s χ 2 test was performed for categorical variables. Good QoL was defined as EHP-5 considered inferior to the lower 25 th percentile, as there is no cutoff existing in literature to define good QoL based on EHP-5. K‐means, a major clustering algorithm, was performed to show structure in data and divide participants into groups [ 10 , 16 ]. Principal compound analysis, mapping high‐dimension data into low‐dimension space, was performed to diminish the primal data into two dimensions. Here we excluded participants without missing data. The main steps in the K‐means algorithm were: (1) Select initial cluster centers with the number of K, (2) Assign each point to its closest cluster center, and (3) Compute new cluster centers. In step 1, K points are defined randomly as initial cluster centers. In step 2, when we assign each point to its closest cluster center, we compute the distance, such as the Euclidean distance, between points and centers. In step 3, the new cluster centers are computed as the mean of all points belonging to each cluster. The optimal number of clusters showing the best fit was selected using the highest cubic classification criterion (CCC), which estimates the number of clusters using Ward’s minimum variance method. Covariables selected for the construction of the clusters were: pain during sexual intercourse, abnormal or heavy menstruation, infertility, pain during urination during periods, pain during bowel movements during periods, other digestive issues (diarrhea, constipation, nausea), worsening pain over time, pain particularly excessive menstrual cramps and other symptoms. Then, the two clusters with 100% of good classification of good or low EHP-5 were compared by multiple logistic regression models computing odds ratios (OR) with 95% confidence interval (95% CI) and adjusted for covariables with p value <0.20 in univariable analysis. Statistics were performed using SAS software (version 9.4; SAS Institute, Carry, NC). A p value < 0.05 was considered statistically significant.

(DOCX)