Optimization of Management Plan with a machine learning model for Ovarian Torsion Cases: Operative versus Conservative

Optimization of Management Plan with a machine learning model for Ovarian Torsion Cases: Operative versus Conservative | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Optimization of Management Plan with a machine learning model for Ovarian Torsion Cases: Operative versus Conservative Alia Alethawy, Omaima Al-Baghdadi, Yauhen Statsenko, Moamar Al-Jefout This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7828438/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background: Ovarian torsion (OT) is a gynecologic emergency requiring prompt and accurate management to preserve ovarian function and fertility. Determining the need for operative intervention versus conservative management remains challenging due to overlapping clinical and imaging features. This study aimed to develop and validate a machine learning (ML)–based prognostic model to assist clinical decision-making in suspected OT. Methods: A retrospective analysis was conducted on 219 females (1 month–75 years) presenting with suspected OT at a tertiary center between 2022 and 2024. Clinical, demographic, laboratory, and imaging variables were analyzed. Predictors of management type (operative vs. conservative) were identified using comparative statistics and Spearman correlation. Principal component and cluster analyses were used for dimensionality reduction and patient stratification. Supervised ML algorithms—including Decision Tree, Random Forest, Neural Network, Gradient Boosting, and Logistic Regression—were trained with oversampling techniques (ROSE, SMOTE, SMOTE-ENN) to address class imbalance. Model performance was assessed using the area under the ROC curve (AUC), sensitivity, and specificity. Results: Operative management was required in 83.6% of cases, while 16.4% were managed conservatively. Significant predictors of conservative management included prior oral contraceptive use (protective; r = −0.199, p = 0.003), absence of a pelvic mass (r = 0.134, p = 0.048), and lower symptom burden (r = 0.148, p = 0.029). PCA identified eight clinical domains, and clustering revealed two distinct phenotypic subgroups. The weighted Decision Tree achieved the best-balanced performance (AUC = 0.76), while Random Forest with ROSE oversampling achieved perfect performance (AUC = 1.00), indicating potential overfitting. Conclusions: Machine learning models can enhance clinical decision-making by stratifying OT patients suitable for conservative management and identifying those requiring urgent surgery. Integration of this model into clinical decision support systems may reduce unnecessary surgeries, optimize fertility-preserving care, and improve Surgery Critical Care & Emergency Medicine Surgical Obstetrics & Gynecology Ovarian torsion Machine learning Prognostic model Conservative management Gynecologic emergencies Fertility preservation Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction Ovarian torsion (OT) is a gynecological emergency that requires prompt diagnosis and management to preserve ovarian function and prevent serious complications. OT, when it is associated with tubal torsion, is termed an adnexal torsion. The annual prevalence is estimated to be around 9.9/100,000 women of reproductive age ( 1 ). The pathophysiology of OT involves torsion of the ovary on its pedicle, leading to compromised venous return, stromal swelling, internal bleeding, and ischemia, which may lead to necrosis. Timely and accurate decisions regarding surgical intervention versus conservative management are crucial for preserving the reproductive function of women with this pathology. Classical diagnostic criteria focus heavily on clinical signs such as abdominal pain, nausea, and vomiting ( 2 ), which are normal clinical presentations, and imaging (Doppler scan usually shows reduced ovarian blood supply), which may not fully capture the complexity of torsion cases. Ovarian torsion is usually managed surgically, normally in the form of diagnostic laparoscopy, which is the gold standard for the diagnosis of OT; furthermore, it enables treatment, either by oophorectomy or detorsion, although recent publications showed a modest decrease in the proportion of oophorectomies for ovarian torsion ( 3 ). However, there is still a risk of serious complications from laparoscopy in approximately 2 per 1000 cases ( 4 ), which can include organ trauma or major vessel injury, both of which harbor significant morbidity and mortality ( 5 ). There is some confusion in the literature about which factors are responsible for the development of ovarian torsion, and what the odds are of a particular clinical feature in determining the likelihood of developing ovarian torsion. Moreover, clinical decision-making is often challenged by nonspecific symptoms and overlapping features with other acute abdominal conditions. Hence, it is vital to develop a prognostic Model Management Plan (MMP) determining either operative or Conservative (non-surgical) management for Cases with Ovarian Torsion. Machine Learning (ML) methods offer the potential to enhance prognostic accuracy by analysing large, multidimensional clinical datasets and uncovering non-linear patterns that may not be eLearning clinicians. The application of ML to ovarian torsion may facilitate the development of predictive models that inform real-time decision-making, guiding whether operative or conservative management is more appropriate for individual patients. This study aims to develop and evaluate ML-based models that predict management outcomes in ovarian torsion using a diverse clinical dataset ( 6 , 7 ). Our study aims to achieve four objectives: Objective 1 is to develop and validate a predictive model that distinguishes between cases of ovarian torsion requiring operative intervention and those suitable for conservative management. Objective 2 : to identify the most informative clinical, demographic, and historical features influencing the management decision. Objective 3 : to assess relationships among studied variables and find their associations reflecting the risk factors of ovarian torsion, clinical symptoms, and disease management. Objective 4 : to build a model prognosticating the suitable disease management plan (operative versus conservative). Methodology This retrospective cohort study was conducted at Tawam Hospital and included all women diagnosed with ovarian torsion between September 1, 2022, and September 1, 2024. A total of 219 patients were identified through electronic medical records. This retrospective study aimed to develop and validate a machine learning–based prognostic model to support clinical decision-making in the management of ovarian torsion. A total of 219 female patients aged one month to 75 years who presented with suspected ovarian torsion to a tertiary care center were included. Data collection spanned clinical, demographic, laboratory, and imaging variables, as well as intraoperative findings and treatment outcomes. Ethical approval was obtained before data collection, and all patient identifiers were removed to ensure confidentiality (Date: 11/21/2024, Ref. No.: HREC SEHA-IRB- 942). Data Collection and Preprocessing Data were extracted from electronic medical records and included over 60 variables: age, BMI, parity, menopausal status, gynecological and surgical history, comorbidities (e.g., PCOS, endometriosis, metabolic syndrome), presenting symptoms (e.g., abdominal pain, vomiting, fever), imaging modalities (ultrasound, CT, MRI), hormonal markers (AMH, FSH, LH), and post-treatment fertility outcomes. Missing values were handled via complete case analysis or imputation as appropriate. All continuous variables were tested for normality using the Shapiro–Wilk test and showed non-Gaussian distribution. Statistical Analysis Comparative analyses between conservatively and operatively managed groups were performed using the Mann–Whitney U test for continuous variables and Pearson’s chi-square test for categorical variables. Spearman’s rank correlation coefficients (r) were calculated to examine associations between individual variables and treatment type. A p-value < 0.05 was considered statistically significant. Dimensionality Reduction and Pattern Recognition Principal Component Analysis (PCA) with Varimax rotation was used to identify latent patterns and reduce dimensionality. Variables with eigenvalues ≥ 1 were retained, and factor loadings ≥ 0.75 were considered strong contributors. PCA revealed eight principal components, each capturing distinct clinical domains such as reproductive history, gastrointestinal symptoms, systemic inflammatory features, and previous surgical history. Cluster analysis using silhouette scores and gap statistics was then conducted to stratify patients into clinically relevant subgroups. Two distinct clusters were identified, representing different phenotypic profiles of ovarian torsion presentations. Predictive Modeling and Machine Learning Supervised machine learning models were developed to classify patients according to the appropriate management strategy (operative vs. conservative). Five algorithms were tested: Random Forest, Decision Tree, Neural Network, CatBoost Gradient Boosting, and Logistic Regression. Due to the significant class imbalance (only 16.4% received conservative management), class weighting and resampling techniques were used to improve model performance. These included Random Over-Sampling Examples (ROSE), Synthetic Minority Over-sampling Technique (SMOTE), and SMOTE-Edited Nearest Neighbors (SMOTE-ENN). Model performance was assessed using receiver operating characteristic (ROC) analysis. The area under the ROC curve (AUC), along with sensitivity and specificity, was calculated for each model. Feature Importance Analysis To identify the most influential predictors, information gain was computed for each variable. The top-ranking features included epigastric pain, prior salpingectomy or oophorectomy, smoking history, fever, dysuria, and bundled symptom clusters (e.g., distension and fainting). These were consistent with findings from PCA and clustering analyses. In summary, our primary outcomes are accuracy metrics of the predictive model in classifying patients with ovarian torsion into operative versus conservative management groups. This includes Area Under the ROC Curve (AUC), Sensitivity and Specificity, and Correct classification rate for conservative management. Secondary Outcomes: Identification of significant clinical, demographic, and historical predictors associated with the type of management (e.g., OCP use, pelvic mass, number of symptoms). Model explainability through feature importance ranking (e.g., information gain, decision tree splits). Performance comparison among machine learning algorithms (Random Forest, Decision Tree, Neural Network, Gradient Boosting, Logistic Regression) in predicting treatment type. Evaluation of post-treatment reproductive outcomes, including spontaneous conception rate, Recurrence of ovarian torsion, Assessment of symptom clusters and phenotypic patterns using principal component and cluster analysis to characterize distinct clinical presentations, and feasibility of integrating the predictive model into a clinical decision support tool (CDST) for use in acute gynecological settings. Results Objective 1: Associations Between Demographic, Clinical Histories, and Clinical Data, and Treatment Type: In the first objective, we assessed the differences between the demographic, clinical histories, and clinical data between the groups with conservative and operative management. To make the comparison, we used the Mann-Whitney U-test correlation due to the non-Gaussian distribution of variables. A total of 219 women were included in the study, of whom 38 (17.4%) were managed conservatively and 177 (80.8%) underwent operative intervention. A subgroup analysis was performed on 179 patients with complete reproductive data. Table 1 compares demographic, clinical, surgical, and outcome variables between two treatment strategies for ovarian torsion: Conservative management (n = 36) and Operative management (n = 183). It also stratifies operative cases into laparotomy (n = 28) and laparoscopy (n = 155). Continuous variables are presented as mean ± SD, categorical variables as % (n). P-values reflect comparisons between groups. Demographic Characteristics: Age differed significantly between groups: Conservative vs. operative (p < 0.001) and laparotomy vs. laparoscopy (p < 0.001). In comparison, operative patients were slightly older on average, with laparotomy patients markedly older (38.4 ± 17.5 y) than laparoscopy patients (26.5 ± 10.7 y). BMI was somewhat higher in operative patients (27.0 ± 6.4) than in those patients with conservative management (25.7 ± 5.5), borderline significant (p = 0.058). Similar borderline difference between laparotomy and laparoscopy (p = 0.058). Menopause was more frequent in operative (11.5%) vs. conservative (5.6%) (p = 0.040), and much higher in laparotomy (32.1%) vs. laparoscopy (7.7%) (p < 0.001). Obstetric History: Gravidity and parity were significantly higher in operative vs. conservative (p = 0.004, p = 0.003), and higher in laparotomy vs. laparoscopy (p = 0.002, p 0.7). Past Surgical/Medical History: No significant differences for prior pelvic surgery, appendectomy, cystectomy, or caesarean section (all p > 0.3). History of endometriosis, OCP use, PID, or PCOS showed no group differences (all p > 0.7). Comorbidities: Overall comorbidities count similarly across groups (p > 0.2). Metabolic syndrome was rare but significantly more frequent in laparotomy vs. laparoscopy (3.6% vs. 0%; p = 0.019). Symptoms: Unilateral lower abdominal pain was the most frequent symptom (91%), with no significant group difference (p > 0.2). Vomiting was more common in operative than conservative (41.5% vs. 30.6%; p = 0.101, NS) and trended lower in laparotomy vs. laparoscopy (25% vs. 44.5%; p = 0.055). Loss of appetite was significantly more frequent in the laparotomy group (10.7%) than in the laparoscopy group (2.6%) (p = 0.040). Diagnostic Modalities: Conservative patients underwent fewer imaging investigations overall (mean number lower; p = 0.012). CT abdomen/pelvis and MRI pelvis were used more often in laparotomy vs. laparoscopy (p = 0.004 and p < 0.001, respectively). Treatment Outcomes: Onset-to-surgery time was longer in laparotomy (3.16 h) vs. laparoscopy (1.57 h) (p = 0.083, NS). Hormonal markers (AMH, FSH, LH) showed no significant differences between groups (all p > 0.09). Fecundity after treatment was similar between conservative (36.1%) and operative (30.6%) (p = 0.768). Torsion recurrence was rare (overall 3.7%) with no significant group differences. In summary, highly significant differences (p < 0.001): Age, menopause, gravidity, parity, diagnostic modalities (CT, MRI). Moderately significant (p < 0.05): Gravidity and parity (between both main groups and surgical subgroups), menopause, metabolic syndrome, loss of appetite. Borderline (p ≈ 0.05–0.1): BMI, vomiting, onset-to-surgery time, abdominal distension. Non-significant: Most comorbidities, surgical history, OCP/PCOS/infertility, recurrence rates, fecundity, hormone levels (Table 1). Objective 2: to identify the most informative clinical, demographic, and historical features influencing the management decision . In the second objective, we looked for possible associations between demographic, clinical histories, clinical data, and patient management type. Since the data studied did not follow the Gaussian distribution, we tested how tightly they are linked by computing Spearman rank correlation criterion. The analysis sought to determine whether there were significant associations between a range of demographic, medical history, and clinical variables and the management approach—operative vs conservative. Given the non-Gaussian distribution of the data, Spearman’s rank correlation coefficient (r) was used, along with associated p-values, to assess the strength and significance of correlations. Out of more than 60 tested variables, only three showed statistically significant associations (p < 0.05) with the treatment type: 1. History of Oral Contraceptive Pills (OCP) use (r = -0.199, p = 0.003) Interpretation: There was a marked negative correlation between the history of OCP use and undergoing operative treatment. This suggests that patients with a history of OCP use were more likely to be managed conservatively. This could imply a protective or modifying role of OCPs in disease progression or symptom control. 2. Presence of Pelvic Mass (ovarian cyst or other masses) (r = 0.134, p = 0.048) Interpretation: A weak positive correlation was observed, indicating that patients with pelvic masses were slightly more likely to undergo operative treatment. This aligns with clinical expectations, where structural pathology (e.g., cysts) often necessitates surgical intervention. 3. Number of Associated Symptoms (r = 0.148, p = 0.029) Interpretation: A strong positive association indicates that having more associated symptoms may tip the balance toward surgical management, likely due to symptom severity or diagnostic ambiguity. All other parameters show no significant associations. Objective 3. To assess the relationships among the studied variables and find their associations reflecting the risk factors of ovarian torsion, clinical symptoms, and disease management Methodology. Objective 3 aimed to explore complex interrelationships among demographic, clinical, and management-related variables in patients with ovarian torsion. A two-phase analytic approach was undertaken: 1. Principal Component Analysis (PCA): After standardization via mean normalization, PCA with Varimax rotation was performed to reduce dimensionality and identify latent structures in the data. Variables with eigenvalues ≥ 1.0 were retained. Loadings ≥ 0.75 were considered strong, 0.50–0.75 moderate, and < 0.50 weak (Table 2). This enabled the identification of key clinical syndromes and background features associated with ovarian torsion. Overall, the PCA structure supports the multidimensional nature of ovarian torsion presentations, capturing elements of age/reproductive history, acute symptomatology, systemic response, and surgical background. This factor structure was essential in informing the subsequent predictive modeling process. Results Principal Component Analysis Eight principal factors explaining 41.84% of the total variance were extracted (Table 2): - Factor 1: Demographic and reproductive history (age, parity, gravidity, menopause). - Factor 2: Acute GI and systemic symptoms (nausea, vomiting, total associated symptoms). - Factor 3: Genitourinary/metabolic symptoms (dysuria, obesity). - Factor 4: Acute abdominal symptoms (abdominal distension, loss of appetite). - Factor 5: Vaginal symptoms (discharge and bleeding). - Factor 6: Fever and systemic inflammatory features (fever, shivering). - Factor 7: Surgical history (previous surgeries, cesarean sections). - Factor 8: Comorbidity and reproductive outcome burden. These findings underscore the heterogeneity of ovarian torsion presentations and support the multidimensional clinical nature of the disease. 2. Cluster analysis: Figure 1 A and 1 B use the silhouette and gap statistic methods to identify the optimal number of clusters for classification, which was determined to be two. Figure 2 displays the corresponding cluster plot, revealing moderate separation based on variables like endometriosis, untreated obesity, and fever. Younger patients with flank pain and fainting dominated this cluster (Table 3). While higher burden of comorbidities, especially endometriosis, obesity, and inflammatory symptoms (fever, shivering, SOB) was observed in this cluster. Also notable was a higher incidence of constipation/diarrhea and a slightly higher number of associated symptoms (p = 0.061). Despite symptomatic differences, no significant variation in management strategies (conservative vs operative) was observed between clusters. Predictive Modeling and Classification Model performance varied substantially (Fig. 3 , Table 4): - Random Forest (ROSE) showed perfect performance (AUC = 1.0), likely due to overfitting. - Decision Tree (Weighted) was the best balanced non-overfitted model (AUC = 0.76), followed by Neural Network and Gradient Boosting (AUC = 0.68 each). Top predictors by information gain (Table 5) included epigastric pain, salpingectomy/oophorectomy, fever, smoking, and dysuria—variables also prominent in PCA and clustering results. Principal Component Analysis and Factor Structure (Table 2) Table 2 summarizes the principal component analysis (PCA), identifying eight factors with eigenvalues greater than 2.0. These factors cumulatively explain 41.84% of the total variance. High loadings on Factor 1 for variables such as age, gravidity, and parity suggest that demographic and reproductive history account for the largest proportion of variance. Medical history (e.g., prior surgeries) load heavily on Factor 7, while flu-like and GI symptoms are explained by Factors 5 and 6, respectively. Principal Component Analysis (PCA) with Varimax rotation extracted eight factors with eigenvalues greater than 2, collectively explaining 41.84% of the total variance in the dataset. The first factor (Factor 1) had the highest eigenvalue (6.76) and accounted for 10.73% of the variance alone, while the remaining factors contributed between 3.6% and 5.9% each. Factor 1, representing age-related reproductive and demographic characteristics, showed strong negative loadings for age (− 0.796), menopause (− 0.686), gravidity (− 0.858), and parity (− 0.821), indicating that these variables cluster strongly and may reflect a reproductive aging dimension. Factor 2 was heavily influenced by acute gastrointestinal and systemic symptoms, with strong negative loadings for nausea (− 0.717), vomiting (− 0.786), and the total number of associated symptoms (− 0.935). Factor 3 loaded significantly on variables such as dysuria (− 0.813) and obesity (− 0.507), suggesting a genitourinary or metabolic symptom cluster. Factor 4 was defined by acute abdominal symptomatology, with strong negative associations with abdominal distension (− 0.791), number of major symptoms (− 0.760), and loss of appetite (− 0.814), possibly representing an acute inflammatory or pain-related factor. Factor 5 loaded strongly on vaginal discharge and bleeding (− 0.852), indicating a distinct gynecologic symptom component. Factor 6 captured fever-related systemic symptoms, with fever (0.818) and shivering/shortness of breath (0.834) as dominant indicators. Factor 7 encompassed prior pelvic surgical history, loading highly on the number of previous surgeries (0.904), history of pelvic surgery (0.827), and caesarean sections (0.781–0.802). Factor 8 loaded most prominently on comorbidity burden, including number of comorbidities (0.703) and history of normal pregnancy (0.537), suggesting a general health and reproductive outcome dimension. Table 3 details the inter-cluster comparison. Cluster 2 patients had more comorbidities, including endometriosis and obesity, and a higher number of associated symptoms (e.g., fever, shortness of breath), while cluster 1 patients showed a higher incidence of flank pain and fainting (p < 0.001 for several comparisons). Figure 3 : Shows the ROC curves of various classification models used to predict treatment modality. The Random Forest model with ROSE oversampling achieved an AUC of 1.00, although this is likely due to overfitting. The Decision Tree model with class weighting performed best among non-overfitted models (AUC = 0.76), followed by Neural Networks and Gradient Boosting (AUC = 0.68 each). Table 4 compares the sensitivity and specificity of these models. While Random Forest (ROSE) scored perfect metrics, the Decision Tree and Neural Network models presented more balanced and reliable values. Table 5 Ranks features according to their predictive value using information gain, aligning closely with the visual information in Fig. 6 . Objective 4. Build a model to prognosticate the disease management plan (operative versus conservative) Methodology: Working on the fourth task, we trained machine learning models to predict the disease management plan. The model architecture we used is as follows: Random Forest, Decision Tree, Logistic Regression, CatBoost, and Neural Network. The major challenge was that the cases that necessitated operative treatment outnumbered those that did not require surgery. Training the models: We resorted to class weighting to address the issue of unbalanced data. Besides, we tested other data balancing techniques on the Random Forest model. These were Random Over-Sampling Examples (ROSE), Synthetic Minority Over-sampling Technique, and Edited Nearest Neighbors (SMOTE-ENN). To study feature importance, we computed the information gain value and ranked the predictors by importance. The performance metrics were the area under the receiver operating characteristic curve (ROC AUC), sensitivity, and specificity of detecting the positive class - patients with conservative treatment (Fig. 5 , 6 ). Discussion This study represents one of the first applications of machine learning (ML) to predict optimal management strategies in ovarian torsion cases. The findings provide insights into clinical patterns that influence surgical decision-making beyond traditional symptomatology and imaging. Our finding that OCP use is inversely related to surgical intervention aligns with previous literature highlighting the protective role of hormonal contraceptives in preventing or reducing the recurrence of benign ovarian pathology and endometriosis-related pain ( 8 ). OCPs are known to suppress ovulation and reduce the formation of functional cysts, which may reduce the urgency or need for surgical treatment. The positive correlation between pelvic mass presence and operative treatment is clinically expected and corroborated by existing evidence indicating that the presence of complex or large ovarian masses is a common indication for surgical exploration, particularly when torsion or malignancy is suspected ( 9 , 10 ). The correlation between the number of associated symptoms and surgery is also intuitive and supported by previous studies, which report that symptom burden, especially when involving acute pain, gastrointestinal complaints, or hemodynamic instability, often prompts surgical exploration to confirm or rule out torsion, rupture, or hemorrhage ( 11 ). Interestingly, despite trends seen in other gynecological conditions, factors such as age, BMI, parity, and history of infertility or endometriosis were not significantly associated with treatment decision in our cohort. This suggests that acute clinical presentation, rather than chronic demographic or historical factors, may weigh more heavily in emergent decision-making scenarios. The results suggest a nuanced clinical paradigm: OCP history may be a protective factor and a potential marker for favorable conservative management outcomes, potentially guiding patient counseling, and pelvic mass identification on imaging remains a strong indicator for operative intervention, reaffirming the role of targeted imaging in acute gynecological triage. The cumulative symptom burden appears to play a modest but significant role in determining the management path, suggesting a need for holistic symptom assessment rather than reliance on single indicators. However, the generally low correlation coefficients (r < 0.2) emphasize that these variables, while statistically significant, have limited predictive power individually. Therefore, clinical judgment, including real-time imaging, hemodynamic assessment, and pain severity, remains paramount. The predominance of surgical management is consistent with literature, emphasizing the urgency of operative intervention in torsion to prevent ovarian necrosis ( 3 ). However, conservative management was feasible in 16.4% of cases—primarily among premenopausal women with preserved Doppler flow and less severe clinical features. These findings support the need for nuanced clinical judgment and risk stratification. The high AUC from the Random Forest-ROSE model should be interpreted cautiously due to probable overfitting, a common issue in imbalanced data contexts ( 12 ). Conversely, the Decision Tree and GBM models offered more balanced and interpretable results, aligning with findings from previous clinical ML studies ( 6 , 13 ). Notably, factors such as epigastric pain, metabolic syndrome, smoking, and bundled symptoms (e.g., fainting with distension) emerged as strong predictors of management type. These variables, often underrecognized, may reflect broader inflammatory or ischemic states that influence physician decisions. Bundled symptoms may also serve as proxies for more severe systemic involvement or delayed presentation. Cluster and factor analyses highlighted two main phenotypic patterns: patients with comorbid metabolic and gynecologic conditions (e.g., endometriosis, untreated obesity) and those with more classical pain presentations. While cluster assignment did not significantly predict management choice, it revealed distinct patient profiles, useful for refining triage and diagnostic protocols. Importantly, torsion severity and ovarian viability—assessed intraoperatively—remain key determinants of oophorectomy. The model identified these as critical outcome-related variables, further emphasizing the need for early intervention to avoid irreversible ovarian damage. Our results show that very few variables were significantly associated with treatment type. However, most correlations were not strong even when statistically significant (r < 0.2), suggesting a possible non-linear association. Clinical Insight: The association between OCP use and conservative management may reflect the suppressive effect of hormonal regulation on symptoms or disease severity. The link between pelvic masses and operative treatment reaffirms surgical indications based on structural findings. Symptom burden may influence decision-making, but individual symptoms alone were not strong predictors. Limitations: Correlation does not imply causation, and low correlation coefficients suggest other factors. Hence, we recommend interpreting these findings narratively in the results and interpreting cautiously. They can inform future hypotheses, but are insufficient for building a predictive model or treatment algorithm. This integrated analysis confirmed that ovarian torsion is a multidimensional clinical entity, with symptoms clustering along demographic, gynecologic, metabolic, and inflammatory axes. Our findings align with previous literature highlighting the role of reproductive age, pelvic surgery history, and inflammatory markers in ovarian torsion risk and diagnosis. Age and reproductive history, major contributors to Factor 1, have been consistently associated with torsion risk. Huang et al and Damigos et al. ( 14 , 15 ) noted that torsion occurs more frequently in women of reproductive age and those with ovarian enlargement due to hormonal stimulation. Inflammatory symptoms (fever, shivering), forming Factors 5 and 6 and defining Cluster 2, mirror findings by ( 16 ), who showed elevated CRP and WBC counts in complicated adnexal torsion. Comorbid obesity and endometriosis, highlighted in Cluster 2, echo findings by ( 17 ). Chronic pelvic pathology may predispose to torsion due to altered adnexal mobility. Finally, surgical history, captured in Factor 7, remains a known risk factor for adnexal torsion recurrence or complications, as reported by ( 11 ). Our study further shows that classification algorithms can distinguish symptom clusters and predict management needs, but real-world application requires careful avoidance of overfitting, as seen with the Random Forest + ROSE model. Clinically, the Decision Tree and Neural Network models offer a realistic compromise in prediction utility. The use of PCA and clustering, seldom applied in prior torsion studies, allowed us to better represent underlying clinical dimensions, offering a nuanced approach to early detection, triage, and risk stratification. Clinical Implications and Utilization The prognostic model developed in this study holds substantial promise for enhancing decision-making in the management of ovarian torsion, a condition where timely and appropriate intervention is critical. By leveraging machine learning algorithms trained on real-world clinical data, the model provides a framework for early stratification of patients into those likely to benefit from immediate surgical intervention and those for whom conservative management may be safely considered. In practice, this model could be embedded into electronic health record (EHR) systems or clinical decision support tools (CDSTs) to assist gynecologists, emergency physicians, and surgical teams in triaging patients upon presentation. For example, a patient presenting with abdominal pain, preserved Doppler flow, and no prior pelvic surgery could be algorithmically flagged as a low-risk candidate for conservative observation, while a patient with significant risk features (e.g., epigastric pain, prior oophorectomy, smoking, and fever) could be prioritized for operative assessment. Importantly, the model’s ability to rank predictors by clinical importance allows for transparent, explainable AI applications, thereby supporting physician trust and shared decision-making with patients. Integration of such a tool could also reduce diagnostic delays, prevent unnecessary oophorectomies, and improve resource allocation in acute gynecological care settings. To ensure safe adoption, the model should undergo prospective validation across diverse healthcare systems. Future iterations may incorporate imaging parameters, real-time vitals, and patient-reported symptoms to further refine accuracy and usability. Our study has some strengths: first, it has an Innovative Methodology. This study is among the first to apply a machine learning (ML) framework to predict surgical versus conservative management decisions in ovarian torsion, offering a data-driven approach to an urgent gynecologic condition. Second, we have a comprehensive variable set: A wide range of clinical, demographic, and surgical predictors were considered, including traditionally underexamined features such as epigastric pain and bundled symptom clusters, enhancing model interpretability and clinical relevance. Third, we developed a Model of Comparison and Validation: Multiple ML algorithms were implemented and evaluated using standard metrics (AUC, sensitivity, specificity), allowing for robust comparison of performance and the identification of generalizable models such as Decision Trees and Gradient Boosting. Lastly, the Feature Importance Insights: The use of information gain and ranking provides transparency on influential clinical features, supporting a better understanding of torsion pathophysiology and management pathways. Limitations: Retrospective Single-Center Design: The data were collected retrospectively from a single tertiary center, which may limit generalizability to other populations or healthcare systems with different management protocols. Moreover, sample size and class Imbalance: The relatively small sample size, combined with the inherent imbalance between surgical and conservative cases, necessitated oversampling techniques such as ROSE and SMOTE. While effective, these methods may introduce synthetic bias or model overfitting. In addition, there is a Lack of Imaging Integration. Radiological findings (e.g., Doppler flow patterns or ovarian volume) were not incorporated into the model, potentially limiting predictive accuracy, especially in borderline cases. Unmeasured Confounding: Socioeconomic factors, patient preferences, and surgeon decision-making biases were not accounted for but may have influenced management choices. Finally, there is no prospective validation: external validation on a separate dataset was not conducted, and prospective testing is needed to confirm model utility in real-time clinical decision-making. Despite limitations due to class imbalance and possible data heterogeneity, the study provides a strong foundation for developing a clinically useful decision support tool. Further prospective validation is needed ( 12 ). Future directions should involve prospective multicenter validation, external testing on heterogeneous populations, and integration with imaging features (e.g., MRI data). Development of a real-time clinical decision support tool embedded in electronic medical records may enhance usability and generalization. Feature importance analysis identified epigastric pain, prior pelvic surgeries, fever, smoking, and metabolic comorbidities as key predictors. Surprisingly, bundled symptom patterns—like abdominal distension combined with fainting—also contributed strongly, though their clinical specificity remains questionable. These findings suggest that clinician decisions are influenced not only by classical torsion symptoms but also by patient history and comorbid status. Our findings confirm that surgical intervention remains the cornerstone of OT management, but conservative treatment may be appropriate in carefully selected cases—particularly those with fewer symptoms, OCP use, and absence of a pelvic mass. ML-based models offer an additional decision-support layer, enhancing risk stratification and potentially reducing unnecessary surgeries. Clinically, integrating such models into electronic health records or decision support systems could allow real-time triage, improve diagnostic accuracy, and support shared decision-making with patients. This is particularly relevant for reproductive-aged women, where fertility preservation is paramount. Prospective, multicenter validation is the next step to ensure safe and generalizable application. The identification of novel predictive features—such as epigastric pain, metabolic comorbidities, and prior pelvic surgeries—enhances clinical understanding and supports risk-based triage. While the model shows high predictive performance, particularly with decision tree–based algorithms, external validation and prospective implementation are necessary to confirm its applicability in routine care. Ultimately, this study highlights that combining clinical expertise with data-driven tools can optimize management of ovarian torsion, reduce morbidity, and safeguard reproductive outcomes. Declarations Ethics approval and consent to participate: This retrospective study was conducted in accordance with the principles of the Declaration of Helsinki. Ethical approval was obtained from the Institutional Review Board of the Abu Dhabi Health Authority, Tawam Hospital (Ref. No.: HREC SEHA-IRB-942, dated 21 November 2024). As this was a retrospective analysis of de-identified patient data, individual consent to participate was waived by the IRB. The study does not contain any person’s data in any form. Conflict of Interest Disclosure The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Funding: This study was funded by Abu Dhabi Health Authority at Tawam Hospital. Authors’ Contributions MA: Conceptualization, Methodology, Data Curation, Supervision, Writing – Review & Editing Acknowledgment: We acknowledge that generative AI technologies were used to improve the style, accessibility, and quality of human-generated text and images. Data Availability The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request. References Abraham M, Keyser EA (2019) Adnexal torsion in adolescents. Obstet Gynecol 134(2):E56–E63 Huang C, Hong M-K, Ding D-C (2017) A review of ovary torsion. Tzu Chi Med J 29(3):143–147 Ryles HT, Hong CX, Andy UU, Farrow MR (2023) Changing practices in the surgical management of adnexal torsion: an analysis of the National Surgical Quality Improvement Program Database. Obstet Gynecol 141(5):888–896 Akman L, Erbas O, Terek M, Aktug H, Taskiran D, Askar N (2016) The long pentraxin-3 is a useful marker for diagnosis of ovarian torsion: An experimental rat model. J Obstet Gynaecol 36(3):399–402 Brierley G, Arshad I, Shakir F, Visvathanan D, Arambage K (2020) Vascular injury during laparoscopic gynaecological surgery: a methodological approach for prevention and management. Obstetrician Gynaecologist 22(3):191–198 Breiman L (2001) Random forests. Mach Learn 45(1):5–32 Esteva A, Kuprel B, Novoa RA, Ko J, Swetter SM, Blau HM et al (2017) Dermatologist-level classification of skin cancer with deep neural networks. Nature 542(7639):115–118 Vercellini P, Bandini V, Vigano P, Di Stefano G, Merli CEM, Somigliana E (2024) Proposal for targeted, neo-evolutionary-oriented, secondary prevention of early-onset endometriosis and adenomyosis. Part I: pathogenic aspects. Hum Reprod 39(1):1–17 Jeong Y-Y, Outwater EK, Kang HK (2000) Imaging evaluation of ovarian masses. Radiographics 20(5):1445–1470 Muto MG (2020) Approach to the patient with an adnexal mass. UpToDate, Waltham, MA.(Accessed on October 3, 2023.) Houry D, Abbott JT (2001) Ovarian torsion: a fifteen-year review. Ann Emerg Med 38(2):156–159 Johnson JM, Khoshgoftaar TM (2019) Survey on deep learning with class imbalance. J big data 6(1):1–54 Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O et al (2011) Scikit-learn: Machine learning in Python. J Mach Learn Res 12:2825–2830 Chen Y-C, Lee M-C, Lee C-J, Hsu B-G (2018) Hyperleptinemia is associated with the aortic augmentation index in kidney transplant recipients. Tzu Chi Med J 30(3):152–157 Damigos E, Johns J, Ross J (2012) An update on the diagnosis and management of ovarian torsion. Obstetrician Gynaecologist. ;14(4) Hasson J, Tsafrir Z, Azem F, Bar-On S, Almog B, Mashiach R et al (2010) Comparison of adnexal torsion between pregnant and nonpregnant women. Am J Obstet Gynecol 202(6):536 e1-. e6 Goossens E, Van Saen D, Tournaye H (2013) Spermatogonial stem cell preservation and transplantation: from research to clinic. Hum Reprod. ;28(4) Tables Tables 1 to 5 are available in the Supplementary Files section. Additional Declarations The authors declare no competing interests. Supplementary Files Table1.docx Comparison of demographic, clinical, and surgical characteristics between conservative and operative treatment groups in ovarian torsion. Table2.docx Principal component analysis: Eigenvalues and rotated factor loadings of significant variables. Table3.docx Statistical comparison between Cluster 1 and Cluster 2 on demographic, surgical, and clinical features. Table4.docx Performance metrics (AUC, sensitivity, specificity) of classification models. Table5.docx Feature importance ranking based on information gain. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7828438","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":527763965,"identity":"1409d887-0fe1-4241-ba1c-2dc817bb30a7","order_by":0,"name":"Alia Alethawy","email":"","orcid":"","institution":"Department of Obstetrics and Gynecology, Tawam Hospital, Al Ain, UAE","correspondingAuthor":false,"prefix":"","firstName":"Alia","middleName":"","lastName":"Alethawy","suffix":""},{"id":527763966,"identity":"74c39e56-4286-400e-bc48-3ef9dd533d49","order_by":1,"name":"Omaima Al-Baghdadi","email":"","orcid":"","institution":"Department of Obstetrics and Gynecology, Tawam Hospital, Al Ain, UAE","correspondingAuthor":false,"prefix":"","firstName":"Omaima","middleName":"","lastName":"Al-Baghdadi","suffix":""},{"id":527763967,"identity":"a078ef43-46c9-40b1-8270-8eb55c63f932","order_by":2,"name":"Yauhen Statsenko","email":"","orcid":"","institution":"Department of Radiology, CMHS, UAEU, Al Ain, UAE","correspondingAuthor":false,"prefix":"","firstName":"Yauhen","middleName":"","lastName":"Statsenko","suffix":""},{"id":527763968,"identity":"b014c31f-b6a5-4f78-ada1-402df25eaa53","order_by":3,"name":"Moamar Al-Jefout","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA1klEQVRIiWNgGAWjYDCCA1Can4GBDSbGTJwWyQaStRgcIFYL3+3TiQ9+5tjlG99IfvbgQwWDPD8Dj7EBPi2S53I3G/ZuS7bcdiPN3HDGGQbDmQ08xgn4tBic4d0mwbuN2cDsRoKZNG8bQ4LBAR7jA4S0SP7dVm9gPCP9G/FapHm3HTYwkMhB2ILXYZJneDcby247biBx5k2Z5IwzEoYzm9mK8Xqf7wzvxodvt1Ub8Lenb5P4UGEjz8/evFkCnxYEEAC7RoJgRCIB/gNEKx0Fo2AUjIIRBgA25EReqDx2CgAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0000-0002-3720-3237","institution":"Department of Obstetrics and Gynecology, CMHS, UAEU, Al Ain, UAE","correspondingAuthor":true,"prefix":"","firstName":"Moamar","middleName":"","lastName":"Al-Jefout","suffix":""}],"badges":[],"createdAt":"2025-10-10 15:13:35","currentVersionCode":1,"declarations":{"humanSubjects":true,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":true,"humanSubjectConsent":true,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-7828438/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7828438/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":93956746,"identity":"5755173f-cc04-45e0-941d-a936dac6bc08","added_by":"auto","created_at":"2025-10-20 16:12:05","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":66426,"visible":true,"origin":"","legend":"","description":"","filename":"OvarianTorsionPrognosticModelfinaldraft.docx","url":"https://assets-eu.researchsquare.com/files/rs-7828438/v1/2283c5d4bbcbd3bd280b22b3.docx"},{"id":93956727,"identity":"6cc29cad-d2b8-435d-b0f6-c2a29cec567e","added_by":"auto","created_at":"2025-10-20 16:11:58","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":342,"visible":true,"origin":"","legend":"","description":"","filename":"rs7828438.json","url":"https://assets-eu.researchsquare.com/files/rs-7828438/v1/dfce440e12f661f1064216f2.json"},{"id":93956730,"identity":"bdf90153-de16-4d1e-ab82-f0129579bfb1","added_by":"auto","created_at":"2025-10-20 16:11:59","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":79962,"visible":true,"origin":"","legend":"","description":"","filename":"rs78284380enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7828438/v1/13cb9d4febd5be6771fb4207.xml"},{"id":93956725,"identity":"777ce706-fc68-4c2d-8775-b8b320cf48fd","added_by":"auto","created_at":"2025-10-20 16:11:58","extension":"xml","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":76668,"visible":true,"origin":"","legend":"","description":"","filename":"rs78284380structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7828438/v1/6ceb69437522f5363e5729be.xml"},{"id":93956685,"identity":"1d2727f4-559c-4f53-b337-c2a5a192fce3","added_by":"auto","created_at":"2025-10-20 16:11:55","extension":"html","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":87114,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7828438/v1/0a27b7efa4f0cf43607033f2.html"},{"id":93956726,"identity":"cf5f7879-f2c1-4c16-a70c-7cf539117451","added_by":"auto","created_at":"2025-10-20 16:11:58","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":47073,"visible":true,"origin":"","legend":"\u003cp\u003eHeatmap of p-values for clinical variable comparison between treatment groups.\u003c/p\u003e","description":"","filename":"figure1image.png","url":"https://assets-eu.researchsquare.com/files/rs-7828438/v1/5d314b0df13c8a6f9d981ec9.png"},{"id":93956748,"identity":"4be485db-7c64-4b43-a38a-3d92d2811d11","added_by":"auto","created_at":"2025-10-20 16:12:06","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":132939,"visible":true,"origin":"","legend":"\u003cp\u003e· \u003cstrong\u003e2A \u0026amp; 2B\u003c/strong\u003e. Cluster number determination using the silhouette method (A) and gap statistic method (B).\u003c/p\u003e","description":"","filename":"Figure2image.png","url":"https://assets-eu.researchsquare.com/files/rs-7828438/v1/ddd684c7c821a03d912e2637.png"},{"id":93956752,"identity":"122da9b8-f5a1-43c9-9706-a79ed3f895ea","added_by":"auto","created_at":"2025-10-20 16:12:06","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":51005,"visible":true,"origin":"","legend":"\u003cp\u003eCluster plot showing separation between patient groups based on comorbidities and symptoms.\u003c/p\u003e","description":"","filename":"Figure3image.png","url":"https://assets-eu.researchsquare.com/files/rs-7828438/v1/ada13b052074715c022a4573.png"},{"id":93956735,"identity":"48dd49ad-ddcc-4c91-883c-72b074d27099","added_by":"auto","created_at":"2025-10-20 16:12:00","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":23160,"visible":true,"origin":"","legend":"\u003cp\u003eVisualization of class imbalance in the study dataset for treatment modalities\u003c/p\u003e","description":"","filename":"Figure4image.png","url":"https://assets-eu.researchsquare.com/files/rs-7828438/v1/00568cdd806400489adac1ac.png"},{"id":93956690,"identity":"a9539835-ee2f-476e-af4d-34de6c46ee8a","added_by":"auto","created_at":"2025-10-20 16:11:56","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":101115,"visible":true,"origin":"","legend":"\u003cp\u003eROC curves for machine learning models used in predicting treatment decisions.\u003c/p\u003e","description":"","filename":"Figure5image.png","url":"https://assets-eu.researchsquare.com/files/rs-7828438/v1/7f78f29416898a773774d3a7.png"},{"id":93956708,"identity":"863708db-3a0d-4f3d-bc2e-9c159b1f9a91","added_by":"auto","created_at":"2025-10-20 16:11:56","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":124165,"visible":true,"origin":"","legend":"\u003cp\u003eTop 20 predictive features ranked by information gain for Random Forest (ROSE) model.\u003c/p\u003e","description":"","filename":"Figure6image.png","url":"https://assets-eu.researchsquare.com/files/rs-7828438/v1/ebe62f56b1a563a29b86911a.png"},{"id":93957146,"identity":"a5e3d19b-69da-497b-a513-e37fd66439ac","added_by":"auto","created_at":"2025-10-20 16:13:03","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1189181,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7828438/v1/1d89ecb4-0824-4bb4-bdbc-58c6fad05cc0.pdf"},{"id":93956675,"identity":"ba729b3b-4f5f-412e-b2b9-8228997a7663","added_by":"auto","created_at":"2025-10-20 16:11:53","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":26350,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of demographic, clinical, and surgical characteristics between conservative and operative treatment groups in ovarian torsion.\u003c/p\u003e","description":"","filename":"Table1.docx","url":"https://assets-eu.researchsquare.com/files/rs-7828438/v1/e1941293c82d0268936ef8fa.docx"},{"id":93956677,"identity":"e4e93bd8-7357-4c10-8582-0425bbc2a533","added_by":"auto","created_at":"2025-10-20 16:11:53","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":24170,"visible":true,"origin":"","legend":"\u003cp\u003ePrincipal component analysis: Eigenvalues and rotated factor loadings of significant variables.\u003c/p\u003e","description":"","filename":"Table2.docx","url":"https://assets-eu.researchsquare.com/files/rs-7828438/v1/f82ab2a270f1369f303d48fc.docx"},{"id":93956710,"identity":"6964680c-00d0-4567-8bf5-f82b9cd55435","added_by":"auto","created_at":"2025-10-20 16:11:56","extension":"docx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":17251,"visible":true,"origin":"","legend":"\u003cp\u003eStatistical comparison between Cluster 1 and Cluster 2 on demographic, surgical, and clinical features.\u003c/p\u003e","description":"","filename":"Table3.docx","url":"https://assets-eu.researchsquare.com/files/rs-7828438/v1/67f93f63aa2864d0487c2a93.docx"},{"id":93956722,"identity":"cdd4fa9f-2ebe-4825-aae2-2ca641511f77","added_by":"auto","created_at":"2025-10-20 16:11:57","extension":"docx","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":15454,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance metrics (AUC, sensitivity, specificity) of classification models.\u003c/p\u003e","description":"","filename":"Table4.docx","url":"https://assets-eu.researchsquare.com/files/rs-7828438/v1/1523cbee0549018a06f9715e.docx"},{"id":93956669,"identity":"b26b0c9f-b59e-4127-aea3-2aa0007285d7","added_by":"auto","created_at":"2025-10-20 16:11:52","extension":"docx","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":15365,"visible":true,"origin":"","legend":"\u003cp\u003eFeature importance ranking based on information gain.\u003c/p\u003e","description":"","filename":"Table5.docx","url":"https://assets-eu.researchsquare.com/files/rs-7828438/v1/756c0238e452362114bdf574.docx"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003eOptimization of Management Plan with a machine learning model for Ovarian Torsion Cases: Operative versus Conservative\u003c/p\u003e","fulltext":[{"header":"Introduction","content":"\u003cp\u003eOvarian torsion (OT) is a gynecological emergency that requires prompt diagnosis and management to preserve ovarian function and prevent serious complications. OT, when it is associated with tubal torsion, is termed an adnexal torsion. The annual prevalence is estimated to be around 9.9/100,000 women of reproductive age (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e). The pathophysiology of OT involves torsion of the ovary on its pedicle, leading to compromised venous return, stromal swelling, internal bleeding, and ischemia, which may lead to necrosis. Timely and accurate decisions regarding surgical intervention versus conservative management are crucial for preserving the reproductive function of women with this pathology. Classical diagnostic criteria focus heavily on clinical signs such as abdominal pain, nausea, and vomiting (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e), which are normal clinical presentations, and imaging (Doppler scan usually shows reduced ovarian blood supply), which may not fully capture the complexity of torsion cases.\u003c/p\u003e\u003cp\u003eOvarian torsion is usually managed surgically, normally in the form of diagnostic laparoscopy, which is the gold standard for the diagnosis of OT; furthermore, it enables treatment, either by oophorectomy or detorsion, although recent publications showed a modest decrease in the proportion of oophorectomies for ovarian torsion (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e). However, there is still a risk of serious complications from laparoscopy in approximately 2 per 1000 cases (\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e), which can include organ trauma or major vessel injury, both of which harbor significant morbidity and mortality (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThere is some confusion in the literature about which factors are responsible for the development of ovarian torsion, and what the odds are of a particular clinical feature in determining the likelihood of developing ovarian torsion. Moreover, clinical decision-making is often challenged by nonspecific symptoms and overlapping features with other acute abdominal conditions. Hence, it is vital to develop a prognostic Model Management Plan (MMP) determining either operative or Conservative (non-surgical) management for Cases with Ovarian Torsion. Machine Learning (ML) methods offer the potential to enhance prognostic accuracy by analysing large, multidimensional clinical datasets and uncovering non-linear patterns that may not be eLearning clinicians. The application of ML to ovarian torsion may facilitate the development of predictive models that inform real-time decision-making, guiding whether operative or conservative management is more appropriate for individual patients. This study aims to develop and evaluate ML-based models that predict management outcomes in ovarian torsion using a diverse clinical dataset (\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eOur study aims to achieve four objectives: \u003cb\u003eObjective 1\u003c/b\u003e is to develop and validate a predictive model that distinguishes between cases of ovarian torsion requiring operative intervention and those suitable for conservative management. \u003cb\u003eObjective 2\u003c/b\u003e: to identify the most informative clinical, demographic, and historical features influencing the management decision. \u003cb\u003eObjective 3\u003c/b\u003e: to assess relationships among studied variables and find their associations reflecting the risk factors of ovarian torsion, clinical symptoms, and disease management. \u003cb\u003eObjective 4\u003c/b\u003e: to build a model prognosticating the suitable disease management plan (operative versus conservative).\u003c/p\u003e"},{"header":"Methodology","content":"\u003cp\u003eThis retrospective cohort study was conducted at Tawam Hospital and included all women diagnosed with ovarian torsion between September 1, 2022, and September 1, 2024. A total of 219 patients were identified through electronic medical records. This retrospective study aimed to develop and validate a machine learning\u0026ndash;based prognostic model to support clinical decision-making in the management of ovarian torsion. A total of 219 female patients aged one month to 75 years who presented with suspected ovarian torsion to a tertiary care center were included. Data collection spanned clinical, demographic, laboratory, and imaging variables, as well as intraoperative findings and treatment outcomes. Ethical approval was obtained before data collection, and all patient identifiers were removed to ensure confidentiality (Date: 11/21/2024, Ref. No.: HREC SEHA-IRB- 942).\u003c/p\u003e\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eData Collection and Preprocessing\u003c/h2\u003e\u003cp\u003eData were extracted from electronic medical records and included over 60 variables: age, BMI, parity, menopausal status, gynecological and surgical history, comorbidities (e.g., PCOS, endometriosis, metabolic syndrome), presenting symptoms (e.g., abdominal pain, vomiting, fever), imaging modalities (ultrasound, CT, MRI), hormonal markers (AMH, FSH, LH), and post-treatment fertility outcomes. Missing values were handled via complete case analysis or imputation as appropriate. All continuous variables were tested for normality using the Shapiro\u0026ndash;Wilk test and showed non-Gaussian distribution.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003eStatistical Analysis\u003c/h2\u003e\u003cp\u003eComparative analyses between conservatively and operatively managed groups were performed using the Mann\u0026ndash;Whitney U test for continuous variables and Pearson\u0026rsquo;s chi-square test for categorical variables. Spearman\u0026rsquo;s rank correlation coefficients (r) were calculated to examine associations between individual variables and treatment type. A p-value\u0026thinsp;\u0026lt;\u0026thinsp;0.05 was considered statistically significant.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eDimensionality Reduction and Pattern Recognition\u003c/h3\u003e\n\u003cp\u003ePrincipal Component Analysis (PCA) with Varimax rotation was used to identify latent patterns and reduce dimensionality. Variables with eigenvalues\u0026thinsp;\u0026ge;\u0026thinsp;1 were retained, and factor loadings\u0026thinsp;\u0026ge;\u0026thinsp;0.75 were considered strong contributors. PCA revealed eight principal components, each capturing distinct clinical domains such as reproductive history, gastrointestinal symptoms, systemic inflammatory features, and previous surgical history.\u003c/p\u003e\u003cp\u003eCluster analysis using silhouette scores and gap statistics was then conducted to stratify patients into clinically relevant subgroups. Two distinct clusters were identified, representing different phenotypic profiles of ovarian torsion presentations.\u003c/p\u003e\n\u003ch3\u003ePredictive Modeling and Machine Learning\u003c/h3\u003e\n\u003cp\u003eSupervised machine learning models were developed to classify patients according to the appropriate management strategy (operative vs. conservative). Five algorithms were tested: Random Forest, Decision Tree, Neural Network, CatBoost Gradient Boosting, and Logistic Regression. Due to the significant class imbalance (only 16.4% received conservative management), class weighting and resampling techniques were used to improve model performance. These included Random Over-Sampling Examples (ROSE), Synthetic Minority Over-sampling Technique (SMOTE), and SMOTE-Edited Nearest Neighbors (SMOTE-ENN). Model performance was assessed using receiver operating characteristic (ROC) analysis. The area under the ROC curve (AUC), along with sensitivity and specificity, was calculated for each model.\u003c/p\u003e\n\u003ch3\u003eFeature Importance Analysis\u003c/h3\u003e\n\u003cp\u003eTo identify the most influential predictors, information gain was computed for each variable. The top-ranking features included epigastric pain, prior salpingectomy or oophorectomy, smoking history, fever, dysuria, and bundled symptom clusters (e.g., distension and fainting). These were consistent with findings from PCA and clustering analyses.\u003c/p\u003e\u003cp\u003eIn summary, our primary outcomes are accuracy metrics of the predictive model in classifying patients with ovarian torsion into operative versus conservative management groups. This includes Area Under the ROC Curve (AUC), Sensitivity and Specificity, and Correct classification rate for conservative management. Secondary Outcomes: Identification of significant clinical, demographic, and historical predictors associated with the type of management (e.g., OCP use, pelvic mass, number of symptoms). Model explainability through feature importance ranking (e.g., information gain, decision tree splits). Performance comparison among machine learning algorithms (Random Forest, Decision Tree, Neural Network, Gradient Boosting, Logistic Regression) in predicting treatment type. Evaluation of post-treatment reproductive outcomes, including spontaneous conception rate, Recurrence of ovarian torsion, Assessment of symptom clusters and phenotypic patterns using principal component and cluster analysis to characterize distinct clinical presentations, and feasibility of integrating the predictive model into a clinical decision support tool (CDST) for use in acute gynecological settings.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\u003ch2\u003eObjective 1: Associations Between Demographic, Clinical Histories, and Clinical Data, and Treatment Type:\u003c/h2\u003e\u003cp\u003eIn the first objective, we assessed the differences between the demographic, clinical histories, and clinical data between the groups with conservative and operative management. To make the comparison, we used the Mann-Whitney U-test correlation due to the non-Gaussian distribution of variables.\u003c/p\u003e\u003cp\u003eA total of 219 women were included in the study, of whom 38 (17.4%) were managed conservatively and 177 (80.8%) underwent operative intervention. A subgroup analysis was performed on 179 patients with complete reproductive data.\u003c/p\u003e\u003cp\u003eTable\u0026nbsp;1 compares demographic, clinical, surgical, and outcome variables between two treatment strategies for ovarian torsion: Conservative management (n\u0026thinsp;=\u0026thinsp;36) and Operative management (n\u0026thinsp;=\u0026thinsp;183). It also stratifies operative cases into laparotomy (n\u0026thinsp;=\u0026thinsp;28) and laparoscopy (n\u0026thinsp;=\u0026thinsp;155). Continuous variables are presented as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD, categorical variables as % (n). P-values reflect comparisons between groups.\u003c/p\u003e\u003cp\u003eDemographic Characteristics: Age differed significantly between groups: Conservative vs. operative (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) and laparotomy vs. laparoscopy (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). In comparison, operative patients were slightly older on average, with laparotomy patients markedly older (38.4\u0026thinsp;\u0026plusmn;\u0026thinsp;17.5 y) than laparoscopy patients (26.5\u0026thinsp;\u0026plusmn;\u0026thinsp;10.7 y). BMI was somewhat higher in operative patients (27.0\u0026thinsp;\u0026plusmn;\u0026thinsp;6.4) than in those patients with conservative management (25.7\u0026thinsp;\u0026plusmn;\u0026thinsp;5.5), borderline significant (p\u0026thinsp;=\u0026thinsp;0.058). Similar borderline difference between laparotomy and laparoscopy (p\u0026thinsp;=\u0026thinsp;0.058). Menopause was more frequent in operative (11.5%) vs. conservative (5.6%) (p\u0026thinsp;=\u0026thinsp;0.040), and much higher in laparotomy (32.1%) vs. laparoscopy (7.7%) (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001).\u003c/p\u003e\u003cp\u003eObstetric History: Gravidity and parity were significantly higher in operative vs. conservative (p\u0026thinsp;=\u0026thinsp;0.004, p\u0026thinsp;=\u0026thinsp;0.003), and higher in laparotomy vs. laparoscopy (p\u0026thinsp;=\u0026thinsp;0.002, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Miscarriage rates showed no significant differences (p\u0026thinsp;\u0026gt;\u0026thinsp;0.7).\u003c/p\u003e\u003cp\u003ePast Surgical/Medical History: No significant differences for prior pelvic surgery, appendectomy, cystectomy, or caesarean section (all p\u0026thinsp;\u0026gt;\u0026thinsp;0.3). History of endometriosis, OCP use, PID, or PCOS showed no group differences (all p\u0026thinsp;\u0026gt;\u0026thinsp;0.7).\u003c/p\u003e\u003cp\u003eComorbidities: Overall comorbidities count similarly across groups (p\u0026thinsp;\u0026gt;\u0026thinsp;0.2). Metabolic syndrome was rare but significantly more frequent in laparotomy vs. laparoscopy (3.6% vs. 0%; p\u0026thinsp;=\u0026thinsp;0.019).\u003c/p\u003e\u003cp\u003eSymptoms: Unilateral lower abdominal pain was the most frequent symptom (91%), with no significant group difference (p\u0026thinsp;\u0026gt;\u0026thinsp;0.2). Vomiting was more common in operative than conservative (41.5% vs. 30.6%; p\u0026thinsp;=\u0026thinsp;0.101, NS) and trended lower in laparotomy vs. laparoscopy (25% vs. 44.5%; p\u0026thinsp;=\u0026thinsp;0.055). Loss of appetite was significantly more frequent in the laparotomy group (10.7%) than in the laparoscopy group (2.6%) (p\u0026thinsp;=\u0026thinsp;0.040).\u003c/p\u003e\u003cp\u003eDiagnostic Modalities: Conservative patients underwent fewer imaging investigations overall (mean number lower; p\u0026thinsp;=\u0026thinsp;0.012). CT abdomen/pelvis and MRI pelvis were used more often in laparotomy vs. laparoscopy (p\u0026thinsp;=\u0026thinsp;0.004 and p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, respectively).\u003c/p\u003e\u003cp\u003eTreatment Outcomes: Onset-to-surgery time was longer in laparotomy (3.16 h) vs. laparoscopy (1.57 h) (p\u0026thinsp;=\u0026thinsp;0.083, NS). Hormonal markers (AMH, FSH, LH) showed no significant differences between groups (all p\u0026thinsp;\u0026gt;\u0026thinsp;0.09). Fecundity after treatment was similar between conservative (36.1%) and operative (30.6%) (p\u0026thinsp;=\u0026thinsp;0.768). Torsion recurrence was rare (overall 3.7%) with no significant group differences.\u003c/p\u003e\u003cp\u003eIn summary, highly significant differences (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001): Age, menopause, gravidity, parity, diagnostic modalities (CT, MRI). Moderately significant (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05): Gravidity and parity (between both main groups and surgical subgroups), menopause, metabolic syndrome, loss of appetite. Borderline (p\u0026thinsp;\u0026asymp;\u0026thinsp;0.05\u0026ndash;0.1): BMI, vomiting, onset-to-surgery time, abdominal distension. Non-significant: Most comorbidities, surgical history, OCP/PCOS/infertility, recurrence rates, fecundity, hormone levels (Table\u0026nbsp;1).\u003c/p\u003e\u003cp\u003e\u003cb\u003eObjective 2: to identify the most informative clinical, demographic, and historical features influencing the management decision\u003c/b\u003e.\u003c/p\u003e\u003cp\u003eIn the second objective, we looked for possible associations between demographic, clinical histories, clinical data, and patient management type. Since the data studied did not follow the Gaussian distribution, we tested how tightly they are linked by computing Spearman rank correlation criterion. The analysis sought to determine whether there were significant associations between a range of demographic, medical history, and clinical variables and the management approach\u0026mdash;operative vs conservative.\u003c/p\u003e\u003cp\u003eGiven the non-Gaussian distribution of the data, Spearman\u0026rsquo;s rank correlation coefficient (r) was used, along with associated p-values, to assess the strength and significance of correlations.\u003c/p\u003e\u003cp\u003eOut of more than 60 tested variables, only three showed statistically significant associations (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05) with the treatment type:\u003c/p\u003e\u003cp\u003e1. History of Oral Contraceptive Pills (OCP) use (r = -0.199, p\u0026thinsp;=\u0026thinsp;0.003)\u003c/p\u003e\u003cp\u003eInterpretation: There was a marked negative correlation between the history of OCP use and undergoing operative treatment. This suggests that patients with a history of OCP use were more likely to be managed conservatively. This could imply a protective or modifying role of OCPs in disease progression or symptom control.\u003c/p\u003e\u003cp\u003e2. Presence of Pelvic Mass (ovarian cyst or other masses) (r\u0026thinsp;=\u0026thinsp;0.134, p\u0026thinsp;=\u0026thinsp;0.048)\u003c/p\u003e\u003cp\u003eInterpretation: A weak positive correlation was observed, indicating that patients with pelvic masses were slightly more likely to undergo operative treatment. This aligns with clinical expectations, where structural pathology (e.g., cysts) often necessitates surgical intervention.\u003c/p\u003e\u003cp\u003e3. Number of Associated Symptoms (r\u0026thinsp;=\u0026thinsp;0.148, p\u0026thinsp;=\u0026thinsp;0.029)\u003c/p\u003e\u003cp\u003eInterpretation: A strong positive association indicates that having more associated symptoms may tip the balance toward surgical management, likely due to symptom severity or diagnostic ambiguity. All other parameters show no significant associations.\u003c/p\u003e\u003cp\u003e\u003cb\u003eObjective 3. To assess the relationships among the studied variables and find their associations reflecting the risk factors of ovarian torsion, clinical symptoms, and disease management\u003c/b\u003e\u003c/p\u003e\u003cp\u003eMethodology.\u003c/p\u003e\u003cp\u003eObjective 3 aimed to explore complex interrelationships among demographic, clinical, and management-related variables in patients with ovarian torsion. A two-phase analytic approach was undertaken:\u003c/p\u003e\u003cp\u003e1. Principal Component Analysis (PCA):\u003c/p\u003e\u003cp\u003eAfter standardization via mean normalization, PCA with Varimax rotation was performed to reduce dimensionality and identify latent structures in the data. Variables with eigenvalues\u0026thinsp;\u0026ge;\u0026thinsp;1.0 were retained. Loadings\u0026thinsp;\u0026ge;\u0026thinsp;0.75 were considered strong, 0.50\u0026ndash;0.75 moderate, and \u0026lt;\u0026thinsp;0.50 weak (Table\u0026nbsp;2). This enabled the identification of key clinical syndromes and background features associated with ovarian torsion. Overall, the PCA structure supports the multidimensional nature of ovarian torsion presentations, capturing elements of age/reproductive history, acute symptomatology, systemic response, and surgical background. This factor structure was essential in informing the subsequent predictive modeling process.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eResults\u003c/h3\u003e\n\u003cp\u003ePrincipal Component Analysis\u003c/p\u003e\u003cp\u003eEight principal factors explaining 41.84% of the total variance were extracted (Table\u0026nbsp;2):\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003e- Factor 1: Demographic and reproductive history (age, parity, gravidity, menopause).\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e- Factor 2: Acute GI and systemic symptoms (nausea, vomiting, total associated symptoms).\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e- Factor 3: Genitourinary/metabolic symptoms (dysuria, obesity).\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e- Factor 4: Acute abdominal symptoms (abdominal distension, loss of appetite).\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e- Factor 5: Vaginal symptoms (discharge and bleeding).\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e- Factor 6: Fever and systemic inflammatory features (fever, shivering).\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e- Factor 7: Surgical history (previous surgeries, cesarean sections).\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e- Factor 8: Comorbidity and reproductive outcome burden.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003eThese findings underscore the heterogeneity of ovarian torsion presentations and support the multidimensional clinical nature of the disease.\u003c/p\u003e\u003cp\u003e2. Cluster analysis:\u003c/p\u003e\u003cp\u003eFigure \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eA and \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eB use the silhouette and gap statistic methods to identify the optimal number of clusters for classification, which was determined to be two. Figure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e displays the corresponding cluster plot, revealing moderate separation based on variables like endometriosis, untreated obesity, and fever.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eYounger patients with flank pain and fainting dominated this cluster (Table\u0026nbsp;3). While higher burden of comorbidities, especially endometriosis, obesity, and inflammatory symptoms (fever, shivering, SOB) was observed in this cluster. Also notable was a higher incidence of constipation/diarrhea and a slightly higher number of associated symptoms (p\u0026thinsp;=\u0026thinsp;0.061).\u003c/p\u003e\u003cp\u003eDespite symptomatic differences, no significant variation in management strategies (conservative vs operative) was observed between clusters.\u003c/p\u003e\u003cp\u003ePredictive Modeling and Classification\u003c/p\u003e\u003cp\u003eModel performance varied substantially (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, Table\u0026nbsp;4):\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003e- Random Forest (ROSE) showed perfect performance (AUC\u0026thinsp;=\u0026thinsp;1.0), likely due to overfitting.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e- Decision Tree (Weighted) was the best balanced non-overfitted model (AUC\u0026thinsp;=\u0026thinsp;0.76), followed by Neural Network and Gradient Boosting (AUC\u0026thinsp;=\u0026thinsp;0.68 each).\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003eTop predictors by information gain (Table\u0026nbsp;5) included epigastric pain, salpingectomy/oophorectomy, fever, smoking, and dysuria\u0026mdash;variables also prominent in PCA and clustering results.\u003c/p\u003e\u003cp\u003ePrincipal Component Analysis and Factor Structure (Table\u0026nbsp;2)\u003c/p\u003e\u003cp\u003eTable\u0026nbsp;2 summarizes the principal component analysis (PCA), identifying eight factors with eigenvalues greater than 2.0. These factors cumulatively explain 41.84% of the total variance. High loadings on Factor 1 for variables such as age, gravidity, and parity suggest that demographic and reproductive history account for the largest proportion of variance. Medical history (e.g., prior surgeries) load heavily on Factor 7, while flu-like and GI symptoms are explained by Factors 5 and 6, respectively.\u003c/p\u003e\u003cp\u003ePrincipal Component Analysis (PCA) with Varimax rotation extracted eight factors with eigenvalues greater than 2, collectively explaining 41.84% of the total variance in the dataset. The first factor (Factor 1) had the highest eigenvalue (6.76) and accounted for 10.73% of the variance alone, while the remaining factors contributed between 3.6% and 5.9% each.\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eFactor 1, representing age-related reproductive and demographic characteristics, showed strong negative loadings for age (\u0026minus;\u0026thinsp;0.796), menopause (\u0026minus;\u0026thinsp;0.686), gravidity (\u0026minus;\u0026thinsp;0.858), and parity (\u0026minus;\u0026thinsp;0.821), indicating that these variables cluster strongly and may reflect a reproductive aging dimension.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eFactor 2 was heavily influenced by acute gastrointestinal and systemic symptoms, with strong negative loadings for nausea (\u0026minus;\u0026thinsp;0.717), vomiting (\u0026minus;\u0026thinsp;0.786), and the total number of associated symptoms (\u0026minus;\u0026thinsp;0.935).\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eFactor 3 loaded significantly on variables such as dysuria (\u0026minus;\u0026thinsp;0.813) and obesity (\u0026minus;\u0026thinsp;0.507), suggesting a genitourinary or metabolic symptom cluster.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eFactor 4 was defined by acute abdominal symptomatology, with strong negative associations with abdominal distension (\u0026minus;\u0026thinsp;0.791), number of major symptoms (\u0026minus;\u0026thinsp;0.760), and loss of appetite (\u0026minus;\u0026thinsp;0.814), possibly representing an acute inflammatory or pain-related factor.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eFactor 5 loaded strongly on vaginal discharge and bleeding (\u0026minus;\u0026thinsp;0.852), indicating a distinct gynecologic symptom component.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eFactor 6 captured fever-related systemic symptoms, with fever (0.818) and shivering/shortness of breath (0.834) as dominant indicators.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eFactor 7 encompassed prior pelvic surgical history, loading highly on the number of previous surgeries (0.904), history of pelvic surgery (0.827), and caesarean sections (0.781\u0026ndash;0.802).\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eFactor 8 loaded most prominently on comorbidity burden, including number of comorbidities (0.703) and history of normal pregnancy (0.537), suggesting a general health and reproductive outcome dimension.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003eTable\u0026nbsp;3 details the inter-cluster comparison. Cluster 2 patients had more comorbidities, including endometriosis and obesity, and a higher number of associated symptoms (e.g., fever, shortness of breath), while cluster 1 patients showed a higher incidence of flank pain and fainting (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001 for several comparisons).\u003c/p\u003e\u003cp\u003eFigure \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e: Shows the ROC curves of various classification models used to predict treatment modality. The Random Forest model with ROSE oversampling achieved an AUC of 1.00, although this is likely due to overfitting. The Decision Tree model with class weighting performed best among non-overfitted models (AUC\u0026thinsp;=\u0026thinsp;0.76), followed by Neural Networks and Gradient Boosting (AUC\u0026thinsp;=\u0026thinsp;0.68 each).\u003c/p\u003e\u003cp\u003eTable\u0026nbsp;4 compares the sensitivity and specificity of these models. While Random Forest (ROSE) scored perfect metrics, the Decision Tree and Neural Network models presented more balanced and reliable values. Table\u0026nbsp;5 Ranks features according to their predictive value using information gain, aligning closely with the visual information in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e6\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003eObjective 4. Build a model to prognosticate the disease management plan (operative versus conservative)\u003c/h2\u003e\u003cdiv id=\"Sec12\" class=\"Section3\"\u003e\u003ch2\u003eMethodology:\u003c/h2\u003e\u003cp\u003eWorking on the fourth task, we trained machine learning models to predict the disease management plan. The model architecture we used is as follows: Random Forest, Decision Tree, Logistic Regression, CatBoost, and Neural Network. The major challenge was that the cases that necessitated operative treatment outnumbered those that did not require surgery.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003eTraining the models:\u003c/h2\u003e\u003cp\u003eWe resorted to class weighting to address the issue of unbalanced data. Besides, we tested other data balancing techniques on the Random Forest model. These were Random Over-Sampling Examples (ROSE), Synthetic Minority Over-sampling Technique, and Edited Nearest Neighbors (SMOTE-ENN). To study feature importance, we computed the information gain value and ranked the predictors by importance. The performance metrics were the area under the receiver operating characteristic curve (ROC AUC), sensitivity, and specificity of detecting the positive class - patients with conservative treatment (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e5\u003c/span\u003e, \u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e6\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study represents one of the first applications of machine learning (ML) to predict optimal management strategies in ovarian torsion cases. The findings provide insights into clinical patterns that influence surgical decision-making beyond traditional symptomatology and imaging.\u003c/p\u003e\u003cp\u003eOur finding that OCP use is inversely related to surgical intervention aligns with previous literature highlighting the protective role of hormonal contraceptives in preventing or reducing the recurrence of benign ovarian pathology and endometriosis-related pain (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e). OCPs are known to suppress ovulation and reduce the formation of functional cysts, which may reduce the urgency or need for surgical treatment.\u003c/p\u003e\u003cp\u003eThe positive correlation between pelvic mass presence and operative treatment is clinically expected and corroborated by existing evidence indicating that the presence of complex or large ovarian masses is a common indication for surgical exploration, particularly when torsion or malignancy is suspected (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThe correlation between the number of associated symptoms and surgery is also intuitive and supported by previous studies, which report that symptom burden, especially when involving acute pain, gastrointestinal complaints, or hemodynamic instability, often prompts surgical exploration to confirm or rule out torsion, rupture, or hemorrhage (\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e). Interestingly, despite trends seen in other gynecological conditions, factors such as age, BMI, parity, and history of infertility or endometriosis were not significantly associated with treatment decision in our cohort. This suggests that acute clinical presentation, rather than chronic demographic or historical factors, may weigh more heavily in emergent decision-making scenarios. The results suggest a nuanced clinical paradigm: OCP history may be a protective factor and a potential marker for favorable conservative management outcomes, potentially guiding patient counseling, and pelvic mass identification on imaging remains a strong indicator for operative intervention, reaffirming the role of targeted imaging in acute gynecological triage.\u003c/p\u003e\u003cp\u003eThe cumulative symptom burden appears to play a modest but significant role in determining the management path, suggesting a need for holistic symptom assessment rather than reliance on single indicators. However, the generally low correlation coefficients (r\u0026thinsp;\u0026lt;\u0026thinsp;0.2) emphasize that these variables, while statistically significant, have limited predictive power individually. Therefore, clinical judgment, including real-time imaging, hemodynamic assessment, and pain severity, remains paramount.\u003c/p\u003e\u003cp\u003eThe predominance of surgical management is consistent with literature, emphasizing the urgency of operative intervention in torsion to prevent ovarian necrosis (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e). However, conservative management was feasible in 16.4% of cases\u0026mdash;primarily among premenopausal women with preserved Doppler flow and less severe clinical features. These findings support the need for nuanced clinical judgment and risk stratification.\u003c/p\u003e\u003cp\u003eThe high AUC from the Random Forest-ROSE model should be interpreted cautiously due to probable overfitting, a common issue in imbalanced data contexts (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e). Conversely, the Decision Tree and GBM models offered more balanced and interpretable results, aligning with findings from previous clinical ML studies (\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eNotably, factors such as epigastric pain, metabolic syndrome, smoking, and bundled symptoms (e.g., fainting with distension) emerged as strong predictors of management type. These variables, often underrecognized, may reflect broader inflammatory or ischemic states that influence physician decisions. Bundled symptoms may also serve as proxies for more severe systemic involvement or delayed presentation.\u003c/p\u003e\u003cp\u003eCluster and factor analyses highlighted two main phenotypic patterns: patients with comorbid metabolic and gynecologic conditions (e.g., endometriosis, untreated obesity) and those with more classical pain presentations. While cluster assignment did not significantly predict management choice, it revealed distinct patient profiles, useful for refining triage and diagnostic protocols.\u003c/p\u003e\u003cp\u003eImportantly, torsion severity and ovarian viability\u0026mdash;assessed intraoperatively\u0026mdash;remain key determinants of oophorectomy. The model identified these as critical outcome-related variables, further emphasizing the need for early intervention to avoid irreversible ovarian damage.\u003c/p\u003e\u003cp\u003eOur results show that very few variables were significantly associated with treatment type. However, most correlations were not strong even when statistically significant (r\u0026thinsp;\u0026lt;\u0026thinsp;0.2), suggesting a possible non-linear association. Clinical Insight: The association between OCP use and conservative management may reflect the suppressive effect of hormonal regulation on symptoms or disease severity. The link between pelvic masses and operative treatment reaffirms surgical indications based on structural findings. Symptom burden may influence decision-making, but individual symptoms alone were not strong predictors. Limitations: Correlation does not imply causation, and low correlation coefficients suggest other factors. Hence, we recommend interpreting these findings narratively in the results and interpreting cautiously. They can inform future hypotheses, but are insufficient for building a predictive model or treatment algorithm.\u003c/p\u003e\u003cp\u003eThis integrated analysis confirmed that ovarian torsion is a multidimensional clinical entity, with symptoms clustering along demographic, gynecologic, metabolic, and inflammatory axes. Our findings align with previous literature highlighting the role of reproductive age, pelvic surgery history, and inflammatory markers in ovarian torsion risk and diagnosis.\u003c/p\u003e\u003cp\u003eAge and reproductive history, major contributors to Factor 1, have been consistently associated with torsion risk. Huang et al and Damigos et al. (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e) noted that torsion occurs more frequently in women of reproductive age and those with ovarian enlargement due to hormonal stimulation.\u003c/p\u003e\u003cp\u003eInflammatory symptoms (fever, shivering), forming Factors 5 and 6 and defining Cluster 2, mirror findings by (\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e), who showed elevated CRP and WBC counts in complicated adnexal torsion. Comorbid obesity and endometriosis, highlighted in Cluster 2, echo findings by (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e). Chronic pelvic pathology may predispose to torsion due to altered adnexal mobility. Finally, surgical history, captured in Factor 7, remains a known risk factor for adnexal torsion recurrence or complications, as reported by (\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eOur study further shows that classification algorithms can distinguish symptom clusters and predict management needs, but real-world application requires careful avoidance of overfitting, as seen with the Random Forest\u0026thinsp;+\u0026thinsp;ROSE model. Clinically, the Decision Tree and Neural Network models offer a realistic compromise in prediction utility.\u003c/p\u003e\u003cp\u003eThe use of PCA and clustering, seldom applied in prior torsion studies, allowed us to better represent underlying clinical dimensions, offering a nuanced approach to early detection, triage, and risk stratification.\u003c/p\u003e\u003cp\u003eClinical Implications and Utilization\u003c/p\u003e\u003cp\u003eThe prognostic model developed in this study holds substantial promise for enhancing decision-making in the management of ovarian torsion, a condition where timely and appropriate intervention is critical. By leveraging machine learning algorithms trained on real-world clinical data, the model provides a framework for early stratification of patients into those likely to benefit from immediate surgical intervention and those for whom conservative management may be safely considered.\u003c/p\u003e\u003cp\u003eIn practice, this model could be embedded into electronic health record (EHR) systems or clinical decision support tools (CDSTs) to assist gynecologists, emergency physicians, and surgical teams in triaging patients upon presentation. For example, a patient presenting with abdominal pain, preserved Doppler flow, and no prior pelvic surgery could be algorithmically flagged as a low-risk candidate for conservative observation, while a patient with significant risk features (e.g., epigastric pain, prior oophorectomy, smoking, and fever) could be prioritized for operative assessment.\u003c/p\u003e\u003cp\u003eImportantly, the model\u0026rsquo;s ability to rank predictors by clinical importance allows for transparent, explainable AI applications, thereby supporting physician trust and shared decision-making with patients. Integration of such a tool could also reduce diagnostic delays, prevent unnecessary oophorectomies, and improve resource allocation in acute gynecological care settings.\u003c/p\u003e\u003cp\u003eTo ensure safe adoption, the model should undergo prospective validation across diverse healthcare systems. Future iterations may incorporate imaging parameters, real-time vitals, and patient-reported symptoms to further refine accuracy and usability.\u003c/p\u003e\u003cp\u003eOur study has some strengths: first, it has an Innovative Methodology. This study is among the first to apply a machine learning (ML) framework to predict surgical versus conservative management decisions in ovarian torsion, offering a data-driven approach to an urgent gynecologic condition. Second, we have a comprehensive variable set: A wide range of clinical, demographic, and surgical predictors were considered, including traditionally underexamined features such as epigastric pain and bundled symptom clusters, enhancing model interpretability and clinical relevance. Third, we developed a Model of Comparison and Validation: Multiple ML algorithms were implemented and evaluated using standard metrics (AUC, sensitivity, specificity), allowing for robust comparison of performance and the identification of generalizable models such as Decision Trees and Gradient Boosting. Lastly, the Feature Importance Insights: The use of information gain and ranking provides transparency on influential clinical features, supporting a better understanding of torsion pathophysiology and management pathways.\u003c/p\u003e\u003cp\u003eLimitations: Retrospective Single-Center Design: The data were collected retrospectively from a single tertiary center, which may limit generalizability to other populations or healthcare systems with different management protocols. Moreover, sample size and class Imbalance: The relatively small sample size, combined with the inherent imbalance between surgical and conservative cases, necessitated oversampling techniques such as ROSE and SMOTE. While effective, these methods may introduce synthetic bias or model overfitting. In addition, there is a Lack of Imaging Integration. Radiological findings (e.g., Doppler flow patterns or ovarian volume) were not incorporated into the model, potentially limiting predictive accuracy, especially in borderline cases. Unmeasured Confounding: Socioeconomic factors, patient preferences, and surgeon decision-making biases were not accounted for but may have influenced management choices. Finally, there is no prospective validation: external validation on a separate dataset was not conducted, and prospective testing is needed to confirm model utility in real-time clinical decision-making.\u003c/p\u003e\u003cp\u003eDespite limitations due to class imbalance and possible data heterogeneity, the study provides a strong foundation for developing a clinically useful decision support tool. Further prospective validation is needed (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eFuture directions should involve prospective multicenter validation, external testing on heterogeneous populations, and integration with imaging features (e.g., MRI data). Development of a real-time clinical decision support tool embedded in electronic medical records may enhance usability and generalization. Feature importance analysis identified epigastric pain, prior pelvic surgeries, fever, smoking, and metabolic comorbidities as key predictors. Surprisingly, bundled symptom patterns\u0026mdash;like abdominal distension combined with fainting\u0026mdash;also contributed strongly, though their clinical specificity remains questionable. These findings suggest that clinician decisions are influenced not only by classical torsion symptoms but also by patient history and comorbid status.\u003c/p\u003e\u003cp\u003e Our findings confirm that surgical intervention remains the cornerstone of OT management, but conservative treatment may be appropriate in carefully selected cases\u0026mdash;particularly those with fewer symptoms, OCP use, and absence of a pelvic mass. ML-based models offer an additional decision-support layer, enhancing risk stratification and potentially reducing unnecessary surgeries.\u003c/p\u003e\u003cp\u003eClinically, integrating such models into electronic health records or decision support systems could allow real-time triage, improve diagnostic accuracy, and support shared decision-making with patients. This is particularly relevant for reproductive-aged women, where fertility preservation is paramount. Prospective, multicenter validation is the next step to ensure safe and generalizable application.\u003c/p\u003e\u003cp\u003eThe identification of novel predictive features\u0026mdash;such as epigastric pain, metabolic comorbidities, and prior pelvic surgeries\u0026mdash;enhances clinical understanding and supports risk-based triage. While the model shows high predictive performance, particularly with decision tree\u0026ndash;based algorithms, external validation and prospective implementation are necessary to confirm its applicability in routine care.\u003c/p\u003e\u003cp\u003eUltimately, this study highlights that combining clinical expertise with data-driven tools can optimize management of ovarian torsion, reduce morbidity, and safeguard reproductive outcomes.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eEthics approval and consent to participate: This retrospective study was conducted in accordance with the principles of the Declaration of Helsinki. Ethical approval was obtained from the Institutional Review Board of the Abu Dhabi Health Authority, Tawam Hospital (Ref. No.: HREC SEHA-IRB-942, dated 21 November 2024). As this was a retrospective analysis of de-identified patient data, individual consent to participate was waived by the IRB. The study does not contain any person\u0026rsquo;s data in any form.\u003c/p\u003e\u003cp\u003e\u003ch2\u003eConflict of Interest Disclosure\u003c/h2\u003e\u003cp\u003eThe authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eFunding:\u003c/h2\u003e\u003cp\u003eThis study was funded by Abu Dhabi Health Authority at Tawam Hospital.\u003c/p\u003e\u003ch2\u003eAuthors\u0026rsquo; Contributions\u003c/h2\u003e\u003cp\u003eMA: Conceptualization, Methodology, Data Curation, Supervision, Writing \u0026ndash; Review \u0026amp; Editing\u003c/p\u003e\u003ch2\u003eAcknowledgment:\u003c/h2\u003e\u003cp\u003eWe acknowledge that generative AI technologies were used to improve the style, accessibility, and quality of human-generated text and images.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAbraham M, Keyser EA (2019) Adnexal torsion in adolescents. Obstet Gynecol 134(2):E56\u0026ndash;E63\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHuang C, Hong M-K, Ding D-C (2017) A review of ovary torsion. Tzu Chi Med J 29(3):143\u0026ndash;147\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRyles HT, Hong CX, Andy UU, Farrow MR (2023) Changing practices in the surgical management of adnexal torsion: an analysis of the National Surgical Quality Improvement Program Database. Obstet Gynecol 141(5):888\u0026ndash;896\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAkman L, Erbas O, Terek M, Aktug H, Taskiran D, Askar N (2016) The long pentraxin-3 is a useful marker for diagnosis of ovarian torsion: An experimental rat model. J Obstet Gynaecol 36(3):399\u0026ndash;402\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBrierley G, Arshad I, Shakir F, Visvathanan D, Arambage K (2020) Vascular injury during laparoscopic gynaecological surgery: a methodological approach for prevention and management. Obstetrician Gynaecologist 22(3):191\u0026ndash;198\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBreiman L (2001) Random forests. Mach Learn 45(1):5\u0026ndash;32\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eEsteva A, Kuprel B, Novoa RA, Ko J, Swetter SM, Blau HM et al (2017) Dermatologist-level classification of skin cancer with deep neural networks. Nature 542(7639):115\u0026ndash;118\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eVercellini P, Bandini V, Vigano P, Di Stefano G, Merli CEM, Somigliana E (2024) Proposal for targeted, neo-evolutionary-oriented, secondary prevention of early-onset endometriosis and adenomyosis. Part I: pathogenic aspects. Hum Reprod 39(1):1\u0026ndash;17\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJeong Y-Y, Outwater EK, Kang HK (2000) Imaging evaluation of ovarian masses. Radiographics 20(5):1445\u0026ndash;1470\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMuto MG (2020) Approach to the patient with an adnexal mass. UpToDate, Waltham, MA.(Accessed on October 3, 2023.)\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHoury D, Abbott JT (2001) Ovarian torsion: a fifteen-year review. Ann Emerg Med 38(2):156\u0026ndash;159\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJohnson JM, Khoshgoftaar TM (2019) Survey on deep learning with class imbalance. J big data 6(1):1\u0026ndash;54\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O et al (2011) Scikit-learn: Machine learning in Python. J Mach Learn Res 12:2825\u0026ndash;2830\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChen Y-C, Lee M-C, Lee C-J, Hsu B-G (2018) Hyperleptinemia is associated with the aortic augmentation index in kidney transplant recipients. Tzu Chi Med J 30(3):152\u0026ndash;157\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDamigos E, Johns J, Ross J (2012) An update on the diagnosis and management of ovarian torsion. Obstetrician Gynaecologist. ;14(4)\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHasson J, Tsafrir Z, Azem F, Bar-On S, Almog B, Mashiach R et al (2010) Comparison of adnexal torsion between pregnant and nonpregnant women. Am J Obstet Gynecol 202(6):536 e1-. e6\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGoossens E, Van Saen D, Tournaye H (2013) Spermatogonial stem cell preservation and transplantation: from research to clinic. Hum Reprod. ;28(4)\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTables 1 to 5 are available in the Supplementary Files section.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"Tawam Hospital","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Ovarian torsion, Machine learning, Prognostic model, Conservative management, Gynecologic emergencies, Fertility preservation","lastPublishedDoi":"10.21203/rs.3.rs-7828438/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7828438/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground: \u003c/strong\u003eOvarian torsion (OT) is a gynecologic emergency requiring prompt and accurate management to preserve ovarian function and fertility. Determining the need for operative intervention versus conservative management remains challenging due to overlapping clinical and imaging features. This study aimed to develop and validate a machine learning (ML)–based prognostic model to assist clinical decision-making in suspected OT.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods: \u003c/strong\u003eA retrospective analysis was conducted on 219 females (1 month–75 years) presenting with suspected OT at a tertiary center between 2022 and 2024. Clinical, demographic, laboratory, and imaging variables were analyzed. Predictors of management type (operative vs. conservative) were identified using comparative statistics and Spearman correlation. Principal component and cluster analyses were used for dimensionality reduction and patient stratification. Supervised ML algorithms—including Decision Tree, Random Forest, Neural Network, Gradient Boosting, and Logistic Regression—were trained with oversampling techniques (ROSE, SMOTE, SMOTE-ENN) to address class imbalance. Model performance was assessed using the area under the ROC curve (AUC), sensitivity, and specificity.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults: \u003c/strong\u003eOperative management was required in 83.6% of cases, while 16.4% were managed conservatively. Significant predictors of conservative management included prior oral contraceptive use (protective; r = −0.199, p = 0.003), absence of a pelvic mass (r = 0.134, p = 0.048), and lower symptom burden (r = 0.148, p = 0.029). PCA identified eight clinical domains, and clustering revealed two distinct phenotypic subgroups. The weighted Decision Tree achieved the best-balanced performance (AUC = 0.76), while Random Forest with ROSE oversampling achieved perfect performance (AUC = 1.00), indicating potential overfitting.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusions: \u003c/strong\u003eMachine learning models can enhance clinical decision-making by stratifying OT patients suitable for conservative management and identifying those requiring urgent surgery. Integration of this model into clinical decision support systems may reduce unnecessary surgeries, optimize fertility-preserving care, and improve\u003c/p\u003e","manuscriptTitle":"Optimization of Management Plan with a machine learning model for Ovarian Torsion Cases: Operative versus Conservative","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-20 16:01:39","doi":"10.21203/rs.3.rs-7828438/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"87779973-fcf2-4e97-8b5a-31f28a209edb","owner":[],"postedDate":"October 20th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":56104027,"name":"Surgery"},{"id":56104028,"name":"Critical Care \u0026 Emergency Medicine"},{"id":56104029,"name":"Surgical Obstetrics \u0026 Gynecology"}],"tags":[],"updatedAt":"2025-10-20T16:01:40+00:00","versionOfRecord":[],"versionCreatedAt":"2025-10-20 16:01:39","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7828438","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7828438","identity":"rs-7828438","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}