Progress of the application clinical prediction model in polycystic ovary syndrome

Clinical prediction models use specific methodologies to predict individual probabilities of illness, complications, treatment outcomes, and includes a diagnostic model and a prognostic model [1]. There are several basic steps from model creation to model application. It includes the formulation of research questions, assessment design, data collection, establishment and evaluation of clinical prediction models, validation and application, and updating of models [2]. Researchers have created a tool called PROBAST to assess the risk of bias and applicability of a prediction model. It is also used in a specific population and has shown initial success [3‐6]. Clinical prediction models use a small number of predictors to predict disease status and prognosis. These predictors can be easily collected and detected.

Recently, several emerging technologies have been used in clinical prediction models. For example, computer technology, interdisciplinary communication, application of big data, and many advanced methods such as machine learning and artificial intelligence were applied. Combining these new technologies with a clinical prediction model can ensure proper allocation of medical resources. Moreover, it can achieve the goals of the three levels of prevention.

Clinical prediction models have been widely used in the medical field, including prediction models for tumor diseases [7‐9], non-tumor diseases [10‐12], and obstetrics and gynecology diseases [13‐15]. In recent years, clinical prediction models have been extensively used in the field of reproduction. For example, the application of clinical prediction models for polycystic ovary syndrome (PCOS) has been increasingly studied. PCOS is one of the most common reproductive disorders. It has attracted the most attention because of its high incidence, considerable heterogeneity, and complex treatments. Moreover, it can also cause short- and long-term complications that require long-term management [16]. Several studies have focused on PCOS by creating models from different perspectives.

In this study, we reviewed and interpreted the progress of clinical prediction models for PCOS from different perspectives. With the help of this study, we provided a comprehensive review of related research and a better guide for clinical management. Moreover, it can achieve more accurate treatment for individual patients with PCOS.

We conducted a comprehensive literature systematic review up to April 2023, utilizing PubMed, Embase, and Web of Science databases. Our search terms encompassed “polycystic ovary syndrome” or “PCOS” along with “prediction model,” “nomogram,” or “model.” We limited our selection to research papers published in English and prioritized those published since 2020, excluding any lacking full manuscripts.

PCOS is the most common endocrine disorder in women of reproductive age, affecting 6–10% of women of reproductive age [17]. There are many guidelines and diagnostic standards for PCOS, including the National Institutes of Health (NIH) standard in 1990 [18], Rotterdam Standard in 2003 [19], Androgen Excess Society (AES) standard in 2006 [20], PCOS Diagnosis and Treatment Guidelines issued by the Endocrine Society of America in 2012 [21], and recommendations from international evidence-based guidelines for the assessment and management of polycystic ovary syndrome in 2018 [22]. The most widely used standard for PCOS is the Rotterdam 2003 criteria. PCOS is defined as sporadic menstruation, sporadic ovulation or non-ovulation, blood biochemistry or clinical signs of hyperandrogenemia except from other diseases, and ultrasound suggesting polycystic changes, conforming to two out of three contents [19].

As shown in Fig. 1, we have primarily categorized the prediction models for PCOS into two types, encompassing diagnostic models for PCOS, prediction models for PCOS (mainly divided into two categories: Prediction Model of PCOS Complications and Prediction Model of PCOS Treatment Outcomes).

This review summarizes the current PCOS predictive models that have been reported, including the PCOS diagnosis model and predictive models (complications and treatment outcome models). PCOS is a heterogeneous disease that can be influenced by a variety of factors including genetic factors, environmental factors, and endocrine status [83]. The definition of PCOS evolved as knowledge of the disease continued to deepen. Overall, the definition of PCOS has become clearer [84, 85]. Hence, based on this review, we propose future research directions for PCOS predictive models. We live in an era of precision medicine in which PCOS incidence differs among different races, so it is necessary to make related prediction models based on different racial groups [86]. We can optimize the past prediction models and take advantage of the specificity and sensitivity to distinguish PCOS and other related diseases. There have been a handful of studies reporting that PCOS may frequently be concurrent with nonalcoholic fatty liver disease (NAFLD) [87, 88], obstructive sleep apnea syndrome [89, 90], diabetes [91, 92] and the like. Thus, we can also develop models predicting the occurrence of other related diseases in PCOS patients. Additionally, many studies on the relationship between gene polymorphism and PCOS have been reported [93‐95]. We can predict PCOS by adding a genetic related predictor. Following up on the offspring of women with PCOS can help predict how to maximize the benefits in PCOS patients from the perspective of the offspring [96‐98]. It has been reported that PCOS patients are prone to endometrial hyperplasia and endometrial cancer (EC) [99]. Moreover, PCOS may impair endometrial receptivity, which can increase adverse pregnancy outcomes [100, 101]. We can predict PCOS pregnancy outcomes by adding endometrial lesions predictor. By developing a series of predictive models, we can make the definition of PCOS more accurate, which can improve the diagnosis of PCOS and reduce the likelihood of false positives and false negatives. It will also help discover complications earlier and treatment outcomes being known earlier, which can result in better outcomes for women with PCOS.

This study is the first review to summarize the PCOS prediction model, aiming to make more people aware of its wide application. It may enable doctors to counsel patients regarding the most effective and patient-tailored treatment strategy before treatment initiation. Notably, even if the models currently have poor prediction performance by internal and external validation, this does not mean that they are worthless [102]. As we all known, a model that performs well in an internal validation always produce poor predictions in other patients [103]. Hence, in the future, models with poor accuracy can be improved by adding well-known parameters and validations, which will further expand our understanding of this disease in terms of precision medicine.