Introduction: Infertility is a significant public health issue. The work environment of nurses can greatly affect their reproductive health and fertility due to exposure to numerous and diverse diseases. The aim of this doctoral dissertation is to identify risk factors for infertility among nurses in Slovenia based on their job and to develop a predictive model for risk assessment based on the characteristics of their work and work environment. Methods: The research was conducted in two phases. The first phase involved implementing a national cross-sectional study to collect data on nurses' exposure to risk factors. The sample included 1,126 women employed in nursing care. First, we examined the existence of differences in individual work environment factors with in relation to infertility outcomes.Variables that were statistically significantly associated with infertility were included in separate logistic regression models to assess the influence of each set of work environment factors on infertility. In constructing a combined logistic regression model, we included factors that were statistically significant in the individual analyses, while accounting for confounding variables (reproductive diagnoses). To identify more and less risky jobs for nurses in relation to infertility, we performed a cluster analysis. The second phase involved the development, comparison, and validation of predictive models using machine learning methods. The predictive models were developed in Python 3.12 using 50 occupational and demographic characteristics selected through univariate analysis. The models included logistic regression, random forest, XGBoost, and LightGBM with regularisation to reduce the risk of overfitting. Model performance was assessed using 10-fold stratified cross-validation on 746 training samples and tested on 187 independent samples. Results: 274 individuals (24.3%) were classified as infertile. The basic model using classical statistical methods identified seven key factors: employer awareness of reproductive health risks, staff shortages, confirmed reproductive diagnoses, previous miscarriages, malnutrition, job dissatisfaction, and not taking hormonal contraception. Meanwhile, the predictive model successfully identified the risk of infertility in nurses, while machine learning methods revealed 20 key predictors of infertility. SHAP analysis indicates that the most important predictive factors are age at the decision to start a family and a family history of infertility. Discussion and conclusion: The 24.3% prevalence of infertility confirms that nurses in Slovenia are a professional group with a high occupational risk in reproductive health and fertility. The XGBoost model demonstrated with high predictive accuracy that infertility risk in nurses is determined by personal, reproductive, and professional factors. The results confirmed that educating employees about hazards in the work environment and ensuring a better work-life balance for protecting fertility of nurses. The development of a predictive model, useful for both employees and employers, significantly contributes to the creation of safe working conditions in healthcare, as it enables early identification of risks and strengthens awareness of reproductive health in clinical practice.
|