Title: Discovering Dynamic Pathways of Semen Quality Decline: A Longitudinal Cohort Study and Temporal Association Rule Mining in Bandar Abbas, Iran

Background: Male factor infertility is a major global reproductive health concern, accounting for approximately ۵۰% of all cases. Traditional clinical diagnostics rely on static semen analysis guidelines (e.g., World Health Organization reference limits), which fail to capture dynamic, longitudinal fluctuations of semen parameters or map temporal cause-and-effect networks (the 'domino effect') of sperm decay over time. In region-specific clinics such as Bandar Abbas, environmental stressors (such as extreme heat and humidity) and localized systemic inflammation create unique regional subphenotypes that universal cutoffs do not adequately capture. Objective: To develop and evaluate a purely data-driven machine learning framework using Temporal Association Rule Mining (TARM) and local quartile discretization to discover longitudinal pathways and directed causal links in the decline of semen parameters in a regional cohort. Methods: A retrospective longitudinal cohort study was conducted on ۱۸,۲۱۷ clinical records from ۳,۰۱۲ unique male patients (each tracked over ۴ consecutive years with at least ۴ tests) attending a specialized fertility clinic in Bandar Abbas, Iran. Continuous parameters analyzed included volume, viscosity, progressive motility (PR), non-progressive motility (NP), morphology anomalies (head, neck, tail), semen WBC, blood ESR, and Sperm DNA Fragmentation Index (DFI). To bypass standard clinical cut-off biases, data-driven regional quartiles (۲۵th and ۷۵th percentiles) were calculated for local state discretization. An optimized TARM pipeline was deployed to extract sequential transition rules of the form {Antecedent}_t --> {Consequent}_t+۱. Longitudinal causal directions were validated using multivariate Granger Causality models (p < ۰.۰۵). Results: The regional cohort exhibited significant baseline variations, with local quartile thresholds reflecting specific environmental adaptation: PR progressive motility (۱۰.۰۰% to ۴۵.۰۰%), head anomalies (۹۵.۰۰% to ۹۸.۰۰%), and DFI (۴.۰۰% to ۵.۰۰%). Granger Causality confirmed directed, longitudinal pathways from early systemic inflammation (elevated ESR at time t) to progressive motility decline and structural morphological defects at t+۱ (F-statistic, p < ۰.۰۵). Sequential rule mining successfully identified the primary domino cascade of sperm deterioration: severe systemic inflammation acted as a primary temporal trigger ({ESR = High}_t), transitioning to severe structural head anomalies ({Head Anomalies = High}_t+۱) with a Support of ۶.۴۶%, Confidence of ۲۶.۶۹%, and Lift of ۱.۴۶۸. Furthermore, neck anomalies ({Neck Anomalies = High}_t) strongly predicted head anomalies at t+۱ (Support = ۵.۳۱%, Confidence = ۲۸.۲۲%, Lift = ۱.۵۵۲), which subsequently culminated in high genetic fragmentation ({DFI = High}_t+۱) triggered by low progressive motility ({PR = Low}_t --> {DFI = High}_t+۱: Support = ۶.۰۱%, Confidence = ۲۳.۵۰%, Lift = ۱.۲۰۴). Conclusion: This study presents a shift from static cross-sectional semen diagnostic guidelines to predictive patient trajectory forecasting. By utilizing pure data-driven regional thresholds and temporal association rules, clinicians can identify early warning triggers of sperm decay and determine optimal treatment windows, providing a powerful, personalized, and region-sensitive clinical decision-support tool in male reproductive medicine.

کلیدواژه ها:

Male Infertility ، Temporal Association Rule Mining ، Longitudinal Cohort ، Sperm DNA Fragmentation ، Bandar Abbas