Development of a Diagnosis Grading System for Patients Undergoing Intrauterine Inseminations: A Machine-learning Perspective

ObjectiveTo develop an innovative, non-invasive and objective grading system for enhancing clinicians assessment of intrauterine insemination (IUI) patients. DesignPatients who had undergone IUI treatments previously were divided into pregnant (N=4618) and non-pregnant(N=20974) groups. An evaluation index method was developed using collected clinical data from the two groups, particularly on indications of considerable differences between the two groups. The weight of each indicator was determined using random forest machine learning, and the indicators and patients conditions were classified using an entropy-based feature discretization technique. The indices for each indicator were further divided into five grades, from A to E, and given five points to one point, respectively. Effectiveness of the system was tested using the ten-fold cross-validation method. SettingReproductive medicine center, Shanghai, China PatientsInfertile couples who had undergone IUI treatment. InterventionNone. Main Outcome MeasuresWeight of each indicator and grades of infertile patients. ResultsFrom the 25,592 medical records of infertile couples who had IUI, 4618 women were pregnant subsequently, with a mean age of 28.69{+/-}3.34 years. From the collected records, 18 indicators (e.g., body mass indices [BMI], endometrial thicknesses, couples ages, IUI cycle days, and semen situations) were selected to construct our diagnosis scoring system. Among the 18 indicators, BMI (weight, 12.49%), endometrial thickness (11.99%), female age (11.88%), semen density (10.41%), semen volume (8.92%), cycle day (7.38%) and male age (6.96%) were closely related to the pregnancy rates. Among patients with the final scores for > 75.29 individually, the pregnancy rates for them was > 56.35%. The systems stability was 95.1% (95%CI,94.5%-95.7%) according to cross-validation data. ConclusionThis quick and objective machine learning-based approach can be used to simplify and enhance the decision-making process among clinicians, especially to advise and to select patients for better IUI outcomes.