
Analysis of Clinical and Geographical Patterns in Childbirth Data Using Data Mining Techniques: A Hospital-Based Study in Kerman
Kasra Kashani 1, Katayoun Alidousti 2, Zahra Keshavarz 3 ℗, Mohammad Mehdi Ghaemi 4 ©, Amirali Farrokhi 1, Hassan Shokri 1
Abstract
Introduction: With the growing application of artificial intelligence in healthcare, data mining algorithms such as association rule mining have emerged as effective tools for uncovering hidden patterns in childbirth data. These patterns can help identify risk factors and improve clinical decision-making. Moreover, geographical analysis of such data enables the detection of spatial disparities and clustering of obstetric complications, which plays a crucial role in planning and optimizing maternal and child health services. Methods and Materials: Following the approval of the institutional ethics committee, access was granted to de-identified childbirth records from a hospital in Kerman. The study followed a structured analytical approach comprising several key steps. Initially, data preprocessing was performed, including the removal of records pertaining to individuals residing outside the city of Kerman. Subsequently, relevant continuous variables were discretized to facilitate pattern recognition. Geographical distribution of maternal and neonatal variables was then examined through geospatial mapping techniques to identify spatial trends. In the final phase, association rule mining was employed to uncover significant relationships and hidden patterns among the clinical and demographic variables. Results: A total of 866 childbirth records were collected from a hospital in Kerman. After preprocessing steps—including the exclusion of records outside the geographical area, removal of non-analytical variables such as discharge date, and elimination of cases with missing data (e.g., BMI)—528 valid records were retained for analysis. The mean age of mothers in the dataset was 30.16 years. The geographical distribution of premature rupture of membranes (PROM) indicated a higher prevalence in specific urban areas of Kerman, with potential spatial clusters identified through mapping analysis. Using association rule mining, significant relationships were discovered between certain maternal conditions and preterm delivery. Three independent rules were extracted, each characterized by support, confidence, coverage, and strength metrics, as follows: • {Macrosomia} → {Preterm delivery} Support = 0.964 | Confidence = 0.979 | Coverage = 0.985 | Strength = 0.994 • {Preeclampsia} → {Preterm delivery} Support = 0.968 | Confidence = 0.981 | Coverage = 0.987 | Strength = 0.992 • {History of stillbirth} → {Preterm delivery} Support = 0.972 | Confidence = 0.979 | Coverage = 0.992 | Strength = 0.987 These findings suggest that certain maternal conditions may serve as early warning indicators for preterm delivery and can support improved clinical decision-making in perinatal care. Conclusion and Discussion: The findings of this study demonstrated that data mining algorithms and geographical analysis can effectively identify meaningful patterns within childbirth data. These methods hold significant potential for improving clinical decision-making and health planning, particularly in the early identification of preterm birth risk factors.
Keywords: Premature Birth, Delivery, Data Mining