Yenidoğanlarda Total Bilirubinin Makine Öğrenmesiyle Tahmin Edilmesi

Yıl 2025, Cilt: 5 Sayı: 3, 3 - 13, 30.12.2025

Kaan Kefal , Deniz İlhan Topcu , Taha Şahin , Aslihan Abbasoglu

Öz

ÖZET
Amaç:
Yenidoğan döneminde sık görülen hiperbilirubinemi, zamanında tanı ve tedavi edilmediğinde ciddi nörolojik hasarlara neden olabilir. Bu çalışmanın amacı, makine öğrenmesi (ML) algoritmaları ile yenidoğanlarda total bilirubin düzeylerini tahmin eden modeller geliştirmek ve bu modellerin performansını değerlendirmektir.
Gereç ve Yöntem:
İzmir Şehir Hastanesi’nde 318 yenidoğana ait 698 örnek retrospektif olarak analiz edilmiştir. Total bilirubin, hematokrit, doğum ağırlığı, gestasyonel yaş, yaş (gün) ve APGAR skoru gibi klinik-demografik veriler kullanılarak sekiz farklı ML algoritması (Gradient Boosting, Random Forest, Naive Bayes, Lojistik Regresyon, Yapay Sinir Ağı vb.) ile sınıflandırma modelleri oluşturulmuştur. Total bilirubin düzeyleri <12.5 ve ≥12.5 mg/dL olarak iki sınıfa ayrılmış, modellerin başarımı 10 kat çapraz doğrulama ile AUC, doğruluk ve F1 skoru gibi metrikler üzerinden değerlendirilmiştir. Model yorumlanabilirliği Decrease in AUC yöntemiyle analiz edilmiştir.
Bulgular:
Gradient Boosting modeli test veri setinde %92 doğruluk, %0.90 F1 skoru ve 0.89 AUC değeri ile en başarılı model olarak belirlenmiştir. Düşük riskli (<12.5 mg/dL) olgular doğru tahmin edilirken, yüksek riskli (≥12.5 mg/dL) gruplarda hata oranları %90’ın üzerindedir. Değişken önem analizi, yaş (gün), doğum ağırlığı ve gestasyonel yaşın model üzerinde en belirleyici etkiye sahip olduğunu göstermiştir.
Sonuç:
Makine öğrenmesi algoritmaları, özellikle düşük riskli yenidoğanlarda total bilirubin düzeylerini başarılı şekilde tahmin edebilmektedir. Ancak yüksek riskli grupların doğru tespiti için sınıf dengesizliğini azaltacak yöntemlerin (SMOTE, cost-sensitive learning vb.) kullanılması gerekmektedir.

Anahtar Kelimeler

Makine öğrenmesi , yenidoğan sarılığı , total bilirubin , Gradient Boosting , karar destek sistemi

Prediction of Total Bilirubin Levels in Newborns Using Machine Learning

Öz

ABSTRACT
Aim:
Neonatal hyperbilirubinemia is a common condition that may lead to severe neurological damage if not diagnosed and treated promptly. This study aimed to develop machine learning (ML) models to predict total bilirubin levels in newborns and evaluate their performance.
Material and Method:
A total of 698 samples from 318 newborns at İzmir City Hospital were retrospectively analyzed. Clinical and demographic variables, including total bilirubin, hematocrit, birth weight, gestational age, postnatal age (days), and Apgar scores, were used to develop classification models using eight ML algorithms (e.g., Gradient Boosting, Random Forest, Naive Bayes, Logistic Regression, Neural Networks). Total bilirubin levels were categorized as <12.5 mg/dL (low-risk) and ≥12.5 mg/dL (high-risk). Models were evaluated using 10-fold cross-validation and performance metrics such as AUC, accuracy, and F1 score. Model interpretability was assessed using the Decrease in AUC method.
Results:
The Gradient Boosting model demonstrated the best performance on the test dataset with 92% accuracy, 0.90 F1 score, and an AUC of 0.89. While the models accurately predicted low-risk cases, their performance for high-risk (≥12.5 mg/dL) cases was limited, with error rates exceeding 90%. Feature importance analysis indicated that postnatal age (days), birth weight, and gestational age had the highest influence on predictions.
Conclusion:
ML models, especially Gradient Boosting, can effectively predict low-risk total bilirubin levels in neonates. However, to improve the identification of high-risk cases, approaches addressing class imbalance (e.g., SMOTE, cost-sensitive learning) should be considered.

Anahtar Kelimeler

Keywords: Machine learning , neonatal jaundice , total bilirubin , Gradient Boosting , decision support system

Kaynakça

• 1. Stevenson DK, Vreman HJ, Wong RJ. Bilirubin Production and the Risk of Bilirubin Neurotoxicity. Semin Perinatol. 2011;35(3):121-6. doi:10.1053/j.semperi.2011.02.005
• 2. Kirk JM. Neonatal jaundice: a critical review of the role and practice of bilirubin analysis. Ann Clin Biochem. 2008;45(Pt 5):452-62. doi:10.1258/acb.2008.008076
• 3. Hansen TWR, Wong RJ, Stevenson DK. Molecular Physiology and Pathophysiology of Bilirubin Handling by the Blood, Liver, Intestine, and Brain in the Newborn. Physiol Rev. 2020;100(3):1291-346. doi:10.1152/physrev.00004.2019
• 4. Bhutani VK, Vilms RJ, Hamerman-Johnson L. Universal bilirubin screening for severe neonatal hyperbilirubinaemia. J Perinatol. 2010;30(Suppl):S6-15. doi:10.1038/jp.2010.98
• 5. Carbonell X, Botet F, Figueras J, Riu-Godó A. Prediction of hyperbilirubinaemia in the healthy term newborn. Acta Paediatr. 2001;90(2):166-70. doi:10.1080/080352501300049343
• 6. Amin SB, Lamola AA. Newborn Jaundice Technologies: Unbound Bilirubin and Bilirubin Binding Capacity in Neonates. Semin Perinatol. 2011;35(3):134-40. doi:10.1053/j.semperi.2011.02.007
• 7. Moncrieff G. Bilirubin in the newborn: Physiology and pathophysiology. Br J Midwifery. 2018;26(6):362-70. doi:10.12968/bjom.2018.26.6.362
• 8. Hsia DYY, Allen FH, Diamond LK, Gellis SS. Serum bilirubin levels in the newborn infant. J Pediatr. 1953;42(3):277-85. doi:10.1016/S0022-3476(53)80182-4
• 9. Bhardwaj K, Locke T, Biringer A, Booth A, Darling EK, Dougan S, et al. Newborn Bilirubin Screening for Preventing Severe Hyperbilirubinemia and Bilirubin Encephalopathy: A Rapid Review. Curr Pediatr Rev. 2017;13(1):67-90. doi:10.2174/1573396313666170110144345
• 10. Maisels MJ, Kring E. Transcutaneous Bilirubin Levels in the First 96 Hours in a Normal Newborn Population of ≥35 Weeks’ Gestation. Pediatrics. 2006;117(4):1169-73. doi:10.1542/peds.2005-0744
• 11. Büyüktoka RE, Surucu M, Erekli Derinkaya PB, Adibelli ZH, Salbaş A, Koc AM, et al. Applying large language model for automated quality scoring of radiology requisitions using a standardized criteria. Eur Radiol. 2025;1-2. doi:10.1007/s00330-025-11933-2
• 12. Jordan MI, Mitchell TM. Machine learning: Trends, perspectives, and prospects. Science. 2015;349(6245):255-60. doi:10.1126/science.aaa8415
• 13. El Naqa I, Murphy MJ. What Is Machine Learning? In: Machine Learning in Radiation Oncology. Cham: Springer International Publishing; 2015. p. 3-11. doi:10.1007/978-3-319-18305-3_1
• 14. Kononenko I. Machine learning for medical diagnosis: history, state of the art and perspective. Artif Intell Med. 2001;23(1):89-109. doi:10.1016/S0933-3657(01)00077-X
• 15. Deo RC. Machine Learning in Medicine. Circulation. 2015;132(20):1920-30. doi:10.1161/CIRCULATIONAHA.115.001593
• 16. Michel A. Review of the Reliability and Validity of the Apgar Score. Adv Neonatal Care. 2022;22(1):28-34. doi:10.1097/ANC.0000000000000859
• 17. Jepson HA, Talashek ML, Tichy AM. The Apgar Score: Evolution, Limitations, and Scoring Guidelines. Birth. 1991;18(2):83-92. doi:10.1111/j.1523-536X.1991.tb00065.x
• 18. Demšar J, Curk T, Erjavec A, Gorup Č, Hočevar T, Milutinovič M, et al. Orange: Data Mining Toolbox in Python. J Mach Learn Res. 2013;14:2349-2353.
• 19. Dobesova Z. Evaluation of Orange data mining software and examples for lecturing machine learning tasks in geoinformatics. Comput Appl Eng Educ. 2024;32(4). doi:10.1002/cae.22735
• 20. Olusanya BO, Kaplan M, Hansen TWR. Neonatal hyperbilirubinaemia: a global perspective. Lancet Child Adolesc Health. 2018;2(8):610-20. doi:10.1016/S2352-4642(18)30139-1
• 21. Maisels MJ. What’s in a Name? Physiologic and Pathologic Jaundice: The Conundrum of Defining Normal Bilirubin Levels in the Newborn. Pediatrics. 2006;118(2):805-7. doi:10.1542/peds.2006-0675
• 22. Mahesh B. Machine Learning Algorithms – A Review. Int J Sci Res (IJSR). 2020;9(1):381-6. doi:10.21275/ART20203995
• 23. Huang J, Ling CX. Using AUC and accuracy in evaluating learning algorithms. IEEE Trans Knowl Data Eng. 2005;17(3):299-310. doi:10.1109/TKDE.2005.50
• 24. Ling CX, Huang J, Zhang H. AUC: A Better Measure than Accuracy in Comparing Learning Algorithms. In: Proceedings of the Eighteenth International Joint Conference on Artificial Intelligence (IJCAI-03). San Mateo, CA: Morgan Kaufmann; 2003. p. 519-526. doi:10.1007/3-540-44886-1_25
• 25. Mohammed M, Khan MB, Bashier EB. Machine Learning: Algorithms and Applications. Boca Raton: CRC Press; 2016. doi:10.1201/9781315371658
• 26. Hand DJ, Till RJ. A simple generalisation of the area under the ROC curve for multiple class classification problems. Machine Learn. 2001;45(2):171-86. doi:10.1023/A:1010920819831
• 27. Staniak M, Biecek P. Explanations of model predictions with live and breakDown packages. R J. 2018 Apr 5. doi:10.32614/RJ-2018-072
• 28. Antwarg L, Miller RM, Shapira B, Rokach L. Explaining anomalies detected by autoencoders using Shapley Additive Explanations. Expert Syst Appl. 2021;186:115736. doi:10.1016/j.eswa.2021.115736
• 29. Canul-Reich J, Hall LO, Goldgof D, Eschrich SA. Feature selection for microarray data by AUC analysis. In: IEEE Int. Conf. Systems, Man and Cybernetics. 2008; p. 768-773. doi: 10.1109/ICSMC.2008.4811371
• 30. Fernandez A, Garcia S, Herrera F, Chawla N V. SMOTE for Learning from Imbalanced Data: Progress and Challenges, Marking the 15-year Anniversary. J Artif Intell Res. 2018;61:863-905. doi:10.1613/jair.1.11192
• 31. Maldonado S, López J, Vairetti C. An alternative SMOTE oversampling strategy for high-dimensional datasets. Appl Soft Comput. 2019;76:380-9. doi:10.1016/j.asoc.2018.12.034
• 32. Fernández A, García S, Galar M, Prati RC, Krawczyk B, Herrera F. Cost-Sensitive Learning. In: Learning from Imbalanced Data Sets. Cham: Springer International Publishing; 2018. p. 63-78. doi:10.1007/978-3-319-98074-4_4
• 33. Hahn S, Bührer C, Schmalisch G, Metze B, Berns M. Rate of rise of total serum bilirubin in very low birth weight preterm infants. Pediatr Res. 2020;87(6):1039-44. doi:10.1038/s41390-019-0415-7
• 34. Katayama Y, Yokota T, Zhao H, Wong RJ, Stevenson DK, Taniguchi-Ikeda M, et al. Association of HMOX1 gene promoter polymorphisms with hyperbilirubinemia in the early neonatal period. Pediatr Int. 2015;57(4):645-9. doi:10.1111/ped.12591
• 35. Yang H, Lin F, Chen Z-kai, Zhang L, Xu JX, Wu YH, et al. UGT1A1 mutation association with increased bilirubin levels and severity of unconjugated hyperbilirubinaemia in ABO incompatible newborns of China. BMC Pediatr. 2021;21(1):259. doi:10.1186/s12887-021-02726-9
• 36. Daood MJ, McDonagh AF, Watchko JF. Calculated free bilirubin levels and neurotoxicity. J Perinatol. 2009;29(Suppl):S14-19. doi:10.1038/jp.2008.218