Sistem Prediksi Tingkat Risiko Peritonitis Pada Pasien Pediatri Peritoneal Dialysis Dengan Explainable AI

Ismail, Fadhila Kamila (2026) Sistem Prediksi Tingkat Risiko Peritonitis Pada Pasien Pediatri Peritoneal Dialysis Dengan Explainable AI. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5026221026-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Peritoneal Dialysis (PD) adalah terapi pengganti ginjal utama bagi pasien pediatri End-Stage Renal Disease (ESRD), tetapi keberhasilannya dibatasi oleh risiko komplikasi peritonitis yang dapat menyebabkan kegagalan terapi. Selama ini, prediksi risiko peritonitis masih bergantung pada penilaian subjektif tenaga medis yang berpotensi kurang konsisten. Di sisi lain, berbagai penelitian sebelumnya berfokus pada perhitungan statistik dan machine learning tanpa memanfaatkan meta-analisis sebagai sumber pengetahuan, belum diimplementasikan dalam bentuk aplikasi web, serta banyak yang masih black-box sehingga kurang transparan bagi tenaga medis. Oleh karena itu, Tugas Akhir ini mengusulkan pengembangan sistem prediksi tingkat risiko peritonitis pasien pediatri PD berbasis web. Sistem dibangun dengan mengimplementasikan dua model perhitungan, yaitu Weighted Average (skenario baseline dan skenario optimized) dan Multiplicative Risk (skenario Fixed RR dan skenario Randomized RR) yang memanfaatkan nilai Risk Ratio (RR) dan p-value dari studi meta-analisis sebagai knowledge base penentu bobot variabel klinis. Aplikasi web dikembangkan menggunakan Streamlit dan dilengkapi dengan Explainable Artificial Intelligence (XAI) untuk menampilkan komponen faktor protektif, faktor risiko, serta rekomendasi intervensi Modifiable Risk Factors (MRF). Pengujian fungsionalitas menggunakan metode Combinatorial Testing (Pairwise Testing, t=2) dengan ACTS menghasilkan 21 test case yang seluruhnya tepat 100%, membuktikan bahwa source code aplikasi telah selaras dengan perhitungan manual. Evaluasi performa sistem terhadap 51 data rekam medis aktual pasien pediatri di Rumah Sakit Umum Pusat Nasional dr. Cipto Mangunkusumo/Universitas Indonesia (RSCM/UI) menggunakan Confusion Matrix menunjukkan bahwa model Weighted Average Optimized, Multiplicative Fixed RR, dan Multiplicative Randomized RR menghasilkan nilai akurasi identik sebesar 74,51%, di mana sistem cenderung bersifat optimis karena memiliki nilai True Negative (TN) yang tinggi dengan tingkat spesifitas 100,00%. Namun, sistem memiliki kelemahan mengenali pasien yang berisiko tinggi peritonitis dengan nilai sensitivitas yang rendah sebesar 0,00%. Hasil Expert Review bersama Dokter Spesialis Anak Konsultan Nefrologi menyatakan bahwa fitur XAI mudah dipahami, mampu meningkatkan transparansi asal-usul persentase prediksi, serta menilai aplikasi ini layak dan valid sebagai alat bantu keputusan klinis yang objektif dan akuntabel di rumah sakit.
====================================================================================================================================
Peritoneal Dialysis (PD) is the primary renal replacement therapy for pediatric End-Stage Renal Disease (ESRD) patients, yet its success is limited by the risk of peritonitis complications, which can lead to treatment failure. Currently, predicting peritonitis risk still relies on the subjective assessment of medical personnel, which is potentially inconsistent. On the other hand, prior studies have focused heavily on statistical calculations and machine learning without utilizing meta-analysis as a knowledge source, lack implementation in a web-based application, and often remain black-box models that lack transparency for clinicians. Therefore, this Final Project proposes the development of a web-based prediction system for peritonitis risk levels in pediatric PD patients. The system is built by implementing two computational models: the Weighted Average model (baseline and optimized variants) and the Multiplicative Risk model (Fixed RR and Randomized RR approaches), which utilize the Risk Ratio (RR) and p-value from meta-analysis studies as the knowledge base to determine clinical variable weights. The web application was developed using Streamlit and is equipped with Explainable Artificial Intelligence (XAI) to display protective factors, risk factors, and intervention recommendations for Modifiable Risk Factors (MRF). Functional testing using the Combinatorial Testing method (Pairwise Testing, $t=2$) with the ACTS tool generated 21 test cases that achieved 100% accuracy, proving that the application's source code aligns perfectly with manual calculations. Performance evaluation against 51 actual medical record data of pediatric patients at the dr. Cipto Mangunkusumo National Central General Hospital/Universitas Indonesia (RSCM/UI) using a Confusion Matrix showed that the optimized Weighted Average model, Multiplicative Fixed RR, and Multiplicative Randomized RR yielded an identical highest accuracy of 74.51%. The system tends to exhibit an optimistic character due to its high True Negative (TN) rate with a perfect specificity of 100.00%, although it shows limitations in identifying high-risk peritonitis patients, resulting in a low sensitivity of 0.00%. The results of the Expert Review conducted with a Pediatric Nephrology Consultant Specialist indicate that the XAI features are easy to understand, successfully enhance the transparency behind the prediction percentages, and validate the application as a feasible and valid clinical decision support tool that is both objective and accountable in hospital settings.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Peritoneal Dialysis (PD), Peritonitis, Risk Rate, Weighted Average, Multiplicative Risk Model, Explainable Artificial Intelligence (XAI), Peritoneal Dialysis (PD), Peritonitis, Risk Rate, Weighted Average, Multiplicative Risk Model, Explainable Artificial Intelligence (XAI)
Subjects:	T Technology > T Technology (General) > T58.62 Decision support systems
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 57201-(S1) Undergraduate Thesis
Depositing User:	Fadhila Kamila Ismail
Date Deposited:	28 Jul 2026 04:21
Last Modified:	28 Jul 2026 04:21
URI:	http://repository.its.ac.id/id/eprint/138552

Actions (login required)

View Item