Wardani, Rachmawati and dr. Eko Arisetijono, Sp.S (K) and dr. Widodo Mardi Santoso, Sp.S (K) and dr. Holipah, PhD (2024) “Analisis Performa Machine Learning Brain.Js Dalam Memprediksi Kejadian Stroke Iskemik Di RSUD dr. Saiful Anwar Provinsi Jawa Timur”. Magister thesis, Universitas Brawijaya.
Abstract
Latar Belakang: Stroke merupakan penyebab kematian terbanyak kedua dan penyebab kecacatan terbesar keempat di seluruh dunia, yang disebabkan oleh faktor risiko yang dapat dimodifikasi maupun tidak dapat dimodifikasi. Pencegahan stroke dapat dilakukan dengan mengontrol faktor risikonya. Teknologi digital, termasuk smartphone dapat mempermudah akses terhadap layanan pencegahan stroke. Machine Learning Brain.js merupakan teknologi digital yang dapat digunakan sebagai mesin untuk memprediksi kejadian stroke. Mudahnya akses untuk memprediksi risiko stroke bagi masyarakat, diharapkan dapat meningkatkan perhatian masyarakat untuk melakukan tindakan pencegahan stroke dimulai dari melakukan kontrol terhadap faktor risiko yang dimiliki. Tujuan: Penelitian ini bertujuan untuk menganalisis akurasi, sensitivitas, spesifisitas, positive predictive value, dan negative predictive value Brain.js sebagai RetrospectivePredictive Model dalam memprediksi kejadian stroke iskemik dalam 5 tahun paska faktor risiko diketahui. Metode: Penelitian prognostik tes dengan pendekatan cohort-retrospective untuk analisis data klinis dan research and development untuk pengembangan machine learning dengan menggunakan framework Brain.js. Mengambil data dari rekam medis pasien stroke iskemik dan non-stroke pada bulan Maret - Agustus 2024 di RSUD dr. Saiful Anwar. Pasien dievaluasi faktor risiko hipertensi, merokok, dan diabetes mellitus dalam lima tahun, kemudian dilakukan pelatihan dan pengujian data pada Brain.js. Penentuan cut off point Brain.js menggunakan kurva ROC-AUC dan dilakukan uji diagnostik untuk menilai akurasi, sensitivitas, spesifisitas, PPV, dan NPV. Hasil: Terdapat 127 pasien stroke iskemik yang dianalisis dengan jenis kelamin terbanyak adalah pria (54%) dan sebanyak 88% berusia lebih dari 51 tahun. Faktor risiko terbanyak yang dimiliki yakni hipertensi sebesar 52%, diikuti merokok 27%, dan diabetes mellitus 21%. Nilai titik potong confidence level Brain.js dengan nilai AUC 0.849 (p<0.001) yakni 0.55, memiliki sensitivitas 76% dan spesifisitas 80%. Uji diagnostik menunjukkan Brain.js memiliki akurasi sebesar 78%, PPV 79%, dan NPV 77%. Kesimpulan: Brain.js memiliki tingkat akurasi, sensitivitas, spesifisitas, PPV, dan NPV yang baik dalam memprediksi kejadian stroke iskemik dalam 5 tahun paska faktor risiko diketahui.
English Abstract
Background: Stroke is the second leading cause of death and the fourth leading cause of disability worldwide, caused by both modifiable and non-modifiable risk factors. Stroke prevention can be done by controlling the risk factors. Digital technology, including smartphones, can facilitate access to stroke prevention services. Brain.js Machine Learning is a digital technology that can be used as a machine to predict stroke events. Easy access to predict stroke risk for the community is expected to increase public attention to take stroke prevention measures starting from controlling the risk factors they have. Objective: This study aims to analyze the accuracy, sensitivity, specificity, positive predictive value, and negative predictive value of Brain.js as a Retrospective-Predictive Model in predicting the occurrence of ischemic stroke within 5 years after the risk factors are known. Method: Prognostic test research with a cohort-retrospective approach for clinical data analysis and research and development for the development of machine learning using the Brain.js framework. Taking data from medical records of ischemic stroke and non-stroke patients in March - August 2024 at RSUD dr. Saiful Anwar. Patients were evaluated for risk factors for hypertension, smoking, and diabetes mellitus in five years, then training and testing of data on Brain.js were carried out. Determination of the Brain.js cut-off point using the ROC-AUC curve and diagnostic tests were carried out to assess accuracy, sensitivity, specificity, PPV, and NPV. Results: There were 127 ischemic stroke patients analyzed with the most gender being male (54%) and 88% being over 51 years old. The most common risk factors were hypertension at 52%, followed by smoking at 27%, and diabetes mellitus at 21%. The Brain.js confidence level cut off value with an AUC value of 0.849 (p<0.001) of 0.55, had a sensitivity of 76% and a specificity of 80%. Diagnostic tests showed that Brain.js had an accuracy of 78%, a PPV of 79%, and an NPV of 77%. Conclusions: Brain.js has good levels of accuracy, sensitivity, specificity, PPV, and NPV in predicting the incidence of ischemic stroke within 5 years after risk factors are known.
| Item Type: | Thesis (Magister) |
|---|---|
| Identification Number: | 0424060142 |
| Uncontrolled Keywords: | Stroke Iskemik, Machine Learning, AI, Brain.js, Prediktor |
| Divisions: | Profesi Kedokteran > Spesialis Neurologi, Fakultas Kedokteran |
| Depositing User: | Unnamed user with username nova |
| Date Deposited: | 18 Dec 2024 02:51 |
| Last Modified: | 18 Dec 2024 02:51 |
| URI: | http://repository.ub.ac.id/id/eprint/234056 |
![[thumbnail of DALAM MASA EMBARGO]](http://repository.ub.ac.id/style/images/fileicons/text.png) |
Text (DALAM MASA EMBARGO)
Rachmawati Wardani.pdf Restricted to Registered users only Download (14MB) |
Actions (login required)
 |
View Item |