Latar Belakang: Kelahiran preterm merupakan penyebab tertinggi kematian neonatal. Indonesia menduduki posisi tertinggi di ASEAN dan kelima di dunia untuk kelahiran preterm. Rumusan masalah: Belum adanya model prediksi kelahiran preterm yang memperlihatkan prediktor yang berguna untuk mengembangkan program pencegahan. Tujuan: Menemukan model prediksi kelahiran preterm berbasis machine learning untuk deteksi dini kelahiran preterm di Fasilitas Kesehatan Tingkat Pertama (FKTP). Metode: Penelitian ini menggunakan desain studi case control dengan menggunakan data rekam medis Rumah Sakit (RS) di Palembang yaitu RS YK Madira, RSMH, RS Bunda, RS Ar Rasyid, RS Muhammadiyah, dan RS Bhayangkara tahun 2019 dengan jumlah sampel 1758 responden yang terdiri dari 879 preterm dan 879 aterm. Faktor risiko yang digunakan pada penelitian ini didapatkan dari Systematic Literature Review yang terdiri dari faktor sosiodemografi (10 variabel), faktor perilaku/gaya hidup (5 variabel), faktor maternal/kondisi ibu sebelum kehamilan (8 variabel), faktor kehamilan/obstetri ginekologi (21 variabel), faktor biologis (3 variabel), faktor pelayanan kesehatan (2 variabel) dan Faktor Kondisi Janin (4 variabel). Pemodelan dilakukan dengan menggunakan machine learning dengan menggunakan algoritme decision tree, K-Nearest Neighbour (KNN), naïve bayes, logistic regression, Support Vector Machine (SVM) dan neural network (CNN1D, multilayer perceptron dan backpropagation). Hasil: Ditemukan 21 variabel penelitian dari 53 variabel yang dibutuhkan, dan menemukan 6 variabel yang menjadi prediktor utama kelahiran preterm di antaranya pre-eklamsia, perdarahan dalam kehamilan, riwayat ketuban pecah dini, jarak antar dua kehamilan, paritas, dan anemia. Pada penelitian ini ditemukan algoritma terbaik yaitu decision tree dengan nilai akurasi 95% untuk training dan 96% untuk testing dan telah dibuat prototype berupa aplikasi berbasis web untuk deteksi dini di FKTP. Kesimpulan: Ditemukan research novelty yaitu diperoleh model prediksi kelahiran preterm, dimana model ini potensial untuk digunakan di FKTP sebagai upaya deteksi dini. Model prediksi ini akan mendeteksi ibu hamil akan berisiko preterm atau tidak berisiko. Apabila diketahui ibu berisiko kelahiran preterm, maka ibu dianjurkan untuk melakukan pemeriksaan di RS, agar tidak terjadi keterlambatan penanganan yang menyebabkan kematian ibu maupun bayi. Dibandingkan tidak ada model prediksi, maka risiko kelahiran preterm tidak dapat dicegah, sehingga keterlambatan penanganan akan terjadi.

Background: Preterm birth is the highest cause of neonatal death. Indonesia occupies the highest position in ASEAN and fifth in the world for preterm births. Formulation of the problem: There is no predictive model of preterm birth that provides a useful predictor for developing prevention programs. Objective: To find prediction model of preterm birth based on machine learning for early detection of preterm birth in First Level Health Facilities (FKTP). Methods: This study uses a case control study design using medical record data at the Hospital (RS) in Palembang that isYK Madira Hospital, RSMH, Bunda Hospital, Ar Rasyid Hospital, Muhammadiyah Hospital, and Bhayangkara Hospital in 2019 with a total sample of 1758 respondents consisting of 879 preterm and 879 term. The risk factors used in this study were obtained from a Systematic Literature Review consisting of: sociodemographic factors (10 variables), behavioral/lifestyle factors (5 variables), maternal factors/mother's condition before pregnancy (8 variables), pregnancy/gynecological factors (21 variables), biological factors (3 variables), health service factors (2 variables) and fetal factors (4 variables). The modeling is done using machine learning using decision tree algorithms, K-Nearest Neighbor (KNN), nave Bayes, logistic regression, Support Vector Machine (SVM) and neural networks (CNN1D, multilayer perceptron and backpropagation). Results: Found 21 research variables from 53 variables were needed, and found 6 variables that were the main predictors of preterm birth including pre-eclampsia, bleeding in pregnancy, history of premature rupture of membranes, distance between two pregnancies, parity, and anemia. In this study, the best algorithm was found, namely decision tree with an accuracy value of 95% for training and 96% for testing and a prototype was made in the form of a web-based application for early detection in FKTP. Conclusion: It was found that the research novelty obtained a predictive model of preterm birth, which is the main cause of AKN, where this model has the potential to be used in FKTP as an early detection effort. This predictive model will detect pregnant women will be at risk of preterm or not at risk. If it is known that the mother is at risk of preterm birth, the mother is recommended to do an examination at the hospital, so that there is no delay in handling that causes the death of both mother and baby. Compared to no predictive model, the risk of preterm birth cannot be prevented, so that delays in treatment will occur. Keywords: Preterm birth, predictor factors, pre-eclampsia, machine learning, decision tr

Latar Belakang: Adverse Pregnancy Outcomes (APO) diantarannya Berat Bayi Lahir rendah (BBLR), kelahiran preterm dan lahir mati merupakan penyebab utama kematian neonatal.  Indonesia menduduki posisi tertinggi di ASEAN dan ke tujuh di dunia terkait kematian bayi baru lahir. Rumusan Masalah: Belum terdeteksi secara dini faktor risiko APO dengan cepat dan tepat di FKTP yang berdampak pada tingginya kematian bayi baru lahir, stunting dan kecerdasan yang rendah, sehingga perlu adanya pengembangan program pencegahan. Tujuan Penelitian: Menemukan model prediksi APO berbasis machine learning untuk pengembangan ePHR sebagai upaya deteksi dini di FKTP. Metode: Penelitian ini menggunakan Mixed Methods: Sequential Explanatory design melalui empat (4) tahap yaitu: (1) Studi Literatur; (2) Pembuatan model prediksi APO berbasis Machine Learning, (3) Membangun Prototipe ePHR Sebagai Deteksi Dini di FKTP dan (4) Uji penerimaan bidan dan Uji Efikasi terhadap penggunaan ePHR. Hasil: Ditemukan 22 Variabel penelitian dari 51 variabel yang dibutuhkan, ditemukan 3 indikator kritis dan 8 fitur esensial yang menjadi predictor utama kejadian APO diantaranya: Kesehatan ibu (komplikasi kehamilan dan atau persalinan, paritas lebih dari 4); Ketersediaan dan kualitas pelayanan (fasilitas pertolongan persalinan); Profile ibu (pendidikan rendah, status sosial ekonomi rendah, usia terlalu tua atau terlalu muda dan ibu bekerja). Analisis menunjukkan algoritma terbaik adalah random forest dengan target multiclass: AUC 98,4%; Sensitivity 95,1%; F1 Score 94,3%, Diimplementasikan kedalam prototipe ePHR  ‘'e-bayiKusehAt” dengan 10 fitur untuk mendukung proses bisnis baru Deteksi dini APO terintegrasi dalam layanan antenatal di FKTP yang digunakan oleh bidan dan ibu hamil; Uji efikasi: Penggunaan Prototipe ePHR ‘'e-bayiKusehAt” meningkatkan kinerja bidan terutama deteksi dini terintegrasi dalam pelayanan Antenatal Diff-in-Diff tertinggi 1,571 (p<0,001); Kemampuan prediksi APO Diff-in-Diff 0,079 (p = 0,002); Mempercepat keputusan rujukan Diff-in-Diff 0,324 (p = 0,001); dan pengetahuan bidan tentang faktor risiko APO Diff-in-Diff: 0,388 ( p=0,031); Ibu hamil meningkatkan Akses edukasi Diff-in-Diff: 1,680 (p < 0,001) dan Pengetahuan Tanda Bahaya: Diff-in-Diff: 1,443 (p < 0,001). Kesimpulan: Telah diperoleh suatu model untuk prediksi APO, yang didasari dari 22 variabel penting. Model tersebut digunakan untuk membangun prototipe e PHR ‘'e-bayiKusehAt”, dimana hasil uji coba penggunaan prototipe berdampak pada peningkatan kinerja bidan dalam Deteksi Dini APO. Penggunaan prototipe ini ternyata juga efektif dalam meningkatkan kesadaran dan kesiapan ibu hamil terhadap potensi risiko kehamilan yang berdampak pada kejadian APO.

Background: Adverse Pregnancy Outcomes (APO) including low birth weight (LBW), preterm birth and stillbirth are the leading causes of neonatal mortality.  Indonesia has the highest position in ASEAN and seventh in the world regarding newborn mortality. Problem Formulation: APO risk factors have not been detected early and quickly at primary health care facilities, which has an impact on high newborn mortality, stunting and low intelligence, so it is necessary to develop a prevention program. Research Objective: Finding a machine learning-based APO prediction model for the development of ePHR as an effort to early detection in primary care. Methods: This research uses Mixed Methods: Sequential Explanatory design through four (4) stages, namely: (1) Literature Study; (2) Machine Learning-based APO prediction modeling, (3) Building ePHR Prototype as Early Detection in FKTP and (4) Midwife acceptance test and Efficacy Test for the use of ePHR. Results: 22 research variables were found from the 51 variables needed, 3 critical indicators and 8 essential features were found to be the main predictors of APO events including: Maternal health (complications of pregnancy and or childbirth, parity more than 4); Availability and quality of services (delivery assistance facilities); Maternal profile (low education, low socioeconomic status, too old or too young and working mothers). Analysis shows the best algorithm is random forest with multiclass target: AUC 98.4%; Sensitivity 95.1%; F1 Score 94.3%, Implemented into the ePHR prototype “e-bayiKusehAt” with 10 features to support new business processes Early detection of APO integrated in antenatal services at FKTP used by midwives and pregnant women; Efficacy test: The use of the ePHR Prototype “e-bayiKusehAt” improves midwives' performance, especially integrated early detection in antenatal services Diff-in-Diff highest 1.571 (p<0.001); APO Diff-in-Diff prediction ability 0.079 (p = 0.002); Accelerating Diff-in-Diff referral decisions 0.324 (p = 0.001); and midwives' knowledge of APO Diff-in-Diff risk factors: 0.388 (p = 0.031); Pregnant women increase Diff-in-Diff education access: 1.680 (p < 0.001) and knowledge of danger signs: Diff-in-Diff: 1,443 (p < 0,001). Conclusion: A model for APO prediction was obtained, based on 22 important variables. The model was used to build the e PHR prototype “e-babyKusehAt”, where the results of the trial use of the prototype had an impact on improving the performance of midwives in Early Detection of APO. The use of this prototype was also effective in increasing the awareness and readiness of pregnant women to potential pregnancy risks that have an impact on the incidence of APO.