Pemodelan Spatio-Temporal Bayesian Menggunakan BYM-2 Berbasis INLA Pada Pendugaan Risiko Relatif Malaria di Wilayah Papua

Abstract

Malaria masih menjadi salah satu masalah kesehatan masyarakat utama di Indonesia, dengan sekitar 90–95% kasus nasional terjadi di wilayah Papua. Penularan malaria dipengaruhi oleh kondisi iklim dan lingkungan serta adanya keterkaitan antarwilayah dan antarwaktu, sehingga diperlukan pendekatan pemodelan yang mampu mengakomodasi ketergantungan spasial dan temporal tersebut. Penelitian ini bertujuan untuk menduga risiko relatif malaria di wilayah Papua menggunakan model Bayesian Besag–York–Mollié 2 (BYM-2) dengan pendekatan integrated nested Laplace approximation (INLA). Data yang digunakan berupa jumlah kasus malaria tahunan pada 42 kabupaten/kota selama periode 2021–2024, beserta peubah iklim dan lingkungan yang meliputi curah hujan, kecepatan angin, kelembapan relatif, normalized difference vegetation index (NDVI), dan elevasi. Struktur ketetanggaan spasial dibangun menggunakan pendekatan hybrid yang menggabungkan queen contiguity dan k-nearest neighbors (k = 2). Model terbaik dipilih melalui perbandingan beberapa kombinasi hyperprior menggunakan kriteria deviance information criterion (DIC), Watanabe–Akaike information criterion (WAIC), dan log pseudo marginal likelihood (LPML). Model terbaik diperoleh dengan kombinasi hyperprior t~PC(2; 0,05) dan ?~PC(0,1; 0,1). Nilai risiko relatif yang dihasilkan memiliki kesesuaian yang baik dengan standardized incidence ratio (SIR), sehingga menunjukkan bahwa model mampu menggambarkan pola risiko malaria dengan baik. Elevasi merupakan satu-satunya peubah lingkungan yang memiliki hubungan secara signifikan terhadap risiko malaria, sementara keragaman spasial dan temporal yang cukup besar masih tetap ditemukan setelah memperhitungkan seluruh peubah yang diamati. Wilayah dengan risiko relatif malaria tertinggi terkonsentrasi di bagian utara dan selatan wilayah Papua, sehingga perlu menjadi prioritas dalam kegiatan surveilans dan pengendalian malaria secara lebih terarah.

Malaria remains a major public health challenge in Indonesia, with approximately 90–95% of the country's reported cases occurring on Papua region. Malaria transmission is influenced by climatic and environmental conditions, as well as spatial and temporal dependencies, highlighting the need for a modeling approach that can account for these complexities. This study aimed to estimate the relative risk of malaria on Papua Island using the Bayesian Besag–York–Mollié 2 (BYM-2) model implemented through the integrated nested Laplace approximation (INLA) approach. The analysis was based on annual malaria case data from 42 districts/cities during 2021–2024, along with climatic and environmental covariates, including rainfall, wind speed, relative humidity, the normalized difference vegetation index (NDVI), and elevation. The spatial neighborhood structure was defined using a hybrid approach that combined queen contiguity and k-nearest neighbors (k = 2). The optimal model was selected by comparing several hyperprior combinations using the deviance information criterion (DIC), Watanabe–Akaike information criterion (WAIC), and log pseudo marginal likelihood (LPML). The best-performing model employed the hyperprior combination t ~ PC(2; 0,05) and ? ~ PC(0,1; 0,1). The estimated relative risks showed good agreement with the standardized incidence ratio (SIR), indicating that the model adequately captured the spatial pattern of malaria risk. Elevation was the only environmental variable significantly associated with malaria risk, while substantial spatial and temporal variation remained after accounting for all observed covariates. Areas with the highest relative risk were concentrated in the northern and southern parts of Papua Island, suggesting that these regions should be prioritized for targeted malaria surveillance and control efforts.