Rantini, Dwi (2022) Pemodelan Mixture Survival Spasial Conditionally Autoregressive dengan Bayesian Reversible Jump Markov Chain Monte Carlo pada Kasus Demam Berdarah Dengue. Doctoral thesis, Institut Teknologi Sepuluh Nopember.
Text
06211860010004-Dissertation.pdf - Accepted Version Restricted to Repository staff only until 1 April 2024. Download (9MB) | Request a copy |
Abstract
Suatu data survival tidak jarang memiliki informasi spasial, maka ada kemungkinan bahwa data tersebut memiliki efek spasial. Jika data tersebut dimodelkan dengan menggunakan efek spasial maka dapat meminimumkan error dari model dibandingkan model yang tanpa melibatkan efek spasial. Efek spasial yang sudah dikenal untuk memodelkan error dari model adalah conditionally autoregressive (CAR) Normal, kemudian pada perkembangannya digunakan CAR Double-Exponential (DE). Jika data survival memiliki pola multimodal, maka perlu dianalisis dengan menggunakan model mixture. Kedua model CAR yang sudah disebutkan sama-sama digunakan untuk error model yang memiliki bentuk simetri. Pada kenyataannya, di dalam suatu model survival tidak selalu menghasilkan error yang berbentuk tepat simetri. Kemudian salah satu metode yang dapat digunakan untuk menentukan banyaknya komponen mixture adalah dengan metode reversible jump Markov chain Monte Carlo (RJMCMC). Namun demikian, di dalam analisis menggunakan metode RJMCMC hanya dapat digunakan untuk distribusi yang memiliki parameter location-scale. Pada banyak penelitian, kasus-kasus survival mengikuti distribusi Weibull, yang mana tidak memiliki parameter location-scale. Untuk memodelkan error yang tidak simetri maka dikembangkan CAR yang mampu menangkap pola error yang tidak simetri maupun asimetri, sehingga dibangun CAR Fernandez-Steel Skew Normal (FSSN). Untuk dapat menggunakan metode RJMCMC pada distribusi yang tidak memiliki parameter location-scale maka dibangun modifikasi dari metode RJMCMC yaitu non-Normal RJMCMC. Pada penelitian ini digunakan model survival tanpa efek random spasial, model survival dengan efek random spasial, model mixture survival tanpa efek random spasial, dan model mixture survival dengan efek random spasial. Dengan penentuan banyaknya komponen mixture menggunakan metode non-Normal RJMCMC dan efek random spasial dimodelkan dengan CAR Normal, CAR DE, dan CAR FSSN. Pada penelitian ini diambil kasus data demam berdarah dengue (DBD) di Surabaya Timur. Dengan metode non-Normal RJMCMC maka distribusi Weibull diubah menjadi sebuah distribusi yang memiliki parameter location-scale yaitu menjadi distribusi Extreme Value (EV) Type 1 (Gumbel-Type for minima). Hasil analisis menunjukkan bahwa data DBD Surabaya Timur memiliki korelasi spasial antar kecamatan dan terindikasi pola multimodal dengan empat komponen mixture. Diperoleh model terbaik yaitu model mixture survival dengan efek random spasial dimodelkan dengan CAR FSSN. Kemudian, variabel yang signifikan berpengaruh terhadap hazard rate pasien DBD adalah stadium-II pasien, denyut nadi, suhu badan, dan kadar leukosit.
================================================================================================
A survival data often has spatial information, so it is possible that the data has a spatial effect. If the data is modeled using spatial effects, it can minimize the error of the model compared to models without involving spatial effects. A well-known spatial effect for modeling errors from the model is Normal conditionally autoregressive (CAR), then in its development Double-Exponential (DE) CAR is used. If the survival data has a multimodal pattern, it needs to be analyzed using a mixture model. The two CAR models that have been mentioned are both used for error models that have a symmetrical shape. In fact, in a survival model it does not always produce an error that is exactly symmetrical. Then one of the methods that can be used to determine the number of components of the mixture is the method of reversible jump Markov chain Monte Carlo (RJMCMC). However, in the analysis using the RJMCMC method, it can only be used for distributions that have location-scale parameters. In many studies, survival cases follow the Weibull distribution, which has no location-scale parameter. To model the asymmetrical error, a CAR was developed which is able to capture asymmetric and asymmetric error patterns, so the Fernandez-Steel Skew Normal (FSSN) CAR was built. To be able to use the RJMCMC method on distributions that do not have location-scale parameters, a modification of the RJMCMC method is built, namely non-Normal RJMCMC. In this study, a survival model without spatial random effects was used, a survival model with a spatial random effect, a mixture survival model without a spatial random effect, and a mixture survival model with a spatial random effect. By determining the number of mixture components using the non-Normal RJMCMC method and spatial random effects modeled with Normal CAR, DE CAR, and FSSN CAR. In this study, data on cases of dengue hemorrhagic fever (DHF) were taken in East Surabaya. With the non-Normal RJMCMC method, the Weibull distribution is converted into a distribution that has a location-scale parameter, namely the Extreme Value (EV) Type 1 (Gumbel-Type for minima) distribution. The results of the analysis show that East Surabaya DHF data has a spatial correlation between sub-districts and indicated a multimodal pattern with four mixture components. The best model was obtained, namely the mixture survival model with spatial random effects modeled with FSSN CAR. Then, variables that significantly affect the hazard rate of DHF patients are stage-II patients, pulse, body temperature, and leukocyte levels.
Item Type: | Thesis (Doctoral) |
---|---|
Uncontrolled Keywords: | analisis Bayesian, CAR, RJMCMC, Model Mixture Survival Spasial, Demam Berdarah Dengue (DBD), Bayesian analysis, Spatial Mixture Survival Model, Dengue Hemorrhagic Fever (DHF) |
Subjects: | Q Science > QA Mathematics > QA279.5 Bayesian statistical decision theory. |
Divisions: | Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49001-(S3) PhD Thesis |
Depositing User: | Dwi Rantini |
Date Deposited: | 18 Feb 2022 01:32 |
Last Modified: | 18 Feb 2022 01:32 |
URI: | http://repository.its.ac.id/id/eprint/94098 |
Actions (login required)
View Item |