EFEK SPOT SIZE TERHADAP NILAI BIOMARKER DENGAN MEMPERTIMBANGKAN KONDISI LINGKUNGAN MELALUI BAYESIAN GENERALIZED ADDITIVE MODELS

Afiyat Nur Izzah, Dr. Adhitya Ronnie Effendie., S.Si., M.Si., M.Sc.

2025 | Tesis | S2 Matematika

Metode Dried Blood Spot (DBS) kian digunakan karena minim invasif dan mudah secara logistik, namun validitas biomarker sensitif terhadap faktor pra-analitik, termasuk ukuran spot. Penelitian ini menganalisis pengaruh suhu, kelembapan, waktu pengeringan, waktu pengiriman, dan ukuran spot terhadap tujuh biomarker cho_x4, crp_x4, trg_x4, cyc_x4, hdl_x4, thb_x4, a1c_x4 menggunakan Bayesian Generalized Additive Model (Bayesian GAM) dengan family Gaussian atau log-normal sesuai karakter data, menggabungkan komponen parametrik dan \textit{smoothing spline}, serta diestimasi melalui NUTS MCMC (3-4 rantai). Hasil penelitian menunjukkan bahwa Bayesian GAM mampu menghasilkan estimasi yang stabil dengan nilai (R-hat ? 1.00) serta credible interval yang relatif sempit, sehingga mengindikasikan konvergensi yang baik. Evaluasi model berdasarkan WAIC dan LOOIC memperlihatkan kecocokan yang memadai, khususnya pada biomarker hdl_x4 dan a1c_x4. Selain itu, variabel lingkungan seperti suhu (t_temp) serta ukuran spot darah tertransformasi logaritmik (log1p(t_sizeA)) terbukti konsisten memberikan pengaruh signifikan terhadap sebagian besar biomarker. Temuan ini menegaskan bahwa pendekatan Bayesian GAM efektif dalam mengakomodasi variabilitas dan ketidakpastian data DBS, serta relevan untuk mendukung analisis biomarker dalam konteks penelitian epidemiologi berbasis risiko.

Dried blood spot (DBS) methods are increasingly used because they are minimally invasive and straightforward to implement, yet biomarker validity can be sensitive to pre-analytic factors such as spot size. This study evaluates how storage temperature, humidity, drying time, shipping time, and spot size influence seven biomarkers (cho_x4, crp_x4, trg_x4, cyc_x4, hdl_x4, thb_x4, a1c_x4) using a Bayesian Generalized Additive Model (Bayesian GAM). The models combine parametric terms with smoothing splines under Gaussian or log-normal families and are estimated with NUTS MCMC using 3–4 chains. The results indicate stable estimates (R-hat ? 1.00) with relatively narrow credible intervals, reflecting good convergence. Model fit assessed via WAIC and LOOIC was adequate overall, with especially strong performance for hdl_x4 and a1c_x4. Environmental factors—most notably temperature (t_temp) and the log-transformed spot size (log1p(t_sizeA))—consistently showed significant effects. These findings underscore the value of Bayesian GAM for accommodating variability and quantifying uncertainty in DBS data, making it a strong approach for biomarker modeling in risk-based epidemiological studies.

Kata Kunci : Dried blood spot, Biomarker, Bayesian generalized additive models