Identifikasi dan Evaluasi Interaksi Obat Antihipertensi Pada Pasien Geriatri Dengan Hipertensi Menggunakan Platform Artificial Intelligence Dibandingkan Dengan Database Uptodate Lexidrug

Januarius Tomi Kiik, Dr. apt. Soni Siswanto, M.Biomed; Prof. Dr. apt. Agung Endro Nugroho, M.Si

2026 | Tesis | S2 Mag.Farmasi Klinik

Latar Belakang: Ketersediaan teknologi AI seperti ChatGPT-4.5 dan Meta AI WhatsApp sangat mendukung praktik pelayanan farmasi klinis, khususnya dalam hal mendeteksi interaksi obat. AI berpotensi membantu apoteker mengidentifikasi interaksi secara cepat, namun akurasinya tetap perlu divalidasi. Namun demikian, berbagai penelitian menunjukkan bahwa terdapat perbedaan tingkat akurasi dari masing-masing platform AI dalam mendeteksi interaksi obat. Hal ini menunjukkan bahwa meskipun AI menawarkan potensi besar, evaluasi lebih lanjut tetap diperlukan untuk menilai keandalan dan kesesuaian penggunaannya dalam konteks klinis. Oleh karena itu, penelitian dan pengembangan yang berkelanjutan sangat diperlukan guna memastikan bahwa penerapan AI khususnya ChatGPT-4.5 dan Meta AI WhatsApp dibidang kesehatan berjalan dengan aman, efektif dan sesuai dengan standar praktik profesional.

Tujuan: Mengetahui perbedaan akurasi, sensitivitas, spesifisitas, nilai prediksi positif dan nilai prediksi negatif dari masing-masing platform AI, yaitu ChatGPT-4.5 dan Meta AI WhatsApp dalam mendeteksi interaksi obat antihipertensi dengan UpToDate Lexidrug sebagai pembanding.

Metode: Penelitian ini merupakan jenis penelitian observasional dengan rancangan cross-sectional yaitu membandingkan antara ChatGPT-4.5 dengan Meta AI WhatsApp model Llama 4 Scout. UpToDate Lexidrug digunakan sebagai standar referensi. Data pada penelitian ini dianalisis menggunakan statistik deskriptif yaitu identifikasi nilai true positive, true negative, false positive dan false negative. Selanjutnya data di evaluasi dengan menghitung nilai akurasi, spesifisitas, sensitivitas, nilai prediksi positif dan nilai prediksi negatif. Alat yang digunakan meliputi lembar pengumpul data dan software pengolah data (Microsoft Excel dan SPSS).

Hasil: Hasil penelitian menunjukkan bahwa ChatGPT-4.5 memiliki performa diagnostik yang lebih unggul dibandingkan Meta AI WhatsApp dalam mendeteksi interaksi obat antihipertensi. ChatGPT-4.5 menunjukkan nilai sensitivitas sebesar 0,862, spesifisitas 0,779, nilai prediksi positif 0,628, nilai prediksi negatif 0,929 serta akurasi 0,804. Sebaliknya, Meta AI WhatsApp memperoleh nilai sensitivitas 0,750, spesifisitas 0,686, nilai prediksi positif 0,509, nilai prediksi negatif 0,683, serta akurasi 0,705. Tingkat kesepakatan tertinggi diperoleh UpToDate Lexidrug - ChatGPT-4.5 dengan nilai cohen's kappa 0,367 yang termasuk kategori fair agreement. Interaksi kategori no known interaction memiliki nilai AUC tertinggi yang tergolong cukup baik pada seluruh analisis. Secara keseluruhan, ChatGPT-4.5 menunjukkan keunggulan dibandingkan dengan Meta AI WhatsApp.

Background: The availability of AI technologies such as ChatGPT-4.5 and Meta AI WhatsApp greatly supports clinical pharmacy practice, particularly in detecting drug interactions. AI has the potential to help pharmacists identify interactions more quickly, but its accuracy still needs to be validated. However, various studies indicate differences in the accuracy of AI platforms in detecting drug interactions. This indicates that while AI offers great potential, further evaluation is still needed to assess its reliability and suitability for use in clinical contexts. Therefore, continuous research and development are essential to ensure that the application of AI—specifically ChatGPT-4.5 and Meta AI WhatsApp—in the healthcare field is safe, effective, and in accordance with professional practice standards.

Objective: To determine the differences in accuracy, sensitivity, specificity, positive predictive value, and negative predictive value among the AI platforms—namely ChatGPT-4.5 and Meta AI WhatsApp—in detecting interactions between antihypertensive medications, using UpToDate Lexidrug as a comparator.

Methods: This study is an observational, cross-sectional study comparing ChatGPT-4.5 with Meta AI WhatsApp (Llama 4 Scout model). In this study, UpToDate Lexidrug was used as the gold standard. Data in this study were analyzed using descriptive statistics, specifically identifying true positives, true negatives, false positives, and false negatives. The data were then evaluated by calculating accuracy, specificity, sensitivity, positive predictive value, and negative predictive value. Tools used included data collection forms and data analysis software (Microsoft Excel and SPSS).

Results: The study results show that ChatGPT-4.5 outperforms Meta AI WhatsApp in detecting interactions involving antihypertensive medications. ChatGPT-4.5 demonstrated a sensitivity of 0.862, specificity of 0.779, positive predictive value of 0.628, negative predictive value of 0.929, and accuracy of 0.804. In contrast, Meta AI WhatsApp achieved a sensitivity of 0.750, specificity of 0.686, positive predictive value of 0.509, negative predictive value of 0.683, and accuracy of 0.705. The highest level of agreement was achieved by UpToDate Lexidrug – ChatGPT-4.5 with a Cohen’s kappa value of 0.367, which falls into the “fair agreement” category. The “no known interaction” category had the highest AUC value, which was considered quite good across all analyses. Overall, ChatGPT-4.5 demonstrated superiority compared to Meta AI WhatsApp

Objective: To determine the differences in accuracy, sensitivity, specificity, positive predictive value, and negative predictive value among the AI platforms—namely ChatGPT-4.5 and Meta AI WhatsApp—in detecting interactions between antihypertensive medications, using UpToDate Lexidrug as a comparator.

Methods: This study is an observational, cross-sectional study comparing ChatGPT-4.5 with Meta AI WhatsApp (Llama 4 Scout model). In this study, UpToDate Lexidrug was used as the gold standard. Data in this study were analyzed using descriptive statistics, specifically identifying true positives, true negatives, false positives, and false negatives. The data were then evaluated by calculating accuracy, specificity, sensitivity, positive predictive value, and negative predictive value. Tools used included data collection forms and data analysis software (Microsoft Excel and SPSS).

Results: The study results show that ChatGPT-4.5 outperforms Meta AI WhatsApp in detecting interactions involving antihypertensive medications. ChatGPT-4.5 demonstrated a sensitivity of 0.862, specificity of 0.779, positive predictive value of 0.628, negative predictive value of 0.929, and accuracy of 0.804. In contrast, Meta AI WhatsApp achieved a sensitivity of 0.750, specificity of 0.686, positive predictive value of 0.509, negative predictive value of 0.683, and accuracy of 0.705. The highest level of agreement was achieved by UpToDate Lexidrug – ChatGPT-4.5 with a Cohen’s kappa value of 0.367, which falls into the “fair agreement” category. The “no known interaction” category had the highest AUC value, which was considered quite good across all analyses. Overall, ChatGPT-4.5 demonstrated superiority compared to Meta AI WhatsApp.

Kata Kunci : ChatGPT-4.5, Meta AI WhatsApp, UpToDate Lexidrug, Interaksi Obat, Hipertensi