Development and evaluation of a diagnostic aiding tool for differentiating tropical fevers using artificial intelligence approach: a study protocol from tertiary care hospital in South India

Introduction Application of artificial intelligence (AI) tools in the healthcare setting gains importance especially in the domain of disease diagnosis. Numerous studies have tried to explore AI in the diagnosis of various diseases, including tropical fevers such as dengue and malaria. However, there is a lack of standard guidelines to develop the AI models, the gap between clinical and engineering expertise and clinical validation of the models, and hence there is a critical need for the development of an integrated diagnostic tool which uses demographical, laboratory variables and epidemiological parameters of patient and provides early prediction. Methods and analysis The present study aimed to develop and evaluate a machine-learning (ML) prediction tool for differential diagnosis of tropical fevers for adult patients (>18 years) using a three-phase approach in a tertiary care centre in South India by January 2026. Phase Ⅰ involves identification of the prevalent tropical fevers and associated clinical parameters to develop the AI model through a retrospective audit and qualitative interview. Phase Ⅱ involves retrospective data collection from hospital medical records for finalised diseases (1000 cases per disease) and clinical parameters, with data being used for model development using the Python language. Support vector machine, logistic regression, K-Nearest Neighbors, Naïve Bayes and ensemble models such as decision tree and Random Forest will be employed along with explainable AI techniques. They are used as they are easy to understand and interpret, well established, most effective for structured data, enhancing the transparency and interpretability of the predictive machine learning models, and their use has been widely supported in previous studies across various contexts. Suitable statistical parameters like specificity, sensitivity and area under receiver operating characteristic (AUROC) will be applied to evaluate model performance. In phase Ⅲ, the developed model will be implemented prospectively to assess the feasibility of model implementation. Model performance such as specificity, sensitivity and AUROC will be calculated, and the finally developed model will be implemented in a single tertiary care hospital to evaluate its overall performance. Ethics and dissemination Ethical approval for the study has been obtained from the institutional ethics committee of the Kasturba Medical College and Kasturba Hospital, Manipal (IEC number: 6/2024). Informed consent will be taken for obtaining the data of the patient for the evaluation of the model in the third phase of the study, and data will be kept confidential. The study results will be disseminated by publishing them in a peer-reviewed journal.