{
  "abstract": "Background Dengue is a significant global health threat, with nearly half of the world’s population at risk. Approximately 75% of the global dengue burden is concentrated in the Southeast Asia and Western Pacific regions. While most symptomatic dengue cases present as acute febrile illnesses, 5–20% progress to severe dengue, which is associated with considerable morbidity and mortality.The critical phase of dengue typically develops 3–7 days after the onset of symptoms and lasts 24–48 hours. During this time, a subset of patients may experience sudden and severe deterioration, making timely identification and management crucial. However, the complex and overlapping clinical profiles of dengue patients, coupled with the multifaceted nature of the disease, make it challenging for clinicians to accurately predict which patients will develop severe dengue. As a result, clinicians often adopt a cautious approach, assuming that all hospitalized patients are at risk of progressing to severe dengue. This leads to intensive monitoring of all patients, contributing to increased workloads and resource strain in hospital settings.Objective This study aimed to develop and validate a machine learning-based risk scoring system, the Dengue Severity Prognostication (DeSProg) system, to predict the progression of dengue patients to severe dengue (SD) in clinical settings. The system was designed to reduce overdiagnosis and unnecessary interventions while ensuring timely management of high-risk cases.Methods The project was conducted in two phases at a tertiary hospital in Malaysia. Phase I: Development of the DeSProg system, comprising a dengue e-clerking form and an SD prediction model. Data from dengue patients were analyzed using logistic regression, random forest, and one-class support vector machine (SVM) models. These models were trained on selected clinical features to determine the optimal prediction model. DeSProg was built using Python libraries, with SQL queries extracting and visualizing dengue-related data. The prediction model with the highest balanced accuracy was integrated into the system’s dashboard. Phase II: Piloting the DeSProg system in clinical settings. New dengue cases admitted between November 2021 and November 2022 were included. System-predicted outcomes were compared to clinicians’ diagnoses to assess its performance.Results The logistic regression model achieved the highest balanced accuracy (80.15%) using 10 clinical features. In the pilot phase, the DeSProg system demonstrated an overall accuracy of 90.74%, with high specificity (93.08%) and a negative predictive value (NPV) of 97.19%. These results highlight its strong performance in identifying patients unlikely to progress to SD, thereby minimizing unnecessary interventions. However, the system’s sensitivity was 30.00%, with a positive predictive value (PPV) of 14.29%, indicating room for improvement in detecting patients at higher risk.Conclusion The DeSProg system showcases the potential of artificial intelligence in mitigating overdiagnosis in dengue management. By accurately identifying low-risk patients, the system can prevent unnecessary treatments and hospitalizations, optimizing healthcare resources. However, the limited sensitivity underscores the need for further refinement to enhance its capability to identify high-risk cases. This study emphasizes the role of AI-powered tools in striking a balance between timely intervention and the avoidance of overdiagnosis in infectious disease management.",
  "authors": [
    {
      "affiliations": [
        "University Malaya, Kuala Lumpur, Malaysia"
      ],
      "name": "Rafdzah Ahmad Zaki"
    },
    {
      "affiliations": [
        "University Malaya, Kuala Lumpur, Malaysia"
      ],
      "name": "Shier Nee Saw"
    },
    {
      "affiliations": [
        "University Malaya, Kuala Lumpur, Malaysia"
      ],
      "name": "Hang Cheng Ong"
    },
    {
      "affiliations": [
        "University Malaya, Kuala Lumpur, Malaysia"
      ],
      "name": "Wai Lam Hoo"
    },
    {
      "affiliations": [
        "University Malaya, Kuala Lumpur, Malaysia"
      ],
      "name": "Siao Hui Law"
    },
    {
      "affiliations": [
        "University Malaya, Kuala Lumpur, Malaysia"
      ],
      "name": "Jovi Wei Chiang Koh"
    },
    {
      "affiliations": [
        "University Malaya, Kuala Lumpur, Malaysia"
      ],
      "name": "Pui Li Wong"
    },
    {
      "affiliations": [
        "University Malaya, Kuala Lumpur, Malaysia"
      ],
      "name": "Norimichi Hirahara"
    },
    {
      "affiliations": [
        "University Malaya, Kuala Lumpur, Malaysia"
      ],
      "name": "Sanjay Rampal"
    },
    {
      "affiliations": [
        "University Malaya, Kuala Lumpur, Malaysia"
      ],
      "name": "Anjana Kukreja"
    },
    {
      "affiliations": [
        "University Malaya, Kuala Lumpur, Malaysia"
      ],
      "name": "Lucy Lum"
    },
    {
      "affiliations": [
        "University Malaya, Kuala Lumpur, Malaysia"
      ],
      "name": "Sharifah Faridah Syed Omar"
    }
  ],
  "title": "007 Leveraging artificial intelligence to prevent overdiagnosis: a machine learning approach for predicting risk of severe dengue",
  "uid": "81936001-57c0-530d-955a-4c6ea8df47aa"
}