2nd Edition of Public Health World Conference (PHWC) 2026

Abstract

Artificial Intelligence (AI) is increasingly adopted for clinical decision support, disease surveillance, and health system management; however, healthcare data quality and trustworthy AI deployment remain significant barriers to real-world implementation. This study aimed to evaluate the suitability of routine outpatient healthcare data for AI-enabled clinical intelligence and public health applications.

Exploratory descriptive analysis and data quality assessment were performed on approximately 10,000 anonymized outpatient encounters obtained during a healthcare innovation challenge. The dataset contained routine clinical observations, symptom information, presenting complaints, and clinician-assigned severity indicators.

Analysis identified substantial data quality deficiencies, including physiologically impossible observations such as systolic blood pressure of 2147483647 mmHg, diastolic blood pressure of 10080 mmHg, pulse rate of 821 beats per minute, oxygen saturation of 999 percent, and random blood sugar of 100130 mg/dL.

Simultaneously, clinically significant events were severely underrepresented: no recorded temperatures above 104 degrees Fahrenheit, only 3 pulse rate observations above 130 beats per minute, 38 cases with systolic blood pressure greater than or equal to 180 mmHg, 89 cases with oxygen saturation less than or equal to 90 percent, and only 43 records at the highest severity level.

These findings demonstrate that routine outpatient datasets may contain both extreme documentation errors and severe class imbalance, creating significant risks for AI model development, validation, and deployment. The findings reinforce the importance of clinical data governance, explainable and human-in-the-loop AI approaches, and workflow-centered validation frameworks for healthcare AI systems.

This study contributes to Digital Health and Public Health Informatics by highlighting robust data quality assessment, clinical oversight, and disease surveillance analytics as prerequisites for reliable clinical workflow intelligence and population-level health insights.