스마트 홈 기반 요로감염 조기 탐지를 위한 불확실성 인식 임상 지원 시스템
📝 Abstract
Urinary tract infection (UTI) flare-ups pose a significant health risk for older adults with chronic conditions. These infections often go unnoticed until they become severe, making early detection through innovative smart home technologies crucial. Traditional machine learning (ML) approaches relying on simple binary classification for UTI detection offer limited utility to nurses and practitioners as they lack insight into prediction uncertainty, hindering informed clinical decision-making. This paper presents a clinician-in-theloop (CIL) smart home system that leverages ambient sensor data to extract meaningful behavioral markers, train robust predictive ML models, and calibrate them to enable uncertainty-aware decision support. The system incorporates a statistically valid uncertainty quantification method called Conformal-Calibrated Interval (CCI), which quantifies uncertainty and abstains from making predictions (“I don’t know”) when the ML model’s confidence is low. Evaluated on realworld data from eight smart homes, our method outperforms baseline methods in recall and other classification metrics while maintaining the lowest abstention proportion and interval width. A survey of 42 nurses confirms that our system’s outputs are valuable for guiding clinical decision-making, underscoring their practical utility in improving informed decisions and effectively managing UTIs and other condition flare-ups in older adults.
💡 Analysis
Urinary tract infection (UTI) flare-ups pose a significant health risk for older adults with chronic conditions. These infections often go unnoticed until they become severe, making early detection through innovative smart home technologies crucial. Traditional machine learning (ML) approaches relying on simple binary classification for UTI detection offer limited utility to nurses and practitioners as they lack insight into prediction uncertainty, hindering informed clinical decision-making. This paper presents a clinician-in-theloop (CIL) smart home system that leverages ambient sensor data to extract meaningful behavioral markers, train robust predictive ML models, and calibrate them to enable uncertainty-aware decision support. The system incorporates a statistically valid uncertainty quantification method called Conformal-Calibrated Interval (CCI), which quantifies uncertainty and abstains from making predictions (“I don’t know”) when the ML model’s confidence is low. Evaluated on realworld data from eight smart homes, our method outperforms baseline methods in recall and other classification metrics while maintaining the lowest abstention proportion and interval width. A survey of 42 nurses confirms that our system’s outputs are valuable for guiding clinical decision-making, underscoring their practical utility in improving informed decisions and effectively managing UTIs and other condition flare-ups in older adults.
📄 Content
With the aging of the global population, individuals, families, and communities face the growing challenge of caring for older adults who are diagnosed with multiple chronic health conditions. In 2023, 78.8% of older adults reported multiple chronic conditions (Watson et al. 2025). Chronic conditions account for 90% of the amount the US spends on healthcare yearly (U.S. Department of Health and Human Services 2025). Living with chronic conditions is linked to greater functional decline, disability, diminished quality of life, higher likelihood of hospitalization, and increased risk of mortality (Newman et al. 2020).
Effective management of chronic health conditions through partnerships with home health nurses can significantly reduce hospital use and care costs while improving quality of life (Sylvia et al. 2008). However, the complexity of disease management and the rapidly decreasing ratio of healthcare professionals to patients are putting demands on health management that cannot be met through traditional mechanisms (Ekstedt et al. 2023). Until more effective treatments are available to halt or reverse many chronic conditions, technology-such as smart home monitoring systems-will be an essential tool for bridging the gap between growing health needs and limited care accessibility.
Smart homes suggest a way to improve management of chronic health conditions and scale nurse capabilities by monitoring behavior patterns and using the information to detect and report condition flare-ups. Currently, to address a flare-up of a chronic condition, caregivers and clinicians rely on inconsistent and unclear client self-reports or very brief in-person assessments, challenging effective condition management (Taniguchi et al. 2020). The resulting inadequacy of at-home management and lack of awareness of condition flare-ups represent leading causes of condition-related hospitalization (Fan et al. 2020). Smart homes lay the groundwork for automating behavior and health monitoring, which can supplement clinical visits, scale the reach of nurses with limited bandwidth, and improve the effectiveness of healthcare systems. By monitoring behavior patterns and anomalies, researchers have used data collected from ambient sensors in smart homes and other environments to detect cognitive decline (Robben, Englebienne, and Kröse 2016), assess sleep quality (Hasan et al. 2024), diagnose health conditions (Sprint, Schmitter-Edgecombe, and Cook 2024), and sense emergency conditions such as falls (Vaiyapuri et al. 2021).
Despite their successes, smart homes have limitations for managing chronic health conditions. First, the variability of human behavior for even healthy individuals complicates the task of modeling behavior from ambient sensor data. Second, health conditions can impact people in different ways, particularly when they occur in combination. Third, current smart home technology focuses on analyzing and reporting detected situations, rather than acting on them. To harness the power of these technologies, smart homes can partner with clinicians to provide a complete cycle of sensing, detecting, identifying, and acting. We refer to such a system as a clinician-in-the-loop (CIL) smart home.
In such a CIL system, the reliability of the system’s predictions becomes critical. Clinicians often need to decide whether to escalate care, adjust treatment, or monitor further based on smart home alerts. If the system’s output comes without a clear measure of confidence, there is a risk of both false alarms (leading to unnecessary interventions) and missed detections (leading to adverse health outcomes).
Uncertainty quantification (UQ) addresses this gap by providing a principled way to express how confident the system is in its classifications or predictions. In the context of chronic condition management, UQ can help prioritize alerts that are highly reliable, flag borderline cases for closer review, and support clinicians in balancing sensitivity with specificity. Moreover, when UQ methods come with statistical guarantees-such as those offered by conformal prediction-clinicians gain an additional layer of trust, knowing that reported uncertainty levels have rigorous backing.
We hypothesize that clinicians can act more confidently and effectively on smart home-detected condition flare-ups when the technology provides clear information on prediction uncertainty. Additionally, clinicians are more likely to trust and adopt the technology if the uncertainty in its predictions is accompanied by reliable guarantees. The main contributions of this paper are:
• Design a clinician-in-the-loop smart home framework that extracts clinically relevant behavior markers from ambient sensor data to predict UTI flare-ups. • Introduce a novel Conformal-Calibrated Interval (CCI) method to quantify predictive uncertainty, with statistically valid coverage guarantees, and enable abstention from making predictions when confidence is low. • Demonstra
This content is AI-processed based on ArXiv data.