DrHouse: An LLM-empowered Diagnostic Reasoning System through Harnessing Outcomes from Sensor Data and Expert Knowledge

TL;DR

DrHouse introduces an LLM-empowered diagnostic reasoning system that concurrently integrates daily sensor data from wearables with continuously updated clinical resources to support multi-turn medical consultations. It employs a dual knowledge-base architecture, multi-source retrieval (including guideline trees and sensor data), careful knowledge fusion, and a probabilistic, concurrent evaluation of candidate diseases to produce explainable diagnoses with disease likelihoods. Across three public dialogue datasets and real-world user studies with clinicians and patients, DrHouse demonstrates up to 18.8% improvements in diagnostic accuracy and strong user acceptance, highlighting its potential to reduce unnecessary tests and improve home-based medical consulting. The work advances practical deployment of sensor-data-aware LLMs in healthcare, providing a scalable framework for ongoing incorporation of up-to-date guidelines and objective physiological indicators into AI-assisted diagnosis.

Abstract

Large language models (LLMs) have the potential to transform digital healthcare, as evidenced by recent advances in LLM-based virtual doctors. However, current approaches rely on patient's subjective descriptions of symptoms, causing increased misdiagnosis. Recognizing the value of daily data from smart devices, we introduce a novel LLM-based multi-turn consultation virtual doctor system, DrHouse, which incorporates three significant contributions: 1) It utilizes sensor data from smart devices in the diagnosis process, enhancing accuracy and reliability. 2) DrHouse leverages continuously updating medical databases such as Up-to-Date and PubMed to ensure our model remains at diagnostic standard's forefront. 3) DrHouse introduces a novel diagnostic algorithm that concurrently evaluates potential diseases and their likelihood, facilitating more nuanced and informed medical assessments. Through multi-turn interactions, DrHouse determines the next steps, such as accessing daily data from smart devices or requesting in-lab tests, and progressively refines its diagnoses. Evaluations on three public datasets and our self-collected datasets show that DrHouse can achieve up to an 18.8% increase in diagnosis accuracy over the state-of-the-art baselines. The results of a 32-participant user study show that 75% medical experts and 91.7% patients are willing to use DrHouse.

Figures (27)

• Figure 1: Overview of DrHouse. DrHouse incorporates patients' sensor data from smart devices into the multi-turn diagnosis process to enhance accuracy and reliability. DrHouse can provide medical consultations for patients at their homes, or offer diagnostic references to doctors to reduce their workload.
• Figure 2: An example of misdiagnosis due to patients' wrong subjective perception. The words highlighted in green and red represent the patient's subjective descriptions of the symptom and the diagnostic conclusions made by an LLM, respectively. Note that we use GPT-4 as a naive LLM for example.
• Figure 3: An example of sensor data metrics in medical diagnosis guidelines. The left figure shows the latest medical diagnostic guidelines for acute bronchitis on the Up-to-Date database. The right figure shows that accessing patients' daily sensor data can take the diagnostic process one step further and assist virtual doctors in decision-making.
• Figure 4: System overview of DrHouse.
• Figure 5: Examples of inaccurate and inefficient sensor data knowledge retrieval. The words highlighted in red represent the results of sensor data knowledge retrieval. Note that we use GPT-4 as a naive LLM for example.