Around 20% of visits to primary care physicians (PCPs) are made by patients with symptoms that resolve spontaneously, and as much as 72% of visits are prompted by acute respiratory symptoms. Excessive and improper use of diagnostic tests is a well-known problem in primary care that increases random outcomes. The same can be said for antibiotic prescription, particularly for respiratory tract infections, which could lead to increased bacterial resistance.
Clinical guidelines and clinical scoring systems may help to standardize diagnosis and treatment, but PCPs report that their low level of applicability, and clinicians’ own lack of time, are barriers to their use in practice.
Machine-learning models (MLMs) are one area of artificial intelligence (AI) that can achieve similar or even superior performance, compared with doctors in various clinical settings. These models could be a powerful tool for the medical diagnostic process, with broad fields of application and development in general medicine.
Clinical test notes are a written report in the patient’s medical notes. They include the doctor’s interpretation of the patient’s symptoms and signs, as well as the reasoning in support of the clinical decisions made during the appointment and any subsequent action taken (eg, laboratory tests or X-rays).
A retrospective review study, carried out in a primary care setting, assessed the performance of an MLM trained in triaging patients with respiratory symptoms using only the clinical features reported by the patient (symptoms and signs) before a medical visit (triage). Clinical text notes were extracted from 1500 records for patients that received one of the following seven International Classification of Diseases 10th Revision (ICD-10) codes: J00 (common cold), J10 and J11 (influenza), J15 (bacterial pneumonia), J20 (acute bronchitis), J44 (chronic obstructive pulmonary disorder [COPD]), and J45 (asthma).
The model scored patients in two extrinsic data sets and divided them into 10 risk groups, with higher values representing greater risk. The researchers analyzed selected outcomes in each group. For each risk group, they examined the following outcomes:
Mean C-reactive protein (CRP) value.
The proportion of patients re-evaluated in primary care and emergency departments within 7 days.
The proportion of patients referred for a CXR, CXRs with signs of pneumonia, and incidentalomas.
The proportion of patients receiving antibiotic prescriptions.
The MLM, specifically the respiratory symptom triage model (RSTM), performs the triage such that patients in high-risk groups have more severe outcomes than those in lower-risk groups. Importantly, no patient in the lowest five risk groups had a CXR with signs of pneumonia or a pneumonia ICD-10 code.
Interestingly, the RSTM is ignorant of ICD-10 code subtypes but scores J15 (bacterial pneumonia) patients at an increasing rate in groups four through 10, while J00 (common cold) and J20 (acute bronchitis) decrease proportionally. J44 (COPD) was only found in groups two though eight, indicating that the model considers patients with pneumonia (J15) and COPD (J44) most likely to have worse outcomes, which matches reality.
The study authors hypothesized that if the RSTM shows similar performance in clinical settings, it could be implemented as a web-based tool, potentially triaging patients online before they make an appointment. The triage could potentially identify patients with low risk of lower respiratory tract infection, who could be attended to without the need for face-to-face consultations.
The RSTM could eliminate CXR referrals for patients in groups where the probability of positive findings is low or nonexistent, which would remove up to one third of CXRs and possibly one half of the incidentalomas without missing a positive CXR.
Even though all patients in the low-risk groups received diagnoses for which the benefit of antibiotics is debatable, antibiotics were often prescribed, regardless of the risk class. This finding means that, unfortunately, the data cannot be meaningfully assessed. The RSTM score needs no input from clinicians and can be ready when a patient enters the examination room, resulting in an easy-to-use, unambiguous, applicable score with a meaningful effect.
This article was translated from Univadis Italy, which is part of the Medscape professional network.
Source: Read Full Article