It’s time to stop manual counting of Respiratory Rate
We’ve published the first in a series of evidence-based case studies on Remote Patient Monitoring (RPM) data quality focused on Respiratory Rate (RR) measurement. In this report we explore the issues we have identified in collecting patient RR remotely and how that impacts quality of care. We also present findings on how Feebris’ AI-powered system can support users to capture and share more precise RR measurements to enable more effective care for vulnerable patients and better use of clinician's time.
Why RR is so important
Respiratory rate (RR) is a critical vital sign; it helps diagnose acute conditions like respiratory infection and pulmonary embolism and identify the severity of chronic conditions such as asthma or cystic fibrosis. Although a critical measure, it remains the most inaccurately captured vital sign in clinical practice. For the elderly patient populations we serve, accurate RR data is crucial for us to effectively determine a patient’s risk level for severe deterioration and to support efficient triage in the community.
Challenges capturing accurate RR measurements
Despite advancements in automated vital sign capture, RR remains the vital sign most commonly captured through manual counting, even in hospital settings. Clinicians or care workers will watch the patient’s chest movement as they breathe and count breaths for a 60-second period. In hospital settings, there is clear evidence of variability among trained clinicians due to errors in counting, estimations, and counting over an insufficient period.
There is currently limited evidence assessing accuracy in community settings – an evidence gap Feebris is working to fill. Feebris has recorded counting bias in the community across a sample size of 400 patient checkups. These checkups evidence a higher deviation of RR measurement by community care staff vs. the natural distribution of RR values using a clinically validated approach.
How are Feebris improving measurement accuracy for RR
At Feebris, we use AI to augment a community worker to improve the quality of RR measurement and the time it takes to capture it. We have designed algorithms that can extract RR from existing sensors commonly used in remote patient monitoring, such as a pulse oximeter and a digital stethoscope.
We have undertaken an evaluation of the performance of these algorithms comparing both carer manual counting of RR and algorithmically derived RR estimates against a clinically validated RR label. Figure 3a highlights the biases seen in carer manual counts and Figure 3b shows the RR predictions for these same signals using the Feebris algorithm. The results demonstrate a significant increase in accuracy and minimisation of bias: 59% of carer RR counts are within 2 breaths of the clinically validated label, compared to 96% of predictions from the RR algorithm. We show how only 53% of all carer counted RRs would place the patient within the correct NEWS band, but by augmenting the data collection with AI, we increase this to 85%.
Errors in RPM data can cost millions in operational waste and avoidable hospitalisation – this publication further evidences that AI support is critical for the accurate capture of vital signs in RPM. We have conducted a larger clinical study that formally evaluates the efficacy of these AI algorithms, and the impact this has on patient outcomes and healthcare costs, which will be published in a peer-reviewed journal later this year.
Read the full case-study and findings here.