Acoustic monitoring can now detect breathing changes, but clinics still use it only in limited ways between appointments. Continuous respiratory monitoring has been promised for a decade. Pulse oximetry clips, wearable patches, and smart inhalers have each aimed to close that gap. Most have done so only partly, in specific populations and conditions. Acoustic monitoring is the newest addition to this toolkit. This article states what it can and cannot do in 2026, not what it might achieve in three years, but what we can deliver to a COPD.
What "Continuous" Means in Practice
"Continuous" does not mean a clinician watches someone breathe in real time. For home acoustic monitoring, it means the device passively samples airway sounds during defined periods, usually overnight and during quiet rest, then sends compressed feature data to a server several times per day. The process is ambient: the patient makes no special effort, completes no diary, and presses no button.
The clinical dashboard receives no raw audio. Instead, it shows calculated acoustic features: crackle density, wheeze ratio, respiratory rate trend, and the ratio of inspiratory to expiratory phase duration. These are recalculated every 15 minutes during active monitoring windows and combined into a daily score. The score indicates whether the patient's airway profile is moving towards or away from a predefined deterioration threshold. It shows graded direction of travel, not a binary alarm.
What Acoustic Sensors Actually Record
Modern microelectromechanical (MEMS) microphones positioned against the anterior chest wall can capture breath sounds from 100 to 2,000 Hz with enough signal-to-noise ratio for feature extraction, if placement is correct and ambient noise stays below approximately 55 dB. In practice, performance is reliable during sleep and early-morning rest. It is less dependable during activity, so current systems sample mainly in defined quiet windows instead of recording throughout the day.
The signal-processing chain includes noise cancellation, breath-phase segmentation, and feature extraction through machine learning models trained on annotated respiratory audio. Respiratory specialists must label breath sounds for clinical annotation, a time-intensive process. As a result, the field's training corpus remains relatively narrow compared with cardiac auscultation datasets. Models used clinically today perform well for the COPD phenotypes represented in their training. Their performance in patients with major comorbidities, unusual body habitus, or concurrent respiratory infections is still being developed, not considered a solved problem.
Three Things We Can Do Now
Current technology gives acoustic monitoring three clinically useful outputs. First is a daily respiratory risk index for each enrolled patient. An index rising over three to five consecutive days is a stronger reason to review the patient than one cross-sectional measurement, because trajectory matters more than a single value. Second is automated detection of defined acoustic events: new-onset wheeze, increased crackle density, and a prolonged expiratory phase. Each has documented significance in COPD exacerbation physiology, while automated detection reduces the manual burden of telephone surveillance for nursing staff. Third is a longitudinal acoustic baseline for each patient. This makes individual risk scoring meaningful: a wheeze frequency normal for one patient may be a significant change for another. The system needs four to six weeks to establish a reliable individual profile.
In our early pilot work, the baseline period repeatedly proved useful when a patient's profile began to change. A clinician who reviews a seven-day acoustic trend with the medication record and recent contact history can decide proportionately whether proactive intervention is needed. That proportionality is missing from the current system, where the choice is often binary: do nothing or send the patient to A&E.
The Limitations, Clearly Set Out
Acoustic monitoring is not a replacement for spirometry. FEV1 and FVC require patient effort and a calibrated device, and provide diagnostic and staging information that acoustic analysis cannot reproduce. When the clinical need is to confirm a diagnosis or assess COPD severity for GOLD staging, spirometry remains the appropriate tool. The two modalities address different clinical questions.
Acoustic features alone cannot yet reliably separate COPD exacerbation from pneumonia or decompensated heart failure. Breath-sound patterns overlap significantly across these conditions. Any alert from the system should therefore prompt contact with and clinical assessment of the patient, rather than serve as a standalone diagnostic conclusion.
Acoustic monitoring for COPD has not yet been validated in a large randomised controlled trial. Our current evidence comes from pilot data and observational analysis in a small enrolled cohort. We state this plainly because it matters: clinicians adopting early-stage technology need a clear account of what the evidence currently supports and what it does not. Our pilot findings are encouraging, but they are not equivalent to a published, peer-reviewed efficacy study.
Why This Conversation Matters to the NHS Now
Several NHS priorities have aligned to make remote respiratory monitoring timely. The NHS Long Term Plan committed to earlier diagnosis of more patients with respiratory conditions and fewer avoidable emergency admissions. The Hewitt Review of integrated care highlighted the need to bring care closer to the patient. Respiratory medicine also remains near the top of the NHS 30-day readmission dashboard, keeping COPD a priority for integrated care boards managing emergency demand.
In that setting, identifying a deteriorating patient two to three days before an emergency department visit would address a specific, measurable gap. For clinicians, the issue is not whether continuous monitoring is conceptually useful. It is whether the alert workflow can fit an existing team structure without creating unsustainable workload. That is a workflow issue, with practical answers.
What Integration Requires in Practice
Our work with pilot sites suggests that the harder questions concern workflow, not the technology itself. Which staff member receives an alert? What escalation route applies if the patient cannot be reached? How is a RespiroScreen alert recorded in the clinical record? These points require answers before monitoring goes live, whatever the technical capability of the underlying algorithm.
The model that has worked best in our pilot uses a named respiratory nurse or monitoring coordinator. That person reviews alerts each morning, phones patients at risk, and escalates to a prescribing clinician when clinical judgement supports it. For a cohort of 30 to 40 patients, this requires roughly 20 to 40 minutes of structured review each day. Much of that time replaces reactive triage rather than adding to it.
This technology supports clinical decisions. Its value depends on the team's capacity and on a workflow designed to act on the information it identifies. We are building RespiroScreen with that constraint in view, rather than working around it.