Lung Disease
Conditions
Brief summary
To evaluate whether an artificial intelligence decision support software (ArtiQ.Spiro) improves the diagnostic accuracy of spirometry interpreted by primary care clinicians, as measured by Clinician Diagnostic Accuracy (vs Reference Standard).
Detailed description
This is a randomised controlled study to evaluate the effects of AI support software on the performance of primary care clinicians in the interpretation of spirometry. Clinicians will be provided with a clinical dataset of 50 entirely anonymous, previously recorded real-world spirometry records to interpret and will be asked to complete specific questions about diagnosis and quality assessment. The records will be randomly selected from a database comprising spirometry records from 1122 patients undergoing spirometry in primary care and community -based respiratory clinics in Hillingdon borough between 2015-2018. Participating clinicians will be allocated at random to receive either spirometry records alone or spirometry records with the addition of an AI spirometry interpretation eport. The clinical spirometry records will be de-identified (name, date of birth, address, postcode, occupation, GP, medications data removed), by a member of the clinical care team. Study participants (participating clinicians) will independently examine the same 50 spirometry records through an online platform. For each spirometry record, the primary care clinician participant will answer questions about technical quality, pattern interpretation, preferred diagnosis, differential diagnosis and self-rated confidence with these answers. The study statistician will be blinded to treatment allocation up to completion of analysis and interpretation. The reference standards for spirometry technical quality and pattern interpretation will be made by a senior experienced respiratory physiologist but without access to AI report. The reference standard for diagnosis will be made by a panel of three respiratory specialists from the clinical care team with access to medical notes and results of relevant investigations but without access to AI report.
Interventions
OTHERArtificial Intelligence-powered Spirometry Interpretation Report
A report generated by artificial intelligence powered software that assessed technical quality of spirometry and estimates the diagnostic probability of six categories: COPD/Asthma/ILD/ Normal/Other obstructive/Other Unidentified
Sponsors
National Institute for Health Research, United Kingdom
Royal Brompton & Harefield NHS Foundation Trust
Study design
Allocation
RANDOMIZED
Intervention model
PARALLEL
Primary purpose
HEALTH_SERVICES_RESEARCH
Masking
DOUBLE (Investigator, Outcomes Assessor)
Eligibility
Sex/Gender
ALL
Age
18 Years to 99 Years
Healthy volunteers
Yes
Inclusion criteria
1. Clinicians working in primary care (for at least 50% of their job plan) in the UK, who refer for or perform spirometry (typically GP, practice nurse) 2. Able to access spirometry traces on study platform 3. Provide written informed consent via study platform
Exclusion criteria
1\. Clinicians who have completed specialist training in respiratory medicine and recognised by the General Medical Council with a right to practise as a NHS consultant in respiratory medicine
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Preferred Diagnostic Performance | Six months | A correct case is where the preferred diagnosis matches the reference final diagnosis. Units will be percentage of total cases that are correct. |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Diagnostic self-rated confidence | Six months | Diagnostic self-rated confidence will be measured on a visual analogue scale (0-10) where 0 = not confident at all; 10= very confident) |
| Quality Assessment self-rated confidence | Six months | Quality Assessment self-rated confidence will be measured on a visual analogue scale (0-10) where is 0 = not confident at all; 10= very confident) |
| Pattern interpretation | Six months | A correct case is where the participants' selected pattern matches the reference pattern. Options are: Normal, Airflow obstruction, Possible restriction or non-specific pattern, Possible Mixed Disorder. Units will be percentage of total cases that are correct. |
| Quality assessment performance | Six months | A correct case is where the participant's quality grade matches the reference quality grade. Options are: Acceptable (Grade A/B) or Not Acceptable (Grades C/D/E/F/U). Units will be percentage of total cases that are correct. |
| Pattern interpretation self-rated confidence | Six months | Pattern interpretation self-rated confidence will be measured on a visual analogue scale (0-10) where 0 = not confident at all; 10= very confident) |
| Differential diagnostic performance | Six months | A correct case is where the preferred or differential diagnosis matches the reference final diagnosis. Units will be percentage of total cases that are correct. |
Countries
United Kingdom
Contacts
Primary ContactEthaar El-Emir, PhD
Backup ContactGeorge Edwards, MSc
Outcome results
None listed