ADVENT A pilot study to evaluate Automatic-DVT diagnostic software v1
Research type
Research Study
Full title
A multi-centre, prospective, double-blinded, pilot study evaluating artificial intelligence driven automatic detection of proximal deep vein thrombosis (DVT)
IRAS ID
285274
Contact name
Nicola Curry
Contact email
Sponsor organisation
Oxford University Hospitals NHS Foundation Trust
ISRCTN Number
ISRCTN11069056
Clinicaltrials.gov Identifier
CTU 18/111, NHSBT CTU Reference
Duration of Study in the UK
1 years, 3 months, days
Research summary
Summary of Research
Deep vein thrombosis (DVT) is a term which describes the blood clots (thrombi) that can form in the deep veins. The deep leg veins are commonly affected (such as the proximal veins: femoral vein or popliteal vein) or the deep veins of the pelvis. The standard approach to making a diagnosis involves an algorithm combining pre-test probability, a blood test called the D-dimer test, and the patient undergoing an ultrasound of the leg veins. Ultrasound is currently completed by trained radiographers.
However, handheld ultrasound probes have recently become available and they have enabled ‘app-based’ ultrasonography to be performed. ThinkSono has developed software which it is hoped has the same accuracy for diagnosing DVT as the standard ultrasound. If this trial has a positive outcome, it would mean that DVT could be diagnosed at point of care by non-radiographers such as nurses, junior doctors, general practitioners and other healthcare staff. By diagnosing DVT early in the clinical pathway (for example, at GP practices), the technology could reduce emergency department admissions and free up specialists to focus on other clinical tasks. These improvements could also potentially reduce the financial burden of the DVT diagnostic service on the NHS.Summary of Results
Thank you to all the participants in the ADVENT trial, a pilot study assessing new software developed by ThinkSono to assist in the diagnosis of deep vein thrombosis (DVT).
People suspected of having a DVT or blood clot in their leg were invited to take part in the ADVENT trial. Alongside the standard ultrasound scan (USS) they received as part of their care, participants were also scanned with the AutoDVT technology. AutoDVT is a handheld ultrasound scanner and dedicated software bundle designed to interpret the scans for signs of DVT. ADVENT investigated the ability of the AutoDVT technology to correctly diagnose proximal (upper leg) DVT in participants. We used the results of the USS as the true, or reference, DVT diagnosis and compared this to the results given by the AutoDVT technology to work out if AutoDVT correctly diagnosed participants. Of all participants with a positive USS diagnosis for proximal DVT, AutoDVT correctly diagnosed a positive proximal DVT in 68% of cases. We call this the sensitivity of AutoDVT. ADVENT also looked at the proportion of participants with a negative USS diagnosis for proximal DVT who were correctly diagnosed by the AutoDVT technology as not having the condition. This called the specificity of AutoDVT and was calculated at 80%.
The research team designed the study so that AutoDVT had to have a sensitivity of at least 90%, and a specificity of greater than 60% in order for a larger scale trial to go ahead. Both these conditions needed to be met for the study to be successful. AutoDVT did not meet these conditions.
ADVENT also investigated an alternative scenario. Instead of using the automatic diagnosis made by the AutoDVT technology, the data that the device recorded was sent to five reviewers. These reviewers independently examined the data and produced a diagnosis. One of the five reviewers' diagnoses was picked at random, like rolling a dice. The chosen diagnosis was compared against the reference USS diagnosis, and the sensitivity and specificity of AutoDVT were calculated. The sensitivity of AutoDVT with reviewer is 85% and the specificity is 73%.
We also performed a statistical technique called "Bootstrapping" to calculate the sensitivity and specificity of AutoDVT with reviewer hundreds of times, picking a reviewer's diagnosis at random every time. The final "Bootstrap" sensitivity and specificity of AutoDVT were very similar to the 85% and 73% values above. In these alternative scenarios, AutoDVT again failed to meet the conditions set by the research team to support moving to a larger scale trial.
In conclusion, the ability of the AutoDVT technology to correctly diagnose proximal DVT in participants, as measured by its sensitivity, is improved when the diagnosis is produced by a reviewer rather than automatically by the software. Unfortunately no scenario considered in this study found that the technology meets the pre-specified conditions.REC name
East of Scotland Research Ethics Service REC 2
REC reference
21/ES/0070
Date of REC Opinion
2 Aug 2021
REC opinion
Further Information Favourable Opinion