Prospective multicenter observational study of an integrated Artificial Intelligence system with live monitoring
Organizational Data
- DRKS-ID:
- DRKS00027322
- Recruitment Status:
- Recruiting ongoing
- Date of registration in DRKS:
- 2022-03-07
- Last update in DRKS:
- 2023-12-01
- Registration type:
- Retrospective
Acronym/abbreviation of the study
PRAIM
URL of the study
No Entry
Brief summary in lay language
In collaboration with German screening units, a prospective observational study will commence in early 2022 to investigate the effectiveness, safety, and productivity of an AI application and workflow to support breast cancer screening physicians under real-world conditions. The PRAIM study (PRospective, multicenter observational study of an integrated AI system with live monitoring to support breast cancer screening) is led by Prof. Dr. med. Alexander Katalinic (University Hospital Schleswig-Holstein) in collaboration with Vara (MX Healthcare GmbH). PRAIM is a non-interventional, observational, controlled, clinical follow-up study of a CE-certified medical product (Vara) designed to display mammograms and pre-classify findings to assist users in their reporting routine. Women will continue to undergo routine breast cancer screening as usual, consistent with the screening guidelines and independent of the research objective being pursued in this study. No study monitoring visits are planned. No (randomized) allocation of mammography studies to AI or control is performed. The source population for this study will be asymptomatic women presenting to the German national breast screening program for biennial screening with digital mammography at participating breast cancer screening units. The study will aim to prospectively follow approximately at least 400,000 screening-age women, regardless of breast cancer diagnosis and screening history. The study consists of three parts: (I) Assessing the association between AI use and breast cancer detection rate (CDR) and recall rate (RR), the two primary endpoints of the study; (II) Longitudinal measurement of AI impact on CDR, RR, and reader sensitivity/specificity; and (III) Assessing the relationship between AI use and interval cancer rate and breast cancer screening program sensitivity. The primary endpoints will be evaluated in the groups of all participants deemed eligible at the time of inclusion. Secondary endpoints are related to study part (II) and (III). Primary and secondary endpoints are compared between women whose studies are read with Live AI, meaning one or both readers use AI support and the Vara user interface, and women from the same screening units whose studies are read under Shadow mode, in which Vara is evaluated in the background without interacting with readers (control group). The Shadow mode group will be supplemented with data from a retrospective historical screening cohort, composed of studies examined by means of mammography and double-reading without AI support.
Brief summary in scientific language
PRAIM is a non-interventional, observational, clinical follow-up study of a CE-certified medical product (Vara) designed to display mammograms and pre-classify findings to assist users in their reporting routine. Women will continue to undergo routine breast cancer screening as usual, consistent with the screening guidelines and independent of the research objective being pursued in this study. No study monitoring visits are planned. No (randomized) allocation of mammography studies to AI or control is performed. The source population for this study will be asymptomatic women presenting to the German national breast screening program for biennial screening with digital mammography at participating breast cancer screening units. The study will aim to prospectively follow approximately at least 400,000 screening-age women, regardless of breast cancer diagnosis and screening history. PRIMARY OBJECTIVE: The purpose of this observational study is to investigate the real-world effectiveness, safety, and productivity of an AI application and workflow to support breast cancer screening radiologists ("Live AI"); and to compare this to standard guideline-concordant screening without AI assistance ("Shadow mode"). It is hypothesized that AI application is not inferior to the standard screening process. MAIN OBJECTIVES: 1. To assess the safety and effectiveness of the AI algorithm, in terms of the screen-detected cancer detection rate and recall rate for studies read with and without AI support 2. To assess improvements in user performance (in terms of diagnostic measures) over time, and understand the factors that have the greatest influence on improvement in order to understand the most optimized setting for AI 3. To assess productivity in terms of number of exams sent to consensus for studies read with and without AI; and resource utilization and cost-effectiveness in terms of recall rate, diagnostic imaging rate, and biopsy rate for studies read with and without AI support EXPLORATORY OBJECTIVE (where available): To assess the interval cancer rate and overall program sensitivity for studies read with or without AI support if data becomes available as per standard of care, made possible through the routine data linkages between screening units and local cancer registries
Health condition or problem studied
- ICD10:
- C50 - Malignant neoplasm of breast
- ICD10:
- D05 - Carcinoma in situ of breast
- Healthy volunteers:
- Yes
Interventions, Observational Groups
- Arm 1:
- Live AI: Studies read in a double-reader setting, where one or both involved readers use Vara software for workflow and AI predictions.
- Arm 2:
- Shadow mode: Studies read as usual in a double-reader setting, without support from the Vara user interface or AI. Studies will also have AI predictions made in the background, but these will not be reported to the user so as to not influence screening practices.
Endpoints
- Primary outcome:
- I. Screen-detected cancer rate Defined as the number of true-positive examinations divided by the total number of screening mammograms, per 1000 women screened. True-positive mammogram is a positive mammogram (BI-RADS assessment ≥3) followed by the biopsy-confirmed diagnosis of breast cancer within the timeframe as defined by the mammography screening program. II. Recall rate Defined as the number of surveillance imaging examinations with BI-RADS assessment ≥3 at consensus conference and recalled for further imaging, per 1000 women screened.
- Secondary outcome:
- 1. Number of studies sent to consensus conference per 1000 women screened 2. Biopsy recommendation rate: number of diagnostic imaging studies with BI-RADS assessment 4 or 5 per 1000 women screened 3. Biopsy rate: number of women with diagnostic biopsies per 1000 women screened 4. False positive rate: a. Number of negative diagnostic imaging studies, per 1000 women screened b. Number of negative diagnostic biopsies, per 1000 women screened 5. AI metrics: a. Number of times safety net is triggered, per 1000 women b. Number of times safety net is triggered and concordant with user prediction, per 1000 women c. Number of times safety net is triggered, shown and waived, per 1000 women d. Number of times safety net is triggered, shown and accepted, per 1000 women e. Number of times safety net is triggered, shown and accepted, resulting in malignant biopsy, per 1000 women f. Number of triaged negative studies, per 1000 women 6. Interval cancer metrics (where possible according Exploratory objective). a. Number of interval cancer diagnoses within 24 months after a normal screening mammogram, per 1000 women b. Number of times safety net is triggered and waived, and resulted in an interval cancer, per 1000 women c. Number of times study was marked as "normal" by AI and resulted in interval cancer diagnosis, per 1000 women d. AI model scores for subsequent interval cancer diagnoses 7. Screening sensitivity (where possible according to Exploratory objective). Defined as the number of screen-detected cancers, over the total number of cancers detected during 24-month period 8. Screening specificity (where possible according to Exploratory objective). Defined as the number of true negatives, over the total number of negatives. 9. Reader sensitivity. Defined as the number of cases a reader recommended to consensus conference during the first and second read, over the total number of screen-detected cancers. 10. Reader specificity. Defined as the number of cases a reader did not recommend to consensus conference, over the total number of normal cases in the screening round. 11. Reader cancer detection rate. Defined as the number of true-positive examinations found by an individual reader divided by the total number of screening mammograms read by an individual reader, per 1000 women screened.
Study Design
- Purpose:
- Screening
- Retrospective/prospective:
- Prospective
- Study type:
- Non-interventional
- Longitudinal/cross-sectional:
- Longitudinal study
- Study type non-interventional:
- Epidemiological study, Patient Registry
Recruitment
- Recruitment Status:
- Recruiting ongoing
- Reason if recruiting stopped or withdrawn:
- No Entry
Recruitment Locations
- Recruitment countries:
-
- Germany
- Number of study centers:
- Multicenter study
- Recruitment location(s):
-
- University medical center Universitätsklinikum Schleswig-Holstein, Campus Lübeck Lübeck
Recruitment period and number of participants
- Planned study start date:
- 2022-03-01
- Actual study start date:
- 2021-07-01
- Planned study completion date:
- 2024-06-30
- Actual Study Completion Date:
- No Entry
- Target Sample Size:
- 400000
- Final Sample Size:
- No Entry
Inclusion Criteria
- Sex:
- Female
- Minimum Age:
- 50 Years
- Maximum Age:
- 69 Years
- Additional Inclusion Criteria:
- Attending biennial breast cancer screening at a screening unit participating in the German national breast cancer screening program.
Exclusion Criteria
None
Addresses
Primary Sponsor
- Address:
- Vara (MX Healthcare GmbH)Max-Urich-Strasse 313355 BerlinGermany
- Telephone:
- No Entry
- Fax:
- No Entry
- Contact per E-Mail:
- Contact per E-Mail
- URL:
- No Entry
- Investigator Sponsored/Initiated Trial (IST/IIT):
- No
Contact for Scientific Queries
- Address:
- Direktor, Institut für Sozialmedizin und Epidemiologie, Universitätsklinikum Schleswig-HolsteinProf. Dr. med. Alexander KatalinicCampus Lübeck, Ratzeburger Allee 16023538 LübeckGermany
- Telephone:
- +49 451 5005 1200
- Fax:
- No Entry
- Contact per E-Mail:
- Contact per E-Mail
- URL:
- No Entry
Contact for Public Queries
- Address:
- Country Manager Germany, MX Healthcare GmbH (Vara)Fridtjof StordeMax-Urich-Strasse 313355 BerlinGermany
- Telephone:
- +49 151 2084 2538
- Fax:
- No Entry
- Contact per E-Mail:
- Contact per E-Mail
- URL:
- No Entry
Principal Investigator
- Address:
- Direktor, Institut für Sozialmedizin und Epidemiologie, Universitätsklinikum Schleswig-HolsteinProf. Dr. med. Alexander KatalinicCampus Lübeck, Ratzeburger Allee 16023538 LübeckGermany
- Telephone:
- +49 451 5005 1200
- Fax:
- No Entry
- Contact per E-Mail:
- Contact per E-Mail
- URL:
- No Entry
Sources of Monetary or Material Support
Commercial (pharmaceutical industry, medical engineering industry, etc.)
- Address:
- MX Healthcare GmbH (Vara)Max-Urich-Strasse 313355 BerlinGermany
- Telephone:
- No Entry
- Fax:
- No Entry
- Contact per E-Mail:
- Contact per E-Mail
- URL:
- No Entry
Ethics Committee
Address Ethics Committee
- Address:
- Ethik-Kommission der Med. Fakultät der Universität zu LübeckRatzeburger Allee 16023538 LübeckGermany
- Telephone:
- +49-451-5004639
- Fax:
- +49-451-5003026
- Contact per E-Mail:
- Contact per E-Mail
- URL:
- No Entry
Vote of leading Ethics Committee
- Vote of leading Ethics Committee
- Date of ethics committee application:
- 2022-02-10
- Ethics committee number:
- 22-043
- Vote of the Ethics Committee:
- Approved
- Date of the vote:
- 2022-03-03
Further identification numbers
- Other primary registry ID:
- No Entry
- EudraCT Number:
- No Entry
IPD - Individual Participant Data
- Do you plan to make participant-related data (IPD) available to other researchers in an anonymized form?:
- Yes
- IPD Sharing Plan:
- Anonymized participant-related data will be shared to anyone who wishes access to the data immediately after publication. There will be one row for each screening case. Results of the manuscript will be reproducible. Plan is to make data available via https://datadryad.org/.
Study protocol and other study documents
- Study protocols:
- No Entry
- Study abstract:
- No Entry
- Other study documents:
- No Entry
- Background literature:
- No Entry
- Related DRKS studies:
- No Entry
Publication of study results
- Planned publication:
- No Entry
- Publikationen/Studienergebnisse:
- No Entry
- Date of first publication of study results:
- No Entry
- DRKS entry published for the first time with results:
- No Entry
Basic reporting
- Basic Reporting / Results tables:
- No Entry
- Brief summary of results:
- No Entry