fungalAi for Fungal Surveillance & Antifungal Stewardship (fungalAi)

Innovative Use of fungalAi for Antifungal Stewardship in Haematology-oncology Patients

This national Australian study will validate and implement an effective approach to real-time electronic surveillance of fungal infections in patients with blood cancers using technology based on artificial intelligence. It will establish metrics for antifungal stewardship allowing benchmarking of these programs; provide decision support for radiologist interpretation of chest imaging and improve reporting, audit and feedback practices in hospitals where these infections are managed.

Study Overview

• Status: Unknown
• Conditions: Fungal Infections
• Intervention / Treatment: Combination product: fungalAi platform technology

Detailed Description

Invasive fungal diseases (IFD) are rare infections that cause a life-threatening pneumonia in patients with weakened immune systems usually due to cancer chemotherapy and transplantation. Fungal spores are found in air, water and soil making exposure unavoidable in vulnerable patients. In developed countries, molds like Aspergillus are the most challenging type of IFD to diagnose and treat. These infections usually manifest as a culture-negative fungal pneumonia and account for approximately 300K of the 1.9M cases of IFD globally, but estimates are not accurate due to an absence of surveillance systems in hospitals where these infections are managed. Hospitals spend millions on antifungal drugs but are unaware of their patients affected, the effectiveness of their prevention efforts and hospital outbreaks may go unnoticed because surveillance, audit and feedback of fungal infections is not occurring.

Optimising patient outcomes through timely diagnosis and appropriate prescribing of antifungal drugs is the goal of antifungal stewardship programs. Antifungal stewardship is of growing importance to hospitals world-wide because antifungal drugs are few in number, expensive to use and are associated with significant side-effects and drug interactions. Surveillance, audit and feedback are the cornerstones of antifungal stewardship programs that ensure patient care is meeting high standards. However, currently hospitals do not have the mechanisms to detect rare events like fungal infections because it usually presents as a pneumonia buried among hundreds of imaging scans.

"fungalAi™" (fungalAi.com) is a technology based on artificial intelligence (Ai) that uses existing data in hospitals to make real time surveillance of fungal infections possible and assist radiologist interpretation of diagnostic imaging. fungalAi does this through:

1. Natural language processing, a computational method of understanding human language.
2. Deep learning based image analysis of diagnostic imaging and
3. An expert system that integrates clinical data.

What will be the impact?

This project will provide hospitals with the mechanisms for performing real-time surveillance and audit of fungal infections in blood cancer patients through the innovative use of Ai. Strengthening antifungal stewardship through real-time surveillance of fungal diseases will improve patient care by revealing gaps in practice, new patient groups at risk for fungal infections and reduce inappropriate prescribing of antifungal medications through timely audit and feedback. The impact of this project will be:

1. Improved diagnosis and recognition of fungal infections.
2. Enhanced prevention.
3. More appropriate use of antifungal medications.

FungalAi is a scalable technology that will be validated against active manual surveillance of fungal infections in a multi-centre Australian clinical trial. The inclusive approach of fungalAi means that it is of value to many vulnerable patients including neglected groups like children who are included in this project. FungalAi is tuned for detection of fungal pneumonia caused by molds because these infections are more diagnostically challenging than other types of fungal infections. As a result, fungalAi leverages chest computed tomography imaging because it is a critical diagnostic test that is widely available and performed more frequently than invasive tests like lung washings or biopsy. Hence fungalAi natural language processing may miss very rare manifestations like brain infections. Nevertheless, automating detection of fungal pneumonia and improving radiologist recognition of a rare disease using a self-improving system based on neural networks is an important step towards improving the supportive care of patients with cancer. Improving outcomes in cancer is not only about finding a cure. Reducing the impact of infectious threats like fungal diseases is just as important and this can now be achieved by integrating artificial intelligence into patient care.

• Study Type: Observational
• Enrollment (Anticipated): 1000

Contacts and Locations

Study Locations

• Australia
  • Victoria
    • Melbourne, Victoria, Australia
      • Alfred Health

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: ADULT; OLDER_ADULT; CHILD
• Accepts Healthy Volunteers: Yes
• Genders Eligible for Study: All

• Sampling Method: Non-Probability Sample

Study Population

Adults and children with blood cancers under the haematology service at participating sites inclusive of inpatient and ambulatory care patients.

Description

Inclusion Criteria:

• Adults and children
• Under the haematology service at participating sites
• Inpatient and ambulatory patients.

Exclusion Criteria: No exclusion criteria

Study Plan

How is the study designed?

Number of groups / cohorts

2

Cohorts and Interventions

Group / Cohort	Intervention / Treatment
Fungal Cases; Patients with confirmed invasive fungal infections according to internationally accepted criteria identified by active manual surveillance. Clinical data will be sent to fungalAi platform technology for disease classification.	Combination product: fungalAi platform technology; Electronic surveillance and radiologic diagnosis of invasive fungal infections using fungalAi and associated methodologies.
Control patients; Patients without invasive fungal infections.Clinical data will be sent to fungalAi platform technology for disease classification.	Combination product: fungalAi platform technology; Electronic surveillance and radiologic diagnosis of invasive fungal infections using fungalAi and associated methodologies.

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Accuracy of electronic surveillance using fungalAi natural language processing compared to active manual methods for detection of fungal pneumonia	Sensitivity, specificity, ROC, Area under precision-recall curve of Ai assisted surveillance for fungal pneumonia using natural language processing of imaging reports compared to active manual surveillance	12 months

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Accuracy of disease classification of deep learning based image analysis for fungal pneumonia at scan level.	Sensitivity, specificity, ROC of deep learning based image analysis at scan level compared to active manual surveillance.	12 months
Accuracy of feature detection of fungal pneumonia using deep learning based image analysis of chest CT compared to radiologist expertise.	Sensitivity, error rate (false positives, false negatives) at pixel level of deep learning based image analysis compared to radiologist labels.	12 months
Accuracy of disease classification of an expert system integrating microbiology and antifungal drug prescriptions with text and image analysis compared to active manual surveillance.	Sensitivity, specificity, ROC, Area under precision-recall curve of Ai assisted surveillance compared to active manual surveillance that will only be performed at Alfred Health.	12 months

Collaborators and Investigators

• Sponsor: Bayside Health
• Collaborators: Monash University; Western Sydney Local Health District; Eastern Health; Fremantle Hospital and Health Service; SA Health; Monash Health; Singhealth Foundation
• Investigators: Principal Investigator: Michelle Dr Ananda-Rajah, The Alfred

Publications and helpful links

Helpful Links

• fungalAi.com

Study record dates

Study Major Dates

• Study Start (ACTUAL): January 1, 2019
• Primary Completion (ANTICIPATED): December 30, 2020
• Study Completion (ANTICIPATED): December 30, 2020

Study Registration Dates

• First Submitted: December 15, 2018
• First Submitted That Met QC Criteria: January 2, 2019
• First Posted (ACTUAL): January 4, 2019

Study Record Updates

• Last Update Posted (ACTUAL): October 23, 2020
• Last Update Submitted That Met QC Criteria: October 22, 2020
• Last Verified: August 1, 2020

More Information

Terms related to this study

• Keywords: Deep learning; Fungal infections; Electronic surveillance; Antifungal stewardship; Image classification
• Additional Relevant MeSH Terms: Infections; Bacterial Infections and Mycoses; Mycoses
• Other Study ID Numbers: 43127/MonH-2018-152967; 012018 (The Alfred Foundation); 2015-54 (OTHER_GRANT: Monash Institute of Medical Engineering)

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: YES
• IPD Plan Description: De-identified individual participant data will be made available in aggregate form for reports, presentations and publications.
• IPD Sharing Time Frame: Data will be made available within 6-12 months after study completion.
• IPD Sharing Access Criteria: Access to study protocol, SAP, CSR will be made publicly available. Access to IPD including individual labelled data will be reviewed by an external independent review panel to ensure that all ethical issues have been met.
• IPD Sharing Supporting Information Type: STUDY_PROTOCOL; SAP; CSR

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No