People, ideas, machines. @enricocoiera AUSTRALIAN INSTITUTE OF - - PowerPoint PPT Presentation

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

People, ideas, machines.

@enricocoiera

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

“People, ideas, things. And in that order”

John Boyd (1927-1997)

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

2014 (1) 2017 (6) 2018 (23) 2016 (4)

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

Healthcare AI has quickly become a high stakes game. The promise of great reward is encouraging the taking of great risk.

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

We don’t need to wait for technical breakthroughs to make a difference to healthcare today

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

Dozens of RCTs now show clinical benefits

Screening

• Detect diabetic retinopathy NPJ Digit. Med. (2018). 90.7% specificity. Led to

FDA approval.

• Screen head CT images for acute neurologic events Nat. Med.(2018). Time to

diagnosis reduced from minutes to seconds. Diagnosis

• Diagnose fracture in wrist radiographs. Natl. Acad. Sci. USA (2018). Av.

clinician relative reduction in misinterpretation rate of 47.0%.

• Diagnose major depressive disorders. Scientific reports (2017) Sensitivity

increased across all severity levels and reached 73% for severe depression Treatment

• Monitor medication adherence in opioid replacement therapy (AiCure)
• Deliver cognitive behaviour therapy to young adults (Woebot) JMIR Ment.

Health (2017) significantly reduced their symptoms of depression compared to self-help online.

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

We are not designing algorithms. We are designing human-machine systems.

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

AI and Thyroid Cancer Diagnosis

• The DCNN showed improved performance in identifying thyroid

cancer patients versus skilled radiologists:

– specificity was 86·1% (81·1–90·2) versus 59·4% (53·0–65·6; p<0·0001). – sensitivity 93·4% (95% CI 89·6–96·1) versus 96·9% (93·9–98·6; p=0·003)

• But thyroid cancer is over-diagnosed and over-treated (15-fold

increase in thyroid cancer in South Korea largely due to

• verdiagnosis).
• Many lesions will never lead to harm.
• So might improved diagnostic performance by this AI cause more

unnecessary treatment, patient risks, and economic cost?

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

The last mile

First mile

Data capture Cleaning Labeling

Middle mile

Model building and testing e.g. ML/DL

Last mile

Real world implementation Routine use

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

AI enables new “Digital Only” services

• A ‘digital only’ primary care provider might:
• Have as its front door an app which:

– triages your problem using an AI driven symptom checker – Makes a booking using a chatbot – Organises repeat scripts via an online pharmacy

• Provides online tele-consultations for problems needing human

interaction

• Has a bricks and mortar partner where it can refer you when

physical presence is required

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

BGPaH patient journey (from 2019 Independent Evaluation by Ipsos MORI)

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

Australian Institute of Health Innovation, Macquarie University

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

Australian Institute of Health Innovation, Macquarie University

Typically BGPaH patients are healthier than those at other practices even after adjusting for age.

(from 2019 Independent Evaluation by Ipsos MORI)

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

Australian Institute of Health Innovation, Macquarie University

BGPaH patients are historically higher users of NHS 111 and A&E than might be expected.

(from 2019 Independent Evaluation by Ipsos MORI)

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

The inverse care law

"The availability of good medical care tends to vary inversely with the need for it in the population served. This ... operates more completely where medical care is most exposed to market forces, and less so where such exposure is reduced.”

(Julian Tudor Hart, 1971)

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

The Australian Alliance for Artificial Intelligence in Healthcare (AAAiH)

Australian Institute of Health Innovation, Macquarie University

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

A national partnership to transform the health system

Established

Commenced late 2017

Partners

61 formal partners to date

Government

5 State Departments of Health

Industry

Google Health, Accenture, Amazon AWS, IBM Research, Oracle, Orion, Alcidion

National

CSIRO, ADHA, Healthdirect Australia, ACSQHC, AIHW, Standards Australia

State

CEC, ACI, SA-NT Datalink

Academia

Universities: Macquarie, Melbourne, Queensland, Bond, UNSW, Sydney, South Australia, Adelaide, Flinders. MCRI, Australian Genomics, AIML

Professional

CHF, RANZCR, HISA, ACHI, MSIA

International

Stanford, Oxford, Toronto, Michigan, Liverpool

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

AI Alliance Goals

• Support and accelerate the adoption of

AI enabled health services in Australia

• Build a community of shared interest

and pooled resource

• Undertake essential research needed

to demonstrate AI-enabled services are scalable, sustainable, and have positive impacts on clinical and economic outcomes :

• Precision Healthcare Flagship
• Consumer Health Flagship
• Rapidly build the national workforce

needed for success

• To ensure AI is safe and ethical,

develop needed national regulatory and governance frameworks

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

“People, ideas, machines. And in that order”

AUSTRALIAN INSTITUTE OF HEALTH INNOVATION

Thank you @enricocoiera enrico.coiera@mq.edu.au