AI in Cancer Care Stephen Wong, PhD, PE John S Dunn Presidential - - PowerPoint PPT Presentation

AI in Cancer Care

Stephen Wong, PhD, PE

John S Dunn Presidential Distinguished Chair & Chief Research Information Officer, Houston Methodist Hospital Associate Director, Houston Methodist Cancer Center Professor, Radiology, Neuroscience, Pathology & Lab Medicine, Cornell University

Disruption i in H Healthcare

High Costs and Low Quality

Sources: IOM Report:Best Care at Lower Cost, 2010 JAMA Network, Oct 2019 OECD Health Statistics. Centers for Medicare and Medicaid Services, USA

Accelerate the pace of healthcare transformation

Macro-economics: Value-based Care, Drug Pricing, Interoperability, HRRP, CMS waiver, Payment Reform Socio-economics: Patients as Consumers, Chronic Care Management, Social Determinants of Health, Population Health, Patient- Physician Engagement Waste costs the U.S. healthcare system $760B to $935B annually

Estimated ranges of total annual cost of waste in the U.S. Health Care system:

• Administrative complexity: $265.6 billion
• Pricing failure: $230.7 billion - $240.5 billion
• Failure of care: $102.4 billion - $165.7 billion
• Overtreatment: $75.7 billion - $101.2 billion
• Fraud and abuse: $58.5 billion - $83.9 billion
• Failure of care coordination: $27.2 billion - $78.2 billion

Wastes in U.S. Healthcare System

Shrank WH, et al. JAMA Network, Oct 7 2019

• ~24 million mammograms done annually in the U.S.
• ~266,600 new cases of invasive breast cancer diagnosed in U.S.

annually in the U.S.

• False positive rate from mammogram is estimated to be

7% 7%~1 ~10% 0%.1

• 55%–85% of breast biopsies showed benign lesions for BI-RADS 4

(2-95% 95% chance of malignancy). ).

• False positive mammograms is estimated to cost ~$4 b

billion an annual nually.2

• Patients who have received false-positive diagnoses have shown

to exhibit a higher level of anxiety and lower level of self-esteem.

Case Study I: Breast Cancer Risk Assessment

• 1. https://www.cancer.gov/types/breast/hp/breast-screening-pdq
• 2. https://www.healthaffairs.org/doi/full/10.1377/hlthaff.2014.1087

Intelligence-augmented Breast Cancer Risk Calculator (iBRISK)

• AI augmented breast cancer

risk calculator on the cloud.

• Remote upload on images (or

image features) and clinical- demographic parameters.

• Show clinicians the list of all

the inputs as a means of garnering trust in the AI black box

Breast Image Features Clinical Signatures & Patient History/Demographics

Online iBRISK Calculator

Mammography Ultrasound Free Text Reports Imagomics Database Deep Learning Risk Assessment Model Natural Language Processing (NLP) Image Analysis

Patel TJ, et. al, Cancer 2017 Jan 1;123(1):114-121 He T, et. al, JCO Clinical Cancer Inform. 2019 May;3:1-12

iBRISK Technical Architecture

PACS EMR Convergent AI includes natural language processing, image analysis, deep learning, and data mining on big multi-modal BI-RADS patient data.

Current Dataset

Retrospective Study Results

No. Subjects Pathology Findings Category I (Benign) Category II (Malignant)

Benign Atypia LCIS DCIS Carcinoma

11,428 7,967 518 69 1,061 1,813

>14,000 BI-RADS 4 patient case data Validation datasets 2,285 cases 764 cases

Pa Pathologica cal Fi Finding Institution Houston Methodist MD Anderson U Texas San Antonio Total Be Benig ign 1710 (74.84%) 1019 (71.56%) 240 (67.42%) 2969 Ma Mali ligna nant nt 575 (25.16%) 405 (28.44%) 116 (32.58%) 1096 To Total 2285 2285 14 1424 356 356 4065 4065

Number o

• f c

f cases i in t three d different t types o

• f m

f medical i institutions

Houston M Methodist C Cancer C Center Missing Value Rate: 13.3 .30% Actual malignant cases: 575 Predicted malignant cases: 575 Sensitivity: 100% Actual benign cases: 1710 Predicted benign cases: 1406 Specificity: 71% MD A Anderson C Cancer C Center Missing Value Rate: 17.7 .78% Actual malignant cases: 405 Predicted malignant cases: 405 Sensitivity: 100% Actual benign cases: 1019 Predicted benign cases: 695 Specificity: 68% U T Texas S San A Antonio C Cancer C Center Missing Value Rate: 16.0 .03% Actual malignant cases: 116 Predicted malignant cases: 115 Sensitivity: 99% Actual benign cases: 240 Predicted benign cases: 184 Specificity: 77%

Multi-center Study Results

Unpublished data In collaboration with MD Anderson Cancer Center and U Texas San Antonio Cancer Center

Case Study 2: Failure of Cancer Care Coordination

connect.curaspan.com

Silver Tsunami of Cancer Survivors

Bluethmann SM, et. al, Cancer Epidemiology, Biomarkers & Prevention, 2016 Jul;25(7):1029-36.

Estimated cancer prevalence by age in the US population from 1975 (216 M) to 2040 (380 M)

Designed together with HM cancer center with behavior economics and clinical workflow to improve cancer patient survival and provide digital therapeutics

Digital Therapeutics and Cancer Care Coordination App

MO MOCHA A (Methodist Hospital Cancer Health App)

• R. Stubbins, et. al, JCO Clinical Cancer Inform. 2018 Dec;2:1-11.

• 9
• 7
• 5
• 3
• 1

1 3 5 5 10 15

r=- 0.399

Post-study survey questions

Summary of MOCHA use, weight loss, and SUS score Correlation of weight change & average daily use of MOCHA

Clinical trial on ~40 Breast cancer survivors with a body mass index (BMI) over 25 At least 6 months post active treatment (surgery, chemotherapy, or radiation).

Post Hospitalization Cancer Care Coordination using MOCHA

• T. He, R. Stubbins, et al. ACO Clinical Cancer Informatics, Dec, 2018

Personalized Music Therapy + Digital Therapeutics

Hypothesis: : Use of personalized music therapy treatment supplemented with digital therapeutics to improve mental health symptoms and behaviors in cancer patients and survivors.

Recruit cancer patients and survivors

• MT-BC identify depression/

stress behaviors & symptoms