Data Science Meets Healthcare: The Advent of Personalized Medicine Jacomo Corbo Canada Research Chair in Information Management, University of Ottawa Research Affiliate, The Wharton School of Business, University of Pennsylvania Chief Scientist, QuantumBlack April 17, 2013
26
Embed
Data Science Meets Healthcare: The Advent of Personalized Medicine - Jacomo Corbo
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Data Science Meets Healthcare: The Advent of Personalized Medicine Jacomo Corbo Canada Research Chair in Information Management, University of Ottawa Research Affiliate, The Wharton School of Business, University of Pennsylvania Chief Scientist, QuantumBlack
April 17, 2013
Healthcare spending growth is unsustainable
2
HEALTH EXPENDITURE CONTINUES TO RISE (SOURCE: National Health Expenditure Database, CIHI)
0
50
100
150
200
250
1975 1980 1985 1990 1995 2000 2005 2010
Bill
ions
of D
olla
rs
Actual Spending Inflation-Adjusted Spending ($1997) Forecast
– Major complications of Tx: Impotence [40/1,000], MI [2/1,000] DVT [1/1,000]
• How effective is screening in reducing prostate cancer deaths? To prevent one death over a 10-year period:
– Number needed to screen: 1,410
– Number needed to treat: 48
• USPSTF: Recommends against screening: "moderate or high certainty that the service has no net benefit or that the harms outweigh the benefits,”
• AUA: Favors Screening: “The American Urological Association (AUA) is outraged at the USPSTF’s failure to amend its recommendations on prostate cancer testing to more adequately reflect the benefits of the prostate-specific antigen (PSA) test in the diagnosis of prostate cancer.”
NO FREE LUNCH FOR SCREENING: Example 2: Breast Cancer & Mammography
Bleyer & Welsch: “We estimated that breast cancer was overdiagnosed (i.e., tumors were detected on screening that would never have led to clinical symptoms) in 1.3 million U.S. women in the past 30 years. We estimated that in 2008, breast cancer was overdiagnosed in more than 70,000 women; this accounted for 31% of all breast cancers diagnosed.” [NEJM, Nov 2012]
USPSTF: “recommends biennial screening mammography for women aged 50 to 74 years. The decision to start regular, biennial screening mammography before the age of 50 years should be an individual one and take patient context into account, including the patient's values regarding specific benefits and harms.”
ACOG: “Due to the high incidence of breast cancer in the US and the potential to reduce deaths from it when caught early, The American College of Obstetricians and Gynecologists (The College) today issued new breast cancer screening guidelines that recommend mammography screening be offered annually to women beginning at age 40.”
17
DEVELOPING A DATA-DRIVEN SCREENING POLICY w/ N Marko (MD Anderson Clinic), P Ardestani (U of Ottawa), O Koppius (Rotterdam School of Management) Hypertension Onset from the Framingham Heart Study Dataset:
– A machine learning (ML) model with only 6 covariates yields an average error of 2.7 years for the onset of hypertension
– Yields a simple screening that ‘catches’ hypertension in 98.9% of the overall population, 100% of most ‘at risk’ patients, and saves ~$275M USD annually (against the CTFPHC & USPSTF’s prescriptions)
Stroke Prediction from the Cardiovascular Health Dataset:
– ML model with 11 covariates predicts strokes with an average error of 2.3 years
– Yields a 16% error reduction over best structural models
– ML model includes features heretofore unrecognized as risk factors in literature (e.g. total medications)
Case 2: More e!cient hospitals
19
ORGANIZATIONAL EFFECTS AND LEARNING RATES ON OR UNIT PERFORMANCE
• Establish that individual, team, organizational experience matters
• Establish evidence for organizational learning-curve heterogeneity
• Moore and Lapré (2012) establish that 1) individual, team, and organizational experience, (2) learning-curve heterogeneity (actors learning at different rates), and (3) workload all simultaneously matter
20
SURGICAL TEAM PERFORMANCE w/ S Toms (Geisinger Health System)
• Data: 381K surgeries at 16 hospitals over 5 years.
• Analyze data about surgical team members, how and with whom they work, to forecast team productivity and patient outcomes, optimize team assignment.
Highlights:
– Dispute conventional wisdom: Inconclusive support for the importance of individual experience; the only team experience measure that is significant is tightly-coupled team experience
– Discover what matters: Most significant variable is dyadic team experience between chief surgeon and head nurse in knee replacement procedures; triadic experience between chief surgeon, head nurse, and anesthesiologist for hip replacements
– Make better predictions: We can also predict ~93% of surgeries to within 15 minutes
And Big Science applications
Addressing the growing chasm between the art of the possible and reality