PREDICTIVE ANALYTICS: PROVIDING NOVEL APPROACHES TO ENHANCE OUTCOMES RESEARCH LEVERAGING BIG AND COMPLEX DATA IMS Symposium at ISPOR at Montreal June 2 nd, 2014 Agenda Topic Presenter Time Introduction: Extending the statistical toolkit to make the most of big data Academic perspective: Acquiring insight from health data using machine learning: methodological overview and case studies Michael Nelson PharmMD (chair) 12:15 12:20 12:20 12:40 Using predictive analytics to drive improvements in health and costs: the case of NorthShore University Jonathan Silverstein MD, MS, FACS, FACMI NorthShore University 12:40 12:55 Predicting the key areas for predictive analytics in real-world evidence 12:55 13:05 Q & A 13:05 13:15
Background and Objectives Background: The volume and complexity of health data is growing rapidly due to the digitization of patient, hospital, prescription, biological, and other vast data streams This presents exciting opportunities to help drive improvements in health care But making the most of this data requires embracing analytical ideas going beyond traditional statistical methods This includes innovative techniques from the fields of predictive analytics - machine learning algorithms and data mining Symposium objectives: To review state-of-the-art methodologies in predictive analytics and machine learning To describe case studies of successful implementations of predictive analytics solutions To identify the areas where predictive analytics may offer most value in realworld evidence Agenda Topic Presenter Time Introduction: Extending the statistical toolkit to make the most of big data Michael Nelson PharmMD (chair) 12:15 12:20 Academic perspective: Acquiring insight from health data using machine learning: methodological overview and case studies Using predictive analytics to drive improvements in health and costs: the case of NorthShore University Jonathan Silverstein MD, MS, FACS, FACMI NorthShore University 12:20 12:40 12:40 12:55 Predicting the key areas for predictive analytics in real-world evidence 12:55 13:05 Q & A 13:05 13:15
Jonathan Silverstein MD, MS, FACS, FACMI NorthShore University Agenda Topic Presenter Time Introduction: Extending the statistical toolkit to make the most of big data Michael Nelson PharmMD (chair) 12:15 12:20 Academic perspective: Acquiring insight from health data using machine learning: methodological overview and case studies Using predictive analytics to drive improvements in health and costs: the case of Northshore University Jonathan Silverstein MD, MS, FACS, FACMI NorthShore University 12:20 12:40 12:40 12:55 Predicting the key areas for predictive analytics in real-world evidence 12:55 13:05 Q & A 13:05 13:15
Key messages Predictive analytics can: Play a central role in the evolution of real-world evidence (RWE) Produce robust RWE in highly complex data settings Apply RWE through technology based on predictive algorithms Predictive analytics must first overcome challenges, including: Reluctance to embrace innovation by key stakeholders The need for greater collaboration between stakeholders Predictive analytics versus traditional statistical methods Study objective Traditional statistical methods Inference / hypothesistesting / description Predictive Analytics Data exploration / prediction / generalisation Scientific philosophy Deductive - variables & model determined a priori Inductive variables & model determined empirically Model representation Model evaluation Includes logistic regression, OLS (ordinary least squares) Based on same sample as estimation Machine learning algorithms e.g. support vector machines, random forests Based on different sample to estimation i.e. hold-out sample
Producing RWE insight through predictive analytics Non-conventional data settings: Predictive analytics provides insight in areas where traditional methods are not well-suited e.g. High dimensional data e.g. genomics, drug discovery, image processing Unstructured data e.g. medical notes, social media Conventional data settings e.g. claims, EMR, prescribing: Standard methods good for regular inference / hypothesistesting Predictive analytics good where results to be used for generalization or prediction Predictive analytics good at determining important associations and avoiding overfitting when: Determining nature of non-linearities and interactions Controlling for complex confounding Ascertaining important variables from large pool of variables Predictive analytics provides value in many RWE areas 1) Patient profiling e.g. Identifying patients who respond most positively to treatment Identifying which treatment is most appropriate for a specific patient profile 2) Detecting undiagnosed disease e.g. Identifying patients likely to be undiagnosed for a given disease Estimating the prevalence of nondiagnosis for a given disease Early disease detection 3) Risk stratification, including identifying patients most likely to: Non-adhere to treatment Experience disease onset or progression 4) Propensity scoring Machine learning good in presence of complex confounding bias 5) Making better use of data assets through imputation e.g. Instead of only using intersection between large claims data and small clinical data, use all patients with imputations for selected variables
Applying RWE through technology Prospering in a world of value-based healthcare: A recent white paper published by the Economist Intelligence Unit identified 7 things pharma companies need to do to prosper in a climate of value-based healthcare, including: Improve the value proposition through more accurate targeting. This may include use of screening or companion diagnostics to identify those patients who would benefit the most. Get involved in partnerships with providers, in order to improve real world patient outcomes. This may include help with ironing out variations in care, and improving adherence. Both areas involve predictive analytics Examples of technology embedding predictive algorithms Example applications Clinical pathways o Recommendation of treatments based on patient profiles Hospital readmissions o Identification of patients most likely to be readmitted to hospital Computer-Aided Diagnostics o Indication of diagnostic classification based on MRI scan Adoption of predictive analytics technology Still embryonic Must overcome substantial challenges before mainstream integration
Two big challenges to adoption of predictive analytics 1 Reluctance to embrace innovation by key stakeholders Regulators are cautious to embrace novel statistical methods e.g. data mining often suspected of suffering from spurious correlation Physicians are often wary about new technology May be seen as undermining clinical independence and judgment Use cases vital in promoting awareness of value of predictive analytics 2 Need for greater collaboration Technology is the vehicle by which predictive analytics can often achieve changes in health care Large-scale analytics technology often held back by: The need for substantial, on-going investment Failure to align objectives between providers, payers and pharmas Lack of data availability, integration and harmonisation Partnerships between providers, payers and life science can help through: Aligning objectives / overcoming market failures Facilitating return on investment to all stakeholders Ensuring solution is product neutral Agenda Topic Presenter Time Introduction: Extending the statistical toolkit to make the most of big data Michael Nelson PharmMD (chair) 12:15 12:20 Academic perspective: Acquiring insight from health data using machine learning: methodological overview and case studies Using predictive analytics to drive improvements in health and costs: the case of NorthShore University Jonathan Silverstein MD, MS, FACS, FACMI NorthShore University 12:20 12:40 12:40 12:55 Predicting the key areas for predictive analytics in real-world evidence 12:55 13:05 Q & A 13:05 13:15
Q & A Availability of symposium materials If you are interested in receiving an abridged copy of the symposium materials: Please leave your business cards when leaving the room or at our exhibition stand Email Angelika Boucsein at: aboucsein@de.imshealth.com or
Affiliations of Speakers Michael Nelson PharmMD (chair) Senior Principal and Regional Leader, Americas, Departments of Biostatistics & Medical Informatics and Computer Science, Jonathan Silverstein MD, MS, FACS, FACMI Vice President and Davis Family Chair of Informatics, Head, Center for Biomedical Research Informatics (CBRI) NorthShore University ; Research Associate (Professor of Surgery), University of Chicago Pritzker School of Medicine Senior Manager, Advanced Analytics,