Big Data in Healthcare: Current Possibilities and Emerging Opportunities Andrew Bate Senior Director, Epidemiology Group Lead, Analytics 23 th March 2015
The Long Road In Developing a New Medicine Clinical Data Analysis Registration Studies in 100-300 Patients (Phase II) Candidate Medicine Tested in 3-10,000 Patients (Phase III) Full Development Studies in Healthy Volunteers Phase I Exploratory Development Large Amounts of Candidate Medicine Synthesized Formulations Developed Extensive Safety Studies Candidate Project Team and Plans Synthesis of Compounds Screening Discovery Early Safety Studies 2
Sensors Facilitate R&D Via Remote Monitoring Miniaturized Clinical Chemistry Devices Hypertension Management Heart Rate Variability (HRV) & Electrodermal Activity (EDA) Activity Tracking ECG Rhythm Detection Non-Adherence Smart Pills, Containers Respiratory Disease 3
Real-World Data/Analytics in a Pharma Context Insights on diseases, products, and patient populations derived from the analysis of real world data beyond controlled trials Data Methodologies Insights Anything other than data from a randomized controlled trial that permits longitudinal observation Structured approach to data analysis and generation of meaningful impacts Innovative, value adding information about products, patients or competitive landscape Claims databases Registries/observational trials Prospective cohort Lab data Clinical records Genomic data Prospective Non-Interventional Research (NIR) Retrospective database analysis Econometric modelling Comparative effectiveness Safety Health economics Care pathways Competitor insights and many more 4
Hundreds of Sources of Different Real World Data: Some Examples THIN Database Country Characteristic Population Size UK GP primary care database 10.5 M 1 Danish National Health Service Register Database Premier Normative Health Information (NHI) Database Denmark Healthcare registry of care 5.5 M 2 US US Clinical data from the hospitals Transactional claims records of a commercial health insurer 130 M+ patient discharges 3 60 M+ 4 Health Insurance Review and Assessment Service (HIRA) Korea Insurance Claims from near universal national system 48 M 5 1 Blak et al Generalisability of The Health Improvement Network (THIN) database: demographics, chronic disease prevalence and mortality rates. Informatics in Primary Care 2011;19:251 5. 2 Furu K. et. al. The Nordic Countries as a Cohort for Pharmacoepidemiological Research. Basic & Clinical Pharmacology &Toxicology 2009; 106: 86-94. 3 Fisher BT et al. In-hospital databases In Pharmacoepidemiology 5 th Edn 2011 pp 244-258 4 Seeger J, Daniel GW. Commercial Insurance Databases. In Pharmacoepidemiology 5 th Edn 2011 pp 189-208 5 Kimura T et al. Pharmacovigilance systems and databases in Korea, Japan and Taiwan. Pharmacoepidemiology and Drug Safety. 2011; 20: 1237 1245 5
Real World Data Use is Standard for Regulated Safety Studies, Now Near-Real Time Surveillance Exploration Product Approval & Launch Hypothesis-free Lead Detection Lead Refinement Hypothesis Evaluation Rapid Detect the unexpected Less persuasive Time Consuming Test the anticipated Convincing 6
Example: Demonstrated Use of EMR Data for Early Identification of Drug Side Effects IC Δ * shows unexpected frequent recording of outcome after terbinafine prescription Angioedema was labelled In January 2004 However extensive testing of all such methods show imperfect performance * IC Δ is the difference in IC before and after prescription on a logarithmic scale 7
Pfizer partners for innovation to advance the Value of RWD for Safety Focus area Case example Organization Challenge Use of data Business impact Value of CDM Humana How to best structure data to analyse across networks of databases Evaluation of two most established Common Data models (CDMs) Access approach can effect results and interpretation, requires consideration in routine analyses Big data network execution capability Harvard- Pilgrim Can open ended surveillance be conducted across a network of databases? Analysis of pioglitazone and 3 other drugs simultaneously across 9 health plans Surveillance for unknown effects is tractable. Lack of access to underlying data can impede necessary follow up NLP for richer insights Humedica How to leverage rich data not collected systematically in EMR data Natural language processing can be used to detect patients with acute liver injury based on narrative data in their records Determined acute liver injury can be detected more effectively and earlier with Natural Language Processing (NLP) narrative access 8
Case Study: RWD Augmentation to Study Psoriasis and Chronic Kidney Disease Hypothesis for testing: Association between Psoriasis and Chronic Kidney Disease (CKD)? Psoriasis severity not recorded directly in UK EMR data Conducted primary data collection on embedded subset of psoriasis patients within the UK EMR THIN GP directly measured Psoriasis extent by body surface area Study showed increasing association of CKD with psoriasis severity Reference: Wan et al (2013) Risk of moderate to advanced kidney disease in patients with psoriasis: population based cohort study. BMJ 347 9
Pfizer Example - Neuroscience Gaming improves multitasking skills: study reveals plasticity in age-related cognitive decline Scientists use videogames to improve older brains September 2013 Akili announces partnership with Pfizer to test video game in people at risk of Alzheimer s disease January, 2014 10
Conclusions Big Data Strategies including RWD use are employed in Pharma R&D, including Safety With advances in technology, data availability and linkage ever greater innovation is possible Big Data will never be a panacea Analyzing more data does not necessarily lead to accurate conclusions Primary data collection such as Randomized Clinical Trials will remain essential Responsible iterative transparent learning systems are essential to link and augment secondary data use Show and communicate value on a population level, and benefits to individual patients 11