big data, smart data, fast data – a paradigm shift · • i am a member of the ibm watson health...

Big Data, Smart Data, Fast Data – a paradigm shift

Andrew Bate Senior Director, Epidemiology Group Lead, Analytics

Keynote ISOP annual conference, Agra 16 October 2016

Disclosures

•  I am a full time employee of Pfizer and hold stocks and stock options

•  I am a member of the IBM Watson Health Advisory Council – non exclusivity, unfunded

•  The company 23andme has partnered with Pfizer for some funded Big Data projects that are mentioned in this presentation

Overview

•  Traditional Pharmacovigilance –  And the wider usage of longitudinal observational

databases •  Changes in the type of data, access models and

analytic approaches •  Examples of the use of ‘Big Data’

Digital data in a rapidly revolving and evolving world

1000100010010010011110000111

Digital data streaming all around the world

Focus and hope of enhanced data analysis and insights than ever before

The Long Road In Developing a New Medicine

Discovery

Exploratory Development

Full Development

Registration

Large Amounts of Candidate Medicine Synthesized

Project Team and Plans Synthesis

of Compounds

Early Safety Studies

Candidate Formulations Developed

Extensive Safety Studies

Screening

Studies in Healthy Volunteers Phase I

Candidate Medicine Tested in 3-10,000 Patients (Phase III)

Studies in 100-300 Patients (Phase II)

Clinical Data Analysis

5

Standard use of spontaneous reports for signal detection – although incremental improvement

Cij i A1202 A0116 A0725 A0154 A0092 A0791 A0163 A0091 A0093 A0224 A0151 A0210 A0043 A0280 A0576 A0156 A0507j Ci/Cj 723 585 517 357 348 270 217 174 174 145 143 125 108 108 92 66 60A1202 723 - 109 171 23 29 126 17 20 11 39 24 27 22 8 43 6 3A0116 585 109 - 121 67 43 88 26 20 13 16 40 24 26 19 17 8 6A0725 517 171 121 - 33 38 109 18 25 14 47 30 43 35 9 38 3 4A0154 357 23 67 33 - 24 9 8 6 11 11 12 8 20 8 3 5 1A0092 348 29 43 38 24 - 25 39 7 9 14 10 8 11 5 10 16 2A0791 270 126 88 109 9 25 - 7 13 5 25 14 19 11 5 47 5 5A0163 217 17 26 18 8 39 7 - 6 6 5 10 5 5 4 2 3 2A0091 174 20 20 25 6 7 13 6 - 19 9 2 5 6 2 1 5 6A0093 174 11 13 14 11 9 5 6 19 - 3 8 8 1 3 1 1 7A0224 145 39 16 47 11 14 25 5 9 3 - 6 29 18 2 7 1 1A0151 143 24 40 30 12 10 14 10 2 8 6 - 1 5 4 3 2 5A0210 125 27 24 43 8 8 19 5 5 8 29 1 - 4 2 6 3 1A0043 108 22 26 35 20 11 11 5 6 1 18 5 4 - 5 3 2 1A0280 108 8 19 9 8 5 5 4 2 3 2 4 2 5 - 1 1 2A0576 92 43 17 38 3 10 47 2 1 1 7 3 6 3 1 - 2 1A0156 66 6 8 3 5 16 5 3 5 1 1 2 3 2 1 2 - 2A0507 60 3 6 4 1 2 5 2 6 7 1 5 1 1 2 1 2 -

A1202 = NMS A0116 = Hypertonia A0725 = Fever A0154 = Tremor A0092 = Confusion A0791 = CPK incr. Ref Orre et al 2005 International Journal of

Neural Systems

Outcome clustering using a recurrent Bayesian neural net

7

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

Evaluate Product Risks

Pfizer Confidential │ 8

Approval Active Surveillance Monitor and detect signals in defined patient cohorts using innovative analytic methods

Risk Minimization Evaluate the effectiveness of risk minimization measures (e.g., product label/education)

Standing Cohorts

Characterize Patient Risk Profile

EMRs

Claims

Registries

Harnessing the Power of Real World Evidence for Safety

Post Approval Safety Studies Compare medication risks in the real world, as prescribed and taken during routine clinical practice

Real World Data now has a role throughout the drug development lifecycle

│ 9

Early development

Full development Discovery Lifecycle

management

How many people suffer from the condi6on and also have co-‐morbidi6es x and y? What drugs are currently used in the treatment of the condi6on and to what extent are clinical guidelines being followed?

Given efficacy and tolerability results from the early trials, how might current treatment pathways be affected with our new drug? How costly are the specific areas of unmet need that a drug with this TPP might address?

Registration/ market access

In designing the PhIII trial, what are the underlying rates of adverse events we expect to see in the trial popula6on? Where can we modify the eligibility criteria in the PhIII protocol to reduce possible recruitment problems?

What is the likely budget impact of introducing the new drug across different pa6ent segments? What poten6al safety issues do we see with the early use of the drug in prac6ce?

How can we run a large clinical trial using EMRs to show the rela6ve effec6veness of our drug? In which pa6ent groups are there compliance issues with the drug?

From: Bate A et al. Designing and incorporating a Real World Data approach to international drug development and use - what the UK offers. Drug Discovery Today. In Press

PROTECT received support from the Innovative Medicine Initiative (IMI) Joint Undertaking

(www.imi.europa.eu), resources of which are composed of financial contribution from the

European Union's Seventh Framework Programme (FP7/2007-2013) and EFPIA

companies’ in kind contribution.

10

Background and Main Objectives of WP2

12

•  Increased use of large-scale, “real world” electronic healthcare databases

•  Studies in same database generate different results

–  e.g., oral bisphosphonates-esophageal cancer, or statins-fractures

•  Studies in different databases generate different results

–  e.g., antibiotics-sudden death, or NSAIDs-cardiovascular risk

•  Debate about value of epidemiology for understanding medicines’ benefits and risks

•  Explain differences in drug-adverse event associations due to choices in methodology and databases (WP2)

•  Testing focused on specific drug event pairs •  Antidepressants/benzodiazepines and hip fracture

•  Inhaled long-acting B2-agonists and acute myocardial infarction •  Anti-epileptics and suicide / suicide attempts

•  Antibiotics and acute liver injury

•  Calcium channel blockers and cancer

Abbing, V et a. Bridging differences in findings from pharmacoepidemiological studies: The PROTECT project. Current Clin Pharmacol. 2014 May;9(2):130-8.

Characteristics of healthcare databases

13

Data

base

Select results

14 http://www.imi-protect.eu/results.shtml

Ref Klungel OH 2015 “Outcomes of studies for six adverse event-drug pairs and five databases: what did we learn?” IMI PROTECT Symposium, London http://www.imi-protect.eu/symposium.shtml

Antibiotics –Acute Liver Injury – across multiple databases with multiple study designs

A “three- tiered” RWD strategy “Ad-hoc” use data sets

Remote access databases

Centralized licensed in-house data

16

Novel Use of Claims & EMRs for signal detection/refinement

Signal Detection • Any Medical Event • Designated Medical Events

Signal Refinement

Signal Evaluation

Rapid Detect the unexpected Less persuasive

Time Consuming Test the anticipated Convincing

Product Approval &

Launch

How to best utilise the wealth of Real World Data and does its value change depending on purpose?

Ref Bate A. 2010 Panel B- Emerging Data Sources -Institute of Medicine (IOM) Committee Meeting, Washington DC

US FDA Sentinel Initiative

•  Large Claims and EHR databases for analysis of drug outcomes, linked in “distributed network”

•  Mandated by Congress: FDA Amendments Act of 2007 •  Full Sentinel System now in routine use

–  Sole FDA use Mini-Sentinel Pilot project ran from 2009-2014

•  Distributed database: data from 18 health plan data partners that retain physical and operational control over its own data

•  Data on 193 million members •  Rapid analysis capability

Sources: 8th Annual Sentinel Initiative Public Workshop 2016 and http://mini-sentinel.org and accessed 22nd February 2016

FDA’s Sentinel Initiative Partner Organizations

Institute for Health

Lead – HPHC Institute

Data and scientific partners

Scientific partners

18

Common data model role in distributed network- the OMOP model

Source 1 Source 2 Source 3

OMOP Analysis results

Analysis method

Transformation to OMOP common data model

Diagram reference: OMOP

Use of a Common Data Model facilitates fast analysis of multiple databases, and allows analyses across a distributed network. Use of data converted to common denominator can be problematic

The work of the US public private partnership OMOP has now transitioned to IMEDS

Innovation in Medical Development and Surveillance (IMEDS)

•  IMEDS is a program within the Reagan-Udall Foundation for the US FDA and is a public private partnership created to build upon the significance progress made of research methodology by FDA’s Sentinel Initiative and the Observational Medicines Outcomes Partnership (OMOP)

•  Primary objective is to advance the science and tolls necessary to support post-market evidence generation on regulated products, including safety surveillance and evaluations, to facilitate utilization of a robust electronic healthcare data platform for generating better evidence on regulated products in the post-market settings

•  See: imeds.reaganudall.org

IMEDS-‐Methods Facilitates methods research in support of using

electronic health data for safety surveillance conducted by FDA and the broader research community.

IMEDS, part of the Reagan Udall Foundation

IMEDS-‐Evalua+on Leverages Sen6nel tools and capabili6es to create a na6onal resource for evidence genera6on accessible by inves6gators

outside the FDA.

IMEDS-‐Educa6on Trains scien6sts in how to conduct methods research using electronic

healthcare data.

IMEDS works with the FDA, industry, and clinicians to improve pa+ent care and the safety of medical products by focusing on three areas:

3 2

1

21 imeds-reaganudall.org

First non-FDA use of Sentinel network: Pfizer sponsored IMEDS Evaluation Pilot conducted

Distributed network surveillance: Recording of angioedema for lisinopril users compared to non-users: 2000-2005

Lisinopril Versus Unexposed

0

10

20

30

40

50

60

70

80

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71

Months

Cum

ulat

ive

Adv

erse

Eve

nts

(ang

ioed

ema)

0

10

20

30

40

50

60

Rel

ativ

e R

isk

Observed AEs Expected AEs RR

Unpublished data based on work in Brown et al., (2007, 2009) in PDS). Contact: [email protected]

Note: Base-case analysis. Outcome: Angioedema. Adjusted for age, sex, and health plan.

Signal of Disproportional Recording at month 13; 3 observed and 0.06 expected

Data from US Health Maintenance Organization research network

Performance characteristics of hypothesis-free surveillance on UK EMR THIN

Measure Threshold Sensitivity Specificity PRR PRR 95% LBCI >1 0.67 0.68 USCCS OR >1 and LBCI >1

(a=0.05) 0.78 0.59

HDPS RR >1 and LBCI >1 (a=0.05)

0.50 0.76

Comparison against the OMOP reference set of established drug-event combinations1

1 Stang et al (2010). "Advancing the science for active surveillance: rationale and design for the Observational Medical Outcomes Partnership." Annals of Internal Medicine 153(9): 600-606.

From Zhou et al 2013

In IMI PROTECT WP3, conducted prospective hypothesis-free signal detection and of 891 pairs, 91 were adjudged to merit further investigation. (ref Cederholm et al 2014)

Analysis in OMOP Common Data Model

-30 -20 -10 0 10 20 30 -5

0

5 Omeprazole - Acute Pancreatitis

IC

-30 -20 -10 0 10 20 30 0

20

40

60

Months relative to prescription

Num

ber o

f eve

nts

Observed Expected


Spontaneous reports valuable, but give limited insights in such situations

Information Component (IC)* shows unexpected recording of outcomes relative to time of prescription

Novel Visualization of Longitudinal Patient Records UK EMR (THIN): LODEX software

* IC is a Bayesian shrinkage observed-to-expected ratio on a logarithmic scale

Variability in longitudinal data capture for healthcare analyses – getting richer data?

25

Patient Time

A

B

C

D

E

F

Structured data in a given data source (e.g., drug prescriptions, diagnosis codes)

Structured data from another source (e.g., inpatient data as well as primary care data)

Unstructured data - info solely gleaned from free text narrative (e.g. lab tests, clinical text observations) e.g. using Natural Language Processing

Information not routinely captured in records (e.g., herbal administration, prescriber and patient attitudes)

Prospective monitoring/collection and capture of new/more accurate data (e.g., DNA information, walked steps per day)

Data that only can be gleaned by interactive, often iterative discussion with healthcare provider, leading to further investigations and tests, and subsequent data capture (e.g., record specific triggering further lab tests)

Other challenges: Ø  Missing patients, and observation time Ø  Accuracy of listed data (and time

stamping) Ø  Duplicated/overlapping information Ø  Inconsistent recording across records

(difficulty linking underlying concepts)

Added value of NLP extracted narrative contents in US EMR

│ 26

STR

UC

TUR

ED

DAT

A U

NSTR

UC

TUR

ED

DATA

Use NLP to capture clinical narrative text

indicative of ALD

Identify IBD-ALD Surveillance Population

Iterative Supervised Learning to Refine and Define the Algorithm to Identify Final

Set of ALD Cases

Assess the Contribution of Unstructured Data in Identifying ALD Cases

Use Predefined Structured and Unstructured Elements to Define ALD

Candidate Cases (Prior Algorithm)

CO

MB

INED

STRU

CTU

RED

A

ND

UN

STRU

CTU

RED

D

ATA

*Developed and executed with WHISCON

Results from study done by Pfizer and WHISCON

Better Performance: Combined NLP Terms and Coded Data

│ 27

Sensitivity Specificity Positive Predictive Value

Final algorithm and clinical expert adjudication

92%

(11/12)

92%

(35/38)

79%

(11/14)

Final algorithm with NLP terms suppressed and clinical expert adjudication

83%

(10/12)

87%

(33/38)

67%

(10/15)

*Fourth round review: N=50

NLP Data Contributed to Defining ALD Onset Dates and Getting Earlier ALD Onset Dates

ALD Cases N=4026

Onset date due to inclusion NLP

N=348 (9%)

Earlier Onset date due to NLP N=220 (5%)

Same onset date with or without NLP N=3458 (86%)

Earlier Onset within one month

N=183 (83%)

Earlier Onset b/w 1 to 6 months

N=10 (5%)

Earlier Onset >6 months

N=27 (12%)

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

29

23andMe & Pfizer Lupus Community Collaboration


•  23andMe is a direct to consumer genetic testing company that returns ancestry information and genetic data (both raw and interpreted)

23andMe & Pfizer Lupus Community Collaboration


•  A cohort of 5,000 Lupus patients are being recruited comprising a research community

–  Participants are consented for longitudinal research of: –  Target recruitment to include a minimum of 7% African Americans –  Participants have a customized 23andMe experience within their online

accounts –  About 3K recruited and genotyped to date

•  Demographics •  Family History •  Lupus symptoms •  Treatment History •  Environmental exposures

23andMe & Pfizer IBD Community Collaboration


•  A cohort of 10,000 IBD patients are being recruited comprising a research community

–  Participants are consented for longitudinal research of” –  Target recruitment to include a minimum of 5% African Americans –  Participants have a customized 23andMe experience within their online

accounts –  About 8K recruited and genotyped to date

•  Demographics •  Family History •  IBD symptoms •  Treatment History •  Environmental exposures

Spontaneous reports •  Manual recording on a form submitted by ‘snail

mail’

•  Limited capability for computerized analysis

•  Quantitative approaches had no routine role

Longitudinal healthcare observational databases

•  Some pharmacoepidemiological studies conducted in single databases

o  Primarily primary data collection, although later some secondary use of individual EHR and Insurance Claims data

•  Studies Limited capability for computerized analysis

Social media data/ mobile data streams •  None

Spontaneous reports •  Huge electronic reporting and resulting

electronic repositories of data

•  Well established limitations but well recognized strengths and deeply embedded in signal management processes

•  Inefficiencies such as known and unknown data duplication limit efficiency of approach

Longitudinal healthcare observational databases

•  Large Networked Databases in US, Canada, Europe and Asia with rapid search capability as well as well established use for formal epidemiological hypothesis testing studies o  Testing of hypothesis free signal detection

capability

Social media data/mobile data streams •  Huge volumes of data, Unclear value for Signal

detection though some foundational research being conducted

Data for Signal Detection Post-Launch Has Transformed Since the Birth of Pharmacovigilance

Need to proactively determine best use of social media data with other data streams for optimal modern pharmacovigilance

1960s 2000s

Evolution of the Watson business

The following slides presented with kind permission of IBM Watson

2011 Jeopardy!

Grand Challenge

Demonstration

2006 – 2010 Research

Project

R&D

2011–2013 Internal

Startup Division

Market Validation

2014–present IBM Watson Group

Commercialization

April 13, 2015

Industry Vertical

34

Understanding Natural Language: wins the quiz show Jeopardy!

The Challenge: Participate in Jeopardy! - America's favorite quiz show

§  Broad knowledge of politics, history, literature, art, entertainment, science required §  No easy questions - irony, subtle differences in meaning, word games, puzzles §  Understanding of the phrase and finding the right answer in the shortest possible time

Jeopardy Example: Keyword matching leads to the wrong result

celebrated

India

In May 1898

400th anniversary

arrival in

Portugal

India

In May

Gary explorer

celebrated

anniversary

in Portugal

Keyword Matching

Keyword Matching

Keyword Matching

Keyword Matching

Keyword Matching

In May, Gary arrived in India after he celebrated his anniversary in Portugal.

arrived in

In May 1898 Portugal celebrated the 400th anniversary of this explorer’s arrival in India.

celebrated

May 1898 400th anniversary

arrival in

In May 1898 Portugal celebrated the 400th anniversary of this explorer’s arrival in India.

Portugal landed in

27th May 1498

Vasco da Gama

On the 27th of May 1498, Vasco da Gama landed in Kappad Beach

Kappad Beach

Para-phrases

Geo-KB

Date Math

explorer

India

Jeopardy Example: Watson DeepQA finds the correct answer

Miniaturized Clinical Chemistry Devices

Sensors Facilitate R&D Via Remote Monitoring

38

Hypertension Management

Activity Tracking ECG Rhythm

Detection

Non-Adherence Smart Pills, Containers Respiratory Disease

Heart Rate Variability (HRV) & Electrodermal Activity (EDA)

(Observational) data revolution - now

SRS LOD Social media

IoT*

Population data coverage XXX XX X

Data granularity on individuals X X X

Accurate, effective & consistent data provision

XXX XXX

Rapidity of data availability for analysis

XX XXX XXX

* Internet of Things (IoT) e.g. wearables, video streaming, device sensor data streaming

(Observational) data revolution – soon?

SRS LOD Social media

IoT*

Population data coverage XXX XX XX XXX

Data granularity on individuals X XX XXX

Accurate, effective & consistent data provision

XXX XXX

Rapidity of data availability for analysis

XX X XXX XXX

•  Internet of Things (IoT) e.g. wearables, video streaming, device sensor data streaming

•  However Major privacy, societal, behavioural concerns and attitudes to be considered and resolved

Pfizer Example - Neuroscience

“Scientists use videogames to improve older brains” September 2013

“Gaming improves multitasking skills: study reveals plasticity in age-related cognitive decline”

“Akili announces partnership with Pfizer to test video game in people at risk of Alzheimer’s disease”

January, 2014

41

BACK UPs


Akili: A Novel, Remote, High-Resolution Cognitive Biomarker in the Form of an iPAD (Mobile) Application

Software assesses divided attention while dual-tasking •  Impact of distraction (“Interference cost”) increases with age •  Attention → core component of executive function •  Training improves performance → malleable brain circuits

Age Group

43

Conclusions

•  Big Data Strategies including ‘Real World Data' use are employed throughout the drug development lifecycle, including Safety –  ‘Big Data’ solutions focusing solely on observational data will

never provide a panacea and must ensure data security and privacy

•  Analyzing more and more data does not automatically lead to more appropriate conclusions, and RCTs and other data types do and will continue to retain a key role

•  IT developments have led to more capability and interest in conducting analyses across networks of many distinct observational ‘real world’ databases

–  Examples, such as the Sentinel Network, use ‘distributed data networks’ –  Efforts to enrich data for analyses, by embedding primary data collection or mining

clinical narratives for additional information are important –  mData is an important develop for the future of Pharmacovigilance

44

Back up

Spontaneous reporting as a tool for Pharmacovigilance

•  Analysis of spontaneous reports of suspected adverse drug reactions (ADRs) is a valuable tool in the detection of previously unknown drug adverse reactions

•  Reports of adverse events (AEs) associated with a drug are not necessarily true ADRs, that is, they may be temporally associated with a drug but not caused by the drug.

•  Hypothesis generation of new possible side effects from such data is referred to as ‘signal detection’ –  An exploratory data analysis task

Ref Bate and Evans 2009 PDS

Methods tested by OMOP for surveillance

•  Disproportionality Analysis (DP) •  Univariate Self-Controlled Case Series (SCCS) •  Observational Screening (OS) •  Multi-Set Case Control Estimation •  Bayesian Logistic Regression (BLR) •  Case Control Surveillance (CCS) •  IC Temporal Pattern Discovery (ICTPD) •  Case-Crossover (CCO) •  HSIU Population-Based Method •  Maximized Sequential Probability Ratio Test (MSPRT) •  High-Dimensional Propensity Score (HDPS) •  Conditional Sequential Sampling Procedure (CSSP) •  Incident User Design (IUD-HOI)

Ref Stang et al 2010 Archives of Internal Medicine

Database model heat map

Database model is that of OMOP CDM

Shows how well different variables convert into a Common Data Model

Ref Zhou et al 2013

Tested two Common Data Models on the Same Data Source

49

Source

Analysis results

MS Analysis method

implementations

Transformation to Common Data Model

OMOP Analysis method

implementations

MS CDM

OMOP CDM

Both CDMs have extensive purpose-built ecosystems of tools and programs for analytic capability and quality assurance

Study result: Outputs similar – but sometimes very different

Issues to consider in selection of RWD

•  Suitability of RWD source to address the question of interest –  Data capture and its structure –  Accessibility

•  Assurance of data security and especially patient privacy •  Demonstrability of data and analysis integrity •  Recency of data available for analysis •  Stakeholder needs

Procedures

51

•  Common protocol for each drug-AE pair –  Extensive sensitivity analyses on main

methodological issues

•  Common standards, templates, procedures –  Detailed data specification including definitions of

exposures, outcomes, and confounders for each database.

•  Blinding of results of analyses within drug– AE teams and across WPs

ICΔ* shows unexpected frequent recording of outcome after terbinafine prescription

Example: Demonstrated Use of EMR Data for Early Identification of Drug Side Effects

* ICΔ is the difference in IC before and after prescription on a logarithmic scale 52

Angioedema was labelled In January 2004

However extensive testing of all such methods show imperfect performance

Observed/ Expected

Observed number cases of combination

Expected number of cases

Strata

43.89 3 0.07 Child,Female

42.29 4 0.09 Child, Male

4.69 5 1.07 Adult, Female

3.02 6 1.99 Adult, Male

Crude Obs/ Exp = 5.82 Adjusted Obs/ Exp* = 5.60 Clearly either overall estimate misleading, subgroup analyses important

Adapted from Bate and Evans PDS 2009

Not always ideal for Exploratory Data Analysis: where would want to consider potential signals of differential risk not solely consider drug-event pair

Loperamide and spontaneous reports of Urinary retention in WHO database

Adjusted estimates to account for basic confounders or subgroup analyses

* Mantel-Haenszel adjustment for age and gender

Now hundreds of Sources of Different Real World Data: Some Examples

Database Country Characteristic Population Size

THIN UK GP primary care database 10.5 M1

Danish National Health Service Register Database

Denmark Healthcare registry of care 5.5 M2

Premier US Clinical data from the

hospitals 130 M+ patient

discharges3 Normative Health Information (NHI) Database US

Transactional claims records of a commercial

health insurer 60 M+4

Health Insurance Review and Assessment Service (HIRA) Korea

Insurance Claims from near universal national

system 48 M5

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 5th Edn 2011 pp 244-258 4 Seeger J, Daniel GW. Commercial Insurance Databases. In Pharmacoepidemiology 5th 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

54

Record linkage efforts over the years to improve safety surveillance - Examples

•  Adverse event prediction for antidepressants using neural net based on chemical structural properties and receptor activity –  Alvager et al 1993 –  Almenoff et al 2011 similar chemical structure and chemical and

biological property linkage •  Linked spontaneous reports to CYP 450 activity information for

drug-drug interaction signal detection –  Strandell et al 2011

•  Complementary value of signal detection across data streams –  Pacurariu et al 2015

•  Basic spontaneous report analysis process linked to EHR data and systems biology, with experimental validation of outputs –  Tatonetti et al 2009, Lorberbaum et al 2015

PROTECT Goal

These methods are being tested in real-world situations

To strengthen the monitoring of benefit-risk of medicines in Europe by developing

innovative methods

to enhance early detection and assessment of adverse drug reactions from different data

sources (clinical trials, spontaneous reporting and

observational studies)

to enable the integration and presentation of data

on benefits and risks

www.imi-protect.eu 56

•  Largest private/public partnership in EU. EMA coordinates PROTECT and a consortium of public and private partners

Other interesting applications of vision mining and data augementation

Sagawa R, Yoshiyasu Y, Alspach A, Ayusawa K, Yamane K, Hilton A.

Analyzing Muscle Activity and Force with Skin Shape Captured by Non-contact Visual Sensor. In Image and Video Technology 2015 Nov 25 (pp. 488-501). Springer International Publishing.

Also with the advent of 3D printing who knows what benefits that bring to

bear

Background reading

•  Bate A et al. Designing and incorporating a Real World Data approach to international drug development and use - what the UK offers. Drug Discovery Today. In Press

•  Behrman RE et al (2011). Developing the Sentinel System—a national resource for evidence development. New England Journal of Medicine, 364(6), 498-499.

•  Brown, JS et al. (2009). Early adverse drug event signal detection within population‐based health networks using sequential methods: key methodologic considerations. Pharmacoepidemiology and drug safety, 18(3), 226-234.

•  Cederholm S et al. 2014 Structured assessment for prospective identification of safety signals in electronic medical records: evaluation in The Health Improvement Network. Drug Safety. 38(1), 87-100.

•  Stang et al (2010) Advancing the Science for Active Surveillance: Rationale and Design for the Observational Medical Outcomes Partnership Annals in Medicine 153(9), 600-6

•  Walker AM, Zhou X, Ananthakrishnan AN, Weiss LS, Shen R, Sobel RE, Bate A, Reynolds RF. Computer-Assisted Expert Case Definition in Electronic Health Records. International Journal of Medical Informatics. In Press

•  Xu Y et al A Comparative Assessment of Observational Medical Outcomes Partnership and Mini- Sentinel Common Data Models and Analytics: Implications for Active Drug Safety Surveillance. Drug Safety. 38(8), 749-765.

58

big data, smart data, fast data – a paradigm shift · • i am a member of the ibm watson health...

Documents