My personal activities on automating evidence synthesis and real world data derived evidence for automated treatment guidelines compilation for precision medicine.
Russian Call Girls in Pune Riya 9907093804 Short 1500 Night 6000 Best call gi...
Â
Leverage machine learning and new technologies to enhance rwe generation and outcomes research 27 jan2020 miami
1. Leverage Machine Learning and
New Technologies to Enhance
RWE Generation and Outcomes
Research
Some personal thoughts of Athula Herath, PhD, MBCS. CEng
Global Head, Real World Evidence Epidemiology, Global Medical Affairs, Novartis Pharmaceuticals
January 27, 2020
Patient Registries, Real World Evidence (RWE) and Health Economic Outcomes Research (HEOR),
Miami
2. Outline
ďˇ
Medicine had been a data science from its inception â
ď observe/experiment, analyze, deduce â intervene (or do nothing).
ďˇ
Evidence synthesis methodology has been pioneered by the medical/epidemiological/statistical
community for ages with the standards established over the last 50 years.
ďˇ
With the astonishing number of therapeutic entities entering the clinical practice a new, and the new
era of combination therapy of the existing/new entities, the medical community is struggling to keep up
to date with the synthesis emerging evidence (and noise).
ďˇ
Formulating objective/coherent and up-to-date medical/treatment guidelines for interventions (in
chronic diseases) is becoming intractable (to humans).
ďˇ
With the emergence of the data science, and resurgence of reinvigorated âmachine learningâ, coupled
with traditional methods may offer us
In this presentation, I will attempt share my personal thoughts (a disclaimer too :) ) and experience in
attempting to automate various stages of evidence synthesis (using Real Word Data and Evidence â
RWD/RWE)
3. ⢠Disease
Stratification
⢠Target to Disease
and Clinical
Outcome Linkage
⢠Objective
assessment of the
"need" in terms of
poor status quo of
clinical outcomes
⢠Clinical outcome
based Combination
Strategy and
Repurposing
Strategy
⢠Disease
Stratification --
establishing target
populations
⢠Stratification Tools
(Clinical outcome
indices for
Diagnostics
Development)
⢠Establishing
appropriate
Clinical Outcomes
for high precision
assessment/demon
stration of
Therapeutic
Effects
(Safety/Efficacy)
⢠High precision
study Designs (in
targetted
populations)
⢠Establishing the
definitive clinical
outcomes to
demonstrate the
efficacy (safety) of
the drug
⢠Assessing/
Demonstrating the
real world
applicability of
clinical study
designs and study
results.
⢠Establishing the
value proposition
⢠Objectively
illustrating the
value proposition
and in specific
populations via
relevant clinical
outcomes/comorbi
dities (countries,
regions) and
comparative
effectiveness
⢠High precision
estimation of
benefit/ risks
⢠Expanding the
label --
characterising
additional
populations/indicat
ions
Contributions by establishing Disease Epidemiology based on Real World Evidence
and Population Centric Evidence Synthesis for the life cycle of the development of
Pharmaceuticals
DiscoveryDiscovery Translational
Medicine
Translational
Medicine
Late
Development/
Registration
Late
Development/
Registration
Post
Marketing
Post
Marketing
5. 5
Landscape of RWD (Big Data in)Healthcare
"..., big biomedical data are
scattered across institutions
and intentionally isolated to
protect patient privacy. Both
technical and social challenges
to linking these data must be
addressed before big
biomedical data can have their
full influence on health care.â
Finding the Missing Link for Big Biomedical
Data
Griffin M. Weber, et. al, JAMA.
2014;311(24):2479-2480.
doi:10.1001/jama.2014.4228
http://jama.jamanetwork.com/article.aspx?articleid=1883026
6. Patient Level Data
ď Well characterised cohorts
ď Randomised Clinical Trials
ď Other studies (e.g.: ad-hoc studies, well/ill designed
biomarker studies)
ď Electronic Health Records (e.g.: CPRD, UK NHS health
records that are being assimilated in CPRD)
Summary Form
ď Publications of all of the above in summary form (e.g.
reported base line characteristics, study
characteristics and efficacy/safety outcomes/results).
7. Inspired by â A Poem
People predict by making up stories
People predict very little and explain everything
People live under uncertainty whether they like it or not
People believe they can tell the future if they work hard enough
People accept any explanation as long as it fits the facts
The handwriting was on the wall, it was just the ink that was invisible
People often work hard to obtain information they already have And
avoid new knowledge
Man is a deterministic device thrown into a probabilistic Universe
In this match, surprises are expected
Everything that has already happened must have been inevitable
Michael Lewis. The Undoing Project: A Friendship that Changed the World (p. 197). 2016,
Penguin Books Ltd
8. Evidence Based Medicine
Our ability to precisely estimate the treatment outcomes is
often impaired by the entanglement of the
Primary disease > treatment/outcomes with
Comorbidities > treatment/outcomes,
and the environmental and operational attributes.
9. Motivation to Invent Disease Evidence Hubs
The emerging disease classification system of treatable traits,
classifying health in terms of Clinical Phenotypes and Endotypes may be
utilized to invent a framework for mining (automating) health and chronic
disease outcomes using Real World Evidence with a view to predicting
local payment regimens for treatment outcomes more precisely
Treatable traits: toward precision medicine of chronic airway diseases
Alvar Agusti, Elisabeth Bel, Mike Thomas, Claus Vogelmeier, Guy Brusselle, Stephen Holgate, Marc Humbert, Paul Jones, Peter G. Gibson, Jørgen Vestbo, Richard Beasl
Ian D. Pavord
European Respiratory Journal 2016 47: 410-419; DOI: 10.1183/13993003.01359-2015
10. Refining the evidence Pyramid
Murad MH, Asi N, Alsawas M, et al, New evidence pyramid,
BMJ Evidence-Based Medicine Published Online First: 23 June 201
doi: 10.1136/ebmed-2016-110401
A)The traditional pyramid.
B) Revising the pyramid:
(1) lines separating the study designs
become wavy (Grading of
Recommendations Assessment,
Development and Evaluation),
(2) systematic reviews are âchopped offâ
the pyramid.
C) The revised pyramid: systematic reviews
are a lens through which evidence is
viewed (by applying dynamically).
11. Revising the Evidence Pyramid to an Evidence Hub
Comorbidity/
Treatment outcomes Primary Disease
Treatment
outcomes --
RWD/RWE
Comorbidity
Treatment
outcomes
RWD/RWE
Evidence Soup
Machine/Deep
learning
Disease Landscapes/
Deep Clinical Phenotyping
Ensemble Outcome
Models
Predictive
Models/Tools
EHR/Continuous Health
/Lifestyle/ Socio
Economic/ Genomes/
Motivation/ Compliance
21st
Century
Evidence
Synthesis
(Epi 2.0)
17. We are a Mixture of Mixtures
https://www.researchgate.net/publication/4119590_Exploring_
Face_Space
18. An algorithm ?
â
Phenotype patients using available patient level clinical,
observational/real word evidence data
â
Map any available molecular expression data to the derived
landscape (e.g.: gene expression, protein expression, any other
mappable molecular expression data)
â
Molecular understanding of the mechanisms involved in the
strata of interest (poor outcomes)
â
Map the current therapies (the entities that are in development by
us and our competitors via the molecular fingerprints
â
Map the published summary data/meta-analysis output of the
current therapies into the landscape
â
Asses/Examine therapeutic effects within the segments of the
populations
19. The Result ...
⢠The âalgorithmâ results a fully parameterized statistical model for a
particular therapeutic area (say severe asthma), we call such
models "Ensemble Outcome Modelsâ.
⢠An "ensemble outcome model" may be useful in more than one
way;
â To compose a novel therapeutic (i.e. pathway or a combination)
at the appropriate level of effect,
â To explore the therapeutic landscape for clinical
development/target validation,
â To estimate the the therapeutic effects we must have in order to
succeed (guide the clinical development),
â To evaluate the therapeutics that are in development outside
(competitors and/or for in licensing etc.)
21. Constructing Clinical Phenotypes â How?
â
Clinically Phenotype patients using available patient level RCT, observational/real
word evidence data
â Use population scale RWD (millions or hundreds of millions of patients,
enriching for the heterogeneity, i.e. across differing populations, i.e. global)
â Use the clinical variables and baseline characteristics of the patients and
longitudinal data that allow us to establish the state of the disease
â Use linear and non linear dimension reduction techniques, including classical
statistical methods (i.e. linear PCA and non-linear PCA, neural networks (non-
linear), and modern deep and machine learning (i.e. word embedding, which
allows us to capture all variables, including the unstructured variables, and
also allow us to handle missing observations, as longitudinal data often will
contain missing values across the patient journeys)
27. 27
âAlice: Would you tell me, please, which way I ought to go from here?
The Cheshire Cat: That depends a good deal on where you want to get to.
Alice: I don't much care where.
The Cheshire Cat: Then it doesn't much matter which way you go.
Alice: ...So long as I get somewhere.
The Cheshire Cat: Oh, you're sure to do that, if only you walk long enough.âÂ
â Lewis Carroll, Alice in Wonderland
Systematically guiding therapeutic development and
assessing them ...
28. Assessing the clinical phenotypes
Data-driven identification of prognostic tumor
subpopulations using spatially mapped t-SNE of
mass spectrometry imaging data
Walid M. Abdelmoula, Benjamin Balluff, Sonja
Englert, Jouke Dijkstra,  View ORCID ProfileMarcel J. T.
Reinders, Axel Walch, Liam A. McDonnell, and Boudewijn
P. F. Lelieveldt
PNASÂ October 25, 2016Â 113Â (43)Â 12244-12249;Â https
://doi.org/10.1073/pnas.1510227113
29. Assess and compare effectiveness, stratified by
the established clinical phenotypes.
30. 30
RWD (Big Data) are used to figure out the right inferences on Small Data.
ď Avoiding the âbase rate fallacyâ â as (Nobel Laureate) Daniel Kahneman
puts it âŚ".. the tendency to predict the outcome that best represents the data,
with insufficient regard for prior probability, has been observed (even) in the
intuitive judgments of individuals who have had extensive training in statistics"
--Amos Tversky and Daniel Kahneman
 http://www.sciencemag.org/content/185/4157/1124
The therapeutic landscape map
ď Is a comprehensive synthesized evidence framework (a data reduction
technique)
ď Is a powerful tool for pinpointing the unmet need in stratified medicine
ď Helps to design new clinical development programmes with high
precision by fully characterising the strata (clusters) in terms of their
clinical characteristics (inclusion/exclusion criteria and diagnostics)
Summary
31. References
â
Defining Phenotypes in Diabetic Nephropathy: a novel approach using a cross-sectional analysis of a single centre
cohort, RM Montero, A Herath, A Qureshi, E Esfandiari, CD Pusey, AH Frankel, Nature Scientific reports 8 (1), 1-8 â
https://www.nature.com/articles/s41598-017-18595-1
â
Moving toward endotypes in atopic dermatitis: identification of patient clusters based on serum biomarker analysisJL
Thijs, I Strickland, CAFM Bruijnzeel-Koomen, S Nierkens, A Herath ...Journal of Allergy and Clinical Immunology 140
(3), 730-737 â https://www.sciencedirect.com/science/article/abs/pii/S0091674917305833
â
Elevated sputum interleukin-5 and submucosal eosinophilia in obese individuals with severe asthma, D Desai, C
Newby, FA Symon, P Haldar, S Shah, S Gupta, M Bafadhel,A Herath ⌠American journal of respiratory and critical
care medicine 188 (6), 657-663 â https://www.atsjournals.org/doi/full/10.1164/rccm.201208-1470OC
â
All publications â Athula Herath â https://scholar.google.co.uk/citations?user=Tg3sJP0AAAAJ&hl=en
32. 12/1/14
Athula Herath, BSc(Hons) PhD, MBCS, CITP, CEng
Global Head of Real World Evidence Epidemiology
Real World Evidence (Evidence and Launch Excellence)
Global Medical Affairs
Athula.Herath@novartis.com
32
Thank YOU!Thank YOU!
Time for Questions?Time for Questions?