\n
\n Alexander Sc hacht\, Lilly
\n
\n Abstract: Can I divide my overall patient popula tion into meaningful segments? Do patients follow different patterns over time? We should ask these questions more often and techniques of unsupervi sed learning\, where the classification of a patient into a group is unkno wn\, answers these questions. We differentiate these approaches from super vised learning techniques in which classification of the patients is known . Typical questions for supervised learnings algorithms include: Can I pre dict patients outcomes given his/her baseline characteristics?
\n
\n Cluster analysis represents a class of approac hes in unsupervised learning. It helps to answer the above questions. Clus ter analysis stands on the determination of metrics\, which measure the di stances between patients in terms of their many different characteristics. In this presentation\, I will present and discuss different approaches av ailable in SAS.
\n
\n The determination o f the number of clusters represents a classical problem of bias-variance t rade-off. The presentation will discuss various heuristics but also practi cal considerations to determine a reasonable choice of clusters.
\n
\n The practical implementation of cluster anal yses comes with various challenges. I will discuss standardization of vari ables\, weighting of variables\, correlated data\, outliers\, finding spur ious small clusters\, and identification of relevant clusters. \; \;
\n
\n Finally\, the communication of c luster analyses has its unique challenges and I will mention various appro aches based on real case studies.
\n
\n < strong>Bio: \;Alexander Schacht (PhD)\, Principal Research Sc ientist\, Global Statistical Sciences leads a group of 5 European based st atisticians driving the statistical activities around launch preparation i ncluding HTA submission to support access and commercialization in differe nt auto-immune diseases. After 2 years at Boehringer Ingelheim\, Alexander joined Lilly in 2004 and held various positions within statistics with a focus on neurosciences working on phase I\, III\, and IV in areas like Alz heimer\, Schizophrenia\, ADHD\, Depression\, and Pain. Alexander received his PhD in Biometrics in 2002 from the University of Gö\;ttingen on wo rk related to non-parametric analysis of covariance. For the publication b ased on this\, he was awarded the 1st. Gustav-Adolf-Lienert Price in 2009 by the German region of the International Biometrical Society. He has publ ished both methodological papers (e.g. on network-meta-analysis\, non-infe riority approaches for time-to-event data) and medical papers including mo re than 60 papers in peer-reviewed biomedical journals. He is a regular sp eaker at both medical and statistical international conferences. As the ch air of the special interest group on benefit-risk of the European Federati on of Statisticians in the Pharmaceutical Industry\, Alexander is leading and promoting research on quantitative assessments of benefit-risk. He is interested in all aspects of launching new treatments.
\n
\n
\n
Overview of meth
ods for subgroup and biomarker identification from clinical data<
/p>\n Abstract: In this talk I will provide a
high-level description of a broad class of statistical methods for subgrou
p/biomarker identification in early and late-phase clinical trials. First\
, I contrast &ldquo\;data-driven&rdquo\; subgroup analysis with a traditio
nal &ldquo\;guideline-driven&rdquo\; approach and describe key elements of
principled data-driven subgroup analysis. Then I review 4 classes of meth
ods for subgroup identification that had emerged recently as a result of c
ross-pollination across machine learning\, causal inference and multiple t
esting (global outcome modeling\, global treatment effect modeling\, model
ing individual treatment regimes\, and local treatment effect modeling). I
also briefly review available software and key features of subgroup ident
ification methods.
\n
\n Bio:&nbs
p\;Ilya Lipkovich is a Sr. Research Advisor at Eli Lilly working
in Real World evidence. He \; received his Ph.D. in Applied Statistics
from Virginia Polytechnic Institute and State University in 2002. He has
more than 15 years of statistical consulting experience in pharmaceutical
industry. Dr. Lipkovich research interests include subgroup identification
in clinical data\, analysis with missing data\, and causal inference from
observational data. He is a chair a Subgroup Analysis Working Group spons
ored by the Society of Clinical Trials. He has published widely including
co-authoring a book &ldquo\;Analyzing Longitudinal Clinical Trial Data. A
Practical Guide.&rdquo\;
\n
\n
\n
\n Andy Nicholls\, GSK
\n
\n Abstract: In 2016 GSK successfully completed the Salford Lung Study\, a 12-month\, open label\, randomised\, effectiven ess study to evaluate fluticasone furoate (FF\, GW685698)/vilanterol (VI\, GW642444) Inhalation Powder delivered once daily via a Novel Dry Powder I nhaler (NDPI) compared with the existing COPD maintenance therapy alone in subjects with Chronic Obstructive Pulmonary Disease (COPD).
\n
Upon completion of the study\, the Scientific Committee expresse
d an interest in using a data-driven approach in order to identify patient
subgroups for which the treatment effect was strongest. \; In this pr
esentation we will look at why SIDES was chosen for this analysis\, the de
sign parameters\, and how it fared. \;
\n
\n
Bio: \;Andy is a Statistician with a strong i
nterest in Data Science\, having previously worked as a specialist R Consu
ltant and Data Scientist for Mango Solutions. \; On re-joining GSK in
2017\, Andy provided support to the Relvar project\, for which he led an e
xploratory cluster analysis using Salford Lung Study data in order to try
to identify patient subgroups that might experience an additional real-wor
ld benefit of Relvar. \; He now works in GSK&rsquo\;s new Statistical
Data Sciences division within BioStats and is Business Systems Owner for t
he BioStats HPC environment for R.