Guided Analytics with Pharma R&D Informatics - Rudi Verbeeck Insights

The best of both worlds Guided analytics in the hands of the SME Pharma R&D IT - Informatics Rudi Verbeeck

Outline Changing world More data More analysis requirements Two worlds Technical skills Subject matter expertise Best of both worlds Exploratory analysis: DIAN study Guided analytics: phase III clinical trial planning Pharma R&D IT Informatics October 2014 2

Evolution of data In the pharmaceutical industry Rich electronic data sources, new linkages, new applications Largely paper based R&D 1990 2000 2010 Computerised records, increased efficiency More data sources, more linkages, more applications Pharma R&D IT Informatics October 2014 3

Pharma R&D IT Informatics October 2014 4

Data challenges for the EMIF consortium Diversity Primary care HIS Administrative Regional record linkage Cohorts Biobanks Scale 52M subject 7 EU countries 25K AD 95K Metabolics Privacy Population data Cohorts Harmonization Common data model Semantic web technology Analysis & research Disease insights Extreme phenotypes Pharmaco-epidemiology Need for efficiency enhancing workflows Pharma R&D IT Informatics October 2014 5

Skills two worlds Identify data sources Preprocessing, QA, cleaning (Exploratory) Analysis Verify conclusions Interpretation, insights Actions Technical skills Domain expertise Pharma R&D IT Informatics October 2014 6

Handshake packaging your analysis Personal Peer Publication Data analysis & exporation Discussions Communication Graphs that allow a rich analysis may not convey a message very well (and vice versa) Pharma R&D IT Informatics October 2014 7

Example: circadian rhythm gene expression Expression profile of typical genes Pharma R&D IT Informatics October 2014 8

Exploratory analysis: data interactions Overview zoom & filter details on demand A table view of the raw data gives an idea of the variables and values Create overview visualizations, investigate distributions and correlations Use filtering to look at subsets of the data Detailed inspection of groups, outliers, anomalies... Brushing & linking show interactions between variables Pharma R&D IT Informatics October 2014 9

Example: DIAN study DIAN = Dominantly Inherited Alzheimer Network Observational study of genetically inherited early onset Alzheimer s disease 73 families Mutations in amyloid precursor protein, presenilin 1, 2 Age at onset estimated from parent > 600 variables per patient Which baseline measurements are correlated with onset of AD? Pharma R&D IT Informatics October 2014 10

Data overview Show raw data in a table view Overview of measurements and visits Pharma R&D IT Informatics October 2014 11

Explore correlations between measurements General overview of measurement groups Detailed correlation for selected variable(s) Individual observations Pharma R&D IT Informatics October 2014 12

Validated statistics: guided analytics Guide the SME through a series of decisions Give freedom to explore, backtrack, what-if Ensure sound & consistent statistical methods Pharma R&D IT Informatics October 2014 13

Example: clinical phase III trial planning What patient inclusion criteria should be used for the Phase III trial of an AD compound? We expect the compound to work best in patients with mild cognitive impairment (MCI), who are on the verge of converting to Alzheimer (AD) We expect the compound to take some time to show clinical effect. We therefore want subjects not to convert early in the trial. Which measurements are realistic in a trial selection setting? Data from ADNI study (Alzheimer s Disease Neuroimaging Initiative) Pharma R&D IT Informatics October 2014 14

Guided analytics In a guided analysis, advanced statistics are packaged into a wizard-like application to guide the subject matter expert through a decision process 1. How is MCI to AD conversion measured? 2. What timeframe corresponds to early conversion? 3. Which (combination of) baseline measurements are predictive for early vs. late conversion? Use logistic regression modelling 4. For selected baseline measures, what cut-off value should be used as a selection criterion? Use survival analysis Pharma R&D IT Informatics October 2014 15

Step 1: define early conversion Text fields explain decision steps Select visit that separates early from late convertors Select conversion criterion (change in diagnosis or change in clinical dementia rating) Graph shows number of early / late convertors Pharma R&D IT Informatics October 2014 16

Step 2: find predictive covariates Select baseline covariates Verify that groups are balanced Verify variables are uncorrelated Calculate logistic regression model in R Verify model diagnostics Find significant covariates by p-value Inspect ROC for full model, stepwise model and cross validation Pharma R&D IT Informatics October 2014 17

Step 3: find cut-off values Select significant covariates from previous step Verify distribution by conversion group and determine cut-off Enter cut-off value Verify time evolution of cut-off groups Kaplan-Meier plot of conversion rates by cut-off group (calculate in R) Pharma R&D IT Informatics October 2014 18

Conclusions Usage patterns should be supported by applications or licensing model. For example, using Spotfire Informatician / biostatistician prepares guided analysis using full client SME follows prepared analysis to draw conclusions using WebPlayer Think about how you present the data. Your chart encoding should be easy to understand. Conclusions from a guided analysis still need to be confirmed with a statistician. Guided analytics is a good communication tool. Distribute workload. Pharma R&D IT Informatics October 2014 19