Analysis of CSRs already published (PhUSE review) Lukasz Kniola, - PowerPoint PPT Presentation

Analysis of CSRs already published (PhUSE review) Lukasz Kniola, Biogen, UK EMA Technical Anonymisation Group (TAG) Meeting 29-30 November 2017

Published CSRs Review Spreadsheet Prepared by PhUSE De-Identification Working Group Will be publically available, searchable. Filters to simplify searches and find similar publications

Categories EMA admin Size Rules Risk Assessment Other Product Name Number of Anonymisation Assessment Data utility considerations Documents method methodology Active Substance Special provision for Number of Studies Narratives Quasi-identifiers: sensitive information Sponsor Overall retained In-text Listings Special provision for non- Product Status Number of Studies Quasi-identifiers: de- subject personal data Subject ID Publication Date with <100 Subjects identified (vendors, PIs, etc.) Dates Procedure Type Number of Studies Population used for AtcCode with 100-<1000 risk calculation Subjects Generic Threshold used Number of Studies Biosimilar with 1000+ Subjects Procedure Number etc.

Publications Over Time 57* 57* 25 60 50 20 40 15 30 10 20 5 10 0 0 Q4-2016 Q1-2017 Q2-2017 Q3-2017 Q4-2017* Publications per quarter Publications overall *) As of 31-Oct-2017

Summary of findings Risk Assessment Methodology Anonymisation Method 45 45 42 40 40 35 35 35 30 30 25 25 20 20 15 15 12 12 10 10 5 4 5 5 3 1 0 0 Redaction Perturbation N/A Fulfilment of Qualitative Quantitative Subjective N/A 3 criteria

Summary of findings – cont.

Summary of findings – cont.

Summary of findings – cont. In-text Listings

Summary of findings – cont.

Common Quasi-Identifiers • Age, sex, race/ethnicity, country, site • Height, weight, BMA, BSA • Medical History • Dates • SAEs, ConMeds, Labs

Quantitative Approach - details General Rules Number of Studies (<100/ Population used Product 100-<1000/ Anonymisation Quasi-identifiers: Quasi-identifiers: de- for risk Threshold Name Sponsor Orphan 1000+ subjects) method Narratives Listings Subject ID Dates retained identified calculation used Alprolix Swedish Orphan Yes 5 (3/2/0) Redaction Fully redacted Redacted Redaction Redaction None Age, race, country, Submission 0.09 Biovitrum AB height, weight, BMI, population SAEs, surgeries, bleeding episodes Zinbryta Biogen Idec Ltd No 11 (4/6/1) Redaction Partially Kept Redaction Year & Study Sex, year of dates, Age, race, country, Submission 0.09 redacted Day coded terms profession, height, population weight Afinitor Novartis No 1 (0/1/0) Perturbation Fully redacted Redacted Perturbation Offset QIs perturbed Gender, age, race, site Similar studies, 0.09 Europharm Ltd ID, date, MH, SAE any sponsor Darzalex Janssen-Cilag Yes 5 (4/1/0) Perturbation Perturbed Perturbed Perturbation Offset QIs perturbed Gender, age, race, site Similar studies, 0.09 International NV ID, date, MH, SAE, any sponsor height, weight, BMI, drinking, smoking habits Imbruvica Janssen-Cilag Yes 3 (0/3/0) Perturbation Perturbed Perturbed Perturbation Offset QIs perturbed Gender, age, race, site Similar studies, 0.09 International NV ID, date, MH, SAE any sponsor

Data utility considerations • Aggregate summaries and analyses have the most scientific value and were kept. • Patient-level data redacted due to: • Rare disease • Sensitive information • Unique information • Unfamiliar process • Presumed very detailed knowledge by plausible adversaries

Questions and Discussion