High-Integrity Health Data Science
MD-trained RWE scientist and statistical programmer with 20+ years translating population-scale healthcare data into peer-reviewed evidence and regulatory-ready outputs. First author on 4 peer-reviewed studies; co-author on 25 publications spanning COPD, asthma, and respiratory health economics. Methodological foundation of the 2025 PRECISE-X COPD exacerbation prediction model (Thorax). Expert in CDISC SDTM/ADaM, causal inference, and real-world data platforms including CPRD and OMOP-CDM.
25 peer-reviewed publications · 20+ years in health data science · CPRD · OMOP-CDM · CDISC
Selected Publications
A selection of methodological contributions and clinical research papers demonstrating rigorous statistical programming and clinical data integrity.
COPD PREDICTION Thorax · 2025
Development and validation of PRECISE-X: predicting first severe exacerbation in COPD
Sadatsafavi M, Miravitlles M, Quint JK, Perugini V, Tavakoli H, et al.; REG-COPD working group.
COPD SURVEILLANCE Ann Am Thorac Soc · 2020
Predicting severe COPD exacerbations: a population surveillance approach with administrative data
Tavakoli H* (first author), Chen W, Sin DD, FitzGerald JM, Sadatsafavi M.
ASTHMA ECONOMICS Allergy · 2016
Ten-year trends in direct costs of asthma: a population-based study
Tavakoli H* (first author), FitzGerald JM, Chen W, Lynd L, Kendzerska T, Aaron S, et al.
71:371–7
RWE Insights
Technical writing on causal inference, survival modeling, and CDISC pipelines for regulatory and research audiences
COPD · PREDICTION MODELING JANUARY 2026
Inside PRECISE-X: How Cox-Lasso Built a Clinically Deployable COPD Exacerbation Model
Preserving Clinical Truth
The PRECISE-X model (Thorax, 2025) traces its statistical core to a Cox-Lasso framework I developed at UBC on CPRD data. I walk through the modeling choices, variable selection under regularization, the bias-variance tradeoff in sparse survival models, and why Lasso outperformed stepwise selection for this application.
CAUSAL INFERENCE · RWE MARCH 2025
A protocol is only as good as the statistical program that executes it. Elegant code makes the science undeniable.
IPTW vs. Propensity Score Matching: When Each Method Belongs in an RWE Study
Both methods adjust for confounding in observational data, but they answer subtly different questions. Drawing from 20 years of population-based respiratory studies, I compare the two approaches on estimand choice, variance behavior, and regulatory acceptability, with worked examples from COPD cost and safety analyses.
CDISC · ADaM · SAS OCTOBER 2024
Traceability by Design: Structuring ADaM Datasets So Reviewers Never Have to Ask
Regulatory reviewers should be able to trace any TLF cell back to raw data without chasing the programmer. I share patterns from oncology ADaM delivery at ICON — from SUPPQUAL design to derivation variable documentation, that make traceability a property of the dataset, not a retrospective audit.
High-Integrity Data Pipelines
Available for elite consulting, technical leadership, and statistical programming advisory for global CROs and pharmaceutical research teams.