Biostatistician with an M.S. in Biostatistics (Arizona State University, GPA 3.97) and a B.Tech in Biotechnology. I work on clinical-trial statistics: designing trials, writing and following statistical analysis plans, fitting the models regulators expect (MMRM, Cox, logistic), and checking results in both SAS and R.
| Stage | What I've built | Project |
|---|---|---|
| Design | Sample size, Lan-DeMets group-sequential boundaries, Bayesian go/no-go interims; every number validated against rpact and gsDesign |
trialdesign |
| SAP and estimands | SAP frozen under a git tag before any outcome was seen; ICH E9(R1) estimand; MMRM with missing-data sensitivity analyses | oasis2-mmrm-trial-simulation |
| CDISC deliverables | SDTM → ADaM → TLFs in SAS, with derived datasets checked against the published PHUSE ADaM files (PROC COMPARE) |
cdisc-pilot-sas-adam-tlf |
| Independent validation | Primary MMRM fitted in R (mmrm) and SAS (PROC MIXED, DDFM=KR); estimates and SEs agree within 1e-6 |
oasis2-mmrm-trial-simulation |
| Evidence synthesis / HTA | Random-effects meta-analysis and meta-regression with heterogeneity, publication-bias and leave-one-out checks | bcg-vaccine-trials |
| Real-world evidence | Target trial emulation with inverse-probability weighting and survival outcomes | causal-survival-analysis |
oasis2-mmrm-trial-simulation: MMRM, estimands and trial simulation · R, SAS
Trial-style longitudinal analysis of brain-volume decline in OASIS-2.
- SAP v1.0 frozen before outcomes, with every later change logged as a deviation
- MMRM primary analysis, robustness checks (covariance, time model, visit windows) and MNAR delta-adjustment and reference-based imputation (
rbmi) - SAS/R double programming that agrees to 1e-6
- ADEMP simulation study (888,000 simulated trials) of type I error, power and sample size: 64 per arm for 80% power, 86 allowing for 30% dropout
- Quarto report,
renv-pinned, CI runs the full pipeline on synthetic data
trialdesign: clinical trial design package · Python
CLI and API for designing two-arm trials before the data exist.
- Fixed designs for continuous, binary and time-to-event endpoints
- Group-sequential designs (O'Brien-Fleming / Pocock spending, non-binding futility) solved by recursive numerical integration
- Bayesian beta-binomial interim with predictive probability of success and assurance
- Exact operating characteristics checked against 200,000-trial simulations; validated against
rpact,gsDesignand base R in CI - Includes a design memo written the way I'd present a design to a clinical team
cdisc-pilot-sas-adam-tlf: CDISC SDTM → ADaM → TLF · SAS
Submission-style programming on the public CDISC pilot study (xanomeline vs placebo, 254 randomized): ADSL, ADAE and ADTTE derived from SDTM with dataset specs; disposition, baseline, exposure and adverse-event tables; Kaplan-Meier/Cox with a proportional-hazards check; ANCOVA and a repeated-measures model for ADAS-Cog. Derived datasets are reconciled against the published PHUSE ADaM files with PROC COMPARE, and outputs go to RTF/PDF with saved logs and a validation checklist.
bcg-vaccine-trials: meta-analysis for evidence synthesis · R
Random-effects meta-analysis of 13 BCG vaccine trials (metafor): pooled RR 0.49 (95% CI 0.34–0.70), I² = 92%, and a latitude meta-regression that explains 76% of the heterogeneity. Includes Knapp-Hartung and permutation tests, six τ² estimators, Egger/Begg tests, trim-and-fill and leave-one-out analyses. These are the methods used in HTA submissions (e.g. IQWiG/AMNOG, NCPE).
aact-trial-outcomes-sql-sas: clinical trial registry analysis · SQL, SAS
What predicts early termination and late or missing results reporting in 44k Phase 2/3 drug trials on ClinicalTrials.gov: PostgreSQL cohort build, then SAS logistic and Cox models with sponsor-clustered standard errors.
causal-survival-analysis: target trial emulation · Python
ACEi/ARB use and cardiovascular death in type 2 diabetes with CKD (NHANES 2005–2018, n = 2,540). Covariate balance after weighting is good (largest SMD 0.051) but the estimate is inconclusive. The repo explains why (residual healthy-user bias, sample size) and validates the methods on simulations with a known truth.
| Project | What it is | Stack |
|---|---|---|
| multi-omics-biomarker-pipeline | Leakage-controlled clinical + 5-gene survival signature for lung adenocarcinoma (TCGA-LUAD), frozen and validated on two external cohorts (Harrell C 0.715 and 0.697) | Python |
| myco-rsm | Response surface / design-of-experiments media optimizer for fermentation | R, Shiny |
| adni-subgroup-calibration-code | In progress: calibration audit of MCI-to-dementia risk models across landmark times and demographic subgroups in ADNI (code only; data cannot be redistributed) | R |
Languages: SAS, R, Python, SQL
SAS: PROC MIXED, PHREG, LIFETEST, LOGISTIC, SURVEYLOGISTIC, FREQ, MEANS, REPORT, COMPARE, SQL, macros, ODS RTF output
R: mmrm, emmeans, nlme, lme4, survival, rbmi, mice, metafor, rpact, gsDesign, tidyverse, Shiny, Quarto
Python: numpy, scipy, pandas, statsmodels, lifelines, scikit-survival, scikit-learn
Clinical standards: CDISC SDTM and ADaM, ICH E9 and E9(R1) estimands, SAP writing, double programming / QC
Methods: group-sequential and Bayesian trial design, mixed models for repeated measures, survival analysis, missing data (MAR/MNAR, reference-based imputation), causal inference, meta-analysis, simulation studies (ADEMP)
Reproducibility: Git, GitHub Actions CI, renv, pytest, synthetic-data pipelines for confidential data
- M.S. Biostatistics, Arizona State University (2025, GPA 3.97/4.00)
- B.Tech Biotechnology, Manipal Institute of Technology (2023)
- Data Science Analyst, ASU × Global Observatory for Physical Activity (GoPA!): built a country-year panel for six African countries, compared naive-drift, ETS and ARIMA forecasts to 2030, and screened predictors with PCA and lagged cross-correlation (confidential data, not shared)
- iGEM, Manipal team: Bronze (2021) and Gold (2022)
- Lab background in microbiology, molecular biology, PCR and bioprocess QC, so I can talk to the scientists whose data I analyse
Writing up the ADNI calibration audit and learning German (A1).
Email is the best way to reach me: rushadisatwork@gmail.com


