Simon Fahn
Running the analysis
Open to Senior and Principal roles, Remote
Senior / Principal Biostatistician

Statistics built
to survive review.

Oncology clinical trials, real-world evidence, and causal inference, across more than 11 years of statistical work. On contract for Eli Lilly and now Jazz Pharmaceuticals, my work covers SAPs, DMC support, interim analyses, and CSR QC. At UC Davis, I have supported 71 studies, including clinical trials.

Start a conversation See the review trail
Survival estimateTwo arms
Time
t = 0
Arm A
1.00
Arm B
1.00
Illustrative curve, not study data. Hover or drag to read S(t).
0
Years of
statistical work
0
Industry SAPs
as primary statistician
0
Studies supported
UC Davis CTSC
0
CSR corrections
before FDA submission
0
Citations
SERRF, first author
The thesis

Every number in a clinical study report should trace back to the plan that asked for it.

When I take over a study, I start by reproducing it. Before one DMC meeting, that habit surfaced efficacy-data errors that could have biased the committee's conclusion. Before one FDA submission, it drove about 50 corrections into a Phase 3 report. Errors rarely live in the model; they live in the handoffs between plan, dataset, table, and text. Working principle, contract oncology work since 2024
Signature case

The Traceability Matrix

On a Phase 3 oncology CSR (LAGOON), I built a matrix tracing every planned analysis to its TLF and its citation in the report, then walked each hop. The pass drove about 50 corrections to tables, numbers, and narrative, including an efficacy-data error caught in the final review round before FDA submission. Select a row to see what gets checked at each hop.

Analysis family
SAP
ADaM
TLF
CSR
Caught
~50
Corrections to tables, numbers, and narrative in a Phase 3 oncology CSR before FDA submission.
Selected row
Efficacy: PFS and OS
    Rows show representative analysis families, not a per-row error count. The flag marks where the late efficacy-data error was caught.
    Pre-DMC
    Took over a study from a departing biostatistician just before a DMC meeting and caught several efficacy-data errors. The corrected data supported the decision to continue the study.
    Final round
    The LAGOON efficacy-data error surfaced in the last review round before FDA submission, the point where a missed number becomes a submitted one.
    Health Canada
    Supported safety and country-level PFS/OS sensitivity analyses for IMforte. Health Canada approved ZEPZELCA plus atezolizumab without conditions in June 2026.
    About

    Two lanes.
    One discipline.

    My work runs in two lanes: oncology clinical trials, and real-world evidence built on causal inference. Both come down to defining the question precisely and proving the answer holds.

    On contract for Eli Lilly and Jazz Pharmaceuticals, I author and revise statistical analysis plans, support Data Monitoring Committees, run interim analyses, and review CDISC SDTM and ADaM specifications. As primary statistician, I authored and signed off the SAP for a Phase 2 single-arm oncology maintenance trial, and revised a pediatric Phase 1/2 SAP with protocol-specified BOIN dose finding.

    At the UC Davis Clinical and Translational Science Center, I bring difference-in-differences, change-in-changes, propensity-score weighting, and time-varying survival models to Epic EHR, OMOP, and national registry data. Before that, as sole biostatistician at the West Coast Metabolomics Center, I led analysis for more than 350 projects and created SERRF, a random-forest normalization method for large-scale lipidomics.

    My publications appear under the name Sili Fan.

    The method

    How a number earns its place

    Four steps, in order, behind the DMC catch, the CSR corrections, and every monitoring package I deliver.

    01 / Specify
    Write the estimand before the code
    For a Phase 2 maintenance trial, that meant defining the 6-month PFS rate estimand, RECIST 1.1 criteria, and censoring rules, then amending the SAP when a protocol change moved the dosing schedule.
    Phase 2 SAP, signed off
    02 / Derive
    Turn CRF fields into rules
    On a pediatric Phase 1/2 SAP with BOIN dose finding, I translated CRF fields into dose-delay, dose-reduction, response, and adverse-event derivation rules a programmer can apply without guessing.
    Pediatric Phase 1/2, BOIN
    03 / Reproduce
    Rebuild it from raw data
    A Phase 2 monitoring package went from raw Medidata Rave extracts to 9 tables, 1 listing, and a 17-slide deck in R, checked by a separate Python QC layer.
    R build, Python QC
    04 / Reconcile
    Resolve every difference
    Independent outputs are compared against the external team's SAS package, and each difference is resolved at its source. On that monitoring package, the programming team accepted the findings.
    Findings accepted
    Career

    Experience

    Oct 2025 / Present
    Senior Study Biostatistician (Contract)
    Jazz Pharmaceuticals via DataTek
    Contract engagements with Jazz and, before it, Eli Lilly were held alongside the UC Davis staff role.
    • Took over a study from a departing biostatistician just before a DMC meeting and caught several efficacy-data errors that could have biased the DMC's conclusion; the corrected data supported continuing the study.
    • Drove about 50 corrections to tables, numbers, and narrative in a Phase 3 oncology CSR, including an efficacy-data error in LAGOON caught in a final review round before FDA submission; built a matrix tracing every planned analysis to its TLF and CSR citation.
    • Authored and signed off the SAP for a Phase 2 single-arm oncology maintenance trial as primary statistician, defining the 6-month PFS rate estimand, RECIST 1.1 criteria, and censoring rules; amended it after a dosing-schedule protocol change.
    • Supported safety analyses answering a Health Canada request for IMforte, plus exposure-adjusted safety and country-level PFS/OS sensitivity analyses; Health Canada approved ZEPZELCA plus atezolizumab without conditions in June 2026.
    • Parallel-programmed a Phase 2 monitoring package in R from raw Medidata Rave extracts (9 tables, 1 listing, 17-slide deck) with an independent Python QC layer; the external programming team accepted the findings.
    • Delivered monthly efficacy and safety monitoring reports (6 cycles on one study, 3 on another) to clinical development, safety, operations, and data-management teams.
    • Reviewed CDISC SDTM and ADaM specifications, ADaM datasets, and the clinical sections of the ZEPZELCA NDA annual report; reconciled PBRER and DSUR totals.
    DMC supportPhase 3 CSRPhase 2 SAPBOIN dose findingHealth Canada
    ~50
    CSR corrections
    before FDA submission
    Oct 2024 / Aug 2025
    Senior Study Biostatistician (Contract)
    Eli Lilly via TechData
    • Parallel-programmed analysis outputs to independently review the external programming team's SAS package.
    • Reviewed CDISC SDTM and ADaM specifications.
    Parallel programmingSDTM / ADaM
    SAS
    Independent
    package review
    Apr 2022 / Present
    Bioinformatics Programmer 3, Biostatistics
    UC Davis Clinical and Translational Science Center (CTSC), Davis, CA
    • Evaluated an ambulatory-access initiative across 229,343 new-patient visits: median wait fell from 26 to 18 days, and change-in-changes attributed an additional 3.38-day reduction (95% CI 2.42 to 4.34) vs comparison patients.
    • Showed a 1.38-minute drop in documentation time per appointment (p<0.001) for 31 physicians piloting an AI documentation tool vs peer controls, using weighted difference-in-differences.
    • Estimated an 8.23-point greater reduction in PROMIS fatigue T-score at week 8 (95% CI 1.88 to 14.59) for qigong vs wait-list in a 39-participant randomized Long COVID trial, using SAS PROC MIXED.
    • Developed LASSO models for pediatric cardiac arrest with 10-fold cross-validation, reaching 0.949 test AUROC on 9,420 later records after training on 11,957.
    • Wrote protocol statistics for a Phase 1 study, contributed design and power justification for a multicenter burn-rehabilitation trial, and contributed to an SAP specifying a planned target-trial emulation.
    Change-in-changesDifference-in-differencesPropensity scoresTarget-trial emulation
    71
    Studies supported,
    incl. clinical trials
    2019 / 2022
    Graduate Student Researcher
    University of California, Davis (CTSC statistical projects)
    • Supported the analysis of 9,758 burn patients from two hospitals, using logistic regression, Cox models, and ROC analysis to evaluate mortality risk.
    • Supported the analysis of a 47-patient randomized trial of virtual-reality distraction during trigger-point procedures, using SAS and baseline-adjusted models.
    • Held a UC Davis-funded research appointment during doctoral coursework in biostatistics.
    Cox modelsROC analysis
    9,758
    Burn patients
    analyzed
    Jun 2015 / Aug 2019
    Principal Biostatistician
    UC Davis West Coast Metabolomics Center (sole biostatistician)
    • Led statistical analysis for more than 350 multidisciplinary projects and contributed statistical input to NIH grant proposals.
    • Created SERRF, a random-forest normalization method for large-scale lipidomics, owning the algorithm, code, validation, and manuscript as first author.
    • Developed metabolomics software adopted by life sciences companies, and trained non-statistician professionals in statistics.
    SERRFNIH grantsSole statistician
    350+
    Projects led
    Evidence

    Estimates, with intervals.

    Three effect estimates from my UC Davis work, each drawn on its own scale from zero. Every interval excludes zero.

    Point estimate and 95% confidence interval
    Additional reduction in median new-patient wait
    Ambulatory-access initiative, 229,343 visits, change-in-changes vs comparison patients
    3.38days, CI 2.42 to 4.34
    Greater reduction in PROMIS fatigue T-score at week 8
    Randomized Long COVID trial, qigong vs wait-list, 39 participants, SAS PROC MIXED
    8.23points, CI 1.88 to 14.59
    More fluid in the first 24 hours with dexmedetomidine
    170-patient burn cohort, ATT propensity-score weighting
    0.61mL/kg/%TBSA, CI 0.07 to 1.15
    0.949
    Test AUROC
    Pediatric cardiac arrest LASSO models, 9,420 later records
    Prediction
    0.778 to 0.576
    AUROC across horizons
    ED admission model, which shaped time-specific performance reporting
    Model evaluation
    26 to 18 days
    Median new-patient wait
    229,343 visits in an ambulatory-access evaluation
    RWE
    22 to 2 days
    Request-to-contact time
    294 pharmacy referrals after an expanded technician role
    Operations
    1.38 min
    Documentation time saved
    Per appointment, 31 physicians, AI documentation pilot, p<0.001
    Diff-in-diff
    9 / 1 / 17
    Tables, listing, slides
    Phase 2 monitoring package rebuilt in R from raw Rave extracts
    Parallel programming
    6 + 3
    Monthly monitoring cycles
    Efficacy and safety reports across two studies
    Jazz
    415
    Google Scholar citations
    SERRF, first author, Analytical Chemistry 2019
    Methods research
    Toolkit

    Methods, standards,
    and data.

    Clinical trials

    SAPsICH E9(R1) estimandsInterim analysesDMC supportSample size and powernQueryEASTPhase 1 to 3 oncologyRECIST 1.1BOIN dose findingStratified randomization

    Clinical deliverables

    CDISC SDTMCDISC ADaMParallel programmingTLF QCCSR QC and traceabilityNDA annual reportPBRERDSUReCRF / UATQTL / KRI plans

    Methods

    Survival analysisMixed modelsPropensity scoresDifference-in-differencesChange-in-changesTarget-trial emulation designBayesian predictive probabilityMultiple imputationLASSO

    Tools

    SAS (macros, PROC MIXED, PROC SQL)RPythonSQLMedidata Rave

    Real-world data

    Epic EHROMOPCARES registryABA National Burn RepositoryHCUP National Inpatient SampleEQ-5D-5LQ-TWiST

    Therapeutic areas

    OncologyInfectious diseaseCardiometabolicCardiovascularRespiratoryNeurologyNephrology and transplantCritical care
    Credentials

    On paper

    Education

    • M.S., BiostatisticsUC Davis, 2015
    • Doctoral coursework, BiostatisticsUC Davis, 2019 to 2022, no degree conferred
    • B.S., MathematicsBeijing Normal University, 2013

    Certifications

    • SAS Certified Advanced Programmer for SAS 9SAS
    • SAS Certified Base Programmer for SAS 9SAS
    • English and MandarinLanguages

    Selected publications

    • Systematic error removal using random forest for normalizing large-scale untargeted lipidomics data. Analytical Chemistry, 2019;91(5):3590-3596.First author, 415 citations
    • Metabox: A toolbox for metabolomic data analysis, interpretation and integrative exploration. PLOS ONE, 2017;12(1):e0171046.Co-first author
    Get in touch

    Have a study that
    has to hold up?

    Open to Senior and Principal Biostatistician, statistical programming, real-world evidence, and data science roles. Based in Davis, California and open to remote. U.S. permanent resident; no sponsorship needed.

    Email Review the work