Work with me

Biostatistics consulting for medical researchers, small companies, and nonprofits — and who’s behind it.

I’m a physician and biostatistician, and I take on biostatistics consulting for medical researchers, small companies, and nonprofits, on the methodological pieces of clinical and health-services research, including the parts that have to hold up under journal-club critique, IRB review, peer review, and regulatory scrutiny.

Every recommendation a clinician follows sits at the top of a chain that runs down through a decision rule, a trial, an estimate, and a model, to a raw measurement. Walking that chain is what I call a trace. The four surfaces below are where I work on it for clients, whether you are building the chain from the study up or stress-testing someone else’s. Each can be a one-off engagement (a methods review, a sample-size memo, a methodology section, a sensitivity-analysis design) or part of a longer-term advisory relationship through study design, analysis, and write-up.

Study design

What I help with. Research-question framing, target trial emulation, study population definition, inclusion / exclusion criteria, sample size and power calculation, treatment-comparator-outcome (PICO / PECO) specification, primary versus secondary endpoint selection, protocol writing, pre-registration drafting.

On the pathway. The top rungs: turning a clinical question into a recommendation-worthy design before any data exist.

Typical engagements. A resident designing their first observational study and needing a defensible analysis plan before data collection. A small company scoping a clinical or observational trial and trying to translate a clinical hypothesis into a statistical one. An NGO running a program evaluation that needs to be designed for credible causal claims rather than just descriptive reporting.

Applied trace. The Medicare Part D insulin DiD case study walks through how a clean difference-in-differences design is constructed: simultaneous treatment, honest control group, parallel-trends defense, placebo tests, leave-one-out robustness.

Analysis methods

What I help with. Causal inference (difference-in-differences, regression discontinuity, instrumental variables, synthetic control, target trial emulation), propensity score methods (matching, weighting, doubly-robust estimators), survival analysis, longitudinal data methods (mixed effects, GEE), Bayesian methods for clinical and health-services applications, Monte Carlo simulation for sample size or bias quantification.

On the pathway. The estimate and the model under it: producing an effect size that means what it is taken to mean.

Typical engagements. An attending whose RCT was disrupted by an external event and now needs causal-inference repair work to salvage the analysis. A company validating a real-world evidence claim against regulatory expectations. A researcher whose existing analysis needs strengthening before submission to a methods-rigorous journal.

Applied trace. The Medicaid outliers case study walks through robust statistics on heavy-tailed data, BH-FDR multiplicity correction, isolation-forest triangulation, and the methodological subtleties of applying these tools at scale on claims data.

Sensitivity and robustness

What I help with. Pre-specified sensitivity analyses, placebo and falsification tests, robustness across model specifications, leave-one-out diagnostics, bias quantification (E-values, Rosenbaum bounds), unmeasured-confounding sensitivity, missing-data sensitivity (multiple imputation, pattern-mixture models), structural-overlap diagnostics for comparator analyses.

On the pathway. Stress-testing the estimate and its assumptions, so the rungs above it can hold.

Typical engagements. A manuscript revision where reviewers requested additional sensitivity analyses and the authors need design help on which to run and how to report them. A regulatory submission where the headline result needs defensible robustness work. A methods peer-review job for a journal or funder.

Applied trace. The NHANES cardiometabolic case study walks through calibration over discrimination as the load-bearing metric, case-definition sensitivity bands, and the structural-overlap trap in comparator analyses.

Methods writing

What I help with. Methods sections for manuscripts, grant applications, IRB protocols, and regulatory submissions; reviewer responses on methodology; methods primers and statistical analysis plan (SAP) drafting; methods appendices and supplementary methods writing; methodology frameworks for emerging areas (such as clinical AI and real-world evidence).

On the pathway. Making the whole trace legible to a reviewer, an IRB, or a regulator.

Typical engagements. A resident writing up their first manuscript and needing a methods section that survives peer review. An attending facing reviewer methods queries that require a careful written response. A company assembling a regulatory submission whose methods documentation needs to meet FDA or EMA expectations. A methodology-focused position paper or framework that needs to be drafted, structured, and reviewed.

Portfolio example. The Risk-of-Bias appraisal for AI training corpora framework is an example of structured methods writing: six bias domains with inline signaling questions, a stylized worked example end-to-end, explicit limitations and open questions. The voice and structure of that piece is the kind of methods writing I produce for client engagements.

How an engagement starts

Every engagement starts with a free 20-minute discovery call to scope the question, the timeline, and the deliverable. The call is non-committal; you leave it knowing whether the work is a fit and what a proposal would look like.

After the call, you receive a written proposal with scope, methodology, deliverable, timeline, and fee. Each proposal is written to the specific study; there is no flat-rate price list, and the fee depends on the scope you actually need rather than a tier you have to fit into. Most one-off engagements (methods review, sample-size memo, methodology section, sensitivity-analysis design) complete in two to six weeks. Longer advisory relationships (study design through analysis through write-up) are scoped as multi-month proposals, typically over three to six months.

If you want the methodology reference in addition to (or instead of) a scoped engagement, the From Data to Bedside pathway covers study design, populations and sample sizes, causal inference, sensitivity analysis, and Monte Carlo simulation. The pathway and the first chapter are free to read; the full long-form chapter behind each rung is available to subscribers, alongside dispatches on new research.

About

By day I’m a Clinical Data Scientist at the Institute for Health Metrics in Manchester, MA, building real-world-data analyses (EHR, claims, and social-determinants measures) across 24.6M+ patient encounters from 50+ community hospitals. Outside that role I take on the consulting work above.

My MPH is in epidemiology and biostatistics from Johns Hopkins Bloomberg School of Public Health, with a Public Health Economics graduate certificate. Before moving full-time into data science I practiced general and occupational medicine in the Philippines, led a systematic review whose findings shaped the Philippine Department of Health’s Wilms tumor chemotherapy guidelines, and ran a national advocacy program that integrated medical certification of cause of death into the country’s medical school curriculum. I keep one foot in each world (clinic, dataset, policy) because the questions worth modeling are the ones clinicians and patients actually live. It is also what lets me trace a recommendation in both directions: down from the bedside guideline to the datapoint it rests on, and back up. Methodologists tend to own the bottom of that chain and clinicians the top; the work I care about is walking the whole thing.

What I’m working on now

Toolkit

I work fluently in R (tidyverse, data.table, survminer, Shiny), Python, and SQL on AWS, with Stata and SAS in reach. I’m comfortable across ICD-10, RxNorm, LOINC, and SNOMED-CT, and methods I reach for routinely include difference-in-differences, target trial emulation, propensity score methods (matching, weighting, doubly-robust estimators), negative-binomial and ordinal-logistic regression, Markov ICERs, PCA-based SDoH composites, predictive modeling for mortality and readmission risk, and Monte Carlo simulation for bias quantification and sample size. On the synthesis side: GRADE, PRISMA, Cochrane RoB 2, ROBINS-I, AMSTAR, TRIPOD-AI, PROBAST, decision-curve analysis, calibration.

The full record (publications, training, professional memberships) is on the CV.

Contact

The fastest path is a brief note over email. I respond within two business days.

For consulting inquiries and discovery calls, send a brief note. For a general methodology question, the same address works; put “Methods question” in the subject line so it triages cleanly.

What an inquiry should include

To make the discovery call efficient, please include:

  1. The research question or methodological problem in one or two sentences.
  2. The current stage of the work (literature review, study design, data collection, analysis, manuscript draft, reviewer revision, regulatory submission).
  3. The timeline (when the work needs to be done by, or whether there is an external deadline such as a grant cycle or a journal revision window).
  4. The format you’re hoping for (one-off methods review, sample-size memo, full study-design engagement, methods section for a paper or grant, ongoing methodological advisor relationship).

A scoped inquiry of this shape is enough to scope the engagement on a 20-minute call. Underspecified inquiries are welcome too, but the call will spend more time on scoping and less on the methods.