Drug safety, real-world evidence,
and claims analytics — done rigorously.
I work with pharmaceutical and biotech companies, health technology firms, insurers, and legal teams on the questions that turn on epidemiologic judgement: is this signal real, what does the real-world data actually support, and will this evidence hold up under scrutiny?
My training sits in an unusual place. I am a pharmacist by first training and an epidemiologist by doctorate — which means I read a safety signal as both a clinical and a statistical object, and I know where each one misleads.
Who I Work With
Different sectors ask different questions of the same data. These are the four settings where my methods and background line up most directly with what the work actually requires.
Safety, evidence, and the questions regulators will ask
Support across the product lifecycle — from pre-submission evidence planning through post-marketing surveillance commitments and label-change evidence packages.
- Safety signal detection, triage, and causality assessment
- Post-marketing surveillance strategy and PBRER/PSUR analytic support
- Real-world evidence to support label expansion or payer dossiers
- Comparative safety and effectiveness studies in claims and EHR data
- Independent methodological review before regulatory submission
Making clinical AI defensible, not just accurate
Most clinical AI is validated once and deployed indefinitely. Drug safety solved this problem decades ago with continuous post-market surveillance; diagnostic and predictive models need the same discipline.
- Validation study design for diagnostic and predictive models
- Continuous performance monitoring and drift detection frameworks
- Evidence standards for regulatory and payer review
- Bias, generalisability, and external validity assessment
- Post-deployment safety surveillance for deployed models
What administrative data can and cannot tell you
Claims data is abundant and treacherous. The methods that make it trustworthy — new-user designs, active comparators, high-dimensional confounding control — are the core of my applied work.
- Cohort construction and algorithm validation in claims data
- Utilisation, cost, and cost-effectiveness analysis
- Risk adjustment and population segmentation
- Medicare, Medicaid, and commercial claims study design
- Policy impact evaluation using quasi-experimental methods
Epidemiologic evidence that survives cross-examination
My legal epidemiology fellowship focuses precisely on how law and evidence interact — quantifying policy effects on population outcomes, and assessing whether a causal claim is defensible.
- General and specific causation assessment
- Independent review of opposing epidemiologic analyses
- Policy impact quantification using difference-in-differences designs
- Literature synthesis and weight-of-evidence evaluation
- Written reports and technical consultation
How I Can Help
Each engagement has a defined question, a defined output, and a defined end point. If your problem doesn't match any of these, describe it and I'll tell you honestly whether I'm the right person — and if I'm not, I'll usually know who is.
Safety Signal Assessment
Disproportionality analysis in FAERS or VigiBase, followed by the part most analyses skip — a structured evaluation of whether the signal survives confounding by indication, channelling, notoriety bias, and reporting artefacts. Screening statistics are treated as hypothesis-generating, never as evidence.
Deliverable: written signal evaluation with methods appendix and recommendationReal-World Evidence Study Design
Protocol development for observational studies in claims, EHR, or registry data — target trial emulation, new-user active-comparator design, DAG-based confounder selection, and a pre-specified analysis plan. Designed so the result is defensible before the data is touched, not rationalised afterward.
Deliverable: full study protocol and statistical analysis planClaims & Health Economics Analytics
End-to-end analysis in Medicare, Medicaid, or commercial claims — cohort definition, algorithm validation, utilisation and cost outcomes, and cost-effectiveness modelling. Includes explicit assessment of what the data cannot support, which is often the most valuable part of the report.
Deliverable: analytic dataset, results report, and reproducible codeClinical AI Validation & Monitoring
Independent evaluation of a diagnostic or predictive model's clinical evidence base: validation design, subgroup performance, calibration, external validity, and a continuous monitoring framework adapted from pharmacovigilance practice so degradation is detected rather than discovered.
Deliverable: evidence assessment and monitoring framework specificationMethodological Review & Second Opinion
A senior independent read of an existing protocol, manuscript, regulatory submission, or opposing expert report. Fast, focused, and written to be circulated internally — where the analysis is sound, where it is vulnerable, and what a reviewer or opposing counsel will attack first.
Deliverable: written critique with prioritised findings — typically within two weeksOngoing Scientific Advisory
Continuing access for teams that need epidemiologic judgement on a recurring basis — a standing call plus asynchronous review of protocols, analyses, and evidence questions as they arise. Suited to safety, medical affairs, HEOR, and clinical AI teams without in-house senior epidemiology.
Deliverable: scheduled advisory sessions and continuing reviewWhy This Background Fits
Akshaya Bhagavathula, PharmD, PhD, FACE
I trained first as a clinical pharmacist, then completed a PhD in epidemiology and biostatistics. That combination is uncommon and it matters here: drug safety questions are simultaneously pharmacological and epidemiological, and analysts who hold only one half of that tend to either over-read a disproportionality statistic or miss a mechanism that explains it.
Before academic research I worked as a field epidemiologist with WHO, USAID and UNICEF across Africa and the Middle East — surveillance built under real constraints, where the analysis had to support a decision the same week. That shapes how I scope work now: the deliverable is a decision, not a manuscript.
My published work spans pharmacovigilance and GLP-1 safety, telehealth and claims-based policy evaluation, geospatial access analysis, and infodemiology. I am a Legal Epidemiology Research Fellow at the Center for Public Health Law Research, Temple University Beasley School of Law, and a named collaborator on the IHME Global Burden of Disease study.
Consulting is conducted independently through EPIAIDEA and is separate from any academic appointment.
Engagement Models
Advisory call
A single focused session on a specific question. Useful when you need a senior read before committing to a direction.
Fixed-scope project
Defined question, defined deliverable, defined timeline. Most analytic and study design work runs this way.
Retainer
Continuing availability for teams with recurring epidemiologic questions across a portfolio.
Expert & litigation support
Causation assessment, independent review of opposing analyses, and written technical reports.
If you think I can help, please reach out.
You don't need a finished brief or an approved budget to start a conversation. If you have a question that sits somewhere between pharmacology, data, and evidence — describe it in a few sentences and I'll tell you honestly what I think.
If it's a good fit, I'll propose a scope. If it isn't, I'll say so directly, and where I can I'll point you toward someone better suited. I'm glad to help, and there's no obligation in asking.