Research Standards

These are the standards that guide RxEconomics work—economic models, real-world evidence studies, policy analyses, expert reports, and publications. They are also what the phrase Expert-Verified AI for HEOR means in practice: AI accelerates the work, and verification is treated as the product. Engagements are led and reviewed by Dr. M. Christopher Roebuck, PhD, MBA.

Where Does Our Evidence Come From?

Real-world evidence is only as credible as its data. We work with client-provided, licensed, or public-use data—commercial claims such as MarketScan and Optum, Medicare (both fee-for-service and Medicare Advantage), Medicaid (including MAX, SDUD, and TAF/T-MSIS), electronic health records, registries, and national surveys—and we identify the data sources each study relies on.

Published findings are cited to their original source—the journal article, report, regulation, or official record itself—rather than to a secondhand summary. Our own bibliography follows the same approach, linking to the original publication wherever a link is available.

How Is AI-Assisted Work Verified?

A chatbot will draft a protocol that reads well—that does not mean it is correct. A model will produce code that runs—that does not mean it implements the estimand. So the most valuable AI work is not generation; it is comparison. Our verification process is built around checking:

1. The protocol against the statistical analysis plan
2. The statistical analysis plan against the code
3. The code against the tables and figures
4. The tables against the claims made in the final deliverable
5. Cited claims against their cited sources

In a systematic literature review, AI may triage abstracts and extract data elements, while a senior researcher adjudicates inclusion decisions and verifies extracted values against their sources. In claims research, AI-drafted code sets across ICD-10, CPT, HCPCS, and NDC are reviewed so that coding decisions are visible rather than buried. The full workflow is described on our AI-powered HEOR page.

Where Do Humans Stay in the Loop?

Everywhere judgment lives. People—not models—define the research question, the estimand, and the study design; make the methodological judgment calls that materially move answers; adjudicate AI-drafted code sets, cohort definitions, and literature screening decisions; interpret results in their clinical, payer, and policy context; and review deliverables before they go out. AI drafts and accelerates. It does not decide.

Reproducibility and Audit Trails

Evidence that cannot be reproduced is hard to trust. Model assumptions are documented where a reader can find them. Coding decisions are traceable back to the protocol language that produced them. Cohort definitions in code are checked against the cohort described in the protocol. The aim is that another qualified team, given the same inputs, could follow the work and challenge any step of it.

Confidentiality and Data Use

Client data, findings, and engagement details are kept confidential. Data are handled under the terms of each client agreement and any applicable data use agreements, and licensed and public-use datasets are used within their license terms.

Reporting and Disclosure

We document AI use in a manner consistent with emerging reporting guidance, such as the ISPOR Working Group's ELEVATE-GenAI framework for large language model use in HEOR. When our work goes to a payer, an HTA body, or a journal, we document and disclose how AI was used.

We treat that guidance as a floor, not a ceiling. Reporting guidance tells readers what AI did; it does not establish whether the result is right. That is the job of the verification process described above—comparing protocol to analysis plan, analysis plan to code, code to tables, and claims to their sources—which goes beyond what disclosure alone requires.

Public policy work is labeled for what it is. Our comments to CMS on Medicare drug price negotiation, for example, state that they were prepared independently—no client funded, requested, commissioned, or reviewed them. Sponsored work that is published carries the funding and conflict-of-interest disclosures required by the publishing journal.

Peer Review

Peer-reviewed rigor is not a slogan here; it is a track record. Dr. Roebuck's research has passed peer review at journals including Health Affairs, Medical Care, the American Economic Review, and The American Journal of Managed Care, and the same methodological standards inform client engagements. The complete record—120+ publications & research reports—is on our publications page.

Corrections Policy

If you believe something published on this site is in error, email contact@rxeconomics.com. We review reports we receive, and where a correction is warranted, we update the page and note the date of the change.