Methodology · SVRS
Statistically Valid Random Sampling for healthcare claim audits.
AEGIS ISD's Statistically Valid Random Sampling (SVRS) module runs claim-sample selection and overpayment extrapolation aligned with HHS-OIG RAT-STATS conventions. Cochran's-formula sample sizing, Fisher-Yates random selection, and Mean Per Unit extrapolation produce defensible, reproducible audit outputs that meet the methodology expectations of state Medicaid program integrity, CMS RADV/RAC, and federal-program review.
Last updated: April 25, 2026
Why SVRS matters
Audit defensibility from the first sampled claim.
State Medicaid program integrity
State Medicaid Fraud Control Units (MFCUs) and Program Integrity Units must demonstrate statistically valid sampling and extrapolation when projecting overpayments to a provider universe. AEGIS SVRS produces the parameter set, selection trace, and Mean Per Unit projection that state and federal review boards expect to see in the audit work paper package.
CMS RADV / RAC and Medicare Advantage
CMS Risk Adjustment Data Validation, Recovery Audit Contractor reviews, and similar Medicare Advantage activity require methodology disclosures that mirror HHS-OIG RAT-STATS conventions. AEGIS SVRS output formats are designed to drop directly into those disclosures with a methodology PDF attached to every sample run.
False Claims Act exposure mitigation
When investigations escalate to qui tam or FCA referral, methodology defensibility is no longer optional. AEGIS SVRS retains an immutable record of the universe definition, parameters, random seed, selection set, determinations, and extrapolation, supporting cooperation with the U.S. Department of Justice and OIG when appropriate.
Methodology
Three pillars: sizing, selection, extrapolation.
SVRS computes the sample, draws it without bias, and projects results back to the universe with explicit confidence intervals. Every step is parameterized, logged, and reproducible.
Sample-size determination — Cochran's formula
Sample size is computed from the user-supplied confidence level (commonly 90%, 95%, or 99%), margin of error, and expected proportion of error. The result is corrected for finite-population effects (FPC) when the universe is small relative to the unbounded sample. AEGIS surfaces every input value and the derived sample size in the methodology PDF, so auditors can replicate the calculation by hand if needed.
Random selection — Fisher-Yates shuffle
AEGIS uses a cryptographically seeded Fisher-Yates shuffle to draw the sample without bias. The seed is retained alongside the sample run, making the selection fully reproducible by AEGIS and by an external reviewer who has the seed. No iterative or stratified-bias selection patterns are used at this layer; stratification, when desired, is applied to the universe before SVRS runs.
Extrapolation — Mean Per Unit
Once medical-review determinations are recorded against each sampled unit, AEGIS computes the Mean Per Unit point estimate, the standard error, and the lower and upper confidence bounds. The projected overpayment is the point estimate scaled to the universe size, with the lower bound typically used as the conservative recovery target consistent with regulator practice.
RAT-STATS alignment
Built to mirror HHS-OIG RAT-STATS output.
The HHS-OIG RAT-STATS application is the reference sampling and extrapolation tool for federal healthcare audits. AEGIS SVRS is intentionally aligned with RAT-STATS conventions for parameter input, random-seed disclosure, sample-size derivation, and Mean Per Unit output, so AEGIS-produced methodology packages are immediately recognizable to OIG, CMS, and state auditors.
Parameter parity
Confidence level, margin of error, expected proportion, finite population correction, universe size, and stratification metadata are surfaced in the same conceptual structure RAT-STATS uses, so an auditor familiar with RAT-STATS can read an AEGIS SVRS report without translation.
Output parity
Sample size, random seed, selected unit list, point estimate, standard error, lower confidence bound, upper confidence bound, and projected overpayment are all surfaced in the AEGIS-generated methodology PDF in a layout designed to drop into a regulator submission with minimal reformatting.
Workflow integration
SVRS plugs into the investigation lifecycle.
SVRS is not a standalone calculator. It runs inside the AEGIS investigation workspace, taking selected claim universes from SIU case management and fraud detection, routing the sampled subset into medical review, and feeding the extrapolated result into recovery tracking.
1. Define the universe
Investigators or analysts define the claim universe in the AEGIS workspace using filters on provider, service period, claim type, dollar band, or any other indexed attribute. The universe size and key attributes are captured for the audit log.
2. Configure parameters
Confidence level, margin of error, expected proportion, and finite-population correction are configured through a guided form. The form previews the Cochran's-formula sample size before the sample is drawn so reviewers can validate parameters first.
3. Generate the sample
AEGIS draws the sample using a cryptographically seeded Fisher-Yates shuffle. The seed and the ordered selection set are persisted to the immutable sample-run record. The sample is routed to medical review with the case context the reviewer needs.
4. Determine each sampled unit
Reviewers complete medical-review determinations on each sampled unit using the Medical Review module — including the QA Analyst stage for peer verification before final determination — and AEGIS records the dollar impact of each determination.
5. Extrapolate to the universe
AEGIS computes the Mean Per Unit point estimate, standard error, and lower and upper confidence bounds, scaled to the universe size. The projected overpayment is presented to the recovery owner with the lower confidence bound highlighted as the conservative recovery target.
6. Generate methodology PDF
AEGIS generates a methodology PDF v2.0 capturing every input, the derivation, the seed, the selection set, the determinations, the extrapolation, and a step-by-step reproducibility appendix. The PDF attaches to the case, the recovery batch, and the audit log for regulator submission.
Compliance and defensibility
Reproducible, immutable, regulator-ready.
Immutable audit trail
Every SVRS run captures the universe definition, parameter set, derivation, random seed, selection set, determinations, and extrapolation in an immutable audit record. AEGIS does not allow retroactive parameter changes; a methodology revision creates a new sample run with its own seed and audit record.
HIPAA-aligned safeguards
SVRS runs inside the AEGIS investigation workspace under the same role-based access, audit logging, schema-per-tenant isolation, and Business Associate Agreement framework described on our Security and HIPAA pages.
Methodology PDF v2.0
The methodology PDF is generated server-side from the audit record. It is content-identical across regenerations of the same sample run and includes a hash so a reviewer can verify the document was produced from the persisted audit record and not edited.
Regulator-ready output package
The sample-run package — methodology PDF, selection set CSV, determination records, and extrapolation summary — is exportable as a single archive ready for inclusion in a regulator submission, MFCU referral, or DOJ cooperation file.
FAQ
Common questions about AEGIS SVRS.
What sample sizes does AEGIS SVRS support?
Sample sizes are computed dynamically from the user-supplied confidence level, margin of error, and expected proportion using Cochran's formula, with finite population correction for the universe size. Typical SIU and program-integrity audits range from 30 to a few hundred sampled units, but AEGIS supports any sample size the parameters and population justify.
Is the methodology defensible to OIG, CMS, or state Medicaid auditors?
AEGIS SVRS is aligned with HHS-OIG RAT-STATS sampling conventions, producing the same statistical artifacts auditors expect: confidence level, margin of error, sample size derivation, random selection seed, and Mean Per Unit point estimate with lower and upper confidence bounds. A methodology PDF is attached to every sample run for inclusion in the audit work paper package.
Can we use our own confidence and error parameters?
Yes. Confidence level (typically 90%, 95%, or 99%), margin of error, expected proportion, and finite-population correction are all configurable per sample run. AEGIS retains the full parameter set in the immutable sample audit log so the methodology is reproducible.
How does AEGIS handle stratified versus simple random sampling?
AEGIS SVRS supports simple random sampling out of the box. Stratified designs (for example, by claim type, provider, dollar band, or service period) are supported by stratifying the universe before SVRS runs, with per-stratum sample sizes computed independently and per-stratum extrapolations rolled up into the final point estimate.
What does the methodology PDF v2.0 include?
The methodology PDF includes the universe definition, the parameters used (confidence, margin, expected proportion, finite-population correction), the Cochran's-formula derivation, the Fisher-Yates seed and selection trace, the determinations recorded against each sampled unit, the Mean Per Unit point estimate with lower and upper bounds, and a step-by-step reproducibility appendix that mirrors the format of HHS-OIG RAT-STATS output.
How does SVRS connect to medical review and recovery?
SVRS sample runs route the selected universe directly into Medical Review with appropriate clinical context. Once determinations are complete, AEGIS extrapolates the result back to the universe and surfaces the projected overpayment in Recovery Tracking, with the methodology PDF attached to the recovery batch for regulator submission.
See SVRS in action
Tailor a demo to your audit and investigation workflows.
Tell us about your current sampling tools, the program-integrity activity SVRS would support, and the regulators or contracting bodies whose methodology disclosures matter to you. We will tailor a demo to show how SVRS fits the way your team actually runs audits.
Demo includes
- Live walkthrough of universe definition and Cochran's-formula sample sizing
- Fisher-Yates seeded sample selection with reproducibility verification
- Routing the sample into the Medical Review QA-stage workflow
- Mean Per Unit extrapolation with confidence interval surfacing
- Methodology PDF v2.0 generation and regulator-ready output package