BailiwickDx Strategies

Approach

Estimates grounded in data, not in opinion.

Rare disease numbers are small, noisy and easy to get wrong in either direction. An overestimate sends a launch after patients who don’t exist; an underestimate leaves real patients out of the plan. The method below is built to avoid both, and to show its working at every step.

The method

From “how many have it” to “how many can we help find.”

The same five steps underpin every Benchmarking and Blueprint engagement. The worked example shows each one applied to a hypothetical condition.

  1. Triangulate prevalence

    Birth or population prevalence is estimated from several independent lines of evidence, such as exome-cohort shares, case accrual against a better-characterized analog, de novo mutation-rate ceilings, and population variant frequencies. Where they agree, confidence rises. Where they disagree, the range says so.

  2. Anchor on how patients present

    Patients don’t arrive labelled with a gene. They arrive with seizures, short stature, developmental delay, or an abnormal newborn screen. The funnel starts from those presentations and follows each step: recognized, referred, tested, tested with something that can find the variant, and reported as pathogenic.

  3. Separate true from diagnosed

    True prevalence and diagnosed prevalence are modeled separately, in unique patients, by age band. That separates the near-term, findable pool from the long-horizon one, and accounts for when a gene–disease link was discovered and when testing actually reached each age group.

  4. Move one lever at a time

    Each candidate lever (sponsored testing, EHR triggers, reanalysis, adult re-testing, panel content, screening) is modeled by changing one input from the central case, then in combination. The result is a ranked list in diagnosed patients per unit of investment, not a list of good ideas.

  5. Say what is assumed

    Every deliverable separates what is well anchored from what is assumed, flags the assumptions that drive the answer, and names the data that would narrow the range. The model is handed over live, so your team can change any input and see the effect.

Evidence base

The sources behind the numbers

Each estimate draws on whichever of these the condition allows, and every input carries a reference.

Testing & variant data

  • ClinVar submissions & trends
  • gnomAD population frequencies
  • Lab panel content
  • Sponsored-testing yield
  • Exome & genome cohort studies
  • VUS reclassification rates

Population & clinical data

  • Claims & EHR (de-identified)
  • Vital statistics & census
  • Prospective incidence cohorts
  • Natural history & registries
  • Newborn screening programs
  • GeneReviews & literature

Tooling

Built, not just recommended.

Bailiwick runs its own analysis tools, so a first read on a condition takes days rather than weeks and every engagement starts from the same tested base.

  • Diagnostic odyssey atlas. A disease-specific map of where patients leak from the path to diagnosis (no test ordered, wrong test, VUS, lost to follow-up), with each leakage point weighted for the condition and paired with levers and precedents.
  • Gene variant dashboard. Gene-level ClinVar submissions, classifications and trends over time, alongside population frequencies, to see how quickly a gene’s evidence base and VUS burden are changing.
  • Diagnosis-rate model. A reusable, disease-agnostic funnel and projection model with a library of calibrated catalysts, delivered to clients as a live spreadsheet.

Principles

What every deliverable holds to

i.

Unique patients, not visits

Encounters, claims and test orders are not patients. Every count is de-duplicated to people before it is reported.

ii.

Method shown

Inline citations, AMA references, a live model, and the statistical test stated wherever two numbers are compared.

iii.

Honest ranges

Where a condition has fewer than 30 described individuals, the estimate says so, and the range is wide because it should be.

iv.

Patients first

A lever is only worth pulling if it brings a real person to a real answer sooner. Programs that inflate numbers without doing that are not recommended.

Compliance by design

Programs that pass legal review the first time.

Sponsored testing and patient-identification programs sit close to the line between scientific exchange and promotion. OIG advisory opinions have now set out workable guardrails for manufacturer-sponsored testing, and each new opinion refines them.

Bailiwick designs programs inside those guardrails from the start rather than retrofitting them, so your legal and compliance reviewers see a familiar structure.

Typical guardrails built in

  • Clinical eligibility criteria set independently of commercial targets
  • No data flows to the sponsor that could enable promotion to individual patients or prescribers
  • Fixed, fair-market lab fees unrelated to product volume
  • Testing that informs care whether or not the result points to the sponsor’s therapy
  • Genetic counseling access and clear patient consent
  • Outcomes measured in aggregate, in unique patients

Bailiwick advises on design and evidence. It does not provide legal advice; your counsel makes the final call.

See the method in action.

The worked example walks through each step with an interactive model you can adjust.

Open the worked example