BOADICEA vs BRCAPRO: an honest comparison of the two leading BRCA risk models

Short version. BOADICEA and BRCAPRO are both Bayesian models that estimate the probability that an individual carries a pathogenic variant in BRCA1 or BRCA2, and estimate future breast and ovarian cancer risk from that probability plus family history. They differ substantially in what they model: BRCAPRO treats breast and ovarian cancer susceptibility as coming from BRCA1 and BRCA2 alone; BOADICEA models BRCA1 and BRCA2 plus moderate-risk genes (ATM, CHEK2, PALB2, RAD51C, RAD51D, BARD1), a polygenic risk component, lifestyle and hormonal factors, and mammographic density where available. In head-to-head validation studies BOADICEA is generally the better-calibrated model, particularly in populations where polygenic and density information can be captured. BRCAPRO remains widely used, has decades of clinical validation, and gives similar clinical decisions at common decision thresholds. They are best understood as complementary rather than interchangeable.

What each model is

BOADICEA — the Breast and Ovarian Analysis of Disease Incidence and Carrier Estimation Algorithm — was developed by the Centre for Cancer Genetic Epidemiology at the University of Cambridge under Antonis Antoniou and colleagues. The original BOADICEA paper appeared in the British Journal of Cancer in 2004; the model has been extended repeatedly since, notably in the 2019 update (commonly referred to as BOADICEA v5 / v6) that added polygenic risk scores, mammographic density, and moderate-risk genes beyond BRCA1/BRCA2. BOADICEA is now delivered through the CanRisk web interface at canrisk.org, which also hosts the CanRisk-Prostate model. CanRisk is free to healthcare professionals after registration and is endorsed by NICE, NCCN, ESMO, Cancer Care Ontario, eviQ, and the UK Cancer Genetics Group.

BRCAPRO is part of the BayesMendel suite developed at the Johns Hopkins / Harvard BayesMendel Lab (Giovanni Parmigiani and colleagues). It uses Bayesian statistics to compute the probability of carrying a pathogenic BRCA1 or BRCA2 variant given the pattern of breast and ovarian cancers in a family, and to estimate future cancer risk from that probability. It shares its methodological foundation with sibling models: MMRpro for Lynch syndrome and PancPRO for familial pancreatic cancer. BRCAPRO is distributed as open-source R code in the BayesMendel package and is implemented inside multiple clinical pedigree platforms including Evagene.

Both models are Bayesian in the same sense: they combine prior probabilities of variant carriage (derived from population-level gene frequencies and penetrance estimates) with the likelihood of observing the family's cancer history given that carriage status. The output is a posterior probability — the carrier probability — conditional on the recorded pedigree. The difference is in what the models condition on.

What each model actually models

This is the most important section of the article because it determines when the two models will disagree.

The table tells the story. BOADICEA is the more fully specified model: it accounts for most of the heritable and environmental influences on breast cancer risk that we know about. BRCAPRO is a simpler model focused narrowly on BRCA1 and BRCA2. This makes BRCAPRO easier to use with limited data but also means it will systematically miss signal that BOADICEA captures.

What the validation studies actually show

Clinicians often ask "which model is more accurate?" The answer is nuanced and depends on the population, the outcome being predicted, and the data available at the point of prediction.

Calibration for carrier probability. Several head-to-head studies have examined how well each model predicts actual BRCA1/BRCA2 carrier status. In the retrospective Italian cancer-genetics-clinic cohort analysed by Fischer and colleagues, BOADICEA and BRCAPRO performed similarly at the common 10% carrier-probability threshold used clinically for eligibility decisions, but the models diverged at other thresholds. BOADICEA has tended to underestimate mutation risk in some populations, while BRCAPRO has tended to underpredict risk in several cohorts — the direction of bias depends on the population, family structure and reference data used.

Calibration for breast cancer incidence. In validation studies that compared several 10-year and lifetime breast cancer risk models (BOADICEA, IBIS/Tyrer-Cuzick, BRCAPRO, Gail and Claus) in unselected or population-based cohorts, BOADICEA and IBIS have consistently emerged as the best-calibrated models. BRCAPRO has generally performed less favourably for predicting incidence, which is expected: BRCAPRO is primarily a carrier-probability model, and incidence prediction is a secondary output derived from penetrance tables rather than a first-class target of the model.

Ethnic and family-structure specificity. Performance varies across populations. In the Northern California Breast Cancer Family Registry study across racial / ethnic groups, model performance differed meaningfully between groups — BOADICEA and BRCAPRO do not give identical answers across populations, and no single model dominated everywhere. In high-risk French-Canadian families, both models predicted mutation carriers reasonably well, with some differences in penetrance estimation.

At common clinical thresholds. The practical question is usually "does this individual meet the threshold for genetic testing?" At the commonly used 10% BRCA carrier probability threshold, BOADICEA and BRCAPRO often agree on eligibility even when their numerical outputs diverge. For a clinician deciding referral versus no-referral, the disagreement rate between the two models is smaller than the difference in their point estimates would suggest.

The honest summary is that BOADICEA is the better-calibrated model in most unselected populations when its richer inputs are available, but BRCAPRO remains clinically useful, especially when polygenic or lifestyle data is not to hand, and especially in settings where a BayesMendel-style family-history-only estimate is what is wanted. Guidelines that cite BRCAPRO specifically still treat it as an acceptable eligibility model.

When to use which

Use BOADICEA / CanRisk when:

• You have access to polygenic risk score data or mammographic density and want those signals reflected in the estimate.
• Your testing panel includes moderate-risk genes beyond BRCA1/BRCA2 and you want a model that accounts for them.
• You are following UK NICE guidance, NCCN breast/ovarian, ESMO, or other guidelines that recommend BOADICEA explicitly.
• You want hormonal, reproductive, and lifestyle factors incorporated.
• You are generating a risk report to communicate to a patient — CanRisk's graphical output is well-designed for patient-facing discussion.

Use BRCAPRO when:

• Your only input is family history — no PRS, no density, no reliable hormonal data.
• You want a transparent, open-source model that you or your developers can audit.
• You want to use the same BayesMendel family of models for breast/ovarian (BRCAPRO), Lynch (MMRpro), and pancreatic (PancPRO) — one methodological foundation, three conditions.
• You are using a clinical pedigree platform that integrates BRCAPRO directly on the pedigree data with no re-entry.
• You need programmatic, scriptable access for research or batch analysis via R.

Use both when:

• You are in a high-stakes decision (for example, risk-reducing mastectomy or salpingo-oophorectomy) and want to triangulate risk across methodologically distinct models.
• You are building a research cohort and want to compare model outputs.
• You are uncertain and want to see whether the two models agree at your decision threshold.

The practical workflow question

In many clinical genetics services the real question is not "which model" but "how many tools do I have to flip between". CanRisk is excellent but lives at canrisk.org. BRCAPRO lives inside your clinical pedigree platform (in Evagene's case) or in R. Risk assessment outside the pedigree canvas means re-entering family history — a transcription opportunity. Workflow efficiency often drives model choice more than the statistical literature does. This is worth being honest about: the "best" model is the one your team will actually run on every eligible patient, not the one that scores 2% better on calibration in a retrospective cohort.

Services running both usually pick a default model (often BOADICEA) and cross-check with the other in complex or borderline cases. Services running one usually pick BRCAPRO via an integrated pedigree platform for workflow reasons, and refer out to CanRisk for specific BOADICEA-required guidance items.

How Evagene fits in

Evagene integrates BRCAPRO directly on pedigree data, alongside MMRpro for Lynch syndrome and PancPRO for familial pancreatic cancer. This means that when a clinician finishes constructing the pedigree, the BayesMendel models run without any re-entry — a workflow benefit that makes BRCAPRO something that actually gets run on every eligible patient, not something that sometimes gets deferred because switching tools costs three minutes. See our BRCAPRO calculator guide for what the model uses as input and how it presents its output inside Evagene.

Evagene does not currently integrate BOADICEA via the CanRisk web service. The CanRisk team documents a JSON web-services API with auth-token authentication, so integration is technically straightforward; we treat it as a roadmap item driven by user demand. For services that need BOADICEA today, the recommended workflow is: build the pedigree in Evagene, export a JSON or GEDCOM summary, and submit it into CanRisk directly. The pedigree structure round-trips cleanly; polygenic scores and density data are entered inside CanRisk, since Evagene does not currently capture them.

Beyond the models themselves, Evagene's AI clinical interpretation can summarise both outputs into a narrative report — drafting the "what this means for the patient" paragraph that sits between the model's numerical output and the clinician's decision. Using bring-your-own-key LLM access (Anthropic Claude or OpenAI GPT), the clinician controls which model and account process the interpretation text, which matters for clinical-governance purposes.

Evagene's batch risk screening runs all integrated BayesMendel models across a pedigree at once with configurable thresholds, surfacing conditions the clinician might otherwise not have checked. It complements rather than replaces the deliberate, single-model workflow of running a specific model (BOADICEA or BRCAPRO) for a specific clinical question.

A note on Tyrer-Cuzick (IBIS), Gail, and Claus

BOADICEA and BRCAPRO are not the only breast cancer risk models. For completeness:

• Tyrer-Cuzick / IBIS emphasises hormonal and reproductive factors and is widely used for 10-year and lifetime breast cancer risk in population screening settings, particularly through MagView and similar implementations. In validation studies it is often the second-best calibrated model after BOADICEA.
• Gail (NCI Breast Cancer Risk Assessment Tool) is a population-level model that does not incorporate detailed family history; it is useful for general counselling but less appropriate for hereditary-risk assessment.
• Claus is an older family-history model focused on first- and second-degree breast cancer, less commonly used today in populations with access to BOADICEA or BRCAPRO.

For a broader review of these tools alongside BOADICEA and BRCAPRO, see our Tyrer-Cuzick alternative guide and our wider hereditary cancer risk assessment overview.

Frequently asked questions

Related reading

• BRCAPRO calculator — how the model works and when to use it
• CanRisk vs Evagene — integrated pedigree management alongside BOADICEA
• BOADICEA alternatives for teams wanting integrated pedigree management
• Tyrer-Cuzick / IBIS — the other widely-used breast cancer risk model
• Hereditary cancer risk assessment — the wider workflow
• Breast cancer family history calculator
• Ovarian cancer family history calculator