A referral arrives with three entries in the family history: a maternal aunt diagnosed with breast cancer in her forties, the patient’s mother’s sister diagnosed with colon cancer some years later, and no information about the father’s side.
At the counseling visit, the picture changes. The patient explains that the first two entries describe the same aunt. The father was adopted, so little is known about his biological relatives. Both of the aunt’s diagnoses are still family-reported, and no pathology records have been reviewed.
The family is illustrative, but the pattern will be familiar to most hereditary cancer programs. It is also a useful test of structured family history data, because nearly every shortcut a system takes shows up somewhere in this record. Any system that receives it should be able to answer the following:
- Does the record now hold one aunt or two?
- Are both reported diagnoses still attached to her?
- Does the system know the paternal branch is empty because of an adoption, or does it simply appear empty?
- When the history reaches a risk model, is the diagnosis still recorded as in her forties, or has it been converted to an exact age?
Structured family history data is less about the questionnaire and more about what survives after it. Each relative is recorded once, together with their relationships, their reported diagnoses, the confidence behind each detail, and a clear statement of what is unknown. That structure has to hold as the history moves from the patient to the counselor, the EHR, and the risk tool.
Genetics teams are well aware of this. NCI’s guidance on cancer genetics risk assessment describes pedigrees that span generations and both sides of the family, and treats diagnoses, ages, and relationships as clinically important. The difficulty is usually in the software, at the points where the history moves from one system to the next.
Pedigree data management: when two relatives are really one person
Consider the aunt first. Maternal aunt and mother’s sister describe the same relationship, but they are not identifiers. Many patients have more than one aunt on their mother’s side, and merging on wording alone risks attaching a diagnosis to the wrong relative.
In this case, the merge is appropriate because the patient confirmed it. The correct result is one aunt with both reported diagnoses attached. Two affected aunts would overstate the family history. One aunt with a single generic cancer entry, produced by a deduplication script, would understate it.
Good pedigree data management means every merge leaves a clear trail. The record should retain:
- A stable identifier for the aunt within the family record.
- Her position in the pedigree.
- Both original submissions, exactly as received.
- The reason for the merge and who approved it.
That approval should come from a person. The counselor who spoke with the patient has the context to confirm the match; an automated matching rule does not. A similar age or the same cancer type is not sufficient evidence, so where identity remains uncertain, the record should show that uncertainty until it is resolved.
Keeping people and diagnoses separate matters for another reason. NCI’s guidance distinguishes multiple primary cancers from recurrence and metastasis. The patient describes breast cancer followed by colon cancer, and the record should capture that account without presenting it as two confirmed primary cancers, since neither diagnosis has been verified.
When supporting records become available, they should attach to the same aunt, along with their source and review history.
Presentation matters as well. Pedigree management software, or any genetic family history software, should make the following visible to the counselor:
- Relationships that are still uncertain.
- Diagnoses that are reported but not yet confirmed.
- Branches with missing information, and the reason it is missing.
A clean pedigree diagram that hides these gaps is less useful than a less polished one that shows them. The diagram and the risk-tool export should also draw from the same record, so the two cannot diverge.
How should a pedigree record unknown family history?
The paternal side presents a different problem. An empty branch is easy to misread. It indicates that information is unavailable, not that cancer was absent.
NCI identifies adoption, small family size, early deaths, and limited knowledge of relatives as limitations of family-history assessment, and recommends updating the history from both sides of the family.
The USPSTF makes a related point in its BRCA recommendation. Because risk assessment relies so heavily on family history, it notes that it is unclear how to assess people with a limited or unknown family history.
The adoption should therefore remain visible to both the clinician and the assessment process. Exporting the branch as no affected paternal relatives would discard the very limitation the patient explained.
It also helps distinguish different forms of missing information, which a single family history negative flag cannot capture. FHIR R4 already provides for most of them:
| Situation | What it means | How FHIR R4 can carry it |
|---|---|---|
| Not asked | This part of the history was never collected | List.emptyReason: notasked |
| Reviewed, nothing known | The history was taken and nothing was known to record | List.emptyReason: nilknown |
| Relative’s history unavailable | For example, the patient does not know the relative or cannot obtain the history | FamilyMemberHistory.dataAbsentReason, such as subject-unknown or unable-to-obtain |
| Specific cancer reported absent | The question was asked and the answer was no | An agreed profile or coded negative finding, never an empty list |
Unaffected relatives matter as well. A family with many relatives who reached their eighties without cancer provides different information from one whose relatives are young or poorly documented. Current ages, ages at death, and length of follow-up all belong in the record. How much weight they carry is a decision for the model and the clinician.
Adoption, half-siblings, and donor conception also need an agreed representation. Biological and social relationships both matter to a family, but they should not be treated as interchangeable inputs to an inherited-risk calculation.
Finally, an incomplete pedigree should not become a barrier to care. NCI cautions that screening tools can miss people who warrant assessment, and that clinical review remains important. Software should surface these gaps, but it should not prevent a clinician from considering referral or testing.
Approximate ages in family history data: where false precision begins
Return to the aunt’s breast cancer. The patient recalls a diagnosis in her forties. A downstream risk tool requires a number, so an adapter supplies 45.
Under an approved missing-data policy, 45 may be a reasonable modeling assumption. The problem arises if that value returns to the clinical record as a confirmed age at diagnosis, which no one ever reported.
A few practices prevent this:
- Preserve what was reported. Store in her forties, or the equivalent range, exactly as the patient gave it.
- Store assumptions separately. If an assessment requires a number, keep it with the assessment, together with the rule that produced it.
- Use what the model supports. Many models accept an age range or an unknown value. Use that representation wherever the model’s documentation allows.
A single relative can also have several relevant ages: current age, age at last follow-up, age at each diagnosis, and age at death. A single age field next to her name cannot tell an export which one it needs.
Cancer descriptions need the same care. If the family reports abdominal cancer and nothing establishes a primary site, the record should keep that broader description. If a later report changes the diagnosis, or shows that a second cancer was a metastasis, it should be added without erasing the family’s original account.
Patient-reported history remains valuable, provided it is labeled accurately. A detail may come from the patient, another relative, a clinician’s note, or a reviewed report. The record should show which, when it was obtained, and what has been clarified since. A counselor’s review of a history does not make every statement in it pathology-confirmed.
The same applies to a relative’s genetic test result. If the family mentions a positive cancer gene test, record it as reported until the actual test details are reviewed. NCI distinguishes testing for inherited cancer risk from tumor testing, and a family’s description of a positive test is not sufficient to record a known germline pathogenic variant.
When a detail requires follow-up, the task should be specific: which detail is in question, what information is being sought, and who is responsible for reviewing it. The clinical team decides which uncertainties matter for the decision at hand.
FHIR Family Member History needs an agreed pedigree mapping
Once reviewed, the history has to move from the pedigree tool to the systems that use it. Valid FHIR is a necessary starting point, but a resource can pass validation and still lose the family the counselor described.
In FHIR R4, Family Member History records health information about a relative in the context of the patient’s care. It supports a relationship, multiple conditions, approximate ages, deceased status, and reasons a relative’s history may be unavailable. Several implementation details still require deliberate decisions:
- Relationships between relatives. Each FamilyMemberHistory relates a relative to the patient. A complete pedigree also needs the relationships between relatives, so agree how those links will be carried, whether through a profile, extensions, or an internal pedigree model, and confirm what the receiving system can reconstruct.
- Relative identity. The resource identifier identifies the record, not the relative. The implementation needs a separate mechanism to keep the aunt’s identity stable across submissions, and she should not become a new Patient record or be linked to another person’s chart without an appropriate identity and access process.
- Age semantics. A relative’s age and the onset of a condition are separate elements. In R4, onset can be an age, a range, a period, or a descriptive string, so in her forties can be carried as a range. Whether that matches the age at diagnosis a particular model expects is a separate question, so document any profile or extension the mapping relies on.
- Status. A completed record does not mean every diagnosis in it has been verified, and an empty condition list is not, on its own, a negative cancer history.
For histories with nothing to report, HL7’s List resource guidance provides emptyReason codes such as notasked and nilknown, and FamilyMemberHistory.dataAbsentReason explains why an individual relative’s history is unavailable. These distinctions need to survive the interface rather than being inferred from an empty export.
The mapping should be tested against the actual EHR and pedigree system, including updates as well as initial loads. This is the part of EHR integration services that is most often underestimated. An endpoint accepting the resource proves very little. The real test is whether one aunt, two reported diagnoses, an approximate age, and a documented paternal gap all arrive intact.
NonStop has worked on this handoff before. Its published family history risk assessment software work for a genetic counseling network covers AI-driven pedigree and inheritance modeling, structured cascade tracking, and integration back into LIS and EHR systems.
In this example, that is the point at which the aunt’s record leaves the pedigree tool and must arrive in the EHR as one person with two reported diagnoses. Each new program still needs its own data definitions and acceptance tests, and that prior work is not evidence that any risk model has been validated for a different population.
Beneath the integration layer, NonStop’s healthcare data engineering services provide the data modeling, quality controls, and lineage. Before a build begins, the family history data model should define:
- How each relative’s identity is maintained over time.
- How reported and reviewed information coexist.
- How FHIR exchange preserves their meaning.
- Exactly what each risk tool receives.
Which hereditary cancer risk model fits the clinical question?
Even an accurate transfer leaves an open question: which assessment is the program running?
A cancer-risk estimate, the probability of carrying a pathogenic variant, and a referral decision answer different questions. NCI’s discussion of breast and gynecologic risk models describes the purpose of each approach and the inputs it requires.
The USPSTF’s 2019 BRCA recommendation makes one distinction explicit. General breast cancer risk models, such as the Gail-based Breast Cancer Risk Assessment Tool, are not designed to identify BRCA-related risk and should not be used for that purpose. This concerns model purpose rather than a complete testing protocol. The USPSTF also has an update to this recommendation in progress, which programs should monitor.
Before any model runs, the software build should document:
- The intended use and the eligible population.
- The model or rule version.
- The required inputs, and how unknown or approximate data is handled.
- The cancer endpoint and time horizon, where applicable.
Outputs built for different questions should not be presented as interchangeable risk percentages.
In this example, the aunt’s approximate age should appear as a known input limitation. The duplicate should already be resolved, and the paternal gap should be represented as the selected tool requires. If the tool needs an assumption, it belongs with the assessment. The clinical record should never be edited simply to make a tool run.
Retain a copy, or a reference, of everything each calculation used: the inputs and assumptions, the model or guideline version, the result, and the clinical review. When a detail is corrected later, the team can see exactly what the earlier result was based on.
Referral rules need their own approved criteria and version control. A model result does not establish testing eligibility, authorize an order, or determine management. Clinicians remain responsible for applying the relevant guidance to the individual patient.
Keeping family history data current from one visit to the next
Family histories change over time. Relatives age, diagnoses are clarified, and patients learn new information. The CDC recommends updating family health history regularly and sharing it with the healthcare team, even when it is incomplete.
When a later questionnaire reports a third cancer in the same aunt, it should be added to her existing record after review. The patient should not need to rebuild the pedigree because the form has changed, and a new submission should not overwrite information the counselor has already reviewed.
A new diagnosis and a correction are different kinds of change. A newly reported cancer and a corrected cancer site can have different clinical consequences. Keep the source, the review, and the link to the earlier entry, along with a policy for deciding when an assessment should be revisited.
This does not require rerunning the model or contacting the patient after every edit. It requires a defined way to bring relevant changes to the responsible clinician.
Privacy needs clear boundaries too. Appearing in a patient’s pedigree does not make the aunt a patient of the program, nor does it permit anyone to contact her. If she becomes a patient, whether through cascade testing outreach or her own referral, her care follows the organization’s established processes. A shared family relationship should not make any individual’s clinical record visible to other family members.
Form completion rates say little about whether a family history is usable. More informative measures include:
- Duplicate relatives merged after intake.
- Diagnoses recorded without an age.
- Histories carrying an explicit information limitation.
- Model exports that cannot run without an assumption.
Reviewing these by source and workflow stage shows where histories lose quality.
A usable family history record: one aunt, two diagnoses, one documented gap
By the end of this process, the aunt’s record should be straightforward: one person, two reported cancers with their sources, an age still recorded as in her forties, and a paternal branch that is empty because of an adoption, with that reason documented.
New records and updates can continue to arrive without erasing how the history was reviewed. That is the record the next clinician and the next risk tool should receive. It is more reliable than a completed form, and considerably more useful.
Questions genetics and informatics teams ask
How do you structure family history data for hereditary cancer programs?
Record each relative once, separately from their diagnoses, and retain biological relationships, ages, sources, and any gaps in what is known. Then define exactly which reviewed details and assumptions each assessment receives, rather than reducing the history to a count of affected relatives.
How do you record unknown family history in FHIR?
Use the codes FHIR already provides. List.emptyReason separates a history that was never asked (notasked) from one reviewed with nothing known (nilknown), and FamilyMemberHistory.dataAbsentReason explains why an individual relative’s history is unavailable. An empty export should never stand in for a negative history.
Can FHIR FamilyMemberHistory store approximate ages?
Yes. In FHIR R4, a relative’s age can be a range or a string, and condition onset can be an age, a range, a period, or a descriptive string. A remembered decade can therefore be carried as a range rather than converted to an invented exact age.
Why shouldn’t the Gail model be used to assess BRCA risk?
It was not designed for that purpose. The USPSTF’s 2019 BRCA recommendation states that general breast cancer risk models such as the Gail-based Breast Cancer Risk Assessment Tool are not designed to identify BRCA-related risk. Familial risk tools built for that question should be used instead.
How do you merge duplicate relatives in a pedigree?
Only on confirmed evidence, typically the patient confirming to the counselor that two entries describe the same person. Keep one stable identifier for that relative, attach every reported diagnosis to them, and record who approved the merge, why, and what the original submissions said. A similar age or cancer type is not sufficient on its own.

