This post is for marketing and content teams at hospitals and specialty clinics who run a patient-education library: pages on diabetes, kidney stones, endometriosis, cataracts and the like. It is also for the developer who has been asked to "add schema" to those pages and wants a model more considered than one blob of JSON-LD per article.
The question it answers is how to describe a condition page as a small graph. The condition is a node. Its signs, risk factors, tests and treatments are linked nodes. The page is a node that is about the condition and was reviewed by a named clinician. The condition then connects to the procedure and doctor pages in the same cluster, so the site describes a coherent subject and not a pile of articles.
The syntax of JSON-LD is covered in our implementation guide, the complete guide and the advanced markup article. Here we focus on the model, which is where most condition markup goes wrong.
Set expectations before you build
Google's structured data gallery has no feature for medical conditions. Google's general guidelines also say structured data is not guaranteed to produce any display. So this markup is a structured statement of what the page says, not a ticket to a rich result.
Two rules still apply. Mark up only what readers can see, and treat the clinical content as needing review by the provider's clinicians before publication. Health pages are the kind Google's quality guidance calls YMYL, and the visible reviewer and review date are part of earning trust. The markup should repeat them, not create them.
What a MedicalCondition can say
MedicalCondition sits under MedicalEntity. Its own properties include the following, with the expected types as defined by schema.org.
| Property | Expected type | Page section it should mirror |
|---|---|---|
signOrSymptom | MedicalSignOrSymptom | "Signs and symptoms" |
riskFactor | MedicalRiskFactor | "Who is at risk" |
cause | MedicalCause | "Causes" |
typicalTest | MedicalTest | "How it is diagnosed" |
possibleTreatment | Drug, DrugClass, LifestyleModification or MedicalTherapy | "Treatment options" |
primaryPrevention | MedicalTherapy | "Prevention" |
secondaryPrevention | Drug, DrugClass, LifestyleModification or MedicalTherapy | "Preventing recurrence" |
possibleComplication | Text | "Complications" |
associatedAnatomy | AnatomicalStructure, AnatomicalSystem or SuperficialAnatomy | Where in the body |
stage | MedicalConditionStage | Staging, if the page uses it |
differentialDiagnosis | DDxElement | Rarely needed on a patient page |
expectedPrognosis, naturalProgression, pathophysiology, epidemiology | Text | Usually leave out |
The last row is deliberate. Prognosis language drifts into outcome promises, and pathophysiology is clinician-level content. A patient page that does not discuss them should not emit them.
From MedicalEntity the node also gets code (a MedicalCode, for an identifier from a recognised coding system if your clinical team supplies one), relevantSpecialty and guideline.
Respect the expected types
Here is a trap. possibleTreatment expects Drug, DrugClass, LifestyleModification or MedicalTherapy. A MedicalTherapy is a subtype of TherapeuticProcedure, which is a subtype of MedicalProcedure. The reverse does not hold. A surgical operation is a SurgicalProcedure, a sibling branch, not a MedicalTherapy.
Put differently, you cannot honestly list "knee replacement" under possibleTreatment if it is typed as a surgical procedure. You can list physiotherapy or radiotherapy there, because those are therapy subtypes, along with drugs and lifestyle changes. Whether a validator complains is a separate matter; the point is that the vocabulary says what it expects, and a model that ignores that makes consumers guess.
For surgical options, link at the page level instead.
- The condition page
mentionsthe procedure nodes that the page discusses. - Or
significantLinkpoints to the treatment page that covers it. - The treatment page, in turn, is
aboutthe procedure.
That keeps the condition's own properties valid and still creates the graph edge.
Model the page as a graph
Use a single @graph per page, with each node identified by a stable @id. The W3C JSON-LD 1.1 specification says a node is identified by @id and that a node object containing only an @id is a reference to a node described elsewhere in the document. That is exactly the tool you need.
A condition page should contain these nodes.
| Node | Type | @id pattern (hypothetical domain) | Key links |
|---|---|---|---|
| The page | MedicalWebPage | https://www.example-hospital.in/conditions/type-2-diabetes/#webpage | about the condition, reviewedBy a clinician, lastReviewed, medicalAudience, mentions procedures |
| The condition | MedicalCondition | .../conditions/type-2-diabetes/#condition | signOrSymptom, riskFactor, typicalTest, possibleTreatment, relevantSpecialty |
| The reviewer | Physician | .../doctors/dr-example/#physician | hospitalAffiliation, medicalSpecialty |
| The publisher | Hospital | .../#hospital | Referenced as the publisher of the page |
Use MedicalWebPage, which inherits reviewedBy and lastReviewed from WebPage and defines medicalAudience itself. Reference the reviewing doctor by @id so the same doctor node, defined fully on the doctor's profile page, is the one named here.
Build the nodes from the page's structure
A good condition page has headings. Each heading maps to one property, and each list item maps to one small node.
- Each symptom in the "Signs and symptoms" list becomes a MedicalSignOrSymptom with a
name. Keep the name as it appears on the page. - Each risk factor becomes a MedicalRiskFactor with a
name. - Each test in the diagnosis section becomes a MedicalTest with a
name. - Each therapy, drug class or lifestyle change in the treatment section becomes the right typed node with a
name.
If a section is a paragraph, not a list, do not split it into guessed nodes. Either leave the property out or have an editor turn the paragraph into a list first. That is a content improvement for readers too, and it is the same logic as our patient education SEO work.
Do not add a list of symptoms that the page does not show. Do not state that a condition "cures" or that a treatment "works". Descriptions of urgent signs, such as when to seek emergency care, must use wording approved by the clinical team and should be visible on the page before they appear in markup.




