In the previous essay, we argued that one of the most valuable things a machine can do in healthcare is remember and relate: preserve what has happened to a person over time and maintain the relationships among those observations. That raises another question: how should that accumulated knowledge be made useful to a clinician?
One increasingly common answer is summarization. Give an AI access to the medical record and ask it to condense hundreds of pages into something a clinician can quickly understand. That can be enormously useful. But summarization and context solve different problems. A summary helps the clinician see less. Context can help the clinician notice more.
Summarization is compression
The value of a good summary is precisely that it leaves things out. It compresses years of notes, laboratory results, medications, diagnoses, imaging, and encounters into a smaller representation that a clinician can absorb quickly. In a healthcare system overwhelmed by information, that is valuable.
But every summary necessarily makes choices about relevance. Some information is retained, some is compressed, and some disappears. That is not a flaw; it is what makes summarization useful. The limitation is that the significance of a fact may not become apparent until something else happens.
A symptom mentioned months ago, a medication that was discontinued, a temporary change in function, or the response to an earlier intervention may seem unimportant today. A future observation can suddenly make it significant. Relevance changes as the person changes.
Context preserves relationships
Imagine that a person reports increasing fatigue. A summary might appropriately tell the clinician that the person has heart failure, diabetes, and chronic kidney disease, list current medications, and describe recent encounters and laboratory results.
Context asks different questions. When did the fatigue begin? What else changed around that time? Was a medication adjusted? Has activity been declining? Has sleep changed? Has weight moved from the person's usual range? Has adherence changed? Did something similar happen before, and what followed?
None of those observations necessarily matters by itself. Their significance may emerge from their relationship to one another and to the person's history.
Context is not simply what we know about the person. It is how what we know relates across the person and across time.
This is why remembering and relating matter. The machine does not need to decide in advance which pieces of a person's history will matter forever. It can preserve them and continually reconsider their relevance as new observations arrive.
Making complexity usable
Preserving context does not mean putting more information in front of the clinician. A clinician confronted with every available observation, relationship, and historical event would face precisely the cognitive burden we are trying to reduce. The machine can maintain that complexity in the background and make potentially meaningful relationships noticeable when they become relevant.
A summary, for example, might report that a person's weight has increased three pounds. Context might make noticeable that the increase is occurring alongside gradually worsening shortness of breath, lower activity, inconsistent diuretic adherence, and a pattern resembling the days before a previous hospitalization. The clinician still decides what that pattern means. What has changed is what the clinician has the opportunity to notice.
The purpose of AI is not to compress the patient until the case fits within human working memory. It is to preserve the richness of the patient's context while making what matters noticeable when a clinician needs to decide.
An evolving representation of the person
This requires something different from a collection of records or a better summary of them. At Senscio, we think of it as an evolving representation of the person. HealthGraph™ preserves accumulated knowledge and relationships. The Digital Twin for Health™ uses that knowledge to maintain a current representation of the person's changing state.
New observations therefore do not simply accumulate. They enter a longitudinal context in which what happens today can be related to what happened yesterday, what was tried six months ago, how the person responded, and what is known about the person more broadly. A Digital Twin for Health is not simply a condensed version of the medical record. Its purpose is to preserve enough of the person's evolving context that meaningful relationships can become noticeable as circumstances change.
The goal is not to make a complex person appear simple. It is to make complexity usable.