Care Around the Person: Health Information Infrastructure for an Age of Agents

An aging parent experiences their care as one life. They take the same medication from whichever nurse gives it to them, and they are the same person to the paramedic as to the physician. The health care system records that life as a series of institutions instead: a hospital chart, a facility assessment, an agency’s plan of care, an aide’s visit note, and a separate portal for each practice. None of these is currently assembled with the others into one record the family can see.

The family is the only party with a continuous view of the person, yet most of that collective knowledge lives only in a group text. The family learns of a new prescription after it has been started, repeats the same history at every admission, and carries what the parent is ordinarily like from one setting to the next by word of mouth. Each institution keeps the record for which it is responsible, and none of them is responsible for a record of the whole person.

This is why the optimization problem above states its objective as it does. The outcome it maximizes belongs to a person and their family, and it is measured across every setting through which they pass. In contrast, each institution only optimizes over the part of the person inside its own walls. In general, a system whose parts each optimize locally need not reach the optimum of the whole.

Most software for families asks them to do something: log a symptom, confirm a task, or answer a check-in. It works for as long as someone is willing to keep it updated, and it tends to end through attrition.

Most of the information such a tool requires already exists. The aide writes a visit note, the facility sends a message, the pharmacy fills a prescription, and the discharge paperwork lists what changed. What is currently missing is access and someone assigned to read it. The agents we are building read what the care system already produces, record what changed and who knows about it, and cite a source for every entry. They ask nothing of the parent to keep the record current, and they add nothing for the people delivering care to fill in. They currently observe and record; they do not diagnose, decide, or act. We begin with what the paid parties in the household already produce and what the family chooses to tell the record; pharmacy and discharge records follow.

Health systems are currently deploying agents inside their own walls, and many of those agents contact the patient to schedule, to remind, and to follow up. That is a reasonable design for an institution, which needs something from the patient. A family needs the reverse, which is to be told what happened without being asked to log anything, and that is the half of the system we are building. What the family knows enters the same way. A family member can write down what is ordinary for the parent, or reply to what they are told, confirming that something is ordinary or correcting a detail, and either is kept with its source; nothing is asked of them, and the record does not wait on it. The baseline, what is ordinary for the parent, shifts with the person this way rather than being written once and left to go stale.

The number of agents acting on one person’s care is likely to grow quickly. The hospital will have them, and so will the pharmacy, the facility, and the family. Each will need to know who the person is, what is currently true about them, from where each fact came, and who is permitted to see it. There is currently no single place to ask, so each agent will reconstruct the person from whatever its own institution holds, and the fragmentation described in Section 1 will carry over into the agents themselves.

We are therefore building health information infrastructure for agents. It is a record of one person from which any authorized agent can read and to which it can write. Every entry carries its source, its time, and the terms under which it may be used. The record is kept for the person, usually by their family. The person decides who, including which family members, is admitted to it. In cases where they cannot, whoever is legally authorized to decide for them does. Institutions and their agents, once admitted, connect to it alongside the records they already keep, rather than each reconstructing the person from its own fragment.

This layer can be built in one of two ways. The first is inside each vendor’s system, in which case a person’s record remains split across every vendor that has treated them. The second is around the person, in which case the vendors connect to the record rather than each holding a separate copy of the person. The first is the default, because every institution already holds its part. We are building the second, on open standards, so that the record can move with the family between settings and between vendors.

Infrastructure of this kind is only useful if it can be trusted. An agent that reports what it read must also report how confident it is and from where the claim came, and a record that holds conflicting accounts must keep both, attributed, rather than silently choosing between them. Autonomy here is earned the way Waymo earned it: inside one narrow, well-understood domain first, with the safety case proven before anyone steps back.

A record assembled around the person has a second use. Research on aging, and on dementia in particular, depends on three things that are difficult to obtain from any single provider: one person’s care joined across the settings that delivered it, a way to reach that person again when a study is recruiting, and an account of daily life between clinical encounters. The family is the only party positioned to supply all three, which is why we currently think a record held this way could help alleviate the scarcity of that kind of data in health research.

Studies already rely on families in this role. Trials in Alzheimer’s disease typically require a study partner who knows the participant well enough to report on their daily function, and a family member already occupies that role. What the family currently lacks is a record from which to answer. An agent working on the family’s behalf could also look for relevant studies and bring them to the family to consider.

Three commitments follow. First, taking part in research is separate from using the record and is never a condition of it. Secondly, it is asked for explicitly, one use at a time, and withdrawing stops any new use of the record. Thirdly, the person decides which information they do or do not want disclosed, to whom, and for what purpose, and where they cannot, whoever is legally authorized to decide for them does.

We currently think this has the structure of a network. The more families hold a record of this kind, the more useful it becomes to research, and the more studies look for participants through it, the more useful it becomes to families. We intend to begin with the families of aging parents and to generalize from there.

None of this exists yet. The research side is a direction rather than a program, and no family’s record is currently used for research.