The key differences between this operation and simply searching the patient compartment are:

• unless the client requests otherwise, the server returns the entire result set in a single bundle (rather than using paging)
• the server is responsible for determining what resources to return as included resources (rather than the client specifying which ones).

This frees the client from needing to determine what it could or should ask for, particularly with regard to included resources. Servers should consider returning appropriate Provenance and AuditTrail on the returned resources, even though these are not directly part of the patient compartment.

It is assumed that the server has identified and secured the context appropriately, and can either associate the authorization context with a single patient, or determine whether the context has the rights to the nominated patient, if there is one, or can determine an appropriate list of patients to provide data for from the context of the request. If there is no nominated patient (GET /[base]/Patient/$everything) and the context is not associated with a single patient record, the actual list of patients is all patients that the user associated with the request has access to. This may be all patients in the family that the patient has access to, or it may be all patients that a care provider has access to, or all patients on the entire record system. In such cases, the server may choose to return an error rather than all the records. Specifying the relationship between the context, a user and patient records is outside the scope of this specification (though see The Smart App Launch Implementation Guide.

When this operation is used to access multiple patient records at once, the return bundle could be rather a lot of data; servers may choose to require that such requests are made asynchronously, and associated with bulk data formats. Alternatively, clients may choose to page through the result set (or servers may require this). Paging through the results is done the same as for Searching, using the _count parameter, and Bundle links. Implementers should note that paging will be slower than simply returning all the results at once (more network traffic, multiple latency delays) but may be required in order not to exhaust available memory reading or writing the whole response in a single package. Unlike searching, there is no inherent user-display order for the $everything operation. Servers might consider sorting the returned resources in descending order of last record update, but are not required to do so.

The _since parameter is provided to support periodic queries to get additional information that has changed about the patient since the last query. This means that the _since parameter is based on record time. The value of the _since parameter should be set to the time from the server. If using direct response, this is the timestamp in the response header. If using the async interface, this is the transaction timestamp in the json response. Servers should ensure that the timestamps a managed such that the client does not miss any changes. Clients should be able to handle getting the same response more than once in the case that the transaction falls on a time boundary. Clients should ensure that the other query parameters are constant to ensure a coherent set of records when doing periodic queries.