Difference between revisions of "Reporting Worklist Prioritization - Proposal"

From IHE Wiki
Jump to navigation Jump to search
Line 62: Line 62:
 
==5. Discussion==
 
==5. Discussion==
  
''<Include additional discussion or consider a few details which might be useful for the detailed proposal>''
+
Might interact with the AIR Data/Root Results proposal. The Root Results might serve as a useful summary, and if more details are needed it might warrant extending attributes in existing MAP.
:''<Why IHE would be a good venue to solve the problem and what you think IHE should do to solve it.>''
 
:''<What might the IHE technical approach be? Existing Actors? New Transactions? Additional Profiles?>''
 
:''<What are some of the risks or open issues to be addressed?>''
 
  
 
+
How should workitem this be packaged? As an option on SWF.b? As an extension of one of the reporting workflow profiles? As a revision to AIW (given that AI results are a high profile piece of data but not the only data of interest)?
''<This is the brief proposal.  Try to keep it to 1 or at most 2 pages>''
 

Revision as of 16:16, 18 August 2021

1. Proposed Workitem: Reporting Worklist Prioritization

  • Proposal Editor: Antje Schroeder, Kevin O'Donnell, Teri Sippel
  • Editor: TBA
  • Domain: Radiology

2. The Problem

At any point in time, there are N items on the reading worklist when a new reading task arrives, so the Reading Worklist Manager has to choose to prioritize the new task into one of N+1 possible positions on the list. All priorities are relative.

While the logic (and preferences) for which other tasks a task should come before is internal to the reading worklist provider (and hopefully customizable), all the details for each task that may be relevant to prioritization come from outside the reading worklist provider so getting access to them is an interoperability problem.

An AI Application that detects lung nodules or that rules out strokes can't tell the reading worklist provider where to prioritize a new study in the worklist because the AI Application doesn't know what else is on the worklist. But the AI can contribute positive or negative findings that can feed the logic of the worklist provider. Other systems will also need to provide (and update) relevant information. And certain details that the worklist provider wants will need the encoding to be standardized (especially if it comes from multiple implementations)

What findings are a higher priority than a partial stenosis of a coronary artery and what are lower is something that will be encoded into the central prioritizer, not each AI Application (granting that the prioritizer might eventually be an AI itself...)

3. Key Use Case

Consider sitting down with a few chief radiologists and radiology department administrators and asking “If you were manually prioritizing this couple dozen entries in a worklist, what does “optimal ordering” mean to you, what characteristics would drive your relative priority choices, how do you handle tradeoffs between a low confidence high risk finding and a high confidence moderate risk finding." etc.

The profile use case should spell out:

  • What pieces of information does the prioritizer need/use to prioritize worklist items
  • priority of the underlying imaging order
  • agreed turn around times (service level agreements)
  • imaging procedure type
  • patient type/location
  • Reading Physician availability
  • known/suspected risks to the patient
  • admitting diagnosis, reason for procedure
  • anatomy being imaged
  • preliminary findings from one or more AI algorithms
  • stoke is present, stroke is absent
  • severity of hemorrhage (large/small, located in a more critical area or a less critical area)
  • confidence of finding
  • lab findings, pathology findings
  • when is the patient expected to leave the facility
  • whether associated information (e.g. lab results, CAD results) is available yet, and if not, when
  • What system(s) can provide each piece of information and how is it encoded, e.g.
  • some from modality (via PACS) in the image header
  • some from analysis application encoded in DICOM SR content
  • some from the Order Placer in the incoming Order message
  • some from a support system in a Procedure Update OMI, ….
  • What codesets are used for certain concepts, e.g.
  • finding codes, anatomy codes, severity codes, etc.

The profile should define:

  • What system has the relevant information listed above
  • How that system encodes and conveys that information to the prioritizer
  • Whether to proxy/aggregate the information to make it more digestible for a prioritizer

4. Standards and Systems

Systems:

  • Reporting Worklist Manager (RIS/PACS)
  • PACS - study data source
  • AI Applications - presence/absence of various findings
  • Might need adjustments to MAP Profile to get additional details incorporated in the results
  • EMR - admission and patient record information
  • Staff/Scheduling System - expertise and availability of reading staff

5. Discussion

Might interact with the AIR Data/Root Results proposal. The Root Results might serve as a useful summary, and if more details are needed it might warrant extending attributes in existing MAP.

How should workitem this be packaged? As an option on SWF.b? As an extension of one of the reporting workflow profiles? As a revision to AIW (given that AI results are a high profile piece of data but not the only data of interest)?