Automated Follow-up program


Purpose

Detect and extract follow-up recommendations from radiology reports, identify and automatically close the loop when an appropriate follow-up examination has been performed, and notify assigned user(s) when the required follow-up examination has not been done within a pre-specified time frame.

Tag(s)

Non-Interpretative, Follow-up

Panel

Business Facing Panel

Define-AI ID

19100006

Originator

Woojin Kim

Lead

Woojin Kim

Panel Chair

Rich Heller

Non-Interpretive Panel Chairs

Alexander J. Towbin, Adam Prater

Panel Reviewers

Business Facing Panel

License

Creative Commons 4.0 

Status

Public Comment

Clinical Implementation


Value Proposition

Management of follow-up recommendations has become an important topic in radiology due to its impact on patient care, medicolegal implications, and financial consequences. A number of articles have recently highlighted the significance of this issue by demonstrating poor follow-up rates. For example, Blagev and group showed a failure rate of 71% of follow-up of incidental pulmonary nodules (https://www.ncbi.nlm.nih.gov/pubmed/24316231 ). Cook and group demonstrated a 44% follow-up failure rate of patients with suspicious/indeterminate findings on CT, MRI, and US (https://www.ncbi.nlm.nih.gov/pubmed/28325488 ). While both home-grown applications and commercial solutions exist, this is an area where AI can potentially further assist in closing the loop and minimize the number of patients who slip through the cracks when it comes to follow-up recommendations, resulting in improved patient care and ultimately better outcomes.

Narrative(s)

A 65-year-old male with heavy smoking history was found to have a 7 mm solid pulmonary nodule within the right lower lobe. The interpreting radiologist made a recommendation for a 6-month CT follow-up. Seven months later, the patient has not returned for his follow-up CT examination. In addition, there is no follow-up CT examination scheduled in the system.

Workflow Description

The AI algorithm picks up a follow-up recommendation (or multiple if more than one is present) within a radiology report. It should be noted that a radiologist may describe follow-up recommendations using a variety of terminology. Also, follow-up recommendations can occur within any portion of the report (i.e., not just within the IMPRESSION section). For each recommendation within the report, the AI algorithm picks up associated elements, such as modality (if an imaging examination was recommended) or action type (e.g., biopsy), anatomy, time interval, and confidence of the recommendation. Every examination with follow-up recommendations is tracked prospectively. When there is an examination that meets the original recommendation criteria, then the AI algorithm automatically closes the loop while ensuring subsequent follow-up examinations, if necessary, are also tracked afterward. Based on evidence-based guidelines, the algorithm also closes the loop once a pre-specified number of follow-up examinations have been completed within a given time frame (e.g., for a recommendation that says CT at 6-12 months then CT at 18-24 months, terminate the follow-up tracking after 24 months assuming no interval change in the pulmonary nodule).

Technical Specifications


Inputs

Radiologist Report

Procedure

All

Views

N/A

Data Type

Text

Modality

All

Body Region

All

Anatomic Focus

All




Primary Outputs


Follow-up Recommendation

RadElement ID


Definition

Follow-up recommendation

Data Type

Text

Value Set

N/A

Units

N/A


Follow- up recommendation modality type

RadElement ID


Definition

Modality (if present)

Data Type

Text

Value Set

X-ray, CT, MRI, US, PET, etc.

Units

N/A


Follow- up recommendation action

RadElement ID


Definition

Action (if present)

Data Type

Text

Value Set

Biopsy, etc.

Units

N/A



Follow- up recommendation anatomy

RadElement ID


Definition

Anatomy (if present)

Data Type

Text

Value Set

Liver, right lower lobe, etc.

Units

N/A


Follow- up time interval

RadElement ID


Definition

Time interval (if present)

Data Type

Numerical +time unit/Description

Value Set

2 weeks, 6 months, annual, etc.

Units

N/A


Follow- up confidence level

RadElement ID


Definition

Confidence Level

Data Type

Text

Value Set

N/A

Units

N/A

Future Development Ideas


The AI algorithm should alert the radiologist whenever there is a follow-up recommendation that is missing from the radiology report when there should be one. In addition, when there is a recommendation, the algorithm should determine whether it fits one of the evidence-based guidelines. If it does not, then the radiologist should be alerted. 


The AI algorithm should be able to determine the likelihood for the patient compliance with the follow-up recommendation to allow an appropriate level of intervention to increase the likelihood of compliance - similar to a patient no-show algorithm. 


Based on prior history, the AI algorithm should be able to determine the likelihood of a patient going outside the system to receive his/her follow-up. This information can aid in closing the loop or possible intervention to keep the patient within the system. 


For notification, the algorithm should be able to determine the best/most appropriate person of contact based on the finding. Once determined, it should send the notification with integration into other systems to allow monitoring of confirmation of receipt. In addition, the algorithm can also incorporate an escalation procedure in case the first person of contact fails to acknowledge receipt despite multiple attempts. The patients can also be contacted based on local policies. Based on the level of integration with other systems, the proper follow-up exam can even be suggested for ease of ordering.