VA RFI: AI Based Text Classifier for Identifying Suicidal Ideations in Patient Interaction

Notice ID: 36C10B22Q0069

“BACKGROUND

VA has a requirement for an Artificial Intelligence (AI)-based text classifier utilizing Machine Learning (ML), Natural language Processing (NLP), and Chatbot software for patient interaction to identify suicidal ideations or likely self-harm in text responses for Veterans calling into the Veterans Crisis Line (VCL). VA currently processes these text based responses with the Medallia survey rules-based decision engine. Though this creates cases, it results in a significant number of false positives for suicidal ideation. VA requires a more sophisticated NLP for more accurate referrals of suicidal ideation.”

“From December 2020 to June 2021 VA’s National Artificial Intelligence Institute hosted a Tech Sprint. One vendor (SoKat) was rated to have the highest NLP capabilities and most specifically NLP capabilities related to suicidal ideation. VA have reviewed several other AI health applications and chat bots but have not found an NLP AI with similar capabilities. VA is considering issuing a sole source to SoKat unless other off the shelf applications are identified with equal or greater capabilities.”

“SCOPE OF WORK

The Contractor shall provide all resources necessary to accomplish the requirements described in this PWS. The VA shall provide data, data repository, VAEC AWS Arches, tools and other technology VA identifies as appropriate to fulfill this contract. The Contractor shall develop train, validate, test, and report data ingested by the SSIE.”

“The data and SSIE will be operated within the Arches Platform as a Service (PaaS) and the Contractor shall use FedRAMP tools available in Arches following VA DevSecOps, Agile, Data Privacy, Security principles, VA PIV and Security requirements regarding access to VA systems and services. SSIE will be a minor application with an ATO under Arches reported in eMASS managed platform by OI&T…”

SPECIFIC TASKS AND DELIVERABLES

“To address and improve OMH VCL assessment of Veteran survey responses, the SSIE shall utilize an AI, ML, NLP Engine to measure distress/dysregulation and Psych Ache (Pain) in terms of crisis intervention (as a more attainable and significant upstream prevention value) in detecting pain and dysregulation in the Veteran response text analysis model to determine indicators of Suicidal Ideation (SI) and intent within plain vernacular English text data. The SSIE shall interpret the data to determine the degree of distress/dysregulation/psych ache related to SI and intent with more accuracy than a rules-based system. SSIE results shall be captured in a format identified by the VCL staff, which shall be determined during Human Centered Design (HCD) discussions. The goal is to score all survey responses and permit the VCL to identify the threshold they would deem appropriate for VCL staff intervention and consultation and be able to provide the responses, which correlate to the score identified by the SSIE. Once the VCL SME approves the model, this can be made available in any way the VCL prefers (e.g., encrypted report via e-mail, an Application Programming Interface (API), etc.).  VHA will supply the survey data within Arches. The SSIE and SSIE output (e.g., report, alerts) shall be managed within Arches using FedRAMP tools, while employing DevSecOps and Agile product management principles…”

Read more here.

Advertisements

LEAVE A REPLY

Please enter your comment!
Please enter your name here