Tuesday, June 20, 2023

Journal Club: CMS AI Playbook (2022, 38pp)

Did you know there's a CMS "AI Playbook?"  There is, coming in at 38 pages, last updated 2022.  It's more about agency priorities that any specific products (like AI in radiology), but it may still be of interest.


Go straight to AI at CMS dot Gov:  https://ai.cms.gov/

Find the PDF playbook here: https://ai.cms.gov/assets/CMS_AI_Playbook.pdf 

##

HHS also has a "playbook" for Trustworthy AI, dating from 2021:

https://www.hhs.gov/sites/default/files/hhs-trustworthy-ai-playbook.pdf

There's an HHS website for AI and links here:

https://www.hhs.gov/about/agencies/asa/ocio/ai/index.html

See the webpage for HHS AI Strategy here:

https://www.hhs.gov/about/agencies/asa/ocio/ai/strategy/index.html

There's also an HHS "Inventory of Use Case."  Here.  That's a table.  For some use cases in paragraph form, here.

click to enlarge
###

###

Here's a GPT4 summary of the half-dozen AI use cases, distilled from the document just cited:

GPT4: 

This document presents the Department of Health and Human Services' (HHS) use cases of artificial intelligence (AI) for the fiscal year 2022. It provides details of various projects where AI has been used or is planned to be used to improve operational efficiencies and results.

FDA Counterfeit Detection Device Version 5 (CD5): The CD5 is a handheld device that uses LEDs to detect counterfeit pharmaceuticals. AI has been used to improve this device by providing a device-generated conclusion, reducing the need for subjective human interpretation of the results. The aim is to speed up the identification process from tens of minutes to mere seconds.

CMS Center for Program Integrity (CPI) Fraud Prevention System Models: This project aims to more effectively detect, prevent, and prioritize potential cases of Medicare and Medicaid fraud, waste, and abuse using machine learning. The system is expected to identify underperforming or obsolescent fraud identification models for deactivation.

FDA Emerging Chemical Hazard Intelligence Platform (ECHIP) and Warp Intelligent Learning EnginE (WILEE): This project aims to leverage AI to better forecast the emergence of a new chemical hazard by screening a broader range of data sources. The model is trained to identify patterns that can be used to anticipate chemical hazard signal detection over time and across data sources.

CMS Rapid Authority to Operate (ATO): This project's goal is to reduce the burden of security compliance planning and accelerate the time to ATO using AI. It creates a component library of reusable templates of control narratives related to specific technology requirements.

CMS OSFLO Help Desk Chatbot: This project aims to assist the help desk in responding to employee and contractor questions more rapidly and effectively by automating voice and text responses for general security and badging questions. The chatbot is projected to resolve 65% of incoming calls.

HRSA Electronic Handbooks (EHB) AI Chatbot: This project aims to develop a self-service platform with 24/7 availability capable of communicating knowledge and action-based responses to grantees using regular natural conversational expressions. The chatbot resolves 90% of incoming inquiries without any involvement of the customer support staff.

OIG Grants Analytics Portal: This project's goal is to turn grants data into actionable intelligence to support the oversight of HHS grants recipients and grant programs. The model extracts single audit finding text from PDFs and assesses multiple dimensions of risk.

NIH eRA Internal Referral Module (IRM): The IRM aims to eliminate referral bottlenecks in the NIH grant referral process by automatically referring grant applications to Program Officers once they are received. The system operates at a high accuracy rate (92%) and has effectively eliminated the referral bottleneck.