Skip to main content

​
Overview

Elation AI product functionality includes multiple product features leveraging AI in the application. The information included in this page includes requirements for the Certified EHR Technology Program for Predictive Decision Support Interventions which are defined as AI utilization in the product and which Elation Health certified to. Requirements for AI in the product include a summary of Elation Health’s AI Risk Management Practices and predefined Source Attributes included in this page. To turn off Elation AI functionality, contact Elation Health Support (support@elationhealth.com). You can find more information on Elation’s AI Data Use Policy. It is important for You to understand some of the limitations of AI and LLMs generally to utilize the Services safely and responsibly. Because AI technology is still evolving, Outputs may be generated that are factually incorrect. The following Elation Health Help Center Articles describe Product Functionality that include AI:

​
Decision Support Interventions Regulatory Documentation

​
Introduction to Decision Support Interventions

Elation Health offers a certified EHR technology platform to meet regulations per the ASTP ONC for all CEHRT. There are multiple requirements for the CEHRT required criteria Decision Support Interventions (b)(11). All required documentation can be found here.
  • Predictive Decision Support Interventions (Developer Supplied Source Attributes)
  • Predictive Decision Support Interventions (Developer Supplied Risk Management Framework Summary)
  • Evidence-based Decision Support Interventions (Source Attributes)
  • Predictive Decision Support Interventions (User Supplied Source Attributes)

​
Predictive Decision Support Interventions: Elation AI which includes Note Assist

Elation Health offers use of artificial intelligence (AI) in our application as the product feature Note Assist. Note Assist is Elation’s AI-powered scribing and visit note generation solution. This tool is thoughtfully embedded in the visit workflow, seamlessly incorporating ambient scribing capabilities into your encounter documentation process. After purchasing Note Assist, anyone who has a Provider Level User account (e.g. physicians, non-prescribing providers) can use Note Assist. Note Assist will create detailed content for your Visit Note and intelligently identify the appropriate fields to fill based on your transcription. Related information will be organized into their respective sections, reflecting how you would document an encounter. More information on the product feature of Note Assist can be found in this Help Center Article. All content that is transcribed from Note Assist can be edited or deleted before signing the visit note, with all visit note signers maintaining content ownership of the visit note. As part of the requirements for Predictive Decision Support Interventions, Source Attributes for Note Assist are included in the table below, and the Risk Management Summary is included in this page as well. To record and change source attributes users can leverage the feedback form in the application to document those changes and record new source attributes.

​
Source Attributes for Elation AI which includes Note Assist

Source Attributes to be Published and Available to End Users
1. Details and output of the intervention, including: Name and contact information for the intervention developer; Elation Health 530 Divisadero St, #872, San Francisco, CA 94117 415-231-5164 Funding source of the technical implementation for the intervention(s) development; • Same as the intervention developer. • Description of value that the intervention produces as an output; and Visit Note Templates Whether the intervention output is a prediction, classification, recommendation, evaluation, analysis, or other type of output.
4. Description of tasks, situations, or populations where a user is cautioned against applying the intervention; and Known risks, inappropriate settings, inappropriate uses, or known limitations Exclusion and inclusion criteria that influenced the training data set; • Use of variables in paragraph (b)(11)(iv)(A)(5)-(13) as input features; • Description of demographic representativeness according to variables in paragraph (b)(11)(iv)(A)(5)- (13) including, at a minimum, those used as input features in the intervention; • Description of relevance of training data to intended deployed setting. Elation Health uses a contracted Large Language Model (LLM) training data set and then employs that for speech to text transcription. Elation Health does not own the LLM and can not describe the training data set including training the data set. The USCDI data classes and elements that are captured in Note Assist are Patient Demographics for the patient’s preferred gender to be consistent with in the summary provided.
7. Quantitative measures of performance, including: Validity of intervention in test data derived from the same source as the initial training data; • Fairness of intervention in test data derived from the same source as the initial training data; • Validity of intervention in data external to or from a different source than the initial training data; • Fairness of intervention in data external to or from a different source than the initial training data; • References to evaluation of use of the intervention on outcomes, including, bibliographic citations or hyperlinks to evaluations of how well the intervention reduced morbidity, mortality, length of stay, or other outcomes. The Note Assist will write a transcript that would be the same source as initial training data. Additional testing includes manual review before editing discrepancies between accuracy of verbal recordings and transcriptions then editing in test data.

​
AI Risk Management Summary

Elation Health has implemented suitable risk management controls in place to ensure AI use in the product is fair, appropriate, valid, effective, and safe (FAVES). Our implemented risk management controls align with the NIST RMF, specifically, section Govern 6.0. We have implemented specific controls to meet the requirements set forth by the ASTP/ONC for certification of Predictive Decision Support Interventions. Included in this documentation is a summary of policies, procedures, and actions taken to maintain this efficacy. Elation Health has built a product using an established LLM and does not control the LLM training data, nor is data used in Elation contributing to training the LLM. AI as Note Assist in the Elation Health product is subject to the same internal rigor as other privacy and security controls that interact with patient health information (PHI). Note Assist as a feature is able to be turned on or off easily as a system wide feature or as a product feature for user accounts. 1. Auditability of AI Systems We have mechanisms to document the development process of each AI system, from sourcing of training data to deployment. This includes:
  • Detailed development logs that track the creation and modification of algorithms as part of the SDLC processes.
  • Outcome logs capturing both the positive and negative impacts of AI decision-making, which enables continuous evaluation and audit.
2. Third-Party AI Systems and Explainability When acquiring AI systems from third parties, we ensure a level of explainability through:
  • Contractual obligations that include vendors to provide documentation on the underlying logic and reasoning of their AI models.
  • Post-deployment testing to assess model interpretability and transparency as noted above in source attributes.
  • Ongoing internal reviews to validate the clarity of AI outputs and verify system behavior under different scenarios.
3. Vulnerability and Bias Reporting by Third Parties Our organization has developed a clear and accessible process for users to report potential vulnerabilities or biases in the AI system. This includes:
  • The required feedback form mechanism for end users and other stakeholders to submit potential issues.
  • Regular risk assessments based on feedback, with corrective actions documented and communicated back to stakeholders as part of the SDLC process.
  • A monitoring group that actively tracks reports of biases and proposes interventions when necessary.
4. Comprehensive Risk Management Plan We have developed a risk management plan that addresses key risks associated with AI system acquisition, procurement of software, and operational infrastructure. This includes:
  • Vendor risk assessments to evaluate third-party software, ensuring compliance with cybersecurity and AI ethics standards.
  • Cybersecurity controls that safeguard the integrity of computational infrastructure, as well as regular updates to patch vulnerabilities.
  • System failure contingency planning, including fail-safe mechanisms and redundancy in critical AI applications.
Summary of Policies and Procedures 1. Development Policy
  • All AI projects must maintain a comprehensive development log, including details on algorithm design, data usage, and system updates. This ensures a clear lineage for all AI systems.
2. Procurement and Vendor Management Procedure
  • Prior to AI acquisition from third parties, all vendors must undergo a risk assessment to evaluate the explainability, transparency, and security of their AI products. Vendors must also provide ongoing access to audit trails for independent third-party review.
3. AI Vulnerability and Bias Reporting Procedure
  • External stakeholders, including users and vendors, are encouraged to report any issues related to AI system performance, vulnerabilities, or biases via a secure online portal. All reports will be addressed by our development team using our normal development feedback process (SDLC).
The summary of these policies and practices are part of our broader governance framework, which aligns with industry standards and resources such as the NIST AI Risk Management Framework. We are committed to fostering responsible AI practices that prioritize transparency, accountability, and ethical considerations in all our AI initiatives.

​
Source Attributes for Evidence Based Decision Support Interventions

​
Introduction

Evidence Based Decision Support Interventions in Elation’s product use the electronic clinical reminders functionality that remind providers and relevant staff users of clinical interventions that are related and applicable to electronic clinical quality measures (eCQMs). Use of eCQM reminders can be enabled by both the practice and the clinical user in their settings and additional configuration details can be found in this Help Center Article. Each eCQM is developed by an outside recognized agency, such as CMS or NCQA, and each eCQM has specifications which dictate what data and how each are included in each measure including data related to USCDI. To find all relevant information (referred to as source attributes by ASTP/ONC), such as developer information and use of USCDI elements, you may find the information associated in the table below with additional details in each of the individual Help Center Articles. Source Attributes associated to each eCQM reminder can be found in eCQI specifications:
  • Bibliographic citation of the intervention (clinical research or guideline)
  • Developer of the intervention (translation from clinical research or guideline)
  • Funding source of the technical implementation for the intervention(s) development
  • Release and, if applicable, revision dates of the intervention or reference source
  • Use of USCDI Data
eCQMSource Attributes
[CMS69v12] Preventive Care and Screening: Body Mass Index (BMI) Screening and Follow-Up PlanElectronic Clinical Quality Improvement Measure Specification (this measure specification provides information on the measure developer, steward [CMS], annual revision updates, and use of USCDI data in the measure specification. The Help Center Article linked also describes how Elation Health has built the product to the measure specification.
[CMS122v12] Diabetes: Hemoglobin A1c (HbA1c) Poor Control (> 9%) (MIPS 2024)Electronic Clinical Quality Improvement Measure Specification (this measure specification provides information on the measure developer, steward [NCQA], annual revision updates, and use of USCDI data in the measure specification. The Help Center Article linked also describes how Elation Health has built the product to the measure specification.
[CMS124v12] Cervical Cancer Screening (MIPS 2024)Electronic Clinical Quality Improvement Measure Specification (this measure specification provides information on the measure developer, steward [NCQA], annual revision updates, and use of USCDI data in the measure specification. The Help Center Article linked also describes how Elation Health has built the product to the measure specification.
[CMS125v12] Breast Cancer ScreeningElectronic Clinical Quality Improvement Measure Specification (this measure specification provides information on the measure developer, steward [NCQA], annual revision updates, and use of USCDI data in the measure specification. The Help Center Article linked also describes how Elation Health has built the product to the measure specification.
[CMS130v14] Colorectal Cancer ScreeningElectronic Clinical Quality Improvement Measure Specification (this measure specification provides information on the measure developer, steward [NCQA], annual revision updates, and use of USCDI data in the measure specification. The Help Center Article linked also describes how Elation Health has built the product to the measure specification.
[CMS131v12] Diabetes: Eye Exam (MIPS 2024)Electronic Clinical Quality Improvement Measure Specification (this measure specification provides information on the measure developer, steward [NCQA], annual revision updates, and use of USCDI data in the measure specification. The Help Center Article linked also describes how Elation Health has built the product to the measure specification.
[CMS139v12] Falls: Screening for Future Fall Risk (MIPS 2024)Electronic Clinical Quality Improvement Measure Specification (this measure specification provides information on the measure developer, steward [NCQA], annual revision updates, and use of USCDI data in the measure specification. The Help Center Article linked also describes how Elation Health has built the product to the measure specification.
[CMS149v12] Dementia: Cognitive Assessment (MIPS 2024)Electronic Clinical Quality Improvement Measure Specification (this measure specification provides information on the measure developer, steward [American Academy of Neurology], annual revision updates, and use of USCDI data in the measure specification. The Help Center Article linked also describes how Elation Health has built the product to the measure specification.
[CMS155v12] Weight Assessment and Counseling for Nutrition and Physical Activity for Children and Adolescents (MIPS 2024)Electronic Clinical Quality Improvement Measure Specification (this measure specification provides information on the measure developer, steward [NCQA], annual revision updates, and use of USCDI data in the measure specification. The Help Center Article linked also describes how Elation Health has built the product to the measure specification.
[CMS165v12] Controlling High Blood Pressure (MIPS 2024)Electronic Clinical Quality Improvement Measure Specification (this measure specification provides information on the measure developer, steward [NCQA], annual revision updates, and use of USCDI data in the measure specification. The Help Center Article linked also describes how Elation Health has built the product to the measure specification.

​
Source Attributes for User Supplied Predictive Decision Support Interventions

User Supplied Predictive Decision Support Interventions are associated with Elation Health’s Caregaps API functionality. This functionality allows a user with subscription access to the Developer Platform to establish and load into Elation their own predictive decision support interventions and create a clinical intervention for users in their instance to meet a specific care gap. Source attributes, definitions of the interventions, are all established by the user. Users that create the Caregap are responsible for identifying the source attribute information required for Predictive Decision Support Interventions, including accessing, recording, and changing source attributes related to the predefined PDSI. All of these source attributes can be included in loading information into the API and additional details on the functionality, including accessing the functionality can be found here.