Clinical Decision Support
Medplum enables building and delivering custom clinical decision support tools for a variety of applications. This guide focuses on the regulated ONC Criteria for Clinical Decision support which is criteria (a)(9) of the HTI criteria on predictive CDS.
The guide will walk through the three major categories of Clinical Decision Support (CDS), as defined by the regulations, and how to enable said CDS on Medplum.
Medplum is not currently certified for (a)(9) but is pursuing certification. Contact us at info@medplum.com for details.
Predictive
Predictive clinical decision support technology is "intended to support decision-making based on algorithms or models that derive relationships from training or example data and then are used to produce an output or outputs related to, but not limited to, prediction, classification, recommendation, evaluation, or analysis.”
Large language model-based clinical decision support tools, as well as tools that use algorithms for risk assessment or triage, fall in the predictive clinical decision support category.
Medplum enables many implementations with predictive clinical decision support. Per the HTI final ruling, predictive clinical decision support systems will become regulated in December 2024. Any system certified to g10, b2, f1 or e1 will be required to provide Insight Reports as part of maintaining their certification.
The following sections describe best practices to prepare for a predictive clinical decision support certification.
Training Data
Demonstrating which training data was used to train an algorithm (and keeping a record of the versioning of said data) is part of the certification process. In the context of a Medplum implementation, be prepared to keep all of your training data in a Medplum project which will show which dataset was used to train the models and that the data is updated (feedback loops).
Code Systems
It is recommended that data is tagged with UMLS code systems, including LOINC, SNOMED and RxNorm. For certification, data must conform to USCDI profiles. Implementations of predictive clinical decision support sometimes include annotating data with code systems using an algorithm or large language model.
Insights Reporting
Electronic health records that support predictive clinical decision support will be required to report on the usage of their product. Prepare the following basic statistics as part of certification:
- Number of times the decision support was used
- Number of unique clinicians who used it
- Number of times it was updated
- Number of complaints received