The Patterns That Every AI/ML Platform Missed
Artificial intelligence (AI) and machine learning (ML) are expected to transform healthcare. Clinical decision support (CDS) tools are expected to streamline physician workflows while intelligently personalizing patient care, decreasing physician burden, and enhancing patient care with precision and accuracy.
However, patients currently do not have access to their own intelligent summaries or data outputs. While patients do not see documentation in their medical records if their physicians are utilizing CDS outputs to guide their care, they should be able to submit a medical record request to request a copy of the CDS outputs being used at the point of care. Patients need to be granted access to CDS outputs that are used to guide patient care and care coordination. This is a matter of patient safety, transparency, and patients’ HIPAA Right of Access if the output is being used to make decisions about an individual.
It is concerning that many patients are unaware of the range of digital health AI/ML tools that their providers may be leveraging to guide their care. Most patients have no idea these outputs exist and currently have no way to request, inspect, access them, or correct the information depicted if errors exist.
Big Tech CDS Outputs Paving the Way
Big tech plays a pivotal role in developing and deploying CDS tools. Here are three high-level examples:
- Google Health’s Care Studio harmonizes patient health information from many sources, weaving an integrated view of patients’ medical history and health information. Results from different electronic health records (EHRs) are depicted together. Physicians and nurses may use features in Care Studio, such as the search function, to generate outputs that will better inform patient care decisions. When such an output guides patient care, such as diagnosis, treatment planning, medication adjustments, and immunization recommendations, these outputs should be documented in the patients’ medical record.
- Care Studio’s newest feature, Conditions, uses natural language processing (NLP) to extract key portions of patients’ records to produce summaries of individual health conditions. A patient’s care team may then further explore critical information related to the management of these conditions, such as medications, lab work, vital signs, etc.
- Best Care for My Patient is a tool that will harness the power of over 140 million patients’ records to help clinicians understand the most effective care practices for their own patients. According to an interview with Dr. Jackie Gerhart, a physician at Epic, conducted at HIMSS22, Epic’s Best Care for My Patient is anticipated to go live sometime in 2023.
CDS Outputs: The Road to the Right of Access
CDS outputs should be documented in the medical record, such as in clinical notes that are shared with patients. The United States Core Data for Interoperability (USCDI) is a standardized set of health data classes and data elements utilized for interoperable health information exchange in the United States. Clinical notes are one of the data classes featured in USCDI. As of April 5, 2021, the 21st Century Cures Act mandated that eight types of clinical notes must be shared with patients, a well-recognized movement known as OpenNotes. The OpenNotes movement has a robust body of research demonstrating the positive impact on patients accessing and reading their clinician’s notes.
In the same way that OpenNotes has transformed patient care and patient empowerment, this is a call to action to ensure all patients also have access to CDS outputs. CDS outputs referenced by providers could be documented in one of the existing eight clinical notes or through the creation of a new CDS note. These notes should summarize the information from CDS outputs used to make decisions about individuals. Another option is the creation of a new data class specifically for the exchange of CDS outputs as data. Our standards development community should be future-facing and prioritize the development of standards that will support the seamless exchange of CDS outputs.
Building Trust and Transparency Through Patient Access
At a time when the digital ecosystem grapples with building trust between health providers, patients, and consumers, there is a significant unmet need for transparency on the use of AI/ML technology that informs patient care. There are patient right-of-access barriers as more sophisticated, electronically available, structural data is generated at the point of care that most patients have no idea may influence their or their loved one’s care and consequent cost of care.
Patients should have a right to access, inspect, and correct errors that may be detected. It is well documented that errors are frequently reported in medical records. Correcting errors in AI/ML generated outputs builds trust between clinicians and patients, protects patient safety, and improves data quality and hygiene. Many patients want to curate their medical records and ensure that the information their care team is using is correct, comprehensive and actionable.
Why Does It Matter?
Individuals’ right of access must be protected. We must recognize that we are creating new digital information barriers by actively barring patients from having access to these outputs. Individual access to outputs must be a priority, especially if the outputs are being utilized by physicians to drive decisions about individuals. Only then will we have the potential to unlock the true power of AI/ML to augment not only patient behavior but also to empower their life with their diagnosis.
Making CDS outputs available to patients will reduce patient administrative burden (PAB), or the work that individuals need to do to get the care they need. Patients may finally be able to understand decades of medical records that have been intelligently summarized into a few key phrases, sentences, or paragraphs. These outputs may be strong assets for improving health literacy, patient education, and equity. Patients may be able to use outputs to fight insurance denials for care that was recommended by their board-certified physician yet denied by their payor as medically unnecessary. Outputs from AI/ML may help patients provide critical information for expediting the application for Social Security disability benefits by more efficiently demonstrating medical necessity. Access to outputs may streamline connections to social community supports as well. This is equity by design.
Placing CDS Outputs in Patients’ Hands: Real-World Use Cases
CDS outputs can do more than augment clinician decision-making. CDS outputs can augment patient decision-making as individuals navigate their diagnoses.
Imagine a patient is being treated for advanced cancer and their doctor at their local community cancer center in New Jersey uses Best Care for My Patient to guide their treatment. Imagine the patient hits a juncture where their treatment is no longer working, and they have disease progression. If the patient wants to pursue a second opinion for a new treatment plan, do they go to another doctor at another facility that also uses Best Care for My Patient to guide their cancer care? Would this give the patient the same recommendations? Should the patient opt to specifically see a doctor who is known for leveraging another form of CDS, such as xCures xDecide, and compare the recommendations? Or should the patient intentionally go to another physician and cancer center that also uses Best Care for My Patient to see if the recommendations match? Epic currently lists healthcare delivery organizations state by state as participating in Cosmos, the patient records database that powers Best Care for My Patient.
In the above example, in order to make the best treatment decision, the patient needs a copy of all of the data that has guided their care to this point, including CDS outputs. With additional access to all of the data, patients will likely need more guidance on navigating their care and coordination of care with AI/ML CDS tools.
Many patients are part of peer health support networks. Some of these networks have patients with similar characteristics, such as the same genetic mutations, like Huntington’s disease or sickle cell anemia, or tumor markers, like the ROS1ders in lung cancer. Peer health support networks are likely to share credible insights with patient communities. The outputs of a tool like Best Care for My Patient could be transformative for the way patients go about their care, especially in cases where that are no standards of care, where all treatment options have been exhausted. In many circumstances, access to CDS outputs could be a matter of life or death.
Brand strategies and healthcare delivery organizations’ workflows cannot continue to prioritize getting CDS outputs solely to clinicians. If we truly want equity by design, patient engagement, transparency, ethical use of AI/ML, and shared decision-making, these gray areas need to be addressed or we are turning a blind eye to augmented information blocking.
The Call to Action: Open the Data Up
There needs to be explicit guidance from the Office of the National Coordinator for Health Information Technology on CDS outputs, confirming that:
1. CDS outputs used to make decisions about an individual patient’s care are part of the designated record set (DRS) as defined by Standards for Privacy of Individually Identifiable Health Information, where it is stated “individuals have a right of access to any protected health information that is used, in whole or in part, to make decisions about individuals.”
2. CDS outputs used to make decisions about an individual patient’s care are EHI, where EHI is defined as electronic protected health information (ePHI) that would be included in a DRS.
2A. Prior to October 6, 2022, as per the Information Blocking Rules, the information blocking definition is limited to the EHI identified by data elements represented in USCDI v1. Until the October 6, 2022 deadline, only CDS outputs that align specifically to the data classes and elements defined by USCDI v1 would, in theory, be considered EHI.
2B. Come October 6, 2022, the definition of information blocking would then include the entire scope of EHI that falls within the DRS.
3. Any actor who does not provide access to CDS outputs, as per the guardrails described in 2A and 2B above, is committing information blocking, unless one of the 8 Information Blocking Exceptions applies.
4. Patients are at risk of experiencing significant harm if outputs from AI/ML-enabled innovations have errors. Actors leveraging these technologies must have mechanisms in place for patients to correct and amend their health information. The Patient Request for Corrections Implementation Guide driven by the HL7 Patient Empowerment Work Group needs to be prioritized.
Outputs from AI/ML-powered tools that clinicians use to make decisions about individual patient care are part of the designated record set, and all patients have a right to access that information about them.
Grace Cordovano (firstname.lastname@example.org) is a board-certified patient advocate and the founder of Enlightening Results, a patient advocacy practice, and co-founder of Unblock Health, which empowers patients and their care providers to reduce and eliminate information blocking practices and barriers to patient care. Cordovano is a member of the HITAC Interoperability Standards Work Group, the HIMSS Public Policy Committee, the HL7 Patient Empowerment Workgroup, and the co-chair of the Protecting Privacy to Promote Interoperability (PP2PI) patient sub-group.
Read the AHIMA Policy Statement on Individual Access to Health Information.
Take the CE Quiz