Clinical Decision Support for Better Care
Providing guidelines for radiologists at the point-of-care has measurable impact on compliance and patient care.
“I think we are closer to providing AI support for radiologists’ decision support than we are for our ordering providers,” said Tarik Alkasab, Massachusetts General Hospital Boston. He described a major problem with guidelines for clinical professionals: They were not appropriately available to the people providing the care, “unless the radiologist taped them right next to the workstation.”
An important step to do better was encoding the guidelines in a computer usable structure. The structure captures
- Features – the elements of a described lesion will be used to determine the output of the algorithm. The elements also include synonyms of those features that might be used in reports.
- Decision Tree – the actual logic that determines the output of the algorithm based on the lesion’s features.
- End points – Templates of the generated text to be inserted into the body, the radiologist’s impression of his findings, and his recommendations on what should be done
The result was an open framework for point-of-care computer-assisted reporting and decision support (CAR/DS) for radiologists, which Alkasab et al. described in detail in the J Am Coll Radiol 2017. “We hope that this initial effort can serve as the basis for a community-owned open standard for guideline definition that the imaging informatics and VRS vendor communities will embrace and strengthen”, say the authors.
Meanwhile, the ACR has been working on turning their content into this new format and calls it ACR Assist™. It is work in progress intended to be a clinical decision support framework for structured clinical guidance to radiologists. ACR Assist™ can be incorporated naturally into the radiology workflow.
Its core clinical components include
- Structured classification and reporting taxonomies such as BI-RADS, LI-RADS, and PI-RADS
- Care pathways and algorithms on how to deal with incidental findings
- Classification and communication needs for actionable findings
„These contents can immediately be plugged into the framework that runs in different vendors’ supporting software”, said Alkasab.
Practical impact: incidental findings
“Obviously, providing this support at the point-of-care has an effect”, he said and gave an example: In abdominal CT, radiologists may see incidental pulmonary nodules in the lung base. Alkasab and colleagues assessed their compliance with the guidelines for making recommendations (J Am Coll Radiol 2016). Their result: When radiologists did not use the Clinical Decision Support (CDS), they were only in compliance with guidelines in about 50 percent of the cases. But when they actually used the tool, 95 percent were in compliance.
“It also matters to our providers how we do this”, explained Alkasab: When recommendations for follow-up were made, there was a much higher likelihood for their decision about follow-up imaging to be compliant. “And this can have direct impact on patient care.”
Motivations to use CDS at point of care
Radiologists can contribute to a better care by providing more consistent reports. “We can also protect ourselves better against medicolegal issues”, said Alkasab. “And we can also demonstrate the quality of our work”, for example when it is about financial incentives from partners or payers.
Additionally, these tools allow incorporating data from various other sources into the radiologist’s environment (upstream). This likely provides a better, more usable output by radiologists for example by structured recommendations or data for registries and research (downstream). “These data can become triggers for new image-driven care pathways”, concluded Alkasab.
“How would you deal with cases in which computer-aided diagnostics recommend a different thing compared to what you think as a radiologist – and you don’t agree with it?”
Ramin Khorasani, Harvard Medical School Boston, answered, “In our setting, you could make an effort in reviewing so that you are aware of what your colleagues think.”
Tarik Alkasab added, when using Clinical Decision Support for radiologists or for ordering providers “it is important that this is not a system that you set up once, but you have to track people and have they use it. The target is not to get 100 percent compliance.”
Presentation Title: Application of Machine Learning in Clinical Decision Support Systems
Speaker: Tarik K. Alkasab, Massachusetts General Hospital Boston, MA/USA
Session code: RCC24D