Achieve True Value-Added Radiology With ‘Real-Time’ EMR Data Mining


EditorMichael Grey’s Note: This article was originally published in the February 2015 edition of DOTmed® HealthCareBusiness News and is republished here with permission.

The concept of value-added radiology (VAR) embraces a number of objectives — from initial involvement with care team physicians in determining the correct study to order, to improving the quality of the interpretation and turnaround time. This VAR concept coincides with the shift in radiology from volume- to value-based reimbursement. Perhaps a more fundamental point of VAR should be elevating the radiologist’s role above merely interpreting the imaging study…basically interpreting the study in a more holistic (and accountable) manner and not based on the images alone.

To be more involved through the entire process, what the radiologist needs is easy and rapid access to the clinical information on the patient that is relevant to the radiology study and therefore context sensitive.

Unfortunately, radiologists are frequently limited from seeing the broad range of clinically relevant information related to the patient’s condition. Most PACS can only provide access to those prior radiology reports and associated images that are stored on the specific PACS being used. The PACS might additionally provide access to the electronic forms or scanned documents containing procedural notes, calculations and measurements that were contributed by the technologist.

The scanned documents typically contain only that clinically relevant information collected during the patient’s visit to the department. In addition to the PACS, the Radiology Information System may provide access to the original order and thus the reasons for the requested study, and perhaps a summary of the patient’s history. That’s it!

Even when the above information is accessible, searching through an unstructured collection of electronic documents in the PACS or the RIS to discover more than the reason for the order is often considered by many radiologists to be too cumbersome and time-consuming to be worth the effort.

In any case, the real source of the patient’s longitudinal clinical information is the EMR, where the radiologist could discover the patient’s complete medical history, the care summaries, results from imaging procedures beyond radiology as well as non-imaging procedures (i.e. pathology), lab results, surgical history, etc.

While the EMR is a vast repository of clinical data, like a three-ring binder of information organized by tabs, this data is frequently not accessible through the radiologist’s diagnostic workstation, and logging into the EMR directly makes the process of accessing and searching for patient-specific information a burden on the radiologist.

Beyond that, most radiologists are used to working in a highly visual and efficient environment (very few clicks). This is in direct contrast to the way the radiologist would have to consume data from an EMR, which is highly textual and a click-heavy user experience.

What the radiologist needs is a concise summary of the patient’s clinical information that is relevant to the specific radiology procedure about to be interpreted. This “clinical summary” should be automatically presented to the radiologist within a display window of the diagnostic workstation. The following two examples illustrate how this would contribute to the concept of VAR, by improving diagnosis and potentially sparing unnecessary exams or an invasive procedure (e.g. biopsy), potential future complications, etc.

A patient is referred to radiology for a CT of the abdomen to rule out liver cancer due to multiple hepatic masses visualized. Having access to the relevant lab results and vital signs would allow the radiologist to discover that the patient has fever and elevated WBC, which changes the diagnosis from cancer to infection and prevents unnecessary costs and discomfort of hospitalization and liver biopsy.

An ED patient with chest pain and a negative ECG is referred to radiology for a chest CT angiography to rule out aortic dissection because of family history of heart disease. Access to the relevant patient record data such as allergies, clinical notes and medications would reveal this patient has active UTI (Urinary Tract Infection) and was given a type of penicillin to treat this condition, although the patient is allergic to penicillin. Having an easy way to consume this data from within the diagnostic radiology application would allow the radiologist to provide a more accurate, high-quality diagnosis.

The ideal solution would be the automated mining of one or multiple EMRs and other potential data sources for predefined clinical information that is relevant. The final goal is to provide the radiologist with all the relevant clinical information, regardless of where it is stored, in an easy to consume manner. The searches are either performed when the study is first opened by the radiologist, or prior to opening the study (as far back as necessary based on the time required to execute the mining process and assemble the summary).

Ideally, the search criteria would be user-defined. This would be as simple as one set of search criteria per procedure type for the entire department, but eventually it should be possible for each radiologist to build their own set of search criteria. In the interest of efficiency, the clinical summary would ideally be limited to a single screen presentation with the option to allow the radiologist to dig deeper.

Of vital importance, the information could be retrieved online, when it is needed from the source of truth (i.e. EMR). This assures that this information is up-to-date, and it eliminates the need from each consuming system (i.e. PACS or applications) to maintain copies of the information. The EMR vendor would provide a Web-based application programmers interface (API) between PACS and EMR that would allow trusted systems to retrieve the information in a secure, standard and reliable manner. The Web API would enable real-time or near real-time retrieval of clinical information from the EMR by other applications (i.e., PACS).

Actually, this concept of tapping the EMR for clinical information is not new. Interviews of radiologists frequently uncover the concept of volume versus value, as in less volume of studies read in favor of more value in each case interpreted.

Radiologists will readily admit that the images are not the whole “picture.” While there is recognition that most of the truly useful clinical information is in the EMR, radiologists will complain that that data is difficult to access. While PACS vendors are now being asked to launch the EMR in patient context, the radiologists are also asking for some methodology to make searching for the relevant data both easy and fast.

There are challenges to bringing this concept to reality. The process has to be largely automated. All radiologists recognize that there is useful clinical information in their EMRs, but they won’t routinely search for that information if they have to work too hard. Launching the EMR in patient context is not going to be a successful approach, because the clinical data is not effectively consumed by simply launching the EMR.

The search process cannot simply be a contextual launch into the binder of clinical data on the patient. What is needed is a search application that assembles information specific to a clinical environment. The search process must be automated and sensitive to relevance. The search process must be based on a rules engine that can convert the clinical setting into search criteria. This effectiveness and the efficiency of this type of search will improve over time.

The process of searching one or more data repositories and assembling the summary information must be very fast. There will be challenges to meeting this performance expectation. This is not like moving image data, but the process requires searching a massive volume of unstructured data, and the EMR is not necessarily good at assembling the clinical data according to clinical context (reasons for ordering a radiology study).

Another complexity, already alluded to, is that we don’t want to have to copy and store the clinical info in the PACS. That would be too slow and run the risk of pulling up old or inaccurate information.

The ideal situation is to be able to communicate very quickly with the EMR and simply present the results of the search as an assembled page of summary information. The solution is to make the information available online and retrieved on-demand, so there is no need to pre-cache and subsequently update the information (keep it synchronized with the EMR). The process must be able to query the EMR for information rather than only get messages pushed to the PACS from the EMR.

This variation on VAR is attainable, but it depends on the radiologist’s ability to effectively consume relevant clinical information from the patient portfolio. To enable that and “unlock” the information, standard APIs should be developed and made available by both the EMRs and PACS vendors because the length of time for the search depends on the API, which in turn determines how far in advance the search must be initiated. It’s definitely time to start thinking about that API.

Stay up-to-date on the latest radiology and medical imaging issues by subscribing to the Medical Imaging Talk Blog today.


Leave a Reply