May 4, 2021

Single-sentence summaries improve EHR data display

INTRO / SUMMARY

The addition of single-sentence summaries can dramatically improve the information communicated when generating summaries of patient data. This example focuses on trying to display that two consultations occurred in the same day.

Watch this article as a video

The previous posts have started to explore how narrative language impacts clinical reasoning and documentation. This post will look at how the addition of small amounts of narrative information can dramatically improve the visual display of graphical information.


(Side not: I thought that some of the ideas presented here might have been new; however, I was re-reading some of Edward Tufte's books last week, and it is no surprise that he discussed these principles already several decades ago.)


Exercise summary:

This exercise aims to generate a visualization in the electronic medical record that communicates the patient's course in the hospital.


Specifically, we will try to display that a patient has had two consultations in one day.

Version 1: data collapse

There are different ways that this information could be visually presented. For instance, we could have a graph with time along the x-axis, and in one of the rows along the y-axis consultations could be indicated. Two dots could be placed into the appropriate spot to indicate that two consultations occurred on a particular day.

If this visualization had many other data points such as when lab investigations, when notes were written, when the patient had outpatient or inpatient medical appointments, it may provide a general overview of the patient's encounters with healthcare and healthcare usage. However, this very high-level visualization does not actually provide much context regarding the patient's clinical narrative.


Version 2: minimal data

Another approach would be to create a small visualization that used words such as “neurology consultation” and “rheumatology consultation” to indicate the specific types of consultations that occurred on that corresponding day. This small addition of real data is a step forward in communicating what is happened to the patient. However, this still lacks a lot of information which would be important if one was trying to review the patient's chart.


2021-April-03: Neurology Consultation
2021-April-03: Rheumatology Consultation


Version 3: useless data overload

In this third approach, we could insert a lot of discrete data points. We could add the date the consultation was requested and the date the consultation was completed. We could add the name of the physician who requested the consultation. We could have the names of who wrote the consultation and who their attending physician was. We could add the location that the consultation was performed in.


A lot of “data” could be added to this visualization, but it would dramatically increase the amount of visual space occupied, while not actually providing narrative information with regards to what occurred with regards to the patient narrative.


2021-April-03: Neurology Consultation
Consultation requested by: Dr. B. Shaw
Date requested: April 3, 2021 14:40 PM
Seen by physician: Dr. L. Smith (M3)
Location assessed: Ward 3.02-1
Attending physician: Dr. C. Clark
Date completed: April 3, 2021 14:40 PM
2021-April-03: Rheumatology Consultation
Consultation requested by: Dr. B. Shaw
Date requested: April 3, 2021 14:40 PM
Seen by physician: Dr. L. Smith (M3)
Location assessed: Ward 3.02-1
Attending physician: Dr. C. Clark
Date completed: April 3, 2021 14:40 PM


Version 4: outsourcing the problem

Continuing our approach to increase the amount of information presented, we could link to the complete textual documentation of the initial consultation request, as well as the final consultation assessment. If one were to read the request for consultation documentation, and the complete consultation itself, for each the neurology and rheumatology consultation, one could develop an accurate assessment of the patient's narrative course in hospital. However, this might be pretty time-consuming based on reviewing the full text of all information. It fails at our goal here to effectively communicate a patient's course in a hospital as fully, yet, as succinctly, as possible.

2021-April-03: Neurology Consultation
Read consult request. Read consultation notes.
2021-April-03: Rheumatology Consultation
Read consult request. Read consultation notes.


Before we move to version five, I think it is helpful to also recognize, that based on the information presented in the above four scenarios, we still don't know what is going on with the patient. We have gathered 'data' and 'facts' about the patient's clinical course, but we don't really have the story.


Let's look at the fifth proposed way to handle this.


Version 5: single-sentence summaries

This approach uses single-sentence summaries, as micro-narratives. When these micro-narratives are attached to discrete pieces of clinical data, such as a lab draw, a summary of a patient's day in the hospital, or in this case, consultation, we can rapidly generate useful summaries of a patient's clinical course.


In this particular example, one could include a single sentence to explain why the consultation was requested. Let us look at an example.


Example A:

2021-April-03: Neurology Consultation
Neurology consulted to evaluate the patient's increasing confusion.
2021-April-03: Rheumatology Consultation
Rheumatology consulted to assist in the inpatient management of the patient's autoimmune medications while in hospital.


These two sentences have added so much more information about our patient, than the large block of factual information presented in Version 3 above.


One of the other important points of single-sentence summaries is that they display the narrative arc of a patient's clinical course. Let us look at the same two data points, a neurology and rheumatology consultation that occurred on the same day, but by changing one of the sentences, we dramatically change the information communicated.


Example B:

2021-April-03: Neurology Consultation
Neurology consulted to evaluate the patient's increasing confusion.
2021-April-03: Rheumatology Consultation
Neurology has requested rheumatology’s assistance in the treatment of a new diagnosis of lupus cerebritis.


We see here, a dramatically different patient narrative with regards to why rheumatology was involved in a patient's course in the hospital. In Example A it seems that they were involved simply as a courtesy to assist in the management of some complex immunosuppressant medications that the patient was already on. In the second example, we see that there is already diagnostic momentum towards consideration of lupus cerebritis as the cause of the increasing confusion and this is why rheumatology is being involved.


Conclusion

The use of single-sentence summaries dramatically improves the information communicated. They do not take up much space. They are very fast to write. They add a new type of narrative linking information into the clinical record that is otherwise not captured in the current electronic medical record.


The ability to succinctly demonstrate a patient’s clinical course in a hospital is not possible in general in contemporary electronic medical records. The reason this is hard is that EHRs only contain factual data in discrete fields, or they contain large unstructured notes.


Single-sentence summaries linked to discrete data can be used in flexible ways to assist in the visual display of information. This type of brief narrative data can easily be displayed within graphical visualizations of a patient's clinical course or be displayed in a more linear text-only format, such as presented above.


One thing in particular that I like about this approach is its ability to demonstrate the reasoning that links a set of actions and facts together. Examples A and B show the same facts, two consultations took place, but it demonstrates a very different reason around these events.


Suppose it turns out a day or two later that lupus cerebritis is, in fact, very much the wrong diagnosis and that an overly zealous junior medical student made a mistake during the initial consultation that led the teams down the wrong path. In that case, the use of brief single-sentence summaries makes it easier to review the chart retrospectively and see exactly where a potential diagnostic reasoning error may have entered into the story.


In the case presented above, once the consultation has occured, it may be better to display in addition to the reason for consultation, a single sentence summary of the conclusion from the consultation.

In the future, we'll look at how single-sentence summaries can improve the display of lab data.

Subscribe for updates

Get an email update weekly, monthly, or quarterly.
Subscribe