Author: John Putzke
The most successful outcomes-oriented patient registries incorporate direct utilization of the data within the day-to-day operations of the clinical setting. A critical component to this approach is the development of clinical reports that facilitate the efficient and accurate comprehension of data. In additional to general organization, the following design techniques are explored: color, spatial relationships, prioritization (high-frequency questions), summation, minimization, standardized timeline, clinic note generation, and key item extraction. A case example using an epilepsy patient registry in Studytrax is presented.
Why A Clinical Report?
Patient registries are a common mechanism used for the advancement and understanding of the natural history and treatment of various disease states. All patient registries are both over-and-under inclusive depending on the aims which will be the topic of multiple upcoming articles. For this article, it is important to note that irrespective of what’s captured in the database (i.e., the content), accurate and complete data capture is critical to ALL patient registries. Work-related incentives and contingencies are powerful data capture mechanisms. Indeed, the most successful patient registries implement data utilization directly into day-to-day operations. That is, staff are much more likely to enter and keep data up-to-date if it helps make their clinical work easier.
Leveraging data by designing a clinical report is a common way to utilize patient registry data. However, a report that merely presents data prompts and associated values in a tabular format is often overwhelming to perceive and time-consuming to fully comprehend, making it largely unhelpful to clinical staff. Add in the additional complexities of longitudinal data and a reflexive tabular approach to clinical report design is rendered untenable in most situations. Thus, one of the biggest challenges is to determine how to design a clinical report that best facilitates the quick, efficient and accurate comprehension of data.
In addition to general organization, some of the most important design techniques to address this challenge are: color, spatial relationships, prioritization, summation, minimization, standardized timeline, clinic note generation, and key item extraction. These are not mutually exclusive techniques, but the concepts within each area are distinct enough for a separation discussion. Each technique is presented below using examples from an epilepsy patient registry in Studytrax. It is hoped this will help facilitate the work of others addressing similar issues. Send comments, questions, suggestions and/or your design tips to firstname.lastname@example.org .
This being a general principle, it’s difficult to provide specific techniques. However, general organization cannot be over-stressed as it’s critical to creating a good clinical report. A spatially well-organized layout, informative headings, bold for emphasis, and grouping related information together (e.g., Demographics, Medical History) are some the most common techniques, but how best to organize the report can vary considerably based on the clinical setting, disease population, and the data content of the registry. Thus, instead of providing specific examples, some of the common questions used to help organize the report are provided below. It’s important to address these questions as early as possible since the answers often end up driving the database content.
- What would be helpful to see or learn from the report?
- What questions will the report answer?
- How are the data related / best grouped together?
- What are the most important data elements?
- What decisions need to be made?
- What information is needed for each decision?
- What are the most / least common issues?
- Are there any safety issues, critical events, etc.?
- What information would help with clinical work-flow?
Humans have excellent vision, even compared to our closest primate relatives. Only birds of prey and a few other species rival our vision (e.g., see Lynne Isbell’s work). Humans also evolved to have color (trichromatic) vision due to various environmental factors (e.g., ability to distinguish foliage / fruit / etc.). Color naturally draws attention and when paired with scaling information makes interpretation of results immediately apparent. For example, the chart below color codes historical (left-side) to current (right-side) seizure frequency, making improvement in the clinical picture patent. Because color is particularly attention grabbing, best to use muted colors. This also makes intentionally un-muted colors to stand out even more when appropriate.
In a series of seminal articles, Hasher and Zacks demonstrated that time (i.e., event sequence), frequency and spatial relationships are automatically encoded into memory. Thus, information conveyed through images, charts, graphs, etc. are readily grasped with minimal effort, freeing up attentional resources for other processing. When paired together with color, the results are striking. Note that some creativity may be needed as data elements don’t always directly map onto an image. For instance, the left side of the image below is a standardized map of the brain. Circles are automatically drawn over areas involved in seizure onset (i.e., red for focal and light-tan for hemispheric area). Note that the concepts “Undetermined,” “Other,” and “Independent” were incorporated into the image by including a gray box around these words. Thus, to include one of these concepts in the display, the circle is merely drawn to touch the box.
Graphs are a common, well-known type of Spatial Relationship display of data, so different variations will not be discussed here. Two things will be mentioned. First, grouping graphs by conceptual domain typically works well. Second, grouping data series by interpretive direction (e.g., higher number = Better) often helps facilitate interpretation across multiple domains and allows more lines to be packed into one graph (see below).
The best clinical reports are typically organized around the most important and/or highest frequency issues. Some examples of high priority issues include safety concerns, cardinal events, and markers of disease progression. Keep in mind, high priority issues may be only indirectly related to disease activity, such as an issue related to clinical workflow or a high frequency question. For example, the text below is automatically generated and placed near the top of a “Seizure Report” since it is needed to answer a common clinical question about driving reinstatement (also note color highlighting, and automated day/month calculation to minimize mental effort).
It’s important to note the priorities may also include those from the patient’s perspective. Certainly, assessment of patient priorities can reveal surprising findings, and are thus given a prominent position in the report. For example, the table below displays the patient’s concerns in rank order. The top three are given emphasis with bold and a larger font.
The nodal/connection-based structure of long-term memory has been well documented. Thus, providing key, summative clinic features within a report often triggers the provider’s memory about a case without having to provide extensive detail. If details are necessary, these are often tucked away in an appendix for look-up as needed. Below is example summative text used from the epilepsy registry report.
It is tempting to have the clinical report include every data element in the database, the reasoning being, “…it is collected, so it must be important.” This approach can quickly lead to a cluttered, dense, unreadable report. Often the most important and difficult work is knowing what NOT to show. Steve Jobs famously responded to a programmer who had just pitched an interface to a CD copy application with multiple button, checkbox, slider controls, ‘There can only be one button, ‘Burn’.”
In a similar fashion, information NOT relevant to a case should be hidden. For example, the image below shows the same table in one report under two different conditions. That is, if a case has Safety Events, a red check is displayed, as well as the individual safety concerns (i.e., top table). Conversely, if a case does NOT have safety concerns, the table has a green checkmark and a note stating, ‘No Events Since Last Visit’ (i.e., lower table).
Most disease registries contain medical history information. Typically there is a long list of diseases along with which family member involved (e.g., sibling, mother, father, etc.). Minimization is an important concern in the display of these data as there is often hundreds of data points, the majority of which are NOT applicable. What to display varies considerably depending on what’s captured (e.g., just the patient’s history vs. extended family members).
The below example is from a registry containing 100+ diseases, diagnoses, and procedures, each capturing which family member(s) involved (i.e., 11 total = sibling, mother, father, etc.). The top is a family pedigree display with positive history counts for each family member. The bottom provides the specifics for each family member. Both parts utilize a color-coded scale based on history counts. Taken together, key elements of the family history can be quickly grasped without the interference from an overwhelming number of items not endorsed.
A special type of Spatial Relationship technique, a Standardized Timeline, deserves a separate discussion. A common, daunting challenge is how to pull together considerably different data (e.g., various conceptual domains, data types and scaling, time frames, etc.) into a clear and concise summary that also accounts for time. A Standardized Timeline does precisely this within a single visual image.
The approach involves:
- Establishment of a consistent time frame (e.g., disease onset to current date)
- Determine key, time related events (often just a single date, but may also be a duration [i.e., start AND stop date])
- Plot events along the timeline, one row per event. Duration is represented across multiple segments.
Another Timeline, this one incorporates duration (from a Melanoma Registry)
In circumstances where each timeline event is defined in a manner that involves a number of details, creating a separate table (often placed in an appendix) provides a helpful way to look-up this information when needed (see below).
Clinic Note Generation
To the extent that dictation or typing can be eliminated, programming logic should be used to generate semi or fully automated text within the clinic report. It is easy to appreciate the potential of this technique with an example. Imagine a standard, boilerplate type phrase:
“The patient’s blood pressure was in the << insert here >> range.”
Conditional logic based on the patient’s systolic blood pressure would then determine whether “normotensive,” “hypertensive,” or “hypotensive” would be inserted. Substantial administrative time can be saved using this technique.
Key Item Extraction
A common limitation of outcome measures, particularly those designed using Classic Test Theory, is that item detail can get lost in overall scores. One approach to ameliorate this issue is to combine the display of total scores with individual items that may be of concern. For example, in the left-hand side of the table below, the total score for various Patient Reported Outcome measures are presented. Note the multiple techniques used: color coding based on cut-scores, automated calculations to minimize effort (i.e., total scores and number of days/months from today), and prioritization of the patient’s concerns.
Below the table is an item display analysis, irrespective of summary scores. That is, an item’s question and response is displayed if endorsement is in the moderate to high range. Conversely, items not in this range are hidden. This makes it easy for clinicians to review areas of concern.
A tabular row/column display is a common mechanism used to display variable values over time. This approach is particularly helpful when there is a large number of variables with discrepant units of measurement (i.e., a case poorly handled by graphs). Below is an example of a traditional table using ‘Last ICU Temperature.’
Dynamic Tables enhance this traditional design by adding the ability to control display attributes using variable values (i.e., any arithmetic expression). The basics of this approach is similar to ‘conditional formatting’ found in spreadsheet applications. However, each cell (e.g., ‘Day 1’ temperature) is best conceptualized as a web page, thus common spreadsheet constraints do not apply (e.g., one value per cell) . Indeed, anything that can be done on the web can be made to conditionally display within the cell, the contents of which is directly driven by registry data. This combination sets up all kinds of exciting possibilities.
In the current iteration of Dynamic Tables, there is support for control of the following five attributes / display characteristics:
- Label font and/or background color
- Display of up to four variable values
- Cell background color
- Cell border color
- Conditional images
- i.e., The result of an expression and/or a variable value determines which image appears.
Looking at a ‘Last ICU Temperature’ example, the image below shows one possible configuration using each of the four options. More specifically:
- The label is red (e.g., based on a initial treatment variable value)
- The temperature value is displayed
- The temperature value sets the background color
- The FiO2 value sets the border color
- A yellow circle appears if the patient is intubated
Taken together, Dynamic Tables offer considerable setup flexibility and can be leveraged to help staff quickly and accurately comprehend clinic metrics and/or the presence of data-related signals based on decision logic. To help generate setup ideas, consider some other examples based on different key concerns:
- Variable value change and/or event presence
- Low/High outlier identification (note: label font color based on last available variable value)
Note also that the data doesn’t have to be patient clinical metrics, but instead related to clinical workflow, staffing responsibilities or other processes. For instance, the table below is used during case conference to indicate needed staff discussion areas (using a Movement Disorders Clinic example).
There are a number of cases that are poorly suited for the display of data points over time, either using graphs or Dynamic Tables. For instance, cases where the main emphasis is on the most recent variable values and/or where there are a large number of visits and associated variable values. Consider the overwhelming nature of the volume of longitudinal tables and graphs needed to display a single case in a cancer registry (see image of visits below).
Summary Tables provide a mechanism to address this issue. The table was designed to:
- Current variable values and percent change
- Data collection timing (i.e., how recent is the data from the report date?)
- Reduce report space compared to listing all longitudinal data points
- Minimize user effort to obtain a snapshot of the patient’s current clinical metrics:
- Leverage visual display characteristics to convey meaning
- Reduce eye movements
- Align variable labels and most recent values
- Shift historical information to the right for access on an as-needed basis
- Provide common variable value summary statistics
- Assess data collection efforts
Each row in the Summary Table contains a variable along with a series of summary characteristics in the columns. The variables are typically separated by domain. Selected items from two domains are shown below.
As a description of the various columns:
- Variable name.
- Most Recent — Date (days ago)
- The most recent variable value, the date obtained and days from report date.
- % Change From Initial / Previous
- Percentage change from the initial and previous visit. The font color lookup table can be as granular as needed and each variable can have its own font color lookup table. For example, the clinical significance a 10% change over three days is different for weight vs. white blood cell count.
- Initial Value Date (days ago)
- The initial variable value, the date obtained and days from report date.
- Previous Date (days ago)
- The previous variable value, the date obtained and days from report date.
- Min / Max (Mean), (Observation Info)
- The variable’s minimum, maximum and mean, and the number of times the variable value was entered and the total number of times it could have been entered.
- Variable value unit of measurement
- Comp %
- Completion percentage (based on ‘Observation info’)
Taken together, these techniques are particularly helpful in building engaging, informative, time-saving clinical reports. By doing so, the clinical report offers a powerful incentive for providers to keep patient registry data accurate and up-to-date in a timely fashion. Send comments, questions, suggestions and/or your design tips to email@example.com .