Lessons for Medicine from Information Visualization
Information visualization is an area of increasing research and development, both in the scientific and business communities. It is closely linked with data mining: a method for knowledge discovery from extremely large, complex data sets. The goals of information visualization specifically germane to medicine include aiding the “discovery of details and relations” and “supporting the recognition of relevant patterns.”9 These relations and patterns may offer new knowledge or understanding that the individual data points do not adequately convey.
Visualization has been important in medical imaging for some time; however, less attention has been given to analysis of numeric and time series data on an individual patient. One example that is widely used is the pediatric growth chart, where the height, weight, and head circumference are mapped to percentages and plotted on a normalized curve to assess the child’s development. This task goes beyond simple display, as data synthesis is used (conversion to a percentage of normal). There are two tasks being performed as well. The first is the initial assessment of the patient in relation to the rest of the population. The second is a longitudinal trajectory of growth, where the points should follow the same line (population percentile growth) even though the actual data points (height, weight, head circumference) changes.
There are numerous examples of information visualization across non-medical disciplines. Taken together, many of these insights can provide a framework for creating improved data displays for clinicians. However, these concepts have not been tested in the clinical setting to determine whether they will increase efficiency of routines, such as acuity ranking. Further, we may need to support hospitalists’ common tasks with separate approaches. The acuity ranking activity might be supported by a summary page showing key outliers and critical values for each patient. The rest of the report could show all abnormal data and (as needed) the details for closer review.
Unfortunately, there has not been much direction in solving this problem from a scientific standpoint. In a review of the literature on the presentation of medical data, Starren and Johnson noted that, “there is a paucity of methods for developing new presentations” in the medical setting.10 Further, they observe that clinical data displays are rarely evaluated quantitatively. Rather they are shown to users to assess acceptance. We need to alleviate that shortcoming.
Next Steps
We believe that a significant amount of research needs to be performed in this area. We also believe that this research should focus on hospital-based specialties—especially hospitalist medicine. Why? Because hospitalists are charged with quickly assessing lots of information on lots of patients, and anything we can do to make that process more efficient will result in better patient care and hopefully, happier hospitalists. So what are next steps?
We can break up the research agenda into two arms: what needs to be displayed, and how do we display it? Although it may seem intuitive, we think it is important to decide the what before the how because the content will really drive the improvements in care.
There has been some emphasis on determining what clinical data are important for physicians. Work on prediction algorithms and scores has led to some estimations of what numbers are important for determining patient acuity and severity. However, an accurate and dependable determination of who is sick, how sick, and who will get sicker is some time off.
For now, it would be helpful to know what data physicians want to see. This will vary by provider and may not always lead directly to a specific outcome, but it is a start. It would be helpful to identify the values that most clinicians would want to know most of the time: high or low white counts, decrease in hemoglobin, decrease in platelets, normalization of creatinine, or other. This would provide the basis for experimenting with how best to display these items.