Interactive Data Visualization: Trajectories of Functional Impairment in Middle Age
This interactive graphic was developed by University of Mississippi Medical Center Data Scientists Radhikesh Ranadive and Rajesh Talluri and Annals Associate Editor Michael Griswold using data provided by the authors of Functional Impairment and Decline in Middle Age: A Cohort Study. It was reviewed by the article authors and the editors.
Functional Impairment and Decline in Middle Age: A Cohort Study
Rebecca T. Brown, MD, MPH; L. Grisell Diaz-Ramirez, MS; W. John Boscardin, PhD; Sei J. Lee, MD; and Michael A. Steinman, MD
Ann Intern Med. [Epub 14 November 2017]. doi:10.7326/M17-0496
Bringing Data to Life: Interactive Visualizations of Complex Data
Rajesh Talluri, PhD; Radhikesh Ranadive, MS; Jaya K. Rao, MD, MHS; Christine Laine, MD, MPH; and Michael Griswold, PhD
Ann Intern Med. [Epub 14 November 2017]. doi:10.7326/M17-2903
colleagues examined trajectories of functional impairments in 6874
middle-aged participants from the Health and Retirement Study
(HRS). The sunburst graphic below provides an interactive
visualization of their data as described in the related editorial.
The innermost ring marks the baseline, and each concentric ring moving outward gives the reported health states from the cohort every 2 years after baseline. The initial sunburst shows that a large percentage of the participants remain healthy at +2 years after baseline (96.9%), at +4 years (90.3%), and so on, with 18.6% remaining healthy continuously throughout the study (concentric green area slices). The white space in each concentric ring represents participants who are lost to the study following a last-known recorded status of death or missing.
Hover over an area slice to see the number and percentage of the cohort with a specific trajectory. Click an area slice to select a group and focus on its subsequent outcome trajectories. Click the innermost ring (baseline) to reset the visualization. See the additional information section for more details.
Sunburst Interactive Data Visualization
Viewing the Number of ADL Impairments
Over a 22-year period, 6874 participants provided repeated measures of 5 basic activities of daily living (ADLs)
- mobility (transferring out of a bed/chair), and
The default view (provided also by clicking on the Number of ADLs tab) shows individuals at each time grouped by their number of reported ADLs: 0 (Healthy), 1, 2, 3, 4, 5, Dead or missing), color-coded in the legend. The concentric bands allow examination of all transitions of study participants through their health states at each additional 2-year HRS report (baseline, +2 years, … +22 years). For example, the snapshot on the right depicts a hover-over of the turquoise slice with the 193 (2.81%) study participants who were healthy at baseline, healthy at the 2-year follow-up, and then reported a single ADL impairment (turquoise grouping in the legend) 4 years after baseline. The trajectory and percentage of these participants are also shown on the timeline breadcrumb at the top of the visualization.
Selecting a Trajectory Group
In addition, one can select and focus on any specific group by clicking on a hover-over to examine that group's subsequent health trajectories. For example, the figure on the left shows that clicking on the earlier hover-over slice for the 193 individuals above expands the turquoise to an entire ring for the +4 year time frame (representing the selection of this subpopulation). Now, all slices in the +6 year ring and beyond correspond to numbers and percentages for the new selected subset of 193 individuals who reported Healthy–Healthy–1ADL initial trajectories. Hovering over the dark green band at +14 years from baseline shows that 32/193 (16.6%) had recovered and remained healthy for a decade after reporting their initial ADL impairment at +4 years after baseline. The trajectory and percentage of these participants are again shown on the timeline breadcrumb at the top of the visualization, now also displaying the selected initial trajectory subset. Clicking the innermost green ring resets the visualization to the original baseline cohort of 6874.
Viewing ADL Impairment Groupings
In addition to viewing numbers of reported ADL impairments, selecting the Any Impairment tab allows one to drill down further to examine the occurrence and co-occurrence of specific impairments. For example, dressing impairments were reported as the first, single ADL at +4 years after baseline by 85 individuals. Note that these 85 individuals belong to the earlier grouping of 193 individuals reporting 1 ADL impairment at +4 years. The "any impairment" view allows one to distinguish the single impairments at +4 years as (dressing = 85; mobility = 57; bathing = 21; toileting = 18; and eating = 12), corresponding to the pooled cohort of 193 individuals with 1 ADL shown in the "Number of ADLs" view. Similarly, co-occurring initial ADL impairments of dressing and mobility (DM in the legend; n = 17); bathing and dressing (BD; n = 10); and mobility and toileting (MT; n = 9) make up the top dually occurring impairments, whereas bathing, mobility, and dressing (BMD; n = 8) and mobility, dressing, and toileting (MDT; n = 8) tie for the top groupings of 3 impairments reported together. As a note, we recognize that the term "mobility" generally connotes walking outcomes in geriatrics and not transferring; however, for the interactive visualization, we needed a single letter to represent each ADL category. Because "T" was taken for toileting, we chose mobility "M" as a surrogate to represent the "transferring out of a bed/chair" ADL outcome. Judicious use of the zoom function in your browser allows examination of these very small slices. Hover-over and selection functional elements are similar in this tab and again, clicking the innermost green ring resets the visualization to the original baseline cohort of 6874.
Viewing ADL Impairment Severities
Lastly, individuals designated as having an ADL impairment could have reported either an ADL "difficulty" (e.g., able to accomplish the ADL alone but with some difficulty) or an ADL "dependency" (e.g., requiring assistance from another to accomplish the ADL). Choosing the Difficulties and Dependencies tab in the interactive sunburst graphic enables further drill downs into impairment severity groupings over time. For example, 21 of the dressing impairments at +4 years had dependent (worse severity) dressing impairments. To represent severity levels in co-occurring ADL impairments, we used capitalized words to represent dependency and lowercase words to represent difficulties; thus a slice with a hover over of "dressing EATING MOBILITY" (dEM in the breadcrumb trail) would indicate reported difficulty dressing co-reported with dependencies for mobility and eating. Here, the number of permutations is so large that we chose to remove the legend in favor of hover-over captions only. Hover-over and selection functional elements are similar in this tab and again, clicking the innermost green ring resets the visualization to the original baseline cohort of 6874.
We hope you enjoy interacting with our data visualization and even more so, we hope it leads to new ideas for improving health.
Dr. Michael E. Griswold
*, PhD, is a Professor in the Department of Data Science at the University of Mississippi Medical Center, John D. Bower School of Population Health, and is a member of the Statistical Editor team at Annals
. His research focuses on data science, data visualization, translational biostatistics, missing data, and multilevel/longitudinal models.
, MS, is a Data Scientist in the Department of Data Science at the University of Mississippi Medical Center, John D. Bower School of Population Health. He designs, develops, and maintains databases and open-source interactive visual displays of quantitative information.
Dr. Rajesh Talluri
, PhD, is an Assistant Professor in the Department of Data Science at the University of Mississippi Medical Center, John D. Bower School of Population Health. His research focuses on biostatistics, statistical genetics, Bayesian modeling, and Machine learning. He develops tools to encourage the use of innovative statistical methods by clinicians and researchers.
Disclaimer: The American College of Physicians does not make any representation or warranty, express or implied, as to the accuracy or completeness of the Content in any form. The Content is not to be construed as legal or medical advice, and is offered “as is,” and solely for educational and/or informational purposes and shall not be interpreted as a clinical judgment by a licensed medical doctor.