Principles of Good Clinical Data Visualization

Good visualization begins with the question. The chart type should be chosen after the question is clear. If the question is about counts by category, a bar chart may be suitable. If the question is about a numeric distribution, a histogram or box plot may be more useful. If the question is about change over time, a line chart may be appropriate. If the question is about the relationship between two numeric variables, a scatter plot may be useful. Choosing the wrong chart type can obscure the issue. Clinical research visualizations should show denominators when they matter. A site with ten missing outcomes may look worse than a site with five missing outcomes, but if the first site enrolled 200 participants and the second enrolled 10, the interpretation changes. Counts and percentages often need to be presented together. Similarly, missing data should not disappear from graphs unless the exclusion is explicit and justified. Scales should be honest and readable. Axis labels should include units where relevant, such as age in years, weight in kilograms, or days from enrollment. Colors should have meaning and should be used consistently. For example, a dashboard may use red for overdue or high-priority items, amber for pending items, and green for completed items. However, color should not be the only way information is conveyed because some readers may have color vision deficiencies. Labels, ordering, and grouping also matter. Plots should be simple enough for the intended audience. A statistical audience may understand density plots, facets, and confidence intervals. A site coordinator may need a clear bar chart showing the records that require action. A principal investigator may need a dashboard summary with drill-down capability. The same data can be visualized differently for different decisions.

Clinical data visualization should be reviewed like any other output. The data manager should confirm that the plot uses the correct dataset, filters, denominators, labels, dates, and grouping variables.