A Guide to Medical Visualisation

In the area of pharmaceutical marketing, a revolution is underway, driven by the power of data visualisation. As a Pharmaceutical Marketing Consultant, I've witnessed firsthand how the strategic use of visual data can transform complex medical research findings into clear, impactful narratives that resonate with both healthcare professionals and patients alike.

There are at least 22 distinct types of graphs and visualisations available in Power BI. And those are not enough when it comes to medical statistics graphs.

To try out the Power of Power BI kindly follow the link.

The Opportunity for Visual Clarity

The essence of effective medico-marketing communication lies in the ability to distil vast amounts of data into digestible, visually engaging formats. Graphs and charts not only make data more accessible but also enhance our ability to detect patterns, trends, and outliers. By applying good graphic design principles, we can significantly improve the comprehension and retention of critical medical information.

Choosing the Right Graph

Selecting the appropriate graph is crucial for conveying the right message. Whether it’s understanding the distribution of a single variable or exploring complex relationships in large datasets, the choice of visualisation can make or break the communication process. Techniques like spaghetti plots, lasagna plots, and network visualisations offer sophisticated means to illustrate data from clinical trials, patient outcomes, and drug safety reports.

Graph Subtype: Simple vs. Grouped vs. Multipanel

The subtype of a graph is chosen based on the complexity of the data. Simple graphs (like bar charts, pie charts, line graphs etc.) are used for single variables. Grouped graphs help compare subcategories within a variable, and multipanel graphs are used for multiple subgroup comparisons across several variables.

• Simple Graphs: These are used when there's no need to group data or when the dataset contains only one grouping variable. A simple dot plot is a type of scatter plot that displays individual data points on one axis. It's a clear way to show the distribution of data when there are not too many points to cause overplotting.
• Grouped Graphs: When the data includes one grouping variable, and there's a need to compare subcategories within this variable, grouped graphs are used. A scatter plot where countries are displayed along the y-axis and two different years are represented along the x-axis. This allows the viewer to compare the changes between the years for each country.
• Multipanel of Simple: This subtype involves using multiple panels (or facets) of simple graphs to represent different subsets of the data without grouping variables. Each panel acts as a separate plot for a subset of the data, which is useful when comparing several categories that do not necessarily relate to each other.
• Multipanel of Grouped: When the data contains multiple grouping variables, a multipanel approach to grouped graphs is taken. This allows the data to be divided into more specific categories and compared across different subgroups within each panel.

Embrace these visual aids, for they are the silent ambassadors of your brand's story.

Data-Specific Graphical Storytelling

Low Level of Detail:

• Bars & Error Bars: The mean or median of the data and uses error bars to represent the uncertainty around these measures, such as the interquartile range (IQR), standard error (SE), or 95% confidence intervals (CI). It's noted that these are less preferred than dot plots, which can show all individual data points.
• Symbols & Error Bar: Like the bar and error bar graph but uses symbols, such as dots or lines, to represent data points.
• Dot Plot: Shows individual data points spaced out along an axis, which is useful for highlighting the frequency of values and is preferred for its clarity.
• Line Plot: Illustrates data points connected by lines, helping to show trends over time or categories.
• Forest Plot: Often used in meta-analyses, it displays the point estimates and confidence intervals of individual studies and the overall estimate.

Fine-Grained Distribution:

• Violin Plot: Combines a box plot with a kernel density plot on each side, showing the distribution of the data. It's useful for comparing the distribution of data across several groups.
• Probability Density: A smooth curve representing the distribution of a continuous dataset. It shows the probability of different outcomes.
• Q-Q plot (Quantile-Quantile plot): This plot compares two probability distributions by plotting their quantiles against each other. If the two distributions being compared are similar, the points will approximately lie on the line y = x.
• Box Plot: Also known as a box-and-whisker plot, it shows the distribution of quantitative data in a way that facilitates comparisons between variables or across levels of a categorical variable. The box shows the quartiles of the dataset while the whiskers extend to show the rest of the distribution.

Cumulative Distribution:

• Kaplan-Meier: This plot is used to estimate the survival function from lifetime data. It's a step function that jumps at each event time. It's especially useful in medical research for visualising the probability that a patient will survive for a certain amount of time after treatment.
• Waterfall: Often used in financial or inventory data, a waterfall chart shows a starting value, intermediate increases and decreases, and a final value, providing a cumulative effect of sequentially introduced positive or negative values.

The main takeaway is that the choice of graph should be driven by the level of detail necessary to convey your message effectively. For high-detail raw data analysis, violin plots and probability densities are useful. For summarized data with a focus on communication, bars with error bars are commonly used but less preferred compared to dot plots, which give a clearer picture of the distribution of individual data points.

Visual Element: Pie Chart vs. Bar Plot

Pie charts are often used to show the proportions of a whole, where each slice represents a part of the total. However, the human brain struggles to compare angles in a pie chart, leading to inaccuracies in interpreting the data. In contrast, bar plots use length, which is easier for our brains to compare accurately. Thus, for data where it is important to see precise comparisons between categories, a bar plot may be more effective than a pie chart.

1. Portion of the circle (Pie Chart Type): The pie chart is a recommended option when you're interested in showing how a single category is a portion of the whole. However, it's not recommended when you're comparing more than one category because it's harder to visually compare the sizes of different pie slices than the lengths of bars.
2. Bar Height (Bar Plot Type): The bar plot is a versatile visualisation tool that can represent a range of data types. It's recommended for showing the comparison of different categories or groups (denoted by bar height). When comparing more than one level of interest, a bar plot can become cluttered, and it's suggested to avoid using pie charts in this context.
3. Symbol along scale (Dot Plot Type): The dot plot is recommended and is a good choice for showing individual data points along a scale, making it easy to see distributions and clusters within the data. It's particularly useful for showing all data points when you have several categorical binary variables, as in the example provided for 8 Adverse Events (AEs). This is beneficial for detailed analysis and comparison across multiple categories.

More complex graphs

Bivariate plots for high-density data

Spaghetti & Lasagna Plots

The "spaghetti plot" shown in the figure is a graphical method used to display individual trends over time. Each line represents a single subject's data across time, in this case, the levels of glycated haemoglobin (HbA1c). The left panel shows the control arm with 75 subjects, and the right panel shows the treatment arm, also with 75 subjects. Because each subject's line may cross over or intertwine with others, it resembles a tangle of spaghetti, which is how it gets its name. While spaghetti plots can visually display all individual trajectories, they can become cluttered and difficult to interpret, especially with a large number of subjects.

The lasagna plot concept to a larger group of 150 subjects. The visualisation becomes denser, but it still allows for the observation of overall trends across the group, such as the proportion of subjects falling into each HbA1c category over time. This type of plot helps understand the broader impact of interventions or the natural progression of a condition in a population.

Scatterplot

(a) Scatterplot of Blood Glucose Levels at Baseline and Week 26: This is a standard scatterplot, each point represents an individual's blood glucose measurement. The diagonal line indicates where the Week 26 result is the same as the baseline (no difference line). Points along this line indicate no change from baseline to Week 26. The shaded area represents the normal range for blood glucose levels, allowing for easy identification of values outside the normal range.

(b) Sunflower Density Plot of Blood Glucose Levels at Baseline and Week 26: This plot is a variation of the scatterplot where points are replaced with 'sunflower petals' to represent the number of overlapping data points. One petal represents a single observation, and more petals indicate a greater number of overlapping observations. This method helps visualise the concentration of data points that might otherwise be obscured in a standard scatterplot due to over-plotting.

(c) Bivariate Kernel Density Estimates of Blood Glucose Levels at Baseline and Week 26: The left panel is a kernel density estimate which is a way to estimate the probability density function of a continuous random variable. It's a smoothed version of the histogram and is used for visualising the distribution of two variables.

The right panel shows a zoomed-in view of the densest region, with contour lines that represent different levels of density. The denser the region, the more data points fall within that area. This is useful for identifying the concentration of measurements and the relationship between the baseline and Week 26 glucose levels.

Network visualisations for large datasets

(a) Full Network Visualisation: The first panel displays the full network with various nodes representing different entities like the vaccine, adverse events, and patients. The nodes are connected by lines, which likely represent the relationships or associations between these entities. The centre of the network seems to have a dense region, possibly indicating a strong association or frequent occurrence of certain AEs with the vaccine.

(b), (c), (d), (e) Reduced Network Visualizations: These panels are focused views of the larger network, isolating the most connected nodes to better analyse and understand the patterns within the data.

Violin Plot

A violin plot would typically show the same data as the boxplot but includes a kernel density estimation on each side, which represents the distribution of the data. The width of the "violin" at different levels indicates the density of data points, with wider sections representing a higher density of data points at that level.

Violin plots are useful for showing the full distribution of the data, which can provide insights into the shape of the distribution that boxplots alone do not show, such as multimodality (more than one peak).

Kaplan–Meier curve

Kaplan-Meier curves, which are used to estimate the survival function from lifetime data. In this context, they appear to represent the probability of experiencing certain adverse events over time among different treatment groups in a clinical trial. The four graphs each correspond to different side effects: Headache, Dizziness, Abdominal Pain, and Nausea.

Phew, and yet many more to go.

We've journeyed through a vast landscape of graphs, from the humble pie chart to the intricate network visualisations. In the dynamic field of pharmaceutical marketing, mastering these visual tools is not just beneficial; it's a necessity. As we continually evolve in our approach to medico-marketing communication, the ability to adapt and harness the power of these graphs becomes paramount. They are not mere embellishments but pivotal in shaping the narratives that drive the promotion of pharmaceutical brands.