Adverse events (AEs), particularly with drug therapies, remain a challenge in modern healthcare. The increasing complexity of cancer therapeutics, an ageing population, and rising multimorbidity all continue to contribute to and complicate safety issues for patients. However, while our understanding of AEs should be evolving, and our AE storytelling for patients should be more engaging, reporting of AEs really has not changed. Indeed, AE narratives remain directed at clinicians and regulators, rather than to patients. We believe there is value in being able to tell patients an AE narrative. Many patients on treatments will experience AEs: for those that do, what are those AEs like? Do they have more than one, and will they be symptoms of a larger experience (e.g., fever). Our registry, Melanoma UK, one of several registries on our real-world-evidence platform and co-created with Melanoma UK, collects patient-reported AE data directly using the PRO-CTCAE instrument: specifically, decreased appetite, nausea, vomiting, diarrhoea, abdomen pain, itchy skin, headache, fatigue, shivering, and sweating.  

We decided to explore what this story might look like, using our data, and focusing on severe AEs (Note: since the PRO-CTCAE has no grading algorithm yet, we graded these on a combination of severity and frequency). 

Below is the incidence of each AE in our data – the traditional stopping point, but that is paradigmatic of what we want to change. Instead, let’s think about co-occurrence: how many AEs are experienced by each person, and what are they?  

Table 1. AE incidence reported in the Melanoma UK study

Let’s start with Figure 1. For patients reporting at least one AE, 60% report only one. Nearly 25% report two, and so on – 40% of people reporting AEs experience more than one.  

Figure 1: Co-occurrence of AEs reported in the Melanoma UK study

Now let’s look at how these AEs correlate – Figure 2. Note that we want to add an element to our AE storytelling: visual communication. One of the shortcomings that we find in traditional AE reporting is the dryness of a table of incidence. To us, this is the difference between AEs being treated as incidental vs relational, with the second concept resonating with patients much more.  

Figure 2 therefore shows the correlation matrix as a heatmap. With this we can quickly locate some points of interest: shivering, sweating, and fatigue; nausea, vomiting, and decreased appetite. These appear to be hotspots, suggesting higher co-occurrence – and, we can note that they are of a type.   

Figure 2: Correlation heatmap of AEs reported in the Melanoma UK study

We explore that concept of co-occurrence with Figure 3: Network of reported AEs in the Melanoma UK study, a network diagram. Immediately we see that the network is mapped into two clusters (using Louvain modularity/community detection), corresponding with the correlations heatmap above: one cluster around what is likely to be latent fever; another which is clustering around GI and nausea issues. And the curiosity of itchy skin, which appears outside of any constellation – almost, unless we do believe in latent fever, and we think about it differently.  

This is what our visualization is helping us accomplish, and what patients also can see for themselves: if I get serious AEs, I will probably get one or both of these underlying problems, with these two sets of symptoms, at least one of which will be severe. 

Figure 3: Network of reported AEs in the Melanoma UK study

*Note: Node size – frequency of the particular AEs; Edge width – correlation coefficient between the two nodes; Node colours – clusters 

This continues to additional levels of specificity, e.g., AEs, their incidence, type, and severity, change and grow as disease severity itself worsens.  

We looked at melanoma split into stages 1 and 2 vs 3 and 4 – Figure 4, a very simple attempt to separate out advanced/metastatic melanoma. Immediately, we see the correlation heatmap separate as well: patients with earlier stages of melanoma appear to see a diarrhoea/abdominal pain cluster, and headaches more generally, but not much else with a pattern. For patients with advanced disease, our clustering from before still holds even as all AEs begin to co-occur more – reflecting increased incidence. 

Figure 4. Correlation heatmap, by disease stage, of AEs reported in the Melanoma UK study

 Finally, we tried to visualize these relationships.  We have designed a chord diagram with two groups; one with AEs and another with stages. Figure 5 shows perfectly that these relationships can be visualized. 

 Melanoma Adverse Events vs Stages

Melanoma Stage 1

Melanoma Stage 2

Melanoma Stage 3

Melanoma Stage 4

Figure 5: Relationships between disease state and AEs reported in the Melanoma UK study

One can quickly see and infer from this that, as disease progresses, the bands connected to AEs increase in width and number.  This exercise has been pretty basic, but we wanted to convey a serious point, one that we believe should be treated seriously by researchers. Everything that we have run through, relative to the table of AE incidence at the start, has given us a richer understanding, not only of AEs and how they occur, but also a better sense of what AEs mean from a patient perspective, or what their experience of the AEs might be.  For the patient, being presented with networks, correlation heatmaps, or chord diagrams – any one of those would help them understand how AEs might fit into, or even form, their experience of treatment for their disease.   

We think this is really important. This relates back to our platform and our mission to “give patients a voice through technology.” We develop all of our registries collaboratively with patients and organisations that help us understand critical things such as: how AEs are impactful in their lives, and how they want to be presented with data that they can use.  

By Harshitha Ravindra

Skip to content