The Relationship Between Nursing Students' Microplastic Pollution Awareness and Environmental Literacy: A Cross-Sectional Study.

Personalized head and neck cancer (HNC) care focuses on creating treatments tailored to individual patients based on cohort characteristics from similar patients. Unfortunately, cancer treatment often results in numerous side effects, which differ between patient cohorts and can last for a long time post-treatment. As a result, clinicians are collaborating with data modellers to understand treatment-related symptoms that appear or persist post-treatment, to predict adverse outcomes and to stratify patients into high- and low-risk cohorts. One of the significant challenges in post-treatment research is posed by the scarcity of cohort data, imposed by the patient monitoring protocol []. Patients are closely monitored during treatment, when they come to the clinic to receive the prescribed doses, as opposed to post-treatment, when they come for biannual follow-up checkups []. As a consequence, post-treatment patient data are collected less often, posing a challenge in outcome prediction. Long Short-Term Memory Network (LSTM) methods have demonstrated excellent results for temporal patient outcome prediction, surpassing traditional statistical and machine learning methods, and have also gained attention in HNC symptom risk prediction [, ] Post-treatment symptom risk prediction is a multidisciplinary field where data modellers collaborate with clinicians to model patient outcome risk, but this modelling often suffers from low interpretability. This is especially true when supervised black-box models, such as LSTMs, are used. Visual analytics can support this research; however, it needs to consider the differences in the mental models of the users. For example, clinicians are more interested in the actionable interpretation of the modelling outcomes and in the accuracy of the methods, which can be applied when treating new patients. Data modellers, on the other hand, are also interested in understanding the mechanisms behind the model’s decisions and tools that help them refine and debug modelling approaches. Moreover, post-treatment symptoms can result from the cumulative effects of various factors [, ], such as treatment-related complications or patient-specific health and lifestyle changes following treatment. Symptoms can also be associated with each other, either due to direct influence, or due to shared root causes. Consequently, there is a growing need for analytical tools that support collaborative cohort modelling through workflows that enable experts to interpret machine-derived (modelled) results with real patient data. Although data visualization is a valuable tool for supporting analytical tasks, it must overcome several challenges in the context of post-treatment symptom prediction. To effectively interpret LSTM model behaviour, data visualization must integrate diverse data facets from heterogeneous cohorts and support data modellers’ and clinicians’ analytical tasks. Specifically, data visualization needs to compare multiple cohorts of interest to understand the impact of the modelled risk, support cohort stratifications by levels of risk to better understand prediction results and blend cohort characteristics with results from different models to gain a deeper understanding of risk categories. Notably, LSTM symptom prediction visualization needs to overcome the cognitive burden associated with the high information density of LSTM models.