Source-linked AI summary
Visualizing Patient Trajectories and Disorder Co-occurrences in Child and Adolescent Mental Health
Dipendra Pant, Kaban Koochakpour, Odd Sverre Westbye, Carolyn Clausen, Bennett L. Leventhal, Roman Koposov, Thomas Brox Røst, Norbert Skokauskas, Øystein Nytrø
TL;DR
The paper addresses the difficulty of using multi-patient CAMHS EHR data to understand patient histories and co-occurring disorders for decision-making. It develops clustered patient-trajectory and ADHD co-occurrence visualizations from Norwegian CAMHS data and evaluates them with clinicians. The visualizations revealed age- and group-specific care patterns and were generally found useful for understanding CAMHS care.
Problem
Multi-patient CAMHS EHR information is often difficult to apply in raw form, while trajectory and co-occurrence visualization has received limited focus in CAMHS.
Method
The study preprocesses 35 years of CAMHS EHR data, groups patients by age, gender, and ADHD status, and visualizes trajectories, clustered patterns, and ADHD co-occurrences.
Results
Females without ADHD showed steadily increasing episodes with age, whereas females with ADHD and all males peaked in middle childhood before declining during the teenage years.
Takeaways & Limitations
Clinicians generally found the trajectory plots and co-occurrence graphs understandable and useful for clarifying CAMHS care patterns and supporting decision-making.
Abstract
from arXiv · showhide
Understanding patient trajectories and identifying patterns in episodes of care is critical for effective healthcare decision-making. We present a patient timeline visualization using clustered episodes of care derived from over 35 years of Child and Adolescent Mental Health Services (CAMHS) data. Patients were categorized into 12 groups based on three features: age group (preschoolers, middle childhood, teenagers) at the start of the first episode, gender, and presence or absence of Attention-Deficit Hyperactivity Disorder (ADHD), in order to group similar patients. The patients, timeline with demographics, and episode of care information are displayed in the trajectory to facilitate understanding of the patient and associated events, allowing observation of temporal patterns and variations. These plots reveal similarities and differences in care needs and patterns across groups. Females without ADHD have a steady increase in the number of episodes of care with age. Females with ADHD and all males experienced a peak in the number of episodes during middle childhood, followed by a decline in the teenage years. To compare and understand the intensity and co-occurring disorders with ADHD across different groups, we plotted an ADHD co-occurrence graph, and Tourette's syndrome was co-occurring predominantly in all age groups. We evaluated and refined our visualizations with the involvement of clinicians, who found them useful for understanding the context of CAMHS care. These visual tools make the population data in the Electronic Health Records (EHR) available for decision-making and enhancing the understanding of care and disorder patterns across groups.
I. INTRODUCTION
CAMHS EHR data are difficult to use for multi-patient decision-making in raw form, motivating visualizations of patient trajectories and ADHD co-occurrences. The study examines how these visualizations can support understanding, comparison, and clinical decision-making.
- CAMHS provides specialist secondary-level care through acute need or referral from general practice or municipal services.
- Multi-patient EHR information is often unavailable to clinicians because of privacy concerns and difficult to apply in raw form.
- Patient trajectory graphics are proposed to communicate complex histories clearly and simply, including episodes of care, diagnoses, medications, and symptoms.
- The study uses clustered episode trajectories, ADHD comparisons, and co-occurrence graphs to help clinicians grasp CAMHS patterns quickly.
- The research questions address visualization of CAMHS histories and co-occurring disorders, prevalent patterns across similar-patient groups, and clinicians’ perceptions of understandability and utility.
B. Objectives •
The paper aims to visualize similar CAMHS patient histories and co-occurring disorders, addressing limited attention to these visualizations in CAMHS. It also evaluates their understandability, utility, and probable impact on clinical decision-making.
- The objectives are to visualize groups of similar patient histories as trajectories and co-occurring disorders as graphs for CAMHS clinical decision-making.
- The study explores whether these visualizations are understandable, useful, and likely to affect clinical decision-making.
- Prior research used clustering and trajectory methods to identify psychiatric treatment patterns and support visualization of care pathways.
- The paper addresses a lack of focus on trajectory and co-occurrence visualization specifically in CAMHS.
III. DATA AND MATERIALS
The study uses 35 years of Norwegian CAMHS EHR data, preprocesses patient episodes, and classifies patients by age group, gender, and ADHD status. Existing K-Prototype clustering supplies episode-level cluster labels for the visual analyses.
- 35 years of Norwegian CAMHS EHR data initially contained 22,643 patients and 30,938 episodes; preprocessing produced 19,248 patients and 22,676 episodes.
- The prior study used unsupervised K-Prototype clustering to group patients by similarity of their episodes of care.
- Three clusters were selected as optimal using Calinski-Harabasz and Silhouette Index comparisons, with SI 0.256 and CI 4473.64.
- Patients with any F90 diagnosis were classified as ADHD, while patients without F90 were classified as no ADHD.
- Initial age defined three groups: preschoolers [0, 6), middle childhood [6,12), and teenagers [12,19) years.
IV. METHODOLOGY
The methodology organizes patients into 12 groups using age, gender, and ADHD status, then visualizes their episode timelines and cluster patterns. Timelines encode episode duration, demographics, visits, diagnoses, and medications across patient age.
- The methodology consists of four steps for constructing and analyzing the visualizations.
- Patients were subdivided into 12 groups by age group, gender, and presence or absence of an ADHD diagnosis.
- Each group’s timeline displays episode lengths, patient information, visits, diagnosed disorder codes, and prescribed medications.
- The horizontal axis represents patient age, while episode colors encode cluster labels representing collections of similar episodes.
C. Patient trajectory pattern
The visualization overlays patient episodes and cluster patterns to show how care trajectories vary across grouped CAMHS populations. It also includes F90-centered co-occurrence graphs at multiple ICD-10 hierarchy levels and assesses clinician understandability and utility.
- Patient episodes, clusters, and patients are overlaid to visualize variation within each of twelve groups.
- F90 co-occurrence graphs connect diagnoses to F90 and label edge frequencies using colors.
- The graphs represent original, three-character, and one-character ICD-10 hierarchy levels.
- Clinicians evaluated the visualizations for interpretability and practical value in supporting clinical decisions.
A. Patient Trajectory & Patient Trajectory Pattern Plot
Trajectory plots display episodes, patient attributes, and cluster membership across age-sorted patient timelines. The gender- and ADHD-oriented analysis reveals distinct episode-count patterns across age groups.
- Patient Trajectory & Patient Trajectory Pattern Plot: Females without ADHD show a steady increase in total episodes with age across all clusters.
- Patient Trajectory & Patient Trajectory Pattern Plot: Females with ADHD rise in episode count during middle childhood and decline during the teenage years, except in cluster 0.Cluster 0 continues increasing with age.
- Patient Trajectory & Patient Trajectory Pattern Plot: Males without ADHD increase during middle childhood and decline during the teenage years in total episodes and clusters 1 and 2.Cluster 0 instead increases with age.
- Patient Trajectory & Patient Trajectory Pattern Plot: Males with ADHD peak in episode count during middle childhood and decline during the teenage years.This pattern is reported for the total and cluster-specific counts, including cluster 0.
- Patient Trajectory & Patient Trajectory Pattern Plot: Trajectory plots annotate episodes with gender, age, visits, medication, and diagnosis, while colors indicate episode clusters.The example data are illustrative rather than real patient information.
2) Cluster Oriented Analysis:
Cluster-oriented analysis shows that episode progression differs across the three clusters. Females without ADHD consistently increase with age, while other groups generally rise in middle childhood and decline during the teenage years.
- Cluster Oriented Analysis: Distinct progression across clusters underscores the importance of cluster characteristics when interpreting episode-count trends.
- Cluster Oriented Analysis: Cluster 0 shows age-related episode increases for females with and without ADHD and males without ADHD.Males with ADHD peak in middle childhood and then decrease during the teenage years.
- Cluster Oriented Analysis: Cluster 1 shows increasing episodes with age for females without ADHD, while other groups rise in middle childhood and decline during adolescence.
- Cluster Oriented Analysis: Cluster 2 shows increasing episodes with age for females without ADHD, while the other groups rise in middle childhood and decline during the teenage years.
3) Vizualisation Across Age Groups by Gender and ADHD:
Age-group visualizations use color and intensity to represent episode trends across gender and ADHD categories. Females without ADHD increase steadily through adolescence, while ADHD groups and males generally peak in middle childhood before declining.
- Vizualisation Across Age Groups by Gender and ADHD: Color and intensity identify episode trends and patterns across the different patient groups.
- Vizualisation Across Age Groups by Gender and ADHD: Females with ADHD increase from preschool to middle childhood and decline during the teenage years with reduced color intensity.
- Vizualisation Across Age Groups by Gender and ADHD: Males with ADHD peak in middle childhood with high purple and green intensity, then decline during the teenage years.Green becomes more prominent during the teenage decline.
- Vizualisation Across Age Groups by Gender and ADHD: Females without ADHD increase steadily from preschool through the teenage years, with green dominant and intensifying with age.
- Vizualisation Across Age Groups by Gender and ADHD: Males without ADHD increase through middle childhood, peak there, and slightly decrease during the teenage years, with green remaining prominent.
B. F90 patient co-occurrence graph
The study uses F90-rooted co-occurrence graphs to compare disorders accompanying ADHD across six patient groups. ICD-10 granularity affects interpretability, with level 3 providing an informative middle ground.
- Six of the twelve patient groups were diagnosed with ADHD and included in the co-occurrence graph analysis.
- Root-node size represents the number of F90 episodes, while leaf-node size represents episodes with diagnoses co-occurring with F90.
- The graphs compare co-occurring disorders at the original ICD-10 level, level 3, and level 1, progressively changing diagnostic granularity.
- Level 3 graphs were considered adequately informative and easy to interpret, whereas Figure 5 compares level 3 co-occurrences across six patient groups.
- All patient trajectory plots and F90 co-occurrence graphs are provided in the supplementary material.
C. Clinical insights and evaluation
Clinician feedback indicated that the visualizations supported cross-plot interpretation and revealed clinically relevant patterns, while also exposing readability and background-knowledge challenges. The evaluation favored ICD-10 level 3 co-occurrence graphs as the most useful level.
- Evaluation: Five CAMHS clinicians evaluated visualization understandability and utility through interviews and questionnaires.Three clinicians had participated in preprocessing and visualization design, while two encountered the plots and graphs for the first time.
- Clinical insights: Preschoolers had fewer CAMHS episodes, while older groups showed more intense and consistent CAMHS engagement.
- Clinical insights: Tourette’s syndrome co-occurred predominantly with F90 across preschool patients and middle-childhood and teenage male groups.
- Clinical insights: Preschool groups also showed autism spectrum disorder and reactive attachment disorder, highlighting early multidisorder patterns.
- Evaluation: Clinicians could relate observations within and across plots and graphs because the visualization functionality was clear and easily understood.
- Limitations and refinement: Detailed co-occurrence plots were too crowded, level 1 caused perceived data loss and confusion, and ICD-10 level 3 was judged most useful.Clinicians also requested stronger captions, clearer explanations of visualization elements, and emphasis on significant data points.
VI. CONCLUSION
The study concludes that patient trajectories and F90 co-occurrence graphs can make CAMHS histories and analytic results more understandable for clinicians. However, evaluation involved few clinicians and data from one geographic region.
- Patient trajectories and F90 co-occurrence graphs provided demographic, episode-of-care, and disorder insights in a user-friendly form.
- The visualizations facilitated CAMHS decision-making and helped clinicians understand patient histories and predict outcomes.
- The visual tools clarified data analytics and machine-learning results, including the intricacies of clustering algorithms.
- Evaluation was conducted with a relatively small clinician group, and the data covered only a single geographic region.
- Future research should apply the visualization techniques to more diverse and comprehensive datasets.