From Registration to Insight: How STRONG AYA Transforms Registry Data to Enhance Decision-Support Tools for Adolescent and Young Adult Oncology

The STRONG AYA consortium leverages federated learning to integrate registry data from the UK's Yorkshire Specialist Register of Cancer in Children and Young People with international datasets, transforming it into actionable insights via the PROMPT software to enhance clinical decision-making and patient consultations for adolescents and young adults with cancer.

The Gist

The Big Picture: A Global Team Up for Young Cancer Patients

Imagine a group of teenagers and young adults (ages 15 to 39) facing cancer. This is a tricky age group. They aren't quite children, but they aren't fully independent adults yet. They are trying to finish school, start careers, and build relationships, all while fighting a life-threatening illness.

For a long time, doctors have had a hard time getting a clear picture of what these young patients need. Why? Because most medical data systems are built for either "kids" or "adults," leaving this specific group in a gray area. Plus, doctors often didn't have enough data to know how a patient's mood or quality of life compared to others with the same cancer.

Enter "STRONG AYA." Think of this as a massive, pan-European "super-team" of doctors, researchers, and the young patients themselves. Their goal is to build a shared library of knowledge to help these specific patients.

The Problem: The "Locked Box" Dilemma

In Europe, different countries have different ways of recording medical data. It's like every country has its own language and its own type of filing cabinet.

• The UK has a specific registry for young cancer patients in Yorkshire.
• The Netherlands has its own system.
• Poland and France have theirs.

Usually, to compare data, you'd have to move all the patient files into one giant central computer. But that's a privacy nightmare. You can't just put everyone's private medical records in one place; it's too risky.

The Solution: The "Secret Recipe" Chef (Federated Learning)

This is where the paper gets really cool. The STRONG AYA team uses a technology called Federated Learning.

The Analogy: Imagine you want to know the average height of people in five different cities, but you aren't allowed to take anyone's ID card or leave their house.

• Old Way: You send everyone to one central stadium to measure them all together. (Too risky, privacy violation).
• STRONG AYA Way: You send a tiny robot (the algorithm) to each city. The robot measures the people locally, writes down just the average number, and comes back to you. The robot never takes the people's names or addresses.

In this paper, the "robot" is a piece of software that travels to the local data centers (like the one in Yorkshire, UK). It does the math locally and only sends back the results (like "the average anxiety score is X"), not the actual patient names or details. This keeps everyone's data safe and private.

The Magic Tool: PROMPT and the "Traffic Light" System

The researchers took this safe, aggregated data and plugged it into a tool called PROMPT, which is already used in hospitals in Leeds, UK.

The Analogy: Imagine a young patient is sitting in the doctor's office. The doctor pulls up a screen showing the patient's recent mood scores (specifically anxiety and depression).

• Before: The doctor would look at the score and say, "Hmm, that's high," but they wouldn't know if it was normal for a 20-year-old with leukemia, or if it was an emergency. It was like driving with no speedometer.
• Now: The screen shows a "Traffic Light" system.
  • Green Zone: The patient's score is within the "normal" range for other young people with similar cancers across Europe.
  • Orange/Red Zone: The score is outside that range.

The screen draws a green band on a graph. If the patient's line goes outside that green band, the doctor sees it immediately. It's like having a GPS that says, "You are driving faster than everyone else on this specific road; maybe slow down and check your tires."

What Did They Actually Do?

1. Connected the Dots: They took data from the Yorkshire cancer registry and linked it to the STRONG AYA network without moving the actual files.
2. Translated the Language: They created a "dictionary" so that the UK's way of writing data matches the Polish, Dutch, and French ways.
3. Built the Dashboard: They showed that a doctor can now see a patient's anxiety levels and instantly compare them to a "reference group" of thousands of other young cancer patients across Europe.
4. Tested It: They ran a simulation with a fake patient to prove the system works. The doctor could see the patient's history and the "green band" of normalcy all on one screen.

Why Does This Matter?

• For the Patient: It helps them feel less alone. When a doctor says, "Your anxiety is higher than usual for someone in your situation," it validates their feelings and helps them get the right help faster.
• For the Doctor: It gives them a "superpower." Instead of guessing, they have data from thousands of peers to guide their decisions.
• For the Future: This isn't just about anxiety. The team plans to use this same system to track survival rates, long-term side effects, and quality of life. It turns a local hospital record into a global insight tool.

The Bottom Line

This paper is a "proof of concept." It's like building a prototype car to show that a new engine works. They proved that you can take local data, keep it safe, combine it with data from across Europe, and use it to give doctors a better tool to care for young cancer patients.

It's a shift from "I hope this works" to "Here is the data showing us how to make it work better."

Technical Summary

1. Problem Statement

Adolescents and Young Adults (AYAs, aged 15–39) with cancer face unique developmental and clinical challenges that are often inadequately addressed by standard cancer care. While Population-Based Cancer Registers (PBCR) are vital for epidemiological monitoring, their utility for AYAs is limited by:

• Data Gaps: Lack of AYA-specific data items, particularly regarding Quality of Life (QoL) and Patient-Reported Outcomes (PROs).
• Data Silos: Existing data is often fragmented across different regions and institutions, preventing large-scale comparative analysis.
• Privacy Constraints: Centralizing individual-level patient data across borders for analysis raises significant privacy and legal (GDPR) concerns.
• Static Reference Ranges: Clinical decision support tools often rely on historical or non-AYA-specific reference ranges for PROs (e.g., anxiety and depression scores), failing to provide context relevant to the specific AYA population.

2. Methodology

The authors describe the integration of the Yorkshire Specialist Register of Cancer in Children and Young People (YSRCCYP) into the STRONG AYA pan-European infrastructure. The technical workflow involves several key components:

A. Data Harmonization and Interoperability

• Core Outcome Set (COS): A standardized set of clinical and QoL variables was developed to ensure data comparability across diverse cohorts.
• Semantic Mapping: To overcome bespoke local data formats, a shared terminology map was created. The Flyover tool was utilized to map local data formats onto the STRONG AYA terminology, generating semantically interoperable Resource Description Frameworks (RDF).
• Federated Learning (FL): Instead of centralizing raw data, the system uses a Vantage 6 node. Data remains local to the participating institutions (e.g., Leeds, Netherlands, Poland, France, Italy). Only the results of analyses (aggregated statistics) are exchanged, preserving patient privacy.

B. Clinical Integration (PROMPT)

• Software Bridge: A custom bridging process was implemented between the PROMPT software (an in-house electronic patient record system at Leeds Teaching Hospital NHS Trust) and the STRONG AYA FL infrastructure.
• Dynamic Reference Ranges: The system periodically queries the STRONG AYA infrastructure to retrieve updated descriptive statistics (median, interquartile ranges) for specific PROs.
• Case Study Implementation: The Hospital Anxiety and Depression Scale (HADS) was used as a proof-of-concept. The system queries for Total Anxiety Score and Total Depression Score using semantic codes (ncit:C118111 and ncit:C118112).

C. Visualization and User Interface

• Traffic Light System: Individual patient scores are compared against the federated reference ranges. Scores within the range are marked green; those outside are orange.
• Longitudinal Analysis: Line graphs display patient scores over time overlaid with the federated interquartile range (Q1, Median, Q3) as a shaded band.
• Stratification: The dashboard allows filtering by age, disease, country, and stage, and can display histograms comparing individual scores against population distributions.

3. Key Contributions

• Technical Architecture: Demonstrated a functional pipeline connecting a regional PBCR (YSRCCYP) to a pan-European FL ecosystem using Vantage 6 and semantic interoperability tools (Flyover).
• Clinical Decision Support: Developed a novel method to inject dynamic, population-specific reference ranges into existing Electronic Health Records (EHR), moving beyond static, historical norms.
• Privacy-Preserving Analytics: Validated the use of Federated Learning to generate actionable clinical insights from multi-national datasets without transferring sensitive individual-level data.
• Open Source Implementation: The bridging code and algorithms (descriptive statistics) are open-sourced on GitHub and Zenodo, facilitating reproducibility and adoption by other centers.

4. Results

• Data Aggregation: The FL query successfully aggregated data from seven data stations across five European countries (Netherlands, UK, Italy, France, Poland).
• Scale: The infrastructure currently includes 26,829 AYA recordings from retrospective and prospective sources.
• HADS Analysis: Specifically for the HADS case study, 368 recordings were available for analysis (138 from the Yorkshire registry, 210 from Poland, 20 from the Netherlands).
• Visualization Success: The system successfully visualized a fictitious patient's HADS scores against the federated interquartile ranges in the PROMPT interface, demonstrating the ability to contextualize individual patient data against a European-wide cohort in real-time.

5. Significance and Future Directions

• Enhanced Patient Care: By normalizing patient experiences against peers with similar diagnoses across Europe, clinicians can better identify when a patient's anxiety or depression is atypical, facilitating shared decision-making and timely intervention.
• Policy and Research: The infrastructure supports high-level research, including survival analysis (Cox regression, Kaplan-Meier curves) and policy-making based on robust, harmonized data.
• Scalability: The proof-of-concept suggests that this model can be extended to other PROs and complex predictive models (e.g., prognostic modeling).
• Sustainability: The authors emphasize that future success depends on clinician adoption, ongoing patient feedback, and the expansion of the dataset to improve the representativeness of the reference ranges.

In conclusion, the paper presents a robust technical framework for transforming fragmented regional cancer registry data into a secure, privacy-preserving, pan-European decision-support tool, directly addressing the unique needs of the AYA oncology population.