nSIGHT™: A Data Discovery Platform for Visualization, Integration and Retrospective Analysis of Multimodal Clinical Research Data

nSIGHT™ is an intuitive, web-based platform designed to overcome clinical data interoperability barriers by enabling researchers to easily discover, visualize, integrate, and analyze de-identified multimodal cancer data for cohort construction, feasibility assessment, and retrospective hypothesis generation.

The Gist

Imagine you are a detective trying to solve a complex mystery: Why do some cancer patients respond to treatment while others don't?

To solve this, you need to look at a patient's entire life story—their medical history, the drugs they took, their genetic makeup, and even samples of their tissue. But here's the problem: In a typical hospital, this information is scattered across nine different locked filing cabinets in nine different rooms.

• One cabinet has the doctor's notes (Electronic Health Records).
• Another has the lab test results.
• Another holds the genetic sequencing data.
• Another has the records of clinical trials.

Usually, these cabinets don't talk to each other. To find the answers you need, a researcher would have to hire a team of specialists to manually open every cabinet, copy the papers, and try to glue them together into one story. It takes weeks, costs a fortune, and often leads to mistakes.

Enter nSight™: The "Google" for Cancer Research.

The paper describes a new tool called nSight™ (pronounced "en-sight"), built by researchers at Roswell Park Comprehensive Cancer Center. Think of nSight™ as a super-smart, magical translator and time-traveling map that solves the "locked cabinet" problem.

Here is how it works, broken down into simple concepts:

1. The Great Translator (Data Integration)

Imagine you have a friend who speaks English, another who speaks Spanish, and a third who speaks "Lab Code." They are all trying to tell you the same story, but you can't understand them all at once.

nSight™ acts as a universal translator. It grabs data from all those different hospital systems (the filing cabinets) and translates everything into a single, common language. It takes messy, complicated medical codes and turns them into clear, readable events like "Diagnosis," "Surgery," or "Chemotherapy."

2. The Time-Traveling Map (Visualization)

Once the data is translated, nSight™ doesn't just give you a spreadsheet. It draws a timeline.

Imagine a patient's life as a long train track.

• The Anchor: The train starts at the "Diagnosis Station."
• The Stops: Every time the patient gets a scan, takes a pill, or has surgery, a little station appears on the track.
• The Colors: Different colors tell you what happened (Green for "Alive," Red for "Deceased," Blue for "Radiation").

Because the system uses "relative time" (e.g., "3 months after diagnosis" instead of "January 15th, 2023"), it protects patient privacy. You can see the story of the disease without knowing the patient's name or address. It's like watching a movie of their medical journey without seeing their face.

3. The "Fast-Forward" Button (Speed and Ease)

Before nSight™, finding a group of patients with a specific mutation who survived 5 years might take a researcher months of manual work.

With nSight™, a researcher can type a question into a search bar (like "Find all patients over 65 with Head and Neck cancer who had a specific gene mutation").

• The Result: In about 20 minutes, the tool builds a list of patients, shows their timelines, and even draws a graph comparing their survival rates.
• The Analogy: It's the difference between digging through a library basement with a shovel (old way) versus typing a keyword into Google and having the book appear on your screen instantly (nSight™).

4. The "What If" Lab (Analysis)

The tool also lets researchers play "What If?" games.

• Question: "Do older patients survive longer than younger ones?"
• Action: The researcher clicks a few buttons, and nSight™ instantly draws a Survival Curve (a graph showing who lived longer). It even does the math to tell you if the difference is real or just a coincidence.

Why This Matters

The paper highlights that cancer research is currently stuck in the "silo" era, where data is trapped in separate systems. nSight™ breaks down those walls.

• For Scientists: It means they can test new ideas quickly. If they have a hypothesis, they can check if the data supports it in minutes, not months. This allows for "fast failure"—quickly realizing an idea won't work so they can try a better one.
• For Patients: Faster research means faster discoveries, which leads to better treatments and cures appearing sooner.

The Bottom Line

nSight™ is a user-friendly dashboard that turns a chaotic mess of hospital records into a clear, visual story. It allows doctors and scientists to see the big picture of cancer research without needing to be computer experts or spending years digging through paperwork. It's a tool designed to help us solve the cancer mystery faster.

Technical Summary

1. Problem Statement

Cancer research is hindered by the lack of interoperability between disparate clinical and research data systems.

• Data Silos: Clinical data (Electronic Medical Records, Lab Information Management Systems) and research data (biorepositories, sequencing data) are stored in separate systems with limited standardization.
• Barriers to Translation: Extracting meaningful information from these silos is laborious, time-consuming, and requires specialized informatics expertise. This creates a bottleneck for translating laboratory discoveries into clinical applications.
• Privacy vs. Utility: Regulatory requirements (HIPAA, IRB) necessitate strict patient privacy, making it difficult to create de-identified cohorts for rapid "fast fail" hypothesis testing without compromising data integrity.

2. Methodology

The authors developed nSight™, a web-based data discovery platform designed to integrate, visualize, and analyze multimodal data while maintaining strict de-identification.

A. Data Architecture & Pipeline

• Data Sources: The system aggregates data from nine distinct sources, including the cancer registry, Electronic Health Records (EHR), Laboratory Information Management Systems (LIMS), Radiology Information Systems (RIS), surgical management software, patient registration, clinical trial management, and molecular diagnostic labs.
• ETC Process (Extract, Transform, Combine):
  • Raw data is extracted using parallel Python multiprocessing scripts.
  • Data is transformed into 21 event types across 6 categories (Disease, Patient, Therapeutic, Biospecimen, Diagnostic, Research).
  • Anchoring Strategy: To ensure de-identification, absolute dates are replaced with relative timelines anchored to the date of diagnosis. All subsequent events are calculated as days/months/years relative to this anchor.
• Standardization: The platform adopts OHDSI's standardized vocabularies (specifically the OMOP Common Data Model).
  • ICD-10-CM and ICD-O-3 are mapped to SNOMED CT.
  • Drug names are mapped to RxNorm.
  • This ensures semantic interoperability and facilitates cross-institutional comparison.

B. System Architecture & Security

• Two-Tier Server Model:
  1. Extraction-Transformation Server: A secure Virtual Machine (VM) within the internal network where raw PHI is processed. Access is strictly limited to the design team via SSH/VPN.
  2. Application Server: Hosts the web interface. It receives only de-identified, transformed data via scheduled nightly tasks.
• Access Control:
  • Users must authenticate via Azure Active Directory (SSO).
  • Access is restricted to users with approved IRB protocols and completed CITI training (Human Subjects Research).
  • Institutional "honest brokers" manage user group permissions.
• Technology Stack:
  • Backend: Python (3.13), Django (REST API), SQLite3 (intermediate storage), Elasticsearch (search indexing).
  • Frontend: Angular.
  • Visualization: D3.js (timeline), Chart.js (pie/bar charts).
  • Analytics: Python lifelines (Kaplan-Meier), scikit-learn (statistical tests).

C. Functional Features

• Omni-Search: Users can query cohorts using inclusion/exclusion criteria based on demographics, histology, treatments, biospecimens, and molecular markers.
• Timeline Visualization: A patient-specific timeline displays the sequence of care events relative to diagnosis. Hovering reveals details (e.g., histology, drug dosage) without showing PHI.
• Statistical Analysis: The "Output" tab allows users to:
  • Compare cohorts using Kruskal-Wallis H-tests (continuous) and Chi-square tests (categorical).
  • Generate Kaplan-Meier survival curves with log-rank p-values.
  • Export data (Excel) for further analysis.

3. Key Contributions

• Novel Visualization: A unique timeline interface that visualizes the entire patient episode of care in a de-identified manner, allowing researchers to spot patterns in disease progression and treatment response.
• Self-Service Capability: Enables non-technical researchers to construct complex cohorts and perform retrospective analysis without needing programming skills or direct access to raw databases.
• Modular Design: The architecture allows for the easy inclusion of new data sources and event types.
• Regulatory Compliance: The dual-server architecture and strict access controls provide a robust framework for handling sensitive clinical data while enabling research utility.
• Standardization: Integration with OMOP CDM standards facilitates future data sharing and collaboration with other institutions.

4. Results

• System Adoption:
  • The platform was deployed at Roswell Park Comprehensive Cancer Center.
  • Usage Metrics: Over 100 individual researchers utilized the system. Monthly search queries grew from a baseline of 313 in the first month to a peak of 465 queries in the eighth month.
  • Training was facilitated via embedded videos, documentation, and in-person workshops.
• Use-Case Validation (Head and Neck Cancer):
  • Researchers used nSight™ to analyze Head and Neck Squamous Cell Carcinoma (HNSCC) patients.
  • They stratified patients into two age groups (18–64 vs. >65) and performed survival analysis.
  • Outcome: The system identified significant differences in 5- and 10-year survival rates between the cohorts.
  • Efficiency: The entire workflow (workspace creation, cohort selection, stratification, and statistical analysis) was completed in approximately 20 minutes.
  • Validation: Findings were independently validated against a REDCap database maintained by the Radiation Medicine department.

5. Significance

• Accelerating Research: nSight™ significantly reduces the time and technical barrier required for cohort discovery and feasibility assessment, allowing researchers to rapidly test hypotheses.
• Bridging the Gap: It effectively bridges the divide between clinical care data and research biorepositories, enabling "bench-to-bedside" translation.
• Scalability: The use of OMOP standards and modular architecture suggests the platform can be adapted by other institutions with similar data warehouses.
• Future Directions: The authors acknowledge current limitations, such as the inability to integrate longitudinal imaging data (CT/MRI) and the lack of tumor-instance-level granularity for some data types. Future iterations aim to incorporate AI-driven radiomics and multiomic analysis pipelines.

In summary, nSight™ represents a critical advancement in cancer informatics, providing a secure, standardized, and user-friendly environment for the retrospective analysis of multimodal clinical data, thereby empowering researchers to identify novel study populations and therapeutic strategies more efficiently.