AI-Driven Clinical Trial Transformation with Next-Generation Data Platform

novartis · Healthcare · 2025 · source Agentic: no · promotional bias: med · summary-only (no fetched source) · extracted with prompt 0.5.2 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: Clinical-trial data platform modernization

Novartis needed to integrate fragmented clinical-development data and enable AI use cases that could shorten drug development timelines.

Constraints:

GXP compliance
data governance
clinical safety
fragmented domains

Architecture

Novartis is a summary-only v2 upgrade. The case describes a GXP-compliant, domain-oriented data mesh platform for clinical-trial modernization. Domain data products run through AWS/Databricks processing and domain AWS accounts, practitioners discover and subscribe through an enterprise catalog, approved users consume business-qualified clean data through BI or AI/ML platforms such as SageMaker and Bedrock, and centralized observability monitors ingestion, processing, catalog access, quality, and cost.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  domain-data-products["domain data products"]
  data-mesh-platform["data mesh platform"]
  databricks-processing["Databricks processing"]
  domain-aws-accounts["domain AWS accounts"]
  enterprise-data-catalog["enterprise data catalog"]
  ai-ml-platforms["AI ML platforms"]
  visualization-tools["visualization tools"]
  central-observability["central observability"]
  domain-data-products --> data-mesh-platform
  data-mesh-platform --> databricks-processing
  databricks-processing --> domain-aws-accounts
  domain-data-products --> enterprise-data-catalog
  enterprise-data-catalog --> ai-ml-platforms
  enterprise-data-catalog --> visualization-tools
  data-mesh-platform --> central-observability
  domain-aws-accounts --> central-observability

Drag to pan · scroll or use +/− to zoom · reconstructed from the source's described component topology

Components

Component	Role	Interacts with
`domain data products`	Domain-oriented data products for patient safety, medical imaging, regulatory data, and other clinical-trial domains.	`data mesh platform`, `enterprise data catalog`
`data mesh platform`	GXP-compliant platform integrating fragmented clinical-development data across domains.	`domain data products`, `Databricks processing`, `enterprise data catalog`, `central observability`
`Databricks processing`	Data processing jobs over AWS infrastructure for transformation and product delivery.	`data mesh platform`, `domain AWS accounts`
`domain AWS accounts`	Development, QA, test, and production environments per domain for data products and CI/CD deployment.	`Databricks processing`, `central observability`
`enterprise data catalog`	Subscription entry point where practitioners find datasets, request access, and enter the data-product experience.	`domain data products`, `AI ML platforms`, `visualization tools`
`AI ML platforms`	SageMaker/Bedrock-style platforms consuming business-qualified clean data for protocol generation and other AI use cases.	`enterprise data catalog`
`visualization tools`	QuickSight, Power BI, Tableau-style consumption layer for approved data products.	`enterprise data catalog`
`central observability`	Central monitoring for ingestion events, processing jobs, catalog access, data quality, and cost.	`data mesh platform`, `domain AWS accounts`

Data flow

Clinical-development data from patient safety, imaging, regulatory, and other domains is modeled as domain data products. Data products flow through raw, intermediate, and consumer-specific layers in AWS/Databricks environments, are registered in a searchable catalog, and become available through an access-subscription workflow. Approved users then consume the data through visualization or AI/ML platforms, while centralized observability captures events, jobs, catalog access, data quality, and cost.

4 supporting quotes

“The company partnered with AWS Professional Services and Accenture to build a next-generation, GXP-compliant data platform that integrates fragmented data across multiple domains (including patient safety, medical imaging, and regulatory data)”summary, stated

“The platform implements a data mesh architecture with domain-oriented data products.”summary, stated

“Data products typically flow through three layers: raw ingestion (1:1 copy with potential anonymization), intermediate transformation (creating reusable cross-domain assets), and consumer-specific processing”summary, stated

“Data practitioners search the enterprise data catalog for relevant datasets, request access, and receive approval or rejection from data owners.”summary, stated

Why this architecture

The architecture chooses governed modular data foundations before clinical AI moonshots. GXP compliance, domain ownership, access subscriptions, metadata, CI/CD, and central observability are necessary because clinical-development AI depends on trusted cross-domain data and must support both operational use cases and ambitious simulation/protocol-generation work.

4 supporting quotes

“Modularity is central to their architecture. Rather than building monolithic systems, they designed use cases to be individually valuable but collectively transformative.”summary, stated

“This modular approach enables component upgrades without wholesale platform replacement.”summary, stated

“This enables the generative AI use cases showcased by Novartis leadership, such as protocol draft generation.”summary, stated

“All ingestion events, processing jobs, catalog access requests, data quality results, and cost information route to a centralized observability platform.”summary, stated

Models used: Amazon Bedrock models · SageMaker models

Scale

Novartis targeted at least six months of acceleration per clinical trial.
“Novartis embarked on a comprehensive data and AI modernization journey to accelerate drug development by at least 6 months per clinical trial.”summary, stated
The patient-safety implementation processed 17 TB through 16 data pipelines.
“The initial implementation with the patient safety domain achieved significant results: 16 data pipelines processing 17 terabytes of data”summary, stated
Reported platform results included 72% faster query speeds, 60% storage cost reduction, and more than 160 hours of manual work eliminated.
“72% faster query speeds, 60% storage cost reduction, and over 160 hours of manual work eliminated”summary, stated
Protocol-generation use cases demonstrated 83-87% acceleration for compliance-acceptable protocols.
“protocol generation use cases demonstrated 83-87% acceleration in generating compliance-acceptable protocols.”summary, stated

Alternatives considered (roads not taken)

Monolithic clinical-data platform

Rejected because: Novartis chose modular, individually valuable use cases that can compose and upgrade over time.

“Rather than building monolithic systems, they designed use cases to be individually valuable but collectively transformative.”summary, stated

Distributed observability logs per component

Rejected because: Centralized observability simplified dashboards and cross-account analysis.

“While logs could theoretically remain distributed, centralizing them simplifies dashboard creation and cross-account analysis.”summary, stated

Patterns

`safety-policy-enforcement` (regulated-data)

The platform is designed around GXP compliance and governed clinical-development data.

`structured-output-contract` (clinical-documents)

Protocol-generation use cases produce compliance-acceptable clinical trial documents faster.

“query speeds, 60% storage cost reduction, and over 160 hours of manual work eliminated, while protocol generation use cases demonstrated 83-87% acceleration in generating compliance-acceptable protocols.”summary, stated

Tradeoffs

How to frame AI in clinical development.

Chose augmentation of human intelligence over replacement of clinical experts.

Rationale: Novartis uses human-augmentation language throughout to manage safety and adoption in regulated drug development.

“The company's AI strategy is grounded in what they call "augmentation" - treating machine learning and artificial intelligence as enhancements to human intelligence rather than replacements.”summary, stated

Gotchas (what bit them)

Clinical AI depends on modular governed data foundations before moonshot use cases.

Root cause: Fragmented clinical domains cannot safely support regulatory AI workflows without data products, observability, and compliance controls.

Resolution: Build modular data-platform capabilities first, then layer operational AI and moonshot simulations on top.

“Modularity is central to their architecture.”summary, implied

Evaluation

Novartis measured platform and use-case acceleration metrics.

Metrics: 16 pipelines · 17 TB · 72% faster queries · 60% storage cost reduction · 160+ hours saved

“The initial implementation with the patient safety domain achieved significant results: 16 data pipelines processing 17 terabytes of data, 72% faster query speeds, 60% storage cost reduction, and over 160 hours of manual work eliminated, while protocol generation use cases demonstrated 83-87”summary, stated

Outcomes

Patient-safety implementation improved query speed, storage cost, and manual-work elimination; protocol generation accelerated 83-87%.

Transcript fetch failed; summary-only. Not clearly an agentic extraction despite AI use cases.

“e patient safety domain achieved significant results: 16 data pipelines processing 17 terabytes of data, 72% faster query speeds, 60% storage cost reduction, and over 160 hours of manual work eliminated, while protocol generation use cases demonstrated 83-87% acceleration in generating compliance”summary, stated