[Paper Review] ISTHMUS: Secure, Scalable, Real-time and Robust Machine Learning Platform for Healthcare

Akshay Arora, Arun Nethi+6 more|arXiv (Cornell University)|Sep 29, 2019

Machine Learning in Healthcare10 references4 citations

TL;DR

Isthmus is a secure, HIPAA-compliant, cloud-based machine learning platform designed for healthcare that enables end-to-end deployment of predictive models with real-time data ingestion, data harmonization, and robust monitoring. It accelerates time-to-deploy for clinical AI by standardizing data pipelines, supporting model lifecycle management, and integrating diverse data sources—including EHRs, SDoH, and IoT—demonstrated through three case studies in trauma prediction, population health, and real-time monitoring.

ABSTRACT

In recent times, machine learning (ML) and artificial intelligence (AI) based systems have evolved and scaled across different industries such as finance, retail, insurance, energy utilities, etc. Among other things, they have been used to predict patterns of customer behavior, to generate pricing models, and to predict the return on investments. But the successes in deploying machine learning models at scale in those industries have not translated into the healthcare setting. There are multiple reasons why integrating ML models into healthcare has not been widely successful, but from a technical perspective, general-purpose commercial machine learning platforms are not a good fit for healthcare due to complexities in handling data quality issues, mandates to demonstrate clinical relevance, and a lack of ability to monitor performance in a highly regulated environment with stringent security and privacy needs. In this paper, we describe Isthmus, a turnkey, cloud-based platform which addresses the challenges above and reduces time to market for operationalizing ML/AI in healthcare. Towards the end, we describe three case studies which shed light on Isthmus capabilities. These include (1) supporting an end-to-end lifecycle of a model which predicts trauma survivability at hospital trauma centers, (2) bringing in and harmonizing data from disparate sources to create a community data platform for inferring population as well as patient level insights for Social Determinants of Health (SDoH), and (3) ingesting live-streaming data from various IoT sensors to build models, which can leverage real-time and longitudinal information to make advanced time-sensitive predictions.

Motivation & Objective

• To address the technical and regulatory challenges hindering the deployment of machine learning in healthcare, such as data quality, privacy, and interoperability.
• To reduce time-to-market for clinical AI models by providing a turnkey, cloud-hosted platform with reusable data and model pipelines.
• To enable seamless integration of diverse data sources—clinical, claims, SDoH, and real-time IoT—into a unified, secure environment for predictive modeling.
• To support end-to-end model lifecycle management, including training, deployment, evaluation, and retraining, with consistent feature engineering across environments.
• To demonstrate the platform’s versatility through real-world case studies in trauma survival prediction, community health data integration, and IoT-driven real-time monitoring.

Proposed method

• Isthmus is hosted on Microsoft Azure Cloud with HIPAA-compliant infrastructure, ensuring security, fault tolerance, and disaster recovery.
• A configurable data ingestion pipeline uses Apache NiFi to collect, clean, and store structured and unstructured data from EHRs, claims, SDoH, and IoT sources into PostgreSQL.
• Standardized data transformation packages are reused across training and inference to ensure consistency and detect data or feature drift in production.
• The platform supports model deployment via APIs, enabling real-time scoring and integration with clinical workflows and EHR systems.
• A modular architecture allows project-level isolation and code reusability, reducing redundancy and accelerating development.
• The system enables real-time model updates and future support for A/B and multi-armed bandit testing to ensure model reliability.

Experimental results

Research questions

• RQ1 How can a secure, scalable, and real-time machine learning platform be architected to meet the unique demands of healthcare data and regulatory standards?
• RQ2 What technical and architectural patterns enable consistent feature engineering across training and inference in clinical ML pipelines?
• RQ3 How can heterogeneous data sources—including EHRs, claims, SDoH, and IoT—be harmonized and integrated into a single platform for predictive modeling?
• RQ4 To what extent can a unified platform improve model performance and deployment speed in clinical settings compared to point solutions?
• RQ5 What are the practical implications of using a turnkey platform for real-time model monitoring, drift detection, and retraining in healthcare?

Key findings

• The Isthmus platform reduced time-to-deploy clinical ML models by reusing standardized data pipelines and model deployment environments across multiple projects.
• Integration of Social Determinants of Health (SDoH) data at the block group level significantly improved the predictive performance of a model for preterm birth risk.
• The platform successfully ingested and processed real-time data from indoor air quality IoT sensors, demonstrating readiness for streaming data workloads.
• The use of a centralized, secure, and auditable data pipeline enabled consistent data quality and feature engineering across training and inference, reducing drift-related model degradation.
• The platform’s modular and reusable architecture allowed for rapid prototyping and deployment of multiple models with minimal duplication of effort.
• Case studies confirmed the platform’s ability to support diverse use cases—from trauma survival prediction to population health analytics—within a single, compliant infrastructure.