What challenges did Unnamed AWS healthcare customers face before AI implementation?

Healthcare providers face challenges in effectively using heterogeneous clinical data including structured and unstructured records to predict patient outcomes. Interpreting complex prediction model results in an intuitive way to support clinical decisions is difficult.

What was the impact of the AI solution?

The model achieved an AUC of 0.82 and an F1 score of 0.74, indicating good predictive performance. Healthcare providers can intervene earlier with patients at risk, improving care outcomes and reducing hospital readmissions. Visualization of SHAP values helped clinical teams better understand model decisions and trust AI predictions.

Source

Evidence: Low25/100

Healthcare Patient Outcome Prediction Using Amazon HealthLake and Amazon SageMaker

Use case typeClinical analyticsUpdated Jun 13, 2026

Unnamed AWS healthcare customers developed a deep learning model to predict patient outcomes such as mortality within 90 days after ICU discharge by utilizing both structured and unstructured healthcare data. The solution used Amazon HealthLake to normalize and extract clinical data, combining embedding techniques for richer unstructured data representation. A custom convolutional neural network model was trained on Amazon SageMaker using TensorFlow containers. Visualization of results was provided using SHAP values for interpretability through a custom UI and API Gateway. This approach enabled improved healthcare provider decision-making and early patient intervention based on predictive insights.

Organization: Unnamed AWS healthcare customers
Industry: Healthcare
Location: United States
Published: June 2021

Reported outcomes

Strategic outcomes

Better decisions & insightEnabled earlier patient interventionCustomer experience & trustImproved clinician trust in AI predictionsBetter decisions & insightImproved healthcare decision-making

Why do we believe this deployment?Customer identity, provider attribution, maturity, and source checks

Customer: Unnamed AWS healthcare customers
Provider: AWS
Maturity: Unknown
Linked source: AWS Machine Learning Blog

No explicit deployment-stage evidence found.

Customer identity supportedSource describes one deploymentMaturity evidence evaluated

Primary read

Use case focus

Showing 3 of 3

1Patient Outcome Prediction
2Deep Learning
3Model Interpretability

Healthcare providers face challenges in effectively using heterogeneous clinical data including structured and unstructured records to predict patient outcomes.
Interpreting complex prediction model results in an intuitive way to support clinical decisions is difficult.

Data from Amazon HealthLake was exported and transformed using AWS Glue and crawlers to create a queryable Data Catalog.
Deep learning models were trained on Amazon SageMaker to combine embeddings from unstructured data and structured clinical variables for accurate predictions.
An API Gateway and Lambda serverless functions were developed to deliver visual explanations of predictions via SHAP values in a web UI for model interpretability.

Technologies

Amazon HealthLake Amazon SageMaker Amazon S3 AWS Lambda Amazon API Gateway

The model achieved an AUC of 0.82 and an F1 score of 0.74, indicating good predictive performance.
Healthcare providers can intervene earlier with patients at risk, improving care outcomes and reducing hospital readmissions.
Visualization of SHAP values helped clinical teams better understand model decisions and trust AI predictions.

Architecture

The architecture includes Amazon HealthLake for clinical data normalization, a data catalog with AWS Glue, model training and inference pipelines on Amazon SageMaker with TensorFlow, and visualization through Lambda and API Gateway for SHAP-based interpretability.

Sources & evidence1

Evidence: Low25/100Evidence strength