Medical records processing · Production

Care Access achieves 86% cost reduction and 66% faster data processing with Amazon Bedrock prompt caching

The problem

Care Access processed hundreds of medical records daily for its health screening program, with each analysis requiring multiple separate prompts that each reprocessed substantial portions of the record content, driving high operational costs and slower processing times as participant volume scaled.

First attempt

The initial LLM implementation required every separate analysis question to reprocess the full medical record content, and as participant volume grew, this approach led to significant daily operational costs.

Workflow diagram · grounded in source
1
EHR retrieval from S3
integration
“Individual electronic health records (EHRs) are retrieved from an Amazon S3 bucket, normalized for processing, and prepared for inference with unnecessary data removed”
2
Prompt cache setup
ai_action
“The medical record content becomes the static cached prefix, while specific analysis questions form the dynamic portion that varies with each query”
3
LLM inference via Bedrock
ai_action
“Each cached health record receives multiple analysis questions using Amazon Bedrock. Cache checkpointing activates when the prefix matches existing cache and exceeds the minimum 1,000 token requirement”
4
Output storage and analytics
output
“Results are combined into a single JSON per participant and stored in Amazon S3 for downstream analytics via Amazon Athena”
5
Clinical trial matching
routing
“Participants are then matched to relevant clinical trials”
Reported outcome

Prompt caching in Amazon Bedrock achieved an 86% reduction in Bedrock costs and a 66% reduction in processing time per record, saving 4-8+ hours of processing time daily, while maintaining compliance standards and enabling continued program growth.

Reported metrics
Amazon Bedrock cost reduction86%
Amazon Bedrock cost decrease factor7x decrease
Processing time per record reduction66%
Processing speed improvement factor3x faster
Show all 6 reported metrics
Amazon Bedrock cost reduction86%
Amazon Bedrock cost decrease factor7x decrease
processing time per record reduction66%
processing speed improvement factor3x faster
daily processing time saved4-8+ hours
time to launch LLM solutionsix weeks (instead of several months)
Reported stack
Amazon BedrockAmazon S3Amazon AthenaAWS Lake FormationAmazon CloudWatch
Source
https://aws.amazon.com/blogs/machine-learning/how-care-access-achieved-86-data-processing-cost-reductions-and-66-faster-data-processing-with-amazon-bedrock-prompt-caching?tag=soumet-20
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Prompt caching in Amazon Bedrock achieved an 86% reduction in Bedrock costs and a 66% reduction in processing time per record, saving 4-8+ hours of processing time daily, while maintaining compliance standards and ena…

What tools did this team use?

Amazon Bedrock, Amazon S3, Amazon Athena, AWS Lake Formation, Amazon CloudWatch.

What results were reported?

Amazon Bedrock cost reduction: 86%; Amazon Bedrock cost decrease factor: 7x decrease; Processing time per record reduction: 66%; Processing speed improvement factor: 3x faster (source-reported, not independently verified).

What failed first in this deployment?

The initial LLM implementation required every separate analysis question to reprocess the full medical record content, and as participant volume grew, this approach led to significant daily operational costs.

How is this medical records processing AI workflow structured?

EHR retrieval from S3 → Prompt cache setup → LLM inference via Bedrock → Output storage and analytics → Clinical trial matching.

Related medical records processing cases