Optimizing Medical Record Processing with Prompt Caching at Scale

care_access · Healthcare · 2025 · source Agentic: no · promotional bias: med · source-backed · extracted with prompt 0.5.6 (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: Medical record LLM processing

Care Access needed to run multiple analyses over hundreds of medical records per day without repeatedly reprocessing the same large static record content.

Constraints:

The program processes 300-500+ medical records daily.
Each record requires multiple LLM analyses.
Medical data must remain HIPAA-compliant and privacy-preserving.
Repeatedly sending static record content increased cost and latency.

Architecture

Care Access used Amazon Bedrock prompt caching to cache static medical record content while varying only the analysis questions.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  medical-record-content["Medical Record Content"]
  bedrock-prompt-cache["Bedrock Prompt Cache"]
  analysis-questions["Analysis Questions"]
  screening-workflow["Screening Workflow"]
  medical-record-content --> bedrock-prompt-cache
  bedrock-prompt-cache --> analysis-questions
  analysis-questions --> screening-workflow

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

Components

Component	Role	Interacts with
`Medical Record Content`	Provides static patient record context for screening analyses.	`Bedrock Prompt Cache`
`Bedrock Prompt Cache`	Caches static record content across multiple analysis questions.	`Analysis Questions`
`Analysis Questions`	Vary per screening analysis while reusing cached record context.	`Screening Workflow`
`Screening Workflow`	Connects participants with health resources and clinical trial opportunities.	—

Data flow

Medical record content is cached once in Bedrock, different analysis questions reuse that cached context, and the screening workflow avoids resending large static record payloads for every question.

“caching the static medical record content while varying only the analysis questions”summary, stated

“Each medical record required multiple LLM-based analyses through Amazon Bedrock”summary, stated

Why this architecture

Prompt caching matched the workload shape: large static records reused across multiple small analysis prompts.

“reprocessing substantial portions of medical data for each separate analysis question led to high costs and slower processing times.”summary, stated

Models used: Amazon Bedrock · Prompt caching

Scale

Care Access processes 300-500+ medical records daily.
“processing 300-500+ medical records daily”summary, stated
Prompt caching reduced data processing costs by 86%, a 7x decrease.
“86% reduction in data processing costs (7x decrease)”summary, stated
Processing became 66% faster, a 3x speedup.
“66% faster processing times (3x speedup)”summary, stated
The system saved 4-8+ hours of processing time daily.
“saving 4-8+ hours of processing time daily.”summary, stated

Alternatives considered (roads not taken)

Resend full medical record context for every analysis

Rejected because: It caused high costs and slow processing.

“reprocessing substantial portions of medical data for each separate analysis question led to high costs and slower processing times.”summary, stated

Patterns

`cost-aware-routing` (prompt-optimization)

Prompt caching reduced duplicate token processing and cost.

“86% reduction in data processing costs”summary, stated

`safety-policy-enforcement` (healthcare-compliance)

The system maintained HIPAA compliance and privacy while scaling screening.

“maintaining strict HIPAA compliance and privacy standards”summary, stated

Tradeoffs

Use prompt caching

Chose Cached static record content over Full-context reprocessing for every question.

Rationale: The repeated-record workload produced large cost and time savings from caching.

“66% faster processing times (3x speedup)”summary, stated

Gotchas (what bit them)

Repeated static context can dominate LLM cost.

Root cause: Each analysis question resent substantial medical record content.

Resolution: Cache static record content and vary only questions.

“caching the static medical record content while varying only the analysis questions”summary, stated

Evaluation

Care Access measured processing cost, speed, daily time savings, and compliance constraints.

Metrics: Cost reduction · Processing speedup · Daily hours saved · Medical records per day

“saving 4-8+ hours of processing time daily.”summary, stated

Outcomes

Costs fell 86%.
Processing became 66% faster.
Daily processing saved 4-8+ hours.
The screening program could scale while maintaining HIPAA compliance.

Care Access processes 300-500+ medical records daily.

“enabled the organization to scale their health screening program efficiently”summary, stated