1 of 7

DigiScribe MedChat

YOUR AI ASSISTANT FOR MEDICAL CODING

2 of 7

Idealization

  • Automating the process of getting relevant ICD codes from clinical notes of doctors
  • All diseases have a relevant standard ICD codes defined by WHO
  • Insurance claims need to have only relevant ICD,CPT, HCPCS codes
  • Payment for claims are done based on validation of these codes
  • Healthcare provider has put lot of efforts to have this codes correctly coded so that claims denials will be less
  • Currently extracting diseases from clinical notes of the doctor and finding the relevant ICD codes are done manually by Medical Coders
  • Wrong Code is a important factor of Claims Denials and impacts the financials

3 of 7

ICD-10 Standard Code Structure

  • International Classification of Diseases 

E00-E89: Endocrine, nutritional and metabolic diseases

├── E10: Type 1 diabetes mellitus

│ ├── E10.1: With ketoacidosis

│ ├── E10.2: With kidney complications

│ └── E10.9: Without complications

└── E66: Obesity

├── E66.0: Obesity due to excess calories

├── E66.1: Drug-induced obesity

└── E66.9: Obesity, unspecified

Sample ICD-10 Code

Sample Clinical Note

Joshua is a 19-year-old male with Type 1 Diabetes Mellitus, diagnosed at age 13, presenting for follow-up after hospitalization for DKA one week ago. He reports he had missed several insulin doses while traveling and started feeling fatigued with nausea and abdominal pain. He was admitted for DKA and treated with IV fluids and insulin, then discharged on his basal-bolus regimen.

4 of 7

Technology

  • pre-trained generative large language models (LLMs) to develop a practical solution that is suitable for zero-shot and few-shot code assignment, with no need for further task-specific training
  • simple_icd_10_cm (python library that deals with ICD-10-CM codes)
  • tree-search algorithm to retrieve icd code parent
  • Model - "gpt-3.5-turbo”
  • Streamlit – GUI

​

​

​

5 of 7

Demo

6 of 7

Future Scope

​

  • Automatic Medical Coding replacing hundreds of manual Medical coders saving huge cost for healthcare provider
  • Less chances of medical coding error
  • Even in manual medical coding can be used for Validation before submission for reimbursements reducing claim denials
  • Can be used by insurance companies for Fraud Detection
  • Can be augmented with AI voice agents and use doctors audio clinical notes as input
  • Can be chained with other LLM agents to generate Clinical documentations

7 of 7

​