For IS Security & Privacy Teams

Information Security Overview

This page is intended for hospital information security, privacy and procurement teams evaluating CoderCoach. It covers what the application does, what data is handled, where it is hosted, and how it is protected.

Questions not covered here? Contact us.

Application overview

What CoderCoach does

CoderCoach is a web-based, AI-assisted coding guidance and practice tool for Canadian clinical coding professionals. It provides guidance grounded in the Canadian Coding Standards, practice on fictional cases, and team practice summaries for managers. CoderCoach is advisory: the human coder makes every coding decision.

💬
CoderCoach Q&A

On-demand coding guidance. Coders ask coding questions in plain language, without patient information, and receive guidance with cited sources and a note of what the guidance could not confirm. Coders can also ask by voice and attach a fictional or de-identified document as context for one question.

📚
CoderCoach Education

Practice on fictional ICD-10-CA and CCI cases. Each attempt is compared with an AI-generated reference. The comparison is indicative and is not a measure of coding accuracy or competency. Managers see practice summaries for their team.

🔴
CoderCoach RT

In development. A planned workspace for coaching on de-identified live case documentation. It is not available for hospital use and requires a separate organizational agreement before any deployment.

Data categories

What information is entered

What is entered
  • Coding questions in plain language (Q&A), typed or spoken. A spoken question is recorded in the browser, sent for transcription, and treated exactly like a typed question afterwards.
  • Optional: a fictional or de-identified document attached to one question, and optional feedback (a 1–3 star rating and a comment).
  • Account information: name and work email address, plus province and coder type.
  • Practice activity: codes submitted on fictional practice cases and the comparison feedback.
  • Team practice summaries for managers: practice activity for the coders in their organization.
What is not entered
  • Patient names, dates of birth or other direct identifiers
  • Health card numbers, chart or visit numbers, or provincial identifiers
  • Real coded abstracts from a hospital
  • Hospital system credentials or EHR/EMR data
  • Financial or billing information
Patient identifier screening

How patient identifiers are handled

Users are instructed not to enter patient information, and a notice appears above the question box. As a safety measure, CoderCoach also screens text before it is answered.

Two-stage screening

Pattern check
Looks for labelled or titled names, dates of birth and other full dates, health card, chart and visit numbers, postal codes, email addresses and phone numbers.
AI name check
A second automated check looks for names. This check is performed by an AI model, so the text is sent to the AI provider for that check.

Text that is flagged is not answered, is not kept in the question log, and the user is asked to remove the identifier. Attached documents go through the same screening, and a document that cannot be checked is not used.

Limitation: These checks reduce risk but cannot catch every identifier. Automated screening must not be relied on to de-identify information.

Data flow

Where information goes

Q&A data flow
1

Coder opens CoderCoach in a browser. The web application is served from Vercel (United States).

2

Coder types or speaks a question, and may attach a fictional or de-identified document. A spoken question is sent to OpenAI to be transcribed; the recording is not kept.

3

The text is screened for identifiers (see above), then sent over HTTPS to the CoderCoach API (hosted on Render, United States).

4

The API sends the question, supporting standards text and any attached document to Anthropic to write the guidance. Standards passages are found using search embeddings (OpenAI).

5

Guidance is returned to the coder's browser. If the coder chooses Listen, a short spoken version is created by OpenAI.

6

The question and the first part of the answer are recorded in the CoderCoach database (Supabase, United States) for support, quality and audit. Attached documents and voice recordings are not stored.

RT data flow

CoderCoach RT is in development and is not available for hospital use.

The intended design is that hospital staff de-identify documentation before submission, that nothing is written back to hospital systems, and that documents are not stored by default. The details will be agreed in the organizational agreement before any deployment.

Infrastructure & hosting

Where the application and data are hosted

Hosting location: All CoderCoach infrastructure is hosted in the United States. Data is subject to US law and to the policies of each provider. Organizations with data residency requirements should assess this against their own policies.

ProviderRoleLocationCertification
VercelWeb application and CDNUnited States (primary US region)SOC 2 Type II
SupabaseDatabase and authenticationUnited States (US East, Ohio)SOC 2 Type II
RenderBackend APIUnited StatesSOC 2 Type II
AnthropicLanguage model (coding guidance)United States, under their API terms
OpenAIEmbeddings, speech-to-text, text-to-speechUnited States, under their API terms
ResendTransactional emailUnited States

Encryption

  • Data is encrypted in transit with TLS 1.2 or later (HTTPS only; HTTP is redirected)
  • Data is encrypted at rest by the hosting providers
Access control

Authentication and access

Authentication

  • Each person has an individual account
  • Sign-in is passwordless: the user receives a single-use link by email; CoderCoach does not store passwords
  • Sign-in is managed by Supabase Auth
  • Shared accounts are not supported

Who can access data

  • Each person has an individual account, and the application limits what each role can see
  • Managers can see practice activity and practice summaries for coders in their organization, including by name — they cannot see the text of Q&A questions or answers
  • Authorized CoderCoach administrators can see Q&A logs and feedback for support, quality and security purposes
  • No third party has access to organization data; CoderCoach does not sell it
  • Hospital systems have no access to CoderCoach's database

Sessions

  • Sessions use signed tokens that expire and are renewed while the user stays signed in
  • A user can sign out at any time, which ends the session on that device
  • An administrator can delete an account on request
Network access & integrations

Internet access and integrations

Internet access

CoderCoach is accessible over the public internet via any modern browser. No VPN, firewall rules, network changes or on-premises installation are needed.

  • The application is at coder.coach, over HTTPS. codercoach.ca and codercoach.com are the information websites
  • Compatible with standard browser security controls and content filtering
  • No browser plugins or extensions required

Software integrations

CoderCoach does not integrate with or connect to hospital systems, EHRs, abstracting systems or any third-party clinical software.

  • No integration with hospital EHR or abstracting systems
  • No HL7 or FHIR interfaces
  • No access to hospital networks or internal systems
  • Standalone web application — no installation or integration required
AI model

AI model and data processing

CoderCoach uses large language models from Anthropic and OpenAI through their APIs. Different features use different models.

What is sent to the AI providers

  • The text of the coding question, after screening, and the transcript of a spoken question
  • The text of an attached document, if the user attached one (also sent for identifier screening)
  • Standards text and instructions supplied by CoderCoach
  • Audio recorded for a spoken question (to OpenAI, for transcription)
  • The text of a written answer, in shortened form, when the user chooses Listen (to OpenAI, for audio)
  • No account information (name, email, organization)
  • No patient-identifiable information — flagged text is not answered and not sent for processing
Training

Under the providers' standard API terms, data sent through the API is not used to train their models. CoderCoach does not train or fine-tune models on user data.

Provider retention

Providers may keep API requests for a limited period for abuse and security monitoring under their standard terms. See Anthropic and OpenAI for current terms.

Processing location

Both providers state that requests under the standard API agreements are processed in the United States.

Privacy design

Privacy by design

Patient information not intended to enter

The application asks users not to enter patient information, shows a notice above the question box, and screens text as described above.

Documents and recordings are temporary

An attached document is read in memory for one question and is not stored. A voice recording is transcribed and discarded. Neither is written to the question log, feedback records or emails.

Data used only to run and improve the service

CoderCoach staff may review questions and feedback to improve guidance and to create fictional test cases. Data is not sold, not used for advertising, and not used to train third-party models.

Data minimisation

Accounts need only a work email and name. Province and coder type are also asked for, to show the right standards and practice cases.

Transparent AI

Guidance is labelled as AI-generated. Each answer lists sources consulted with the standard's version, and states that the interpretation is AI-assisted and has not been independently validated.

RT requires de-identification

When RT is deployed, hospital staff are responsible for removing all patient identifiers before submission. The organizational agreement confirms this responsibility.

Data retention & deletion

Data retention and deletion

DataRetention
Attached documents and voice recordingsNot stored
Question text and answer text in the Q&A logCurrently kept until a retention schedule is applied. A schedule is being introduced.
Optional feedback (comment, question and answer)Currently kept until a retention schedule is applied.
Usage and cost records (no question text)Kept for operational purposes
Account and practice dataWhile the account is active, then as below
Account deletion

An administrator can delete an account on request. Deletion requests are handled by CoderCoach staff. Some activity records may remain in logs until the retention schedule applies. When an organization ends its agreement, CoderCoach will export its data on request and delete the rest. Contact us for deletion requests or further detail.

RT deployment

CoderCoach RT deployment

CoderCoach RT is in development. It is not available for hospital use. When it is ready, deployment will require a separate organizational agreement covering:

Data de-identification responsibilities and process — the hospital is responsible for removing all patient identifiers before submission
Organizational data handling and security review — CoderCoach provides this document and answers questions from IS and Privacy teams
User provisioning and access control — who within the organization has RT access
Acceptable use — RT provides coding guidance; the coder makes every coding decision
Applicable licensing, privacy and security obligations under the organization's own policies
Data retention, deletion and audit log access, which are set in the agreement. By default CoderCoach does not store submitted documents.

CIHI disclaimer: This product is not affiliated with, sponsored by, or endorsed by the Canadian Institute for Health Information (CIHI), and CIHI has not reviewed or approved its content or functionality. ICD-10-CA and CCI are published by CIHI and are used under CIHI licence terms.

Questions from your IS or Privacy team?

We are happy to answer detailed questions, provide additional documentation, or arrange a call with our technical team to support your security review.

Contact us