What is MH-TAC?
The Mental Health Text Analytics Cloud (MH-TAC) platform has been developed to date as a prototype ‘proof of concept’ via a collaboration between South London and Maudsley and Cambridgeshire and Peterborough NHS Trusts, along with King’s College London, University of Cambridge, and University of Sheffield.
The purpose has been to set up an NHS domain environment where NLP applications can be hosted (compliant with any security requirements for the applications themselves) and where text from mental healthcare records can be processed via chosen applications (compliant with any security requirements of the data custodian) with the text and associated meta-data returned directly to the data custodian for their further use.
This has required the assembly both of a technical architecture to accommodate processing and a service configuration and documentation to accommodate information governance and data security requirements.
MH-TAC infrastructure consists of two interfaces – an API for real-time processing of relatively small volumes of text, and a facility for bulk uploads and processing of larger volumes. Performance of the bulk processing has been developed and evaluated via successful full text uploads from the South London and Maudsley and further proof of concept has been completed through the recent processing and return of text from Oxford Health NHS Foundation Trust. Performance of the API has been developed and evaluated via successful processing of data feeds from Cambridgeshire and Peterborough since 2019.
NLP functionality rendered available
Via the MH-TAC prototype service, all current CRIS NLP functionality is rendered theoretically available to other mental health services. The key dependencies here are the following:
1. Data Sharing Agreement and other required documentation.
2. An extraction of text to be processed. This requires the text fields and appropriate document-level and individual-level identifiers (encrypted) to be prepared as a table for processing. This is envisaged to be a readily achievable task for any NHS database administrator with back-end electronic record familiarity (e.g., a member of a Trust’s digital services team).
3. A mutually agreed data transfer process for uploading source text and receiving processed output. This is a component of the initial Data Sharing Agreement.
Available CRIS NLP functionality is described in the CRIS Natural Language Processing Applications Library.
MH-TAC Governance and Security
As the service provider, South London and Maudsley NHS Foundation Trust acts as the data processor. Partners are required to sign a Data Processing Agreement (DPA) with SLaM to join. The SLAM service is limited to system administration and user support only.
Data uploaded and/or stored by partners in the Dedicated Document Store (DDS) are not available to SLaM. SLaM does not access or directly process partners’ data in any manner. If/when a partner withdraws or the DPA is ended SLaM will destroy the partner’s DDS infrastructure, entirely.
The service is offered as a secure infrastructure service, run and provided by Microsoft in the Azure Cloud and implementation of security controls, therefore, rests with Microsoft. The Microsoft Cloud is compliant to industry standards such as ISO27001.
The mechanism of data transfer over the internet renders data fully encrypted at rest and transfer between partners’ local premises and GATE cloud.
Routine audits will be carried to assure partners that all operational service and governance commitments are carried according to the standards and schedules described in the DPA.
