Scribe

Problem

Medical Social Workers (MSWs) meet needy patients daily to understand their personal backgrounds, financial circumstances, and social / medical issues, so as to provide an overall psycho-social assessment of the patient and to apply for financial assistance on their behalf.

These meetings are typically 1 hour long, after which MSWs must spend time documenting and consolidating their notes on the patient. Writing these case notes can take between 30 mins to 2 hours, depending on the case complexity and the experience of the MSW. The average MSW may see between 3 to 8 patients in a day, depending on seniority and specialty. There were about 600 Medical Social Workers in Singapore in 2019 ↗ (opens in new tab) and their numbers have only increased since then.

Solution

We built Scribe to help Medical Social Workers (MSWs) synthesise MSW-Patient conversations into structured progress notes. 

Scribe allows MSWs to record their conversations with patients, after which a multi-lingual transcript (with support for English, Malay, Chinese, and Singlish) is generated using state-of-the-art speech recognition models. 

Using this transcript, MSWs can then generate a summary of the transcript with a chosen output type, such as actionables in bullet points or a summary with section headings. A large language model (LLM) processes and summarises this transcript with the desired output type.

Both the speech recognition model and LLM are open-source models running on our infrastructure and we rely on no external APIs for Scribe, so sensitive patient data and conversations never leave our system.

The generated summary can then be exported directly to Care360 as a draft case note. (Care360 is a patient management system used by MSWs and is the key driver for their daily workflows.) On Care360, MSWs can further edit the case note to their satisfaction. By transcribing and summarising MSW-patient conversations and providing MSWs with a solid foundation to swiftly compose their notes, Scribe speeds up the note-writing process and enables MSWs to prioritise what truly matters—fostering meaningful connections and enhancing patient care.

User Testing and Feedback

We tested Scribe in a mock setting (i.e. simulated patient-MSW conversations) with MSWs from CGH, SGH, and SACH, as well as with Social Work students from NUS. 

For the transcription, it was suggested dialect support be added given that many needy patients are elderly. The transcription also did not always indicate which speaker made which utterance, which made it less readable. However, all participants agreed the lack of readability of the transcription was not a concern as long as the quality of the generated summary was high. As one remarked, “If summarised well, I won’t even refer to the transcript”.

Most MSWs were impressed by the quality of the summaries. One MSW remarked that they were “quite close to what we would write”, and another wondered if automatically generated summaries might result in MSWs not thinking for themselves. There was some surprise about the variable nature of summary generation, i.e. that the output generated would be slightly different every time. The summary was occasionally too verbose at times or had minor errors. Overall, they were quite forgiving of errors and pleased that Scribe provided a comprehensive summary that they could quickly refine / shorten to their needs.

The main concern with Scribe is whether patients would be less forthcoming in their conversations when recorded. It may not be appropriate to use Scribe when interacting with patients with more sensitive personal situations. Most participants agreed that they would only use the tool for patients they had more rapport with or if its usage was more normalised in their institution / cluster.

Starting from Mar 2024, we will be working with the healthcare clusters to pilot test Scribe with actual patients.

Team

Scribe was built by a multidisciplinary team of engineers, designers, and product managers.

If you would like to use Scribe in your organisation, please email team@scribe.gov.sg.

Learn more about Scribe
For the original prototype, see the full deck on Google Slides ↗ (opens in new tab). For the current version, visit guide.scribe.gov.sg ↗ (opens in new tab).