Nanchang County People's Hospital
Nanchang County People's Hospital uses a local DeepSeek-R1 box and reports 60% less record-writing time
A county hospital in Jiangxi, China, installed an all-in-one appliance running DeepSeek-R1 to help doctors write inpatient records and suggest diagnoses. A vendor-promoted report says record-writing time fell by 60%, electronic record accuracy reaches 90% and a diagnosis suggestion takes 1 minute.
Key figures
Copied exactly from the sources. Copy any figure with its citation.
Saved 60%
Doctors' time to write inpatient records, after deploying the DeepSeek appliance1
Reaches 90%
Accuracy of electronic medical records1
What it took
Cost, team size and timeline were not disclosed in the sources.
The problem
In a normal inpatient workflow at Nanchang County People's Hospital, medical staff type and tidy up patient information by hand. The article says this is slow and prone to human error, and it held back service quality. Doctors also have to recheck medical terms and diagnosis logic while writing.
What they built
The hospital installed an all-in-one appliance from the vendor 宏杉科技 (Hongshan Technology), which contains the DeepSeek-R1 model, a vector engine for a document library (which lets the model look up documents) and a data-security module. It works inside the electronic record screen. It checks the text a doctor writes, flags term errors and logic conflicts, and offers an assistant that suggests a likely diagnosis from the chief complaint.
How they did it
The article describes a local application of the model and gives no technical detail of the hosting. The assistant is opened from a menu beside the record. It watches the text during writing, which the article describes as moving quality control from after-the-fact review to checks while the record is written. It gives no cost, team size or timeline.
Results
The article reports that record-writing time was saved by 60%, that electronic record accuracy reaches 90%, and that a diagnosis suggestion takes just 1 minute, with accuracy of around 90%. The head of the hospital's information department is quoted saying the real-time error alerts let doctors spend more attention on clinical decisions.
Lessons for your project
Run the tool inside the writing flow. The assistant sits in the record screen and flags term errors and logic conflicts as the doctor writes.
Define what accuracy means before you quote it. The source gives 90% for records and for diagnosis suggestions without saying how either was tested.
Check privacy options for patient data. The appliance includes a data-security module; ask any vendor how patient data is stored and who can see it.
Keep a doctor in the loop. The tool suggests and checks, and the clinician still decides.
Caveats
- The only source is a promotional article for the vendor's product, reposted on an aggregator site from another outlet. The hospital's own statement is limited to one quote.
- No method, sample size or baseline is given for the 60%, the 90% or the 1-minute figures. The article adds that accuracy approaches 100% in some scenarios, without saying which.
- The article says the hospital deployed the appliance 'recently'. The deployed date shown is the article's month.
- The article says this is the first county-level hospital in Jiangxi to deploy a local large model; this claim is the vendor's and is not verified.
Sources
- 1分钟诊断+90%准确率！宏杉科技DeepSeek一体机落地南昌县人民医院 (1-minute diagnosis and 90% accuracy: Hongshan Technology's DeepSeek appliance lands at Nanchang County People's Hospital)
China.com (via Jinri Redian) · 2025-03-26 · Vendor case study · m.tech.china.com
Summarized in our own words from the sources above. Spotted an error? See corrections.