On Wednesday last week (31st October, 2007) there was a demonstration of some interesting software systems developed through a collaboration between the University of Sydney’s IT Department and the Intensive Care Department at Royal Prince Alfred Hospital. A joint press release was prepared and I think it is worth bringing to blog readers.
Begin Release -----
The first Natural Language Processing Systems to be used in Patient Care in an Australian HospitalSydney University’s Eureka Prize winner of 2005, Professor Jon Patrick, is once again leading the field by developing a suite of natural language processing systems to support patient care in the Intensive Care Unit at the Royal Prince Alfred Hospital Sydney. In a successful collaboration with Dr Robert Herkes, the Director of the ICU two new systems have been introduced into operations.
The first system converts doctors’ progress notes of patients into the formal medical encoding ontology SNOMED CT thus enabling more consistent descriptions of patient conditions and allowing large scale retrieval and analysis from the narrative part of the patient record.
The RPAH announced that it has commenced the use of a new Ward Rounds Information System (WRIS) developed by the School of Information Technologies at the University of Sydney. The purpose of the WRIS is
- to improve the efficiency of collecting data about patients during ward rounds, and
- demonstrate the automatic computation of SNOMED CT codes as clinicians write their progress notes.
The second system, a Clinical Data Analytics Language (CDAL) is being launched at the same time. Its purpose is to answer any question that can be answered from the data stored in the clinical information system. It will assist clinicians in the management of the vast amounts of complex information generated during an ICU admission and ultimately improve the quality and efficiency of care. CDAL allows staff to frame any question about their data in their database and get the answer almost immediately. CDAL, as well as giving access to the normal data in the clinical information system, also searches the doctor’s notes to help answer questions, operating somewhat like a Google engine would operate over the text data in the clinical information system. However it is much better than Google in that it understands a great deal about the natural language used in the notes and so it does a much better job of retrieving semantic content based on its context, for example you can ask for records where the “diagnosis is diabetes”, rather than searching for all instances of the word “diabetes”. The CDAL is supported by the SNOMED encodings created by the WRIS system so as to make its retrievals more accurate.
CDAL also has an hypothesis testing capability along with a restricted natural language interface so that you can ask a question of it directly in ordinary medical language, such as “ is there a significant difference in blood sugar levels for patients with diabetes mellitus between those over 50 years old compared to those under 50.” The CDAL engine will check the clinical notes for patients who have diabetes (excluding those records as “no diabetes”) extract their blood sugar levels from the correct records in the database and compute the significance test.
As one doctor said “I can do my research while I am on my ward rounds”. Other staff report that they expect significant time saving by being able to get accurate searching of pages of clinical notes for specific items of information when they need to resolve uncertainties for administering patient care or review a patient case.
One of the first tasks being targeted for CDAL is a collaboration with the Mayo Clinic to build an efficient and effective mechanism to identify Acute Respiratory Distress Syndrome ARDS. The Mayo and the RPAH will use an algorithm designed at the Mayo for identifying ARDS at-risk patients that can be installed in the CDAL in a minute and evaluated routinely every few minutes for all patient in the ward. The task is to assess the extent to which the constant monitoring by a CDAL routine can aid earlier detection of ARDS, and support rapid modification and subsequent evaluation of each version of the algorithm.
CDAL like WRIS, has been researched and developed to by independent of the system it is attached to so it can be moved to any other clinical information system and used in all other departments across the hospital.
WRIS achieves its purposes by two processing steps. It computes a tailored extract of the patient’s clinical record from the ICU’s information system CareVue, relevant to the needs of completing the ward round. This extract includes pertinent haemodynamic and laboratory data which is presented to the clinician on a screen, who then adds the relevant progress notes. After analysing the progress notes, WRIS computes the SNOMED CT codes in real-time, which the clinician then verifies. The correct codes are then able to be stored back into CareVue. The results can also be used to index the records so that when staff are searching for particular cases, or notes within a case, they can be retrieved directly in the same fashion that Google gives us access to relevant content across the Internet.
The WRIS system is the first example in Australia of the use of Natural Language Processing (NLP) with real-time processing at the point of care to support the care of patients.
The system will be of significant advantage to the clinician in their ward rounds. The automatic extraction of relevant content will give considerable time savings, both in terms of duplication and transcription considered to be up to 10 minutes per patient.
SNOMED CT has been introduced by the National E-Health Transition Authority (NEHTA) as Australia’s standard coding scheme for electronic clinical records. The introduction of WRIS is the RPA’s first initiative to systematically record its clinical notes in SNOMED CT codes. Automatically computing the SNOMED CT codes saves the work of the clinician needing to complete the coding.
The advantages of automatic coding will be extended and enhanced in the future with projects to:
- Automatically compute ICD 10AM and DRG codes from SNOMED CT codes.
- Improve data analytics for medical research with a stable representation of the contents of the medical records;
- Improve logistical planning for hospital management with more reliable and more detailed descriptions of the hospitals case mix;
- Engage clinicians in stabilizing their case descriptions around an agreed terminology and so enhance communications between different specialities and with the wider health community outside the hospital setting.
Research between the University of Sydney and the ICU at RPAH is on-going with projects to:
- enhance the accuracy of the text to SNOMED CT converter;
- develop a rich language for performing data analytics on the patient databases;
- automatically compute ICD 10AM codes from the SNOMED CT codes;
- automatically compute Diagnostic Related Group code (DRGs) from the combination of the electronic medical record and the ICD 10 AM codes:
- perform real-time auditing of patient care automatically by computationally comparing the patient’s record of care with the appropriate clinical guidelines.
Dr Robert Herkes
Director
Intensive Care Unit
Royal Prince Alfred Hospital
Camperdown, NSW
Professor Jon Patrick
Health Information Technologies Research
School of Information Technologies
Faculty of Engineering and IT
University of Sydney
Camperdown, NSW
End Release -----
It is really to see this sort of collaboration happening – especially for one who in another life was an ICU specialist – and to have beginning to emerge practical and useful outcomes of the use of clinical terminologies.
Keep up the good work!
I am sure there are other groups doing interesting and useful things around the ‘wide brown land’! I would love to let people know about them. My e-mail address is easy to find on the blog!
David.
0 comments:
Post a Comment