270 likes | 460 Views
Effective Decision Support in the Neonatal Intensive Care Unit. NEONATE. NEONATE - Staff. Aberdeen, Dept. of Computing Science : Jim Hunter (PI) Gary Ewing (Research Fellow)* Aberdeen, Dept. of Psychology Robert Logie (PI) Sue Rudkin (Research Assistant)* * funded by PACCIT.
E N D
Effective Decision Support in the Neonatal Intensive Care Unit NEONATE
NEONATE - Staff • Aberdeen, Dept. of Computing Science: • Jim Hunter (PI) • Gary Ewing (Research Fellow)* • Aberdeen, Dept. of Psychology • Robert Logie (PI) • Sue Rudkin (Research Assistant)* * funded by PACCIT
NEONATE - Staff • Edinburgh, Simpson Maternity Hospital: • Neil McIntosh (PI) • Yvonne Freer (Nursing Research Co-ordinator) • Lindsey Ferguson (Research Nurse)* • Peter Badger (System Developer) • Alan Maxwell (Programmer) * funded by PACCIT
Background • Neonatal ICU: PC beside each cot displaying time series (and other) data. • Clinical trial: no detectable improvement in clinical outcomes. • COGNATE (ESRC funded - ‘96 to ‘99): those spending most time with the babies (junior doctors and nurses) didn’t: • refer to the displays very often, or • extract as much information as possible. • One conclusion: need for decision support. • New system now deployed.
Overall Objective • How can computer-based decision support improve patient care in neonatal intensive care units, • with minimal increase in the load on clinical staff, • and therefore the maximal use of data collected automatically?
Decision Making - Highly Simplified! DATA CLINICIAN ACTION
Sources of Data • Equipment connected to the baby • Monitor, ventilator, incubator, … • Equipment not directly connected • On-ward blood gases, laboratory, X-ray, … • Direct sensory perception • Vision, touch, hearing.
Automatic Data Collection • Currently available • Monitor (one second resolution) : heart rate, blood pressure, O2, CO2, temperatures ... • ‘Real soon now’ • Ventilator : Mode, pressures, FiO2, respiration rate … • Incubator : O2, temperature • On-ward blood gases : pH, pO2, pCO2 ... • Laboratory : Haemoglobin, Na, K, Urea …
Sensory Data • ‘COGNATE’ showed that sensory data was very important. • Skin colour, expression, degree of movement, muscle tone, type of crying, ... • Impossible to collect automatically, either now or in any reasonably foreseeable future.
Observations: Practicalities • Employ a research nurse to record: • all sensory data (‘descriptors’) and actions taken • over a four month period (mid-October to mid-February) • Preparatory work • ontology of actions and descriptors • fast entry of observations • protocol
Questions COMPLETE DATA Was this the ‘best’ action? CLINICIAN Review the actions taken: (i) on the ward; (ii) in ‘off-ward’ re-runs. ACTION
Questions COMPLETE DATA Can we interpret what is going on? PROGRAM Investigate a variety of intelligent signal processing techniques.
Questions COMPLETE DATA PROGRAM Can we help to improve the actions? CLINICIAN ACTION Run a repeat of the ‘off-ward’ experiments.
Questions COMPLETE DATA AUTOMATICALLY ACQUIRED DATA CLINICIAN PROGRAM Can we still help to improve the actions? ACTION
Elicitation of Terms • Need for an agreed lexicon of terms to describe the actions carried out and the relevant sensory data (descriptors). • Desirable to investigate relationships between terms for optimal data input. • Interviews to elicit clinical actions (~70) and patient descriptors (552). • Subjects (32) : 8 JN, 8 SN, 8JD, 8 SD. • SME vetted actions (51) and descriptors lists (132); • Removal of singletons and synonyms.
Card-Sorting • "Concept Sorting" is a well-known technique. • A flexible method to label the basic categories and to reach an agreement on a conceptual structure. • SME sorts cards into piles (values) according to different criteria (attributes). • Cognitive psychology studies have shown that this is very efficient elicitation technique.
Card-Sorting Procedure Merge label cards into higher level categories Continue process until highest level categories reached. Sort cards into piles and label piles Shuffled Set of Cards
Card-Sorting Experiments • “Actions” card-sorts: Subjects (32) : 8 JN, 8 SN, 8JD, 8 SD (completed), • 2 sorts per subject (at different times); • Approx. I hours per session. • “Descriptor” card-sorts: Subjects (32) : 8 JN, 8 SN, 8JD, 8 SD (in progress, 1st sort 2/3 complete), • 2 sorts per subject (at different times)’ • Approx. I.5 hours per session;
Analysis of Results • Card-sort will be analysed by cluster analysis. • Standard stats packages need data in the form of matrices of either distance scores or similarity scores: • Tedious procedures or software to convert; • Output can be hard to interpret. • Exploring, use of special purpose software to analyse and present results; e.g. tree diagrams (dendograms).
Project Plan A: Prepare for data collection. B: Develop schemes for codifying visual and tactile data. C: Collect detailed data on patient state and actions. D: Carry out preliminary analysis and selection. E: Conduct and analyse off-ward experiments. F: Design and implement decision support software. G: Evaluate the effectiveness of the decision support aids. H: Write up and disseminate results.