Using cognitive task analysis to facilitate the integration of decision support systems into the neonatal intensive care unit Gordon D. Baxter a, * , Andrew F. Monk a , Kenneth Tan b , Peter R.F. Dear b , Simon J. Newell b a Department of Psychology, University of York, Heslington, York YO10 5DD, UK b Department of Paediatrics, St James’ University Hospital, Leeds LS9 7TF, UK Received 28 May 2004; received in revised form 17 December 2004; accepted 10 January 2005 Artificial Intelligence in Medicine (2005) 35, 243—257 http://www.intl.elsevierhealth.com/journals/aiim KEYWORDS Decision making; Neonatal intensive care; Artificial respiration; Respiratory distress syndrome; Expert systems Summary Objective: New medical systems may be rejected by staff because they do not integrate with local practice. An expert system, FLORENCE, is being developed to help staff in a neonatal intensive care unit (NICU) make decisions about ventilator settings when treating babies with respiratory distress syndrome. For FLORENCE to succeed it must be clinically useful and acceptable to staff in the context of local work practices. The aim of this work was to identify those contextual factors that would affect FLORENCE’s success. Methods: A cognitive task analysis (CTA) of the NICU was performed. First, work context analysis was used to identify how work is performed in the NICU. Second, the critical decision method (CDM) was used to analyse how staff make decisions about changing the ventilator settings. Third, naturalistic observation of staff’s use of the ventilator was performed. Results: A. The work context analysis identified the NICU’s hierarchical communica- tion structure and the importance of numerous types of record in communication. B. It also identified important ergonomic and practical requirements for designing the displays and positioning the computer. C. The CDM interviews suggested instances where problems can arise if the data used by FLORENCE, which is automatically read, is not manually verified. D. Observation showed that most alarms cleared automa- tically. When FLORENCE raises an alarm staff will normally be required to intervene and make a clinical judgement, even if the ventilator settings are not subsequently changed. Conclusions: FLORENCE must not undermine the NICU’s hierarchical communication channels (A). The re-design of working practices to incorporate FLORENCE, reinforced through its user interface, must ensure that expert help is called on when appropriate * Corresponding author. Tel.: +44 1904 433170; fax: +44 1904 433181. E-mail address: g.baxter@psych.york.ac.uk (G.D. Baxter). 0933-3657/$ — see front matter # 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.artmed.2005.01.004