When a graph is poorer than 100 words: A comparison of computerised natural language generation, human generated descriptions and graphical displays in neonatal intensive care

Abstract

Volunteer staff from a Neonatal Intensive Care Unit (NICU) were presented with sets of anonymised physiological data recorded over approximately 45 minute periods from former patients. Staff were asked to select medical/nursing actions appropriate for each of the patients whose data were displayed. Data were shown in one of three conditions (a) as multiple line graphs similar to those commonly shown on the ward, or as textual descriptions generated by (b) expert medical/nursing staff or (c) computerised natural language generation (NLG). An overall advantage was found for the human generated text, but NLG resulted in decisions that were at least as good as those for the graphical displays with which staff were familiar. It is suggested that NLG might offer a viable automated approach to removing noise and artefacts in real, complex and dynamic data sets, thereby reducing visual complexity and mental workload, and enhancing decision-making particularly for inexperienced staff.

Automatic generation of natural language nursing shift summaries in neonatal intensive care: BT-Nurse

“…Again there are many variations, ranging from controlled psychological experiments in artificial contexts [5] to randomised controlled clinical trials [26].…”

“…BT-45 was developed first. It was evaluated using a controlled experiment [5], where doctors and nurses were shown clinical information in three presentations (computer-generated text, manually written text, and visualisation), and asked to make a treatment decision; this was done off-ward, using historical data from babies who had been in the NICU several years previously. Decision quality was best when doctors and nurses saw the manually written texts; there was no significant difference in decision quality based on computer-generated texts and visualisations.…”

Multitasking: Multiple, Domain-Specific Cognitive Functions in A Virtual Environment

“…Shallice & Burgess, 1991). Many occupations require multitasking, for example emergency medicine and medical decision making (Chisholm, Dornfeld, Nelson & Cordell, 2001;Law et al, 2005;van der Meulen et al, 2010), management (Seshadri &Shapira, 2001), or navigation (e.g. Garden, Cornoldi & Logie, 2002;Law, Logie & Pearson, 2006;Spiers & Maguire, 2006).…”

Comparing Multi-label Classification with Reinforcement Learning for Summarisation of Time-series Data

“…Finally, our domain for feedback generation is motivated by previous studies (Law et al, 2005;van den Meulen et al, 2010) who show that text summaries are more effective in decision making than graphs therefore it is advantageous to provide a summary over showing users the raw data graphically. In addition, feedback summarisation from time-series data can be applied to the field of Intelligent Tutoring Systems (Gross et al, 2012).…”