Whether from first-hand experience or from TV and film, we’ve probably all seen those medical charts that hang at the bottom of hospital beds. A new study makes the surprising claim that it might be better if these graphical charts were replaced or complemented with short passages of text conveying the same information. Marian Van Der Meulen and colleagues say that graphs are prone to misinterpretation by inexperienced, distracted staff and text leads to more accurate courses of action. Of course translation of medical charts into text-based summaries is labour intensive to an impractical degree, as the researchers freely acknowledge. But they say new software that can automatically translate data into text-based summaries could potentially solve this problem.
Van Der Meulen’s team presented 35 nurses and doctors from the neonatal intensive care unit (ICU) at the Royal Infirmary of Edinburgh with real data from 24 infant patients. The participants’ task was to scrutinise the data and decide on what the next course of action should be. The data, which provided information on factors like blood pressure and temperature, as well as previous actions taken by staff, was either presented via time series graphs, in the conventional manner; as text-based summaries translated from the graphs by medical experts; or as computer-generated text. It’s important to note that both forms of text summary provided no clinical interpretation, they merely summarised the salient information in the graphical data.
Remarkably, the participating nurses and doctors chose significantly more appropriate courses of action after looking at the textual summaries written by an expert as compared with looking at the standard time-series graphs. Decisions made after looking at the computer-generated text were poorer than decisions taken after the human-generated text but were just as accurate as decisions made from the graphs.
‘Overall, these results confirm that in a neonatal ICU, human generated descriptions of time series physiological measures are better able to support medical decision-making than graphs with trend lines,’ the researchers said.
These findings will only have relevance to real-life hospital settings if a way can be found to make the computer-generated text as effective as the text written by a human expert. The researchers are confident that this can be achieved. A research paper they have in press has compared the two types of text to look for differences that could help improve the BT-45 software that was used in this study. Such differences include the human text having a more coherent grammatical structure and narrative and a tendency to group physiological measures together.
‘…[F]urther development of this technology is likely to be extremely fruitful in supporting complex real-world cognition,’ the researchers concluded.
van der Meulen, M., Logie, R., Freer, Y., Sykes, C., McIntosh, N., & Hunter, J. (2010). 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. Applied Cognitive Psychology, 24 (1), 77-89 DOI: 10.1002/acp.1545