Harnessing computational intelligence to improve the management of health emergencies

The analysis of thousands of records of patients' physiological values has allowed a researcher in the UPV/EHU's Computational Research Group (GIC) to develop a system of algorithms to determine the degree of seriousness of the condition of the patients.  A system of this type offers an improvement in spotting the more serious cases, and not only do the patients receive better attention, the resources of the healthcare services are also better managed.

Hospitals and all the other centres devoted to healthcare accumulate numerous databases with all the records of a whole host of physiological variables of the patients they treat. The processing and analysis of these data can enable healthcare staff to anticipate and spot those patients at greater risk of deterioration. Asier Garmendia, a researcher in the GIC (Computational Intelligence Group) at the UPV/EHU, has developed a system based on computational intelligence for this purpose.

In his study and subsequent development of the algorithms needed for the system, he used two databases, each from one of two hospitals in Santiago de Chile. One of the databases chosen for this study was on paediatric patients who had been admitted at some point to the intensive care units owing to respiratory problems, and the other on patients who having gone to A&E were discharged but who returned after a few days and were subsequently admitted to hospital. These two databases coincide "with two of the biggest problems in healthcare associated with large cities like Santiago de Chile, which are respiratory diseases caused by pollution and the management of the attention and care of the patients who come in search of medical attention," said Garmendia.

Early detection, better care

In the first of the cases, by using the records of the variables taken from each patient every so often during hospitalisation, the aim was to specify the degree of triage, which is the variable that classifies patients according to seriousness on the basis of the rest of the variables measured such as temperature, oxygen saturation, breathing rate, etc. "By means of computational intelligence algorithms an attempt is made to predict what the triage should be," said the researcher. The final aim of this system would be "to automatically monitor the patients and that an alarm should be sounded whenever the triage worsens". This study has also revealed that the variable that best predicts the triage level is breathing rate. "This is strange, as the doctors say that in their opinion the variable that best predicts this triage is the oxygen saturation in the blood," he added.

In the second of the cases, what they sought was to try and spot the destination that should be given to the patients going to the A&E service, in other words, whether to discharge them or admit them to hospital. "The problem that exists in this aspect is that a proportion of the patients who are discharged during the first consultation return to the A&E service after a few days and then they are in fact hospitalised. Approximately 14% of paediatric cases returning to consultation within a period of time of between 3 and 7 days are admitted to hospital. In the case of adult patients, they are 1 in 3," explained Garmendia.

"Having a system to solve this problem would lead to better care of the patients of course, but it would also signify considerable economic savings. Firstly, the resources of the healthcare services would be better managed, and secondly, a situation that currently arises with insurance would be avoided: insurance companies do not cover the costs arising out of hospitalisation in these cases, as they regard having discharged the patient during the first consultation as hospital negligence", he added. The result given by these algorithms developed for this purpose was a degree of precision of 60%; in other words, "our system was capable of spotting seriousness in six out of every ten patients who, in principle, did not appear to be likely candidates for hospital admission immediately".

The prediction systems developed "could be extended to and applied in any hospital in any country," said Garmendia. But one thing is clear and it is that before that "it is necessary to go on working on the design of the system, expand the number of data, and make the necessary adjustments," he concluded.

Additional information

Asier Garmendia-Mujika (Lazkao, 1969) conducted this study as part of his PhD thesis written up in the Computational Intelligence Group of the Department of Computation Science and Artificial Intelligence of the UPV/EHU's Faculty of Informatics. The thesis is entitled Contributions from computational intelligence to healthcare data processing and was supervised by Manuel Graña-Romay, the group's lead researcher. Garmendia also had the collaboration of the CEINE research group of the University of Chile where he obtained the data for developing the system.