Article Preview
TopIntroduction
Machine learning is a branch of Artificial Intelligence (AI) that constructs self-learning algorithms capable of labeling new data after extracting knowledge from previous examples. Some authors considered the evolution of pattern recognition as one of the main topics nowadays (Wooldridge & Jennings, 1995). Due to extreme importance and data availability, the healthcare community focuses on models that help predict diagnostics, create decision support systems (predict treatment outcome), and increase medical image quality. New and more complex models are used to demand knowledge representation from supervised to unsupervised and reinforced learning. In the present case study, the Artificial Neural Networks (ANN) are nonlinear statistical learning algorithms inspired by the neurons' structure and functionality.
In recent years the increase in computational capacity and data availability brought to the table new and innovative ways to implement and abstract the complex mathematical operations behind machine learning. The TensorFlow (Introdução Ao TensorFlow, n.d.) library and KERAS (Sebastian & Vahid, 2017) are among the top APIs used worldwide. Due to data availability and vital importance, healthcare is receiving attention from machine learning experts and developers worldwide, mainly on treatment predictions, outcomes, treatment guidelines, and without a doubt, medical image processing. No work related to the usage of a model applied to HL7 messages exchanged was identified.
Many healthcare organizations worldwide use Health Level Seven (HL7) standardized messages to exchange information between devices that perform different tasks and “talk” other languages inside an institution. For example, a physician orders a blood exam. That request represents a trigger event that leads to a message creation (request) following HL7 v2 standards. The interface is responsible for intermediating the exchange of data, mainly using sockets. That standard is known as pipe type “|” once every segment on a field is separated by a pipe. HL7 standards constitute the backbone for the healthcare organization's interoperability. Thus, that kind of system must work 24/7 without failures. So, every system that monitors and prevents errors on those interfaces is vital. After research, a gap was identified among the existing HL7 architectures. No ML modules were identified regarding the error prediction or stress hours identification directly related to HL7 interfaces. On the other hand, some works on diagnosis and patient's medical history were found regarding semantic interoperability.
To clarify the subject, interoperability in healthcare is the capability for two or more systems to exchange data and ultimately generate information from that (Garde et al., 2007). Without international standards, it is not possible to accomplish semantic and technical interoperability. HL7 International, OpenEHR, or even SNOMED CT are worldwide recognized organizations for proposing internationally accepted standards (Kalra, 2006). Those standards are used in clinical situations representation, information exchange, or archive/data generation interfaces (Clinical Knowledge Manager (CKM), n.d.). The exchange of information under the HL7 format is essential to a healthcare organization's day-to-day life. Almost all systems are connected using that kind of message, and the failure of one of such web interfaces could cause the complete breakdown of a healthcare institution. Exams are requested, and reports are printed using HL7 as a way to transport the message.
Developing such interfaces for interoperability is complex. Almost perfection and availability are some critical aspects of these interfaces. The other significant part of that specific interface is availability—some seasons with intense work like flu season or different particular situations. The idea is obvious. The healthcare systems do not have a margin for error or misunderstandings.
The present case study intends to infer the possibility of introducing an ANN or LSTM network to predict the number of messages being exchanged under the HL7 format for a given hour inside a healthcare organization. Then that model can be used by other systems to prevent errors and take measures in busy hours. The daily schedule for HL7 agents inside an organization can be approached based on that model as a prediction in a case similar to road traffic on cars or web applications usage.