护理教育中的人工智能和预测分析

Artificial Intelligence (AI) seems to be the new buzzword in healthcare and higher education. It is hailed as one way to solve problems that impact health professionals, patients, students, and educators by trying to improve the speed and accuracy by which health, learning, service, and other outcomes can be predicted and hence decisions made. Although it seems like a relatively recent technological development, the origins of AI can be traced back to the 1950’s and it has undergone consecutive waves of development and dissolution over the preceding years (Boden, 2018). One of its founding fathers defined AI as, “the science and engineering of making intelligent machines, especially intelligent computer programs” (McCarthy, 2007). These computational approaches began to gain traction in the 1990’s, when machine learning came to the fore with a range of supervised learning techniques. This enabled more complex statistical models to be built that were capable of making soft, probabilistic decisions. For example, in the United States, Harvey (1993) used an artificial neural network to support nurses’ decision making around diagnosis. Natural Language Processing (NLP), another branch of AI, also began to flourish around this time due to developments in linguistics and statistics, and improvements in computational power (Manning, 1999). Despite advances in AI, with the emergence of unsupervised learning and reinforcement learning techniques that enable ‘deep learning’ and progress with NLP systems, it was many years later before AI was applied in nursing education.

In 2008, Moseley and Mead (2008) used a machine learning technique called a decision tree to predict the drop-out rate from an undergraduate nursing programme in the United Kingdom. With a relatively small dataset of 3978 university student records, from 528 nursing students, the algorithm utilised this as a training dataset and then a smaller sub-set as a test dataset to forecast student attrition from the programme. They reported this method achieved 84% sensitivity, 70% specificity, and 94% overall accuracy. However, it was several years later before AI was employed again in pedagogical nursing research. A recent empirical study developed a predictive model to gauge if a nursing student would graduate from their university education programme at different points in their professional training (Hannaford et al., 2021). The researchers used eight different machine learning algorithms, i.e., random forest, xgboost, neural networks, support vector machine, C5.0, Naïve Bayes, K-nearest neighbour, and logistic regression, to build a model and compare the accuracy of the various techniques in determining nursing student graduation.

While most research studies focus on developing and testing AI algorithms and their associated predictive models, Swan (2021) administered an online survey (n = 675) with nursing students, nurse faculty, and practising nurses in the United States to explore their level of knowledge, use of, and attitudes towards AI in healthcare. Only 30% reported knowing how AI was used in clinical nursing practice and the majority had only fair or no understanding of technologies used in AI, with some participants highlighting nurses need a range of competencies in this area. Other studies have gone further and implemented an AI based tool in the real-world of nursing. For instance, Narang et al. (2021) compared nurses using an echocardiograph enhanced via a deep learning algorithm to sonographers using a standard echocardiograph, to understand if this type of AI approach could produce the same quality diagnostic medical images with limited training and clinical experience. The authors noted this could be particularly beneficial for both education and clinical practice in low resource settings. A study protocol by Shorey et al. (2019) also described a virtual counselling chatbot that would run on a NLP based system. In theory, this would interact with nursing students as a virtual patient to try and improve their communication skills.

Some nursing researchers have begun to synthesise literature on AI in nursing education. Harmon et al. (2021) undertook a scoping review of AI and virtual reality in clinical simulation for nursing pain education and found only four relevant studies. However, the description of AI used in this review seems to loosely centre on ‘computer assisted instruction’. Furthermore, none of the four included studies mention any specific AI technique in the development or application of the simulation technology, nor are any actual AI approaches noted in the review results. This may indicate a lack of understanding or confusion about what this field encompasses. A recent review of the literature on AI in nursing more broadly included over 100 studies (O’Connor et al., 2021), but highlighted the limited number related to nursing education, as most of the current evidence base focuses on clinical nursing practice in hospital settings. The review also emphasised the need to train and upskill nurses in AI, which others have also noted to prepare students, faculty, and nurses in practice for digitally enabled healthcare now and into the future (Booth et al., 2021, O'Connor and LaRue, 2021).

An international nursing consortium on AI convened in 2019, to identify priority areas for action (Ronquillo et al., 2021). They highlighted a lack of nurse faculty expertise in health informatics and AI that needs to be urgently addressed, so this topic can be taught in undergraduate and postgraduate nursing programmes. They posit that a lack of knowledge of AI, among other issues, could hold back the profession from participating in and leading AI initiatives in healthcare. Drawing on the experiences of other professional colleagues who are starting to utilise AI more in teaching and learning could benefit nursing. Masters (2019) discussed the future of AI in medical education, describing a number of potentially useful applications such as ‘intelligent’ systems that can respond to gaps in student knowledge and provide personalised feedback, virtual facilitators that could support students at university and during clinical training, and algorithms for administrative tasks such as tracking engagement and attendance. A professor at Georgia Institute of Technology deployed a virtual teaching assistant, based on the IBM Watson platform, for an online Masters programme to provide feedback and support to computer science students posting on online forums, which the graduate students seemed to value (Georgia Institute of Technology, 2016). Predictive learning analytics are also becoming popular where past student data is being mined using machine learning techniques to predict how current and future students may behave. These insights could guide university recruitment, course enrolment, and pedagogical strategies to customise learning, assessment, and feedback, as well as identify at-risk students, although Ekowo and Palmer (2016) advise these analytical tools should be used ethically so as not to disadvantage certain student groups.

However, Bayne (2015) argues that automated teaching tools such as virtual teacherbots seem to be driven by “productivity-oriented solutionism”, when pedagogy grounded in humanism should form the foundation of higher education. Popenici and Kerr (2017) also discusses AI more broadly in higher education, emphasising it is more likely these sophisticated computational approaches will extend the capabilities of educators and enhance the learning process and environment rather than replace teaching staff. Nevertheless, they advocate for transparency and oversight in how AI is applied across the university sector, to help reduce the risk of algorithmic bias. This can come from unrepresentative datasets and the corresponding algorithms, predictive modelling, and decision making derived from these, which could exacerbate structural inequalities in higher education. For example, the University of Texas Austin had been using a machine learning system for several years to assess applications to a computer science doctoral programme as it saved faculty administrative time. However, the system was stopped in 2020 because the algorithms were trained using a database of past admission decisions and some argued that historical inequity could introduce racial, gender, and other biases into the process (Burke, 2020).

It remains to be seen if AI will be embraced in nursing education, by faculty, students, university management, and administration. What is clear is that if these computational techniques are going to be applied, then great care needs to be taken to ensure they help us predict a better future in teaching and learning for both students and educators and do not become an unhealthy predilection concerned solely with saving money and time. No doubt AI will continue to evolve and advance, meaning there could be some merit in applying it wisely in nursing education.