AI in Healthcare: Do Not Forget About Allied Healthcare

Abstract

Artificial intelligence, the simulation of human intelligence by computers and machines, has found its way into healthcare, helping surgeons, doctors, radiologists, and many more. However, over 80% of healthcare professionals consists of people working in allied health professions such as nurses, physiotherapists, and midwives. Considering the aging of the general population around the world, the workforce shortages in these occupations are especially crucial. As the COVID-19 pandemic demonstrated, globally, most healthcare systems are strained, and there is a consensus that current healthcare systems are not sustainable with the increasing challenges. AI is often viewed as one of the potential solutions for not only reducing the strain on the healthcare workforce, but also to sustain the current workforce. Still, most AI applications are being developed for the medical community and often allied health is overlooked or not even considered despite comprising a large proportion of the total workforce. In addition, the interest of the private sector to invest specifically in the allied health workforce is low since the financial incentive is low. This paper provides examples of AI solutions for seven important allied health professions. To increase the uptake of AI solutions in allied healthcare, AI companies need to connect more with professional associations and be as patient-oriented as many claim to be. There also needs to be more AI schooling for allied healthcare professionals to increase adoption of these AI solutions.

1. Introduction

The issues and challenges of healthcare are not new and have been increasing significantly in the last decade. Aging populations are dependent on healthcare due to their increasingly more complex diseases. In combination with higher demands for administrative tasks and auditing, in the context of financial oversight, the strain on the healthcare system has become critical [1]. This became clear during the COVID-19 pandemic, which showed that, globally, most healthcare systems are not sustainable and that the socio-economically challenged populations are especially vulnerable [2], thus increasing the social and economic inequity within society. The increasing numbers of healthcare professionals that are burned-out and switching careers in an area that already had workforce shortages are a direct threat to society. A high administrative burden and complex social–economic burdens with increasing disease complexity have complicated the day-to-day practice. It has been recognized that a transition of the healthcare systems around the world is required to sustain care and keep it accessible for all. This transformation of healthcare will have several drivers, but digitalization and Artificial Intelligence (AI) are considered the most important means of accelerating the transformation [1].

AI is “the theory and development of computer systems able to perform tasks that normally require human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages” [3]. Examples of AI techniques are machine learning (ML), neural networks (NNs), deep learning (DL) and generative AI (GenAI). AI is being used in many different industries, and more recently, it has found its way into healthcare as well, guiding and assisting surgeons during surgeries [4], doctors in making better decisions [5], radiologists in locating tumors in magnetic resonance (MR) images [6], and much more. However, most AI solutions focus on high-profile professions such as surgeons, physicians, radiologists, etc., while over 80% [7] of healthcare professionals consists of people working in Allied Health Professions (AHPs) [8] such as nurses, physiotherapists, and midwives. See Figure 1 for a list of the most prominent healthcare professions, including those in allied healthcare. AHPs can also profit from the use of AI but they are often overlooked. In this opinion paper, we will give several examples of how AHPs can be supported by AI solutions, show why this is an important part of the acceptance of AI applications in day-to-day healthcare, and discuss some potential roadblocks and how they can be solved.

2. Applications of AI in Allied Healthcare

2.1. Nursing

Nurses could reduce a large portion of their workload by using AI to automate time-consuming tasks, giving them more time for interaction with the patient [9]. Some examples are

• Patient monitoring: Wearable sensors can be used to monitor vitals (e.g., heart rate, oxygen saturation (SpO₂), and glucose levels) and alert nurses when something is wrong [10]. This can reduce the response times in critical situations.
• Medication management: AI-enabled medication management systems can help nurses by removing repetitive tasks, enabling them to react to critical situations in a timely manner, and providing a comprehensive overview of the patients’ medication status [11].
• Workload optimization: AI can help to identify patient care needs and allocate nursing staff more efficiently [12].
• Wound analysis: Photographs of wounds can be analyzed in combination with data such as temperature, humidity, strain, and pressure to make better treatment decisions [13].
• Training and simulation: In the metaverse, virtual reality (VR) and augmented reality (AR) simulators enhance nursing education [14], providing realistic practice scenarios.
• Fall detection: AI algorithms can prevent falls in elderly people by analyzing risk factors, helping to reduce the number of times nurses need to intervene [15].
• Writing assistance: Generative AI (GenAI) can help nurses write messages to patients, saving time [16]. It can also translate text, create summaries, etc.
• Registration and classification: In the BETERZO project [17], GenAI systems were developed to optimize administrative tasks, classification, and decision support for nurses in the oncological setting. Speech-to-text technology is combined with Multi-Agent Generative AI systems and integrated into clinical nursing frameworks such as NANDA and NOC.

2.2. Physiotherapy

In addition to nurses, physiotherapists can benefit from AI as well, although the number of existing implementations is relatively low [18]. Some examples are

• Rehabilitation: AI-enabled motion sensors and apps can analyze a patient’s movements during exercises, providing real-time feedback to correct their posture [19].
• Tele-rehabilitation: AI platforms enable remote therapy sessions where patients receive guided exercises and automated assessments without needing frequent in-person visits [20].
• Robotics in therapy: Wearable robotic exoskeletons, in combination with sensors and AI, can assist in upper limb rehabilitation [21]. The Motor Learning and Neurorehabilitation Lab [22] develops innovative technology to improve the rehabilitation of neurological patients using robots, AI, and other technologies.
• Injury prediction and prevention: AI systems can analyze the patient’s biomechanics and gait to identify risk factors for injuries and suggest preventive measures [23].
• Disease and fatigue detection: In the ED-DETECT and ACT4FATIGUE [24] projects, machine learning was used for the early detection of rare diseases (Ehlers–Danlos Syndromes) and comorbidity (chronic fatigue) based on common clinical outcomes in youth with chronic diseases.

2.3. Midwifery

Like nurses and physiotherapists, midwives are allied health professionals. Midwives care for mothers and newborns around childbirth. Their profession can take advantage of AI in many ways:

• Pregnancy risk assessment: AI models can be used to predict risks such as preeclampsia [25] or gestational diabetes [26] based on patient data, allowing for earlier interventions.
• Prenatal ultrasonography: AI can assist in interpreting ultrasound images and videos, improving the accuracy of measurements like fetal growth and amniotic fluid levels [27]. It can also be used to detect disorders using images, such as congenital heart diseases [28].
• Prenatal education: AI can be used to create personalized educational content about pregnancy stages, nutrition, and preparations for childbirth [29].
• Prenatal and fetal monitoring: AI algorithms can be used to detect abnormalities in the fetal heart rate and uterine activity during labor, helping midwives to intervene promptly when necessary [30]. In the PregnaDigit project [31], women with high-risk pregnancies were monitored at home (fetal heart rate, blood pressure, and blood sugar levels). AI early warning systems were developed and deployed to reduce hospital stays and reduce complication risks.
• Postnatal education: Chatbots and virtual assistants can guide new mothers on breastfeeding, infant care, and postnatal recovery [32].

2.4. Radiography

Radiographers are another group of allied health professionals. They are technicians trained to take medical images and assist radiologists. Radiographers can benefit from AI algorithms in several ways:

• Image analysis: AI tools can support radiographers by detecting abnormalities in medical images such as X-rays, CT scans, or MRIs, improving diagnostic accuracy [33].
• Patient positioning: AI can automatically ensure optimal patient positioning within a CT or MRI scanner [34].
• Workflow automation: AI can help to automate the scheduling of imaging studies (e.g., X-ray [33] and CT [35]), optimize radiographers’ workloads [36], and consequently minimize patient wait times.
• Radiation dose optimization: AI can reduce patient radiation doses while retaining diagnostic-quality imaging [37].
• Reporting: GenAI can convert handwritten radiography reports into structured reports [38].

2.5. Occupational Therapy

Occupational therapists, another allied health profession, “use everyday life activities (occupations) to promote health, well-being, and your ability to participate in the important activities in your life” [39]. Occupational therapists (OTs) and occupational therapy assistants (OTAs) can help with training, lifestyle changes, and rehabilitation. Some examples of AI-assisted occupational therapy are

• Customized assistive devices: AI-based prosthetic limbs can be tailored to individual needs, improving functionality and comfort [40].
• Cognitive rehabilitation: AI-powered games and apps can help patients regain memory or attention skills after brain injuries or strokes [41].
• Activity analysis: Sensors and AI can be used to track daily activities to identify challenges that patients face in their environment, guiding therapy adjustments; for example, this could be used with cancer patients [42].

2.6. Dietetics and Nutrition

Dietitians are allied health professionals who try to manage the nutrition of (elderly) patients and prevent malnutrition. They can benefit from AI in several ways as well:

• Personalized meal plans: AI can help to evaluate dietary preferences, medical conditions, and allergies to create customized nutrition plans [43].
• Monitoring nutritional intake: AI can analyze photos of food to determine the volume, nutrients, and calories [44].
• Predictive health outcomes: Machine learning can be used to assess the impact of dietary habits on long-term health, enabling preventive care [45].

2.7. Speech Therapy

Finally, we will discuss the allied health profession of speech therapist. Speech therapy is defined as the “assessment and treatment of communication problems and speech disorders, performed by speech-language pathologists (SLPs)” [46]. The following are some examples of AI applications in speech therapy:

• Speech problem detection: Automatic speech recognition systems can improve the early detection of speech disorders (e.g., stuttering, cluttering, and articulation disorders) and interventions [47].
• Speech therapy applications: AI could be used to create automated speech therapy tools for individuals with speech disorders [48], making speech therapy more accessible and affordable. Voiceitt [49] and Project Euphonia [50] have created AI-based applications to help people with speech disorders communicate more easily.

3. Why Allied Healthcare Needs More AI Applications

The above provided many examples of how AI can help people working in AHPs such as nurses, physiotherapists, midwives, radiographers, occupational therapists, and dietitians. This list is not complete and can even be expanded with AHPs working in mental health [51] or dental hygiene [52]. However, most of the attention (and funding) has been on AI solutions that help health professionals such as physicians, surgeons, and radiologists. They often have a higher position in the hospital, have more exposure to the outside world, and have better visibility for companies selling their AI-enabled products. Most articles (both scientific publications and blog posts) discuss how AI can help these medical professions but forget about AHPs. However, with the current aging population and healthcare staff shortage, AHPs also need AI solutions. As stated before, over 80% of the healthcare professionals consists of people working in AHPs. Since these people also perform some administrative and routine tasks, it is relatively easy to make progress with AI. The current growth of generative AI applications can be especially useful in, e.g., speech-to-text analysis of communication with the patient and creating summaries of conversations. In particular, reducing administrative burdens may have a significant effect on the longevity of the healthcare system, as increasing time spend on EHR documentation was found to be associated with burnout [53]. Well-documented and structured EHRs were also found to be associated with better patient outcomes. Thus, the use of AI could reduce the time spent on administrative tasks and enable the people in AHPs to spend more time interacting with the patient. It will also reduce costs, as process and workflows can be greatly simplified with AI solutions. Finally, AI could help to prevent wrong decisions from being made, because the decisions would be based on large historical datasets instead of relying on the experience of the people working in AHPs.

4. Potential Issues (and Solutions) for the Implementation of AI Applications in Allied Healthcare

One of the main issues around the implementation of AI in allied healthcare is bias. Bias can come in many forms. Preexisting bias has its roots in social institutions, practices, and attitudes. Technical bias arises from technical constraints or considerations. Emergent bias arises in the context of use [54]. In healthcare, one of the most common biases is selection bias, where the research participants may differ from the population of interest. Or there might be a gender bias, e.g., datasets with mostly male patients are used to make decisions about female patients. The Catalogue of Biases [55] contains an extensive list of biases. Bias can cause ethical issues, such as reduced quality of care for the patients underrepresented in medical datasets. Therefore, it is important that not only legislations such as the GDPR and the AI Act are being followed, but also that Responsible AI is taken into account. Responsible AI tries to ensure that AI is sustainable, human-centered, inclusive, fair, and transparent [56], which are of crucial importance in healthcare.

Another important issue is privacy and security. Healthcare data are generally very sensitive as they contain, for example, personal data and genetic information. This means that the privacy of the patients needs to be taken care of, and that data are saved and transferred in a secure manner. A possible solution for privacy issues is the use of privacy enhancing techniques (PETs) such as pseudonymization, anonymization, federated learning, multiparty computation, and the use of synthetic data. Pseudonymization is the replacement of personal data with a code, e.g., replacing a name with a randomized patient ID Anonymization goes a step further and tries to change the data in such a way that re-identification is (almost) not possible. Federated learning is a machine learning technique that trains an algorithm across multiple decentralized sources, without having to exchange the data. An example of a federated learning system using medical data is the Personal Health Train (PHT) [57]. Multiparty computation consists of multiple cryptographic techniques, enabling multiple parties to jointly compute data without being able to view each other’s data [58]. Finally, the use of synthetic data is a relatively new method through which new data can be synthesized based on existing data using GenAI techniques [59]. The security of healthcare data can be ensured by implementing safety measures such as data encryption and two-factor authentication.

The future implementation and adoption of AI applications in allied healthcare depends not only on technical aspects, but also on the willingness of people in AHPs to accept AI. There has been some anxiety in the nursing community about “AI taking over their jobs” [60], similar to many other professions. To increase the acceptability of AI in nursing, we need to make it clear to the nurses that they are not going to be replaced by AI. Just like with doctors, nurses will not be replaced by AI, but traditional nurses will be replaced by nurses who know how to work with AI. They can improve their workflows using AI and spend more time communicating with and caring for their patients. However, at this moment, AI implementation seems to be associated with both positive and negative effects on nursing (and other AHPs) workloads [61], as they can streamline their administrative tasks using AI but are also confronted with new challenges and demands, such as a need to understand what exactly the AI algorithms are doing. This brings us to another important aspect: schooling. People need to know how to use AI and understand, on at least a basic level, how it works. In the EU, high-risk AI systems (which includes many systems in the medical field) need to have human oversight. This means that users need to know how to safely use and monitor the system, recognize potential malfunctions, biases, or safety risks, interpret AI outputs correctly, and understand the system’s limitations [62]. They also need to be trained to intervene effectively if the system outputs incorrect or harmful results. Suitable course materials can be found online (such as Stanford’s “AI in Healthcare Specialization” [63] and MIT xPRO’s “Artificial Intelligence in Healthcare: Fundamentals and Applications” [64]), but AHPs should also follow courses specifically tailored to the specific AI systems that they use in their daily work.

The issue of companies not willing to spend money on AI solutions for allied healthcare could be solved by professional associations, such as the American Nurses Association (ANA) [65], the American Physical Therapy Association (APTA) [66], the American College of Nurse-Midwives (ACNM) [67], and the World Confederation for Physical Therapy (WCPT) [68]. They represent a large number of allied healthcare professionals and could help AI companies reach a large customer base. They could be included both in the research phase of AI applications and in the large-scale adoption of these applications. In addition to these professional associations, the World Health Organization (WHO) [69] could play a role as a global organization that supports healthcare. Many companies have a patient-oriented goal: they want to make the lives of patients better. Patients have more interactions with AHPs such as nurses than with doctors, so helping AHPs also benefits the patients in a much more direct way. Companies might be stimulated to work on allied healthcare projects if it is supported by funding agencies through public–private partnerships. Finally, legislation reforms could help to take away some obstacles that companies are facing when they want to invest time and money into allied healthcare projects.

5. Discussion

After taking into account these possible roadblocks for the adoption of AI in allied healthcare and removing them using the proposed countermeasures, we can work towards improved integration of AI into allied healthcare to benefit not only the allied healthcare professionals, but also their patients. GenAI holds great promise as it advances rapidly, and in the near future (within 5 years), it could create new opportunities that cannot be imagined at this moment in time. GenAI is being increasingly used to speed up and improve the work of AHPs, such as creating, summarizing, and translating text reports [70], but it can be used for much more than that. AI agents are the next big thing for the near future: autonomous artificial AI systems performing specific tasks without human intervention, which have great potential in medicine and healthcare [71]. These AI agents can be used to automate specific tasks or even workflows. In the further future (in more than 5 years from now), we might reach the stage of Artificial General Intelligence (AGI), where AI matches a human’s wide range of intelligence and has more autonomy and adaptability, creating even more methods for improving allied healthcare [72]. AGI could fully replace human professional, which could solve the current personnel shortage caused by the aging population and increasing healthcare needs.