*Unlocking the Potential of Artificial Intelligence (AI) for Healthcare DOI: http://dx.doi.org/10.5772/intechopen.111489*

care unit data to develop DeepSOFA, a real-time mortality risk prediction score based on the traditional SOFA score [43]. This model compared the traditional SOFA scores to deep learning technology in augmenting a clinician's decision-making by generating accurate real-time prognostic data relating to mortality [43]. The DeepSOFA model was more accurate than baseline SOFA models for predicting inhospital mortality among ICU patients with baseline SOFA models significantly underestimating the probability of death, especially among non-survivors [43]. Recognition of mortality risk earlier in the disease course has the potential of aiding clinicians in taking preventative measures earlier and with more accuracy resulting in improved outcomes.

The COVID-19 pandemic demonstrated the utility of AI and ML for prehospital and posthospital management of patients. For instance, remote patient monitoring (RPM) came to the forefront during the pandemic as hospital systems became overwhelmed with patients. RPM is a healthcare technology that uses digital devices, wearable sensors, and wireless communication to collect and transmit medical data from patients outside of traditional clinical settings. Traditionally, RPM has been utilized to monitor chronic diseases; however, the pandemic accelerated the use of this technology for acute monitoring and management of patients with COVID-19 infections. RPM is achieved through use of smart devices such as blood pressure meters, thermometers, glucometers, and pulse oximeters utilizing an ecosystem known as the internet of health things (IoHT). IoHT refers to the interconnectivity of medical devices, wearables, and healthcare systems that allow for the exchange of health-related data between patients and healthcare providers.

ML techniques applied to enormous data sets generated through continuous monitoring of cardiac- and respiratory-related signals, coughing, body temperature, and patterns of activity collected from COVID-19 patients help predict the health status of a patient or individual easily [44]. Consequently, based on these measurements, the appropriate medication can be administered, or people can be transferred to the hospital when necessary. Crotty et al. utilized RPM capabilities to monitor 5367 patients with COVID-19 infection and found a substantial reduction in ICU utilization, reduced length of stay, and lower 30- and 90-day mortality when compared to patients who did not participate in active monitoring [45]. RPM has the potential to improve patient engagement and health literacy by providing real-time information that can improve outcomes, such as pruning education, which likely led to improvement in oxygenation requirements and improved outcomes [45].

AI and ML are also improving cardiovascular health through predictive analytics. Predictive analytics is the use of data, statistical algorithms, and machine learning techniques to identify the likelihood of future outcomes based on historical data. In the case of cardiovascular health, predictive analytics can be used to identify patients who are at high risk of cardiovascular disease [46, 47]. By using data from EHR, wearable devices, and other sources, healthcare professionals can identify patterns and trends such as low heart rate variability (HRV) that may indicate a higher risk of subsequent cardiovascular events (14). AI and ML algorithms can be also used to analyze HRV signals to track and evaluate the effectiveness of therapeutic interventions such as HRV biofeedback. Burlacu et al. outlined a systematic review on the beneficial effects of HRV-biofeedback, a slow breathing technique, on different cardiovascular diseases such as arterial hypertension, heart failure, and coronary artery disease. HRV modulation can be implemented in high-risk patients to significantly reduce stress levels and improve autonomic nervous system function and cardiovascular endpoints [48].

### **5.1 Challenges of AI** *in vital* **sign monitoring**

With the increasing utilization of smart medical devices, there is a growing risk of sensitive medical data being accessed [49] or stolen. To address this concern, it is necessary to implement robust security measures such as encryption and access controls to ensure that personal information is properly protected. Rajasekaran proposed that the IoHT must include several key features such as trust ability, low transmission latency, security, confidentiality, integrity, and availability [50]. They proposed a blockchain-based anonymous privacy-preserving authentication scheme to preserve the key features outlined above.

Additionally, the lack of interoperability of the different wearable devices is one of the biggest hurdles that we need to overcome. By enabling different devices to share data, interoperability opens the door to new possibilities for personalized healthcare. For example, a wearable health device that tracks a user's physical activity can share data with another device that monitors their heart rate. This data can then be combined to create a more complete picture of the user's health, helping healthcare providers make more informed decisions about treatment and care. Interoperability will also help generate high-quality data that can be used to train ML algorithms.

For machine learning algorithms to work effectively, they require a large amount of high-quality data to train on. ML algorithms can be biased if the training data contains systematic inaccuracies or overrepresents one group. Straw et al. demonstrated one such bias when they reviewed Indian Liver Patient Dataset (ILPD), which is the open source data set used extensively to create algorithms that predict liver disease. Due to the under representation of females in the data set, the model demonstrated a higher false negative rate in women leading to lower disease detection in females [51]. To minimize the risk of bias, it is important to carefully select training data by using diverse and representative data sets. Additionally, the development and deployment of ML models should be guided by ethical and inclusive principles. Mccradden et al. outlined ethical principles of nonmaleficence, relevance, accountability, transparency, and justice as the foundation for the regulation of healthcare ML algorithms [52].

In inference, AI and ML have the potential to revolutionize the way healthcare is delivered. With the ability to collect vast amounts of patient data in real-time, AI algorithms can provide valuable insights into patients' health, improve the accuracy of diagnosis, detect health issues early, and improve patient engagement and health literacy. While there are still challenges to overcome such as security, interoperability of wearable devices, and ML bias, AI and ML have the potential to significantly improve patient outcomes and transform the way healthcare is delivered. Clinicians and policymakers, however, must ensure that the technology is accessible and affordable for all patients, regardless of their socioeconomic status. While there have been significant advancements in RPM technology in recent years, many patients, particularly those living in rural or underserved areas, may not have access to these tools due to cost or limited availability.
