**5. Technology**

Unfortunately, even in modern medicine, with some of the most advanced medical equipment in the world, it is not an easy task to be able to remove noise, recognize a deteriorating patient in early stages, initiate timely resuscitation maneuvers, correctly identify a differential diagnosis, and orchestrate highly complex multidisciplinary teams to focus on critical short term goals while maintaining a strategic plan to allow the "person-in-the-bed" to recover to a high functioning level, as close as possible to the pre-morbid state and prevent a possible PICS.

Because the ICU is only one segment in the patient's journey, we must be comfortable stepping outside the physical boundaries and create operating conditions that include electronic outreach and monitoring in a control tower fashion, proactive ICU nursing rounding, and rapid response, preventing complications of medical care, enhanced rehabilitation pathways, avoiding transition of care gaps, preparing survivor clinics, readmission prevention with advance care at home, and others yet to be invented.

#### **5.1 Smart alerts**

Changing the ICU framework towards early identification and protocolized management of critically ill patients is perhaps one of the most significant contributions developed in recent years [100]. ICU and Quality Improvement teams participate in local, national, and international research teams to develop processes and decision support tools deployed at the point of care in critically ill patients [101, 102]. Evidence supports that smart alerts improve the care provided in the ICU, some examples include: (a) adherence to basic critical care processes such as the sepsis bundle [103], (b) ventilator bundle with sedation holiday compliance [100, 101], and (c) enhanced risk stratification utilizing severity of disease score calculators [104].

When smart alerts are deliberately deployed using implementation science tools, they can lead to better outcomes such as decreased in ICU, hospital, and 28 days mortality rates from sepsis [103] and alleviated cognitive workload with more accessible navigation through the EMR [105]. Smart alerts also have the potential to add value in intensive care units by lowering cost, decreasing ICU and Hospital LOS, improving disability-adjusted life years, quality-adjusted life years, and incremental cost-effectiveness ratio [106].

To juxtapose the attributes smart alerts have, they can potentiate negative consequences if the alerts are missed or delayed due to technical difficulties, which might translate into delays in care. Some examples include technical failures, firewall blocking, and security issues with smartphone/tablet alert delivery which may present a barrier to optimal alert delivery in the ICU setting [100].

End-user fatigue contributes to disengagement with the alerting mechanisms; it increases human error, information overload, as well as alerts with higher false positive (noise) rates. This can derive from user preferences for specific alert delivery methods, which can affect compliance [107].

#### **5.2 Decision support systems and artificial intelligence**

Artificial Intelligence (AI) tools intended for practice enhancement will have to address these challenges. Based on insights gained from studies of human error [60], understanding of information needs [108], and both cognitive and organizational

ergonomics [109], we need to establish an infrastructure to develop and test advanced analytics centered on clinician's needs and ICU decision support tools applicable both at the bedside and across the hospital of the future. AI applications in critical care are in the early stages of development. Some of the initially published studies showcase applications to predict LOS, ICU readmission, mortality rates, and early identification of complications and risk stratification [110]. The promise of AI enhancing safety in the ICU depends on its successful application to common ICU problems such as pointof-care ultrasound, volume assessment, medication titration, ventilator management, and other smart devices [111]. An essential element to consider when evaluating the role of AI is that scientific evidence still needs to catch up with machine learning algorithms. For example, the FDA has approved 130 AI devices, of which 126 were based only on retrospective research [112]. Also, we must ensure that measures are taken to avoid algorithmic biases such as race, gender, and other social determinants of health. As such, algorithm developers should transparently report the steps taken for the algorithm to be equitable and representative of the entire population or at least report the specific input data sources, population composition, bias assessment validation, and training data location and period [112].

The role of AI-based technologies in critical care is expanding exponentially. Much of the work utilizing AI in intensive care has been around predictive analytics for early identification of deteriorating/at risk patients and machine learning models to predict patient's clinical trajectory [111]. Solutions such as using predictive analytics to provide early insights on whether a patient will develop delirium [113], pressure injuries [114] from prolonged ICU stay, sepsis [115], or have unattended bed exits triggering falls, have the potential to become the gold standard for the healthcare industry. They augment the decision support process, enhance reliability, and accelerate much-needed agility in critical care. These tools are crucial to optimizing the care our patients receive and achieving the goal of zero harm. Halamka and Cerrato describe in *The Digital Reconstruction of Healthcare* that the future belongs to advanced data analytics. Supporting human skills with Big Data and AI-based algorithms can be equated to "giving the best artists the best analytic tools" [116]. It will elevate the potential for decreased ICU stay, and fewer hospital-acquired infections, among many other healthcare-associated conditions. This vision advances the frontiers of patient safety into an ultra-safe space. Another promising feature of AI-based solutions is solidifying healthcare in the high-reliability space where there is less dependence on human behavior and more reliance on systems and structures. Minimizing patient harm while delivering care includes high-reliability practices which lead to a "more integrated picture of operations at the moment and earlier detection of potential threats to safety" [117]. Empirical evidence suggests that advanced predictive analytics must be part of the solution to improve the patient's safety journey in critical care environments.

#### **5.3 Electronic medical record (EMR)**

The transition from handwritten records to EMR has improved efficiency, revenue capture, and billing. Evidence supports using EMRs to enhance patient safety and improve outcomes [118]. Reports exist promoting that EMR saves time during documentation [119]. If critical elements of EMR design and implementation are overlooked, then the positive or negative impact on patient safety and quality could potentially be amplified [108]. To appropriately design and implement an EMR in the ICU, crucial elements must be considered, including information overload, clinical setting, rule development, controlled environment testing, field testing, and

implementation science. As an example, Pickering et al. have demonstrated that after following these steps for introducing a novel Ambient Warning Response and Evaluation (AWARE), EMR was associated with improved efficiency of data access and decreased cognitive overload due to improved presentation format [120].

#### **5.4 Dashboards and feedback**

The amount of information and metrics being tracked by ICU providers and leaders can be overwhelming. It is paramount that critical priorities are distilled at the unit, team, and provider levels. Priorities need to be divided into metrics to induce improvements that can be tracked in dashboards and transparently published and accessed. To sustain the gains, providers should be given factual and non-judgmental feedback as close to real-time as possible. Utilizing this methodology with the sepsis care bundle compliance, for example, some ICUs have been able to improve compliance, [121] sustain gains, and translate them into improved patient outcomes [103, 122].
