One of the many acute challenges during the pandemic has been the departure of approximately 18% of the U.S. healthcare workforce,4 resulting in staffing shortages in hospitals. This has impacted the well-being of the remaining healthcare practitioners 5 as well as their patients – many of whom are deferring care that is vital to disease management and overall maintenance of health,6,7 or who do not have access to timely emergency care.1 Though such challenges may be perceived as insurmountable in the face of overcrowded emergency rooms, burnout of healthcare personnel, and patients who are reluctant to seek care, the incorporation of artificial intelligence (AI) and machine learning (ML) into healthcare systems may alleviate some of these burdens.
AI and ML have been widely used in consumer applications across industries, yet their potential value in healthcare has been largely unrealized. As health systems continue to struggle in the grip of the pandemic, the time has come to push patient risk assessment into the future with novel tools, such as machine learning. By incorporating AI and ML into electronic healthcare record (EHR) systems, hospitals with limited resources – whether it be due to geographic isolation or staffing shortages –may improve patient care and outcomes.
Many prognostic calculators and risk stratification tools require the use of information from patient assessment and/or clinical gestalt, which is, as of late, often being done in clinical settings that lack adequate staffing. Traditional risk scoring systems are also often cumbersome and impede clinical workflow by requiring patient data to be manually input by the examining physician.8–10 In contrast, AI and ML risk stratification tools have the ability to harness the wealth of information contained within EHR data to provide automated risk projections using minimal amounts of patient data. These may be used to supplement clinical assessments and disrupt many of the antiquated practices in healthcare.
Predictions made by these tools use individual health data and are reflections of an individual’s dynamic health processes. They characterize trends and stratify a patient’s risk of experiencing a particular outcome based on the individual. Despite this, AI and ML remain underutilized in healthcare. This may be due to the relative novelty of ML in medicine, a general lack of understanding about how ML works and why certain predictions are made –often called black box predictions – and perceived challenges regarding the implementation of such tools. Realistically, ML tools can be implemented quickly and with minimal training to the end-user and serve as clinical decision support (CDS) tools to optimize limited clinician time.
In the context of the ongoing global health emergency, AI and ML-based CDS tools have the potential to emerge as an invaluable resource to overburdened care teams by facilitating prioritization of medical care for patients based on individual risk levels. In hospitals that are overcrowded or have inadequate staffing, ML-based risk assessments identify patients who are in more immediate need of care and guide the allocation of personnel and health-related resources. They can also aid with determining hospital admission or identify the need to transfer a patient to a higher level of care. Triage, in this manner, could ensure that only patients in immediate or near-immediate need of treatment are under the direct care of a healthcare provider (HCP). With ongoing staffing shortages and hospitals pushed to the brink of capacity, this could enable HCPs to provide more precise care to fewer, sicker patients, as opposed to providing less precise care to a greater number of non-urgent patients.
Aside from staffing shortages among the general U.S. healthcare workforce, the challenges faced by rural hospitals are multi-pronged. In contrast to larger, well-funded hospital facilities, rural hospitals are experiencing a disproportionate dearth in personnel during the COVID-19 pandemic, particularly nurses.11 12 Rural hospitals also frequently have limited financial resources and lack immediate access to acute-care specialists, unlike larger hospital settings where these specialists are often on staff. Health technology is a mechanism for improving health equity among these individuals who are often marginalized in terms of health access, as shown with telestroke13 initiatives. Telestroke is a remote system of providing expert patient assessment and managing patient care with regards to stroke. It has been incorporated into a number of rural healthcare systems and contributed to improved treatment time and outcomes for patients hospitalized with a stroke. 14 Despite these improvements, broad uptake is hindered by the high costs of telehealth interventions and difficulty retaining remote, specialized teams, which may not be reimbursable.15
With autonomous projections based on individual data, ML and AI are critical to unlocking equitable, personalized medicine that can impact individuals in settings where access to advanced healthcare is provided on an as-needed basis. By identifying which patients are in the greatest need of acute care, healthcare providers will be better enabled to focus limited tangible and personnel resources on high-risk patients or request the services of a specialist who may not be immediately available on-site in a timely manner. Enabling healthcare providers to focus their care on high-risk patients results in resources that can be rationed and allocated for patients and procedures that require immediate attention.
It is inarguable that we have witnessed the multifaceted and detrimental impacts that COVID-19 has had on healthcare delivery and patient outcomes in the U.S. and beyond. However, it may have also inadvertently shepherded us into the future, where highly accessible, personalized medical tools are at the fingertips of healthcare providers to guide and empower clinical decision-making.
About Andrew Pucher
Andrew serves as the Chief Executive Officer and a member of the Board of Directors at Dascena, Inc. Andrew is a strategic healthcare leader, with extensive experience in developing new product markets and scaling commercial businesses. Prior to joining Dascena, Andrew served as Chief Corporate Development Officer at Tilray until its $3bn acquisition. Previously, Andrew was a Managing Director at Goldman Sachs, where he worked for over a decade as a healthcare investment banker financing and advising leading biotech, pharma, and medical device companies.
References
1. Emergency Department Crowding: The Canary in the Health Care System | Catalyst non-issue content, https://catalyst.nejm.org/doi/full/10.1056/CAT.21.0217 (accessed 10 January 2022).
2. Committee on Guidance for Establishing Crisis Standards of Care for Use in Disaster Situations, Institute of Medicine. Crisis Standards of Care: A Systems Framework for Catastrophic Disaster Response. Washington (DC): National Academies Press (US), http://www.ncbi.nlm.nih.gov/books/NBK201063/ (2012, accessed 11 January 2022).
3. Hick JL, Hanfling D, Wynia MK, et al. Crisis Standards of Care and COVID-19: What Did We Learn? How Do We Ensure Equity? What Should We Do? NAM Perspect; 2021: 10.31478/202108e.
4. About 1 in 5 healthcare workers have left medicine since the pandemic began — Here’s why, https://www.beckershospitalreview.com/workforce/about-1-in-5-healthcare-workers-have-left-medicine-since-the-pandemic-began-here-s-why.html (accessed 10 January 2022).
5. Burnout in Hospital-Based Healthcare Workers during COVID-19. Ontario COVID-19 Science Advisory Table. DOI: 10.47326/ocsat.2021.02.46.1.0.
6. Delays and Disruptions in Cancer Health Care Due to COVID-19 Pandemic: Systematic Review | JCO Global Oncology, https://ascopubs.org/doi/10.1200/GO.20.00639 (accessed 10 January 2022).
7. Splinter MJ, Velek P, Ikram MK, et al. Prevalence and determinants of healthcare avoidance during the COVID-19 pandemic: A population-based cross-sectional study. PLOS Med 2021; 18: e1003854.
8. Kansal A, Green CL, Peterson ED, et al. Electronic Health Record Integration of Predictive Analytics to Select High-Risk Stable Patients With Non–ST-Segment–Elevation Myocardial Infarction for Intensive Care Unit Admission. Circ Cardiovasc Qual Outcomes 2021; 14: e007602.
9. Ebinger J, Henry T, Kim S, et al. Development and Evaluation of Novel Electronic Medical Record Tools For Avoiding Bleeding After Percutaneous Coronary Intervention. J Am Heart Assoc 2019; 8: e013954.
10. Sahay S, Tonelli AR, Selej M, et al. Risk assessment in patients with functional class II pulmonary arterial hypertension: Comparison of physician gestalt with ESC/ERS and the REVEAL 2.0 risk score. PloS One 2020; 15: e0241504.
11. Rural Hospitals Can’t Find the Nurses They Need to Fight COVID, https://pew.org/3Dw86s9 (accessed 11 January 2022).
12. Grimm CA. Hospitals Reported That the COVID-19 Pandemic Has Significantly Strained Health Care Delivery. 62.
13. Demaerschalk BM, Berg J, Chong BW, et al. American Telemedicine Association: Telestroke Guidelines. Telemed J E Health 2017; 23: 376–389.
14. Lazarus G, Permana AP, Nugroho SW, et al. Telestroke strategies to enhance acute stroke management in rural settings: A systematic review and meta‐analysis. Brain Behav 2020; 10: e01787.15. Zachrison KS, Richard JV, Mehrotra A. Paying for Telemedicine in Smaller Rural Hospitals: Extending the Technology to Those Who Benefit Most. JAMA Health Forum 2021; 2: e211570.
