Traditionally, academic medicine has been upheld by 3 core pillars—clinical care, research, and education—each dedicated to the advancement of its respective domain. Excellence in all 3 pillars—the traditional “triple threat” in academic medicine is increasingly challenging. There are the mounting demands of clinical care, the competitiveness of research funding, and the lack of time and support for continuous lifelong education. This challenge is exacerbated by growing separation of research from clinical operations. A Learning Health System (LHS) has emerged as a model particularly well suited for Critical Care Organizations (CCOs) to integrate clinical and academic missions in a paradigm of mutual reinforcement. An LHS is defined as a “health system in which internal data and experience are systematically integrated with external evidence, and that knowledge is put into practice.” A defining characteristic of an LHS is the continuous cycle of clinical investigations motivated by real-world data from both within and outside of the system and application of knowledge gained from those investigations into clinical practice to improve outcomes or care processes which, in turn, can generate new areas of investigations.

There are several challenges to building an LHS as detailed by Morain and colleagues. These include organizational culture, data systems and data sharing, funding learning activities, limited supply of skilled individuals, managing competing priorities, and regulatory challenges. However, CCOs are particularly well suited to overcome these challenges to adopt an LHS model by providing the foundation by which an LHS can be built upon. A CCO is a freestanding organizational structure in which critical care physicians operate and lead most intensive care units (ICUs) and critical care outreach efforts in a hospital system. This creates a culture of interprofessional collaboration “horizontally” among ICUs, critical care consult teams, and providers, and “vertically” throughout a hospital system. The operational goal of a CCO as an LHS is to provide high-quality, effective critical care and to synergize rather than compete with the academic research on care delivery, quality improvement (QI), and clinical trials. Such alignment of missions allows a CCO to invest funds and personnel support into research aligned with clinical and operational priorities. Furthermore, CCO integration with academics maximizes resource utilization and improves efficiency, providing a strong environment to perform research and disseminate knowledge.

The operationalization of an LHS into a CCO has 4 major interlocking components that integrate with each other. These components are observational real-world evidence, QI, implementation science, and clinical trials ( Fig. 1 ). This cohesive framework allows a CCO to generate focused practical informative research that directly advances patient care, clinical knowledge, and organizational performance ( Table 1 ). In the next sections, we will detail, with real-world examples, how CCO can operationalize these components of an LHS to support the clinical and research missions of academic medicine.

The cyclical process by which a Critical Care Organization becomes a Learning Health System for acute care in the hospital.

Access to data is an integral component of a CCO and serves as the foundation of an LHS. Extraction of data can be obtained from internal and external sources such as electronic health records (EHRs), registries, administrative systems, or claims data. ^, The acute care environment of the CCO is particularly well suited for an LHS given the highly granular longitudinal data of the in-hospital EHR, detailed data on interventions and medications including timing of administration, and outcomes that can be measured with short follow-up times (eg, hospital mortality), and little loss of data. Access to such data and the ability to analyze the data across multiple ICUs and hospital systems is integral to the horizontal integration of a CCO. Such data are also integral to observational research and QI initiatives.

In an LHS, routine clinical care should also generate questions that can be rigorously evaluated from this real-world data to help inform future practice. For example, during the coronavirus disease 2019 (COVID-19) pandemic, the limited number of ventilators raised questions about how to best triage mechanical ventilators to maximize lives saved during the pandemic. To better understand how to prepare for future crisis and disasters, our team used data acquired during the COVID-19 pandemic in simulation models to better understand which policy in high-flow nasal cannula use might result in less mortality under different scenarios of ventilator shortages and volumes of patients with respiratory failure. The use of such real-world data in simulated models can provide guidance for how to approach the next pandemic.

To be most effective in an LHS, CCOs need hospital-wide data from all hospitals covered by the CCO to capture the continuum of critical illness and recovery. For example, we used hospital-wide data from all patients to derive and validate a machine learning algorithm to predict which patients in the hospital may be at higher risk for developing catheter-related blood stream infection, an important quality measure for hospitals and CCOs that extends beyond the ICU setting. Similarly, the Accurate Prediction of Prolonged Ventilation (APPROVE) score was derived from hospital-wide data from 68,775 patients admitted to 4 hospitals to identify patients at high risk for respiratory failure requiring invasive mechanical ventilation or hospital mortality. The APPROVE score was then incorporated into the EHR of a fifth hospital in the CCO system and prospectively validated. Based upon the performance, this score is now incorporated into real-time predictive analytical to help identify patients outside of the ICU who may be at high risk for respiratory deterioration. Such machine learning algorithms for predictive analytics is particularly powerful in the longitudinal data-rich environment of acute care hospitals in a CCO.

Access to multicenter data also allows for participation in research consortiums and networks. Multicenter studies of natural practice variability amid gaps in knowledge can provide the equipoise for future comparative effectiveness trials. For example, the (Observation of Variation in Fluids and Characterization of Vasopressor Requirements in Shock [VOLUME-CHASERS]) study used multicenter observational data to investigate vasopressor exposure in septic shock demonstrating real-world variability in fluid and vasopressor use. The (Crystalloid Liberal or Vasopressors Early Resuscitation in Sepsis [CLOVERS]) trial from the (Prevention and Early Treatment in Acute Lung Injury [PETAL]) Network, recognizing the practice variation in management of shock in real-world practice, compared a fluid liberal versus early vasopressor use in the management of hypotension in septic shock and ultimately found no difference in mortality.

CCOs are often charged with new operational decisions to meet the needs of the institution. In an LHS, there is an opportunity to analyze the patient and institutional level impact of these interventions, which is not only helpful internally for the CCO to understand but it can inform other institutions on future decisions and expectations when faced with similar operational changes. For example, the opening of a medical intermediate care unit in one hospital in our health care system allowed us to analyze the impact of these beds in that hospital compared to another hospital during the same period. We found the addition of intermediate step-down beds can improve medical ICU throughput by reducing time to admission to the ICU (−26.7% reduction compared to baseline 95% CU −44.7% to −8.8%) and ICU length of stay for survivors (−27.5% reduction compared to baseline 95% CU −50.5% to −4.6%) without affecting mortality (adjusted OR 0.81 95% confidence interval [CI] 0.42–1.55).

Similarly, a separate initiative to add a midlevel provider to the rapid response team provided an opportunity to understand the impact on a patient and critical care delivery level. The addition of a midlevel provider to a rapid response team significantly reduced time to transfer to the ICU for critically ill patients (−19.2% change from baseline 95% CI −31.6% to −6.7% change) with a nonstatistically significant trend toward lower mortality in the hospital with the intervention (adjusted OR 0.76 95% CI 0.51–1.15). Interestingly, staffing with a midlevel provider at night or weekends did not provide more benefit than consistent weekday staffing. This information can help other institutions with their staffing priorities if they are considering a similar model.

Expertise in research methodology, statistical analysis, data science, and clinical informatics will be integral to a CCO evolving into an LHS. This does not necessarily require that all such expertise and resources are financed completely by the CCO. A CCO’s integrated structure allows for close collaboration with experts in other departments and institutions to facilitate an LHS.

For CCOs to meet their goal of delivering consistent, high-quality critical care, promotion of evidence-based practices is essential. Local data can provide near real-time analysis of local practice patterns and processes of care. When the data are benchmarked against those of an outside institution or against published standards, it can show gaps in care and opportunities for improvement. Moreover, CCOs must be able to use data to efficiently generate metrics for Plan-do-Study Act (PDSA) cycles for QI, a proven framework for testing interventions in a structure manner.

A CCO is well positioned to support each phase of the PDSA model—from identifying problems and implementing interventions to analyzing outcomes and refining protocols. Furthermore, QI is more powerful in a CCO when it is not episodic but is continuous. A CCO can engage all the stakeholders in the organization to put an intervention into practice across all the organization’s ICUs with constant monitoring of adherence, iterative refinement of best practices, identification of drift, and timely course corrections. This ultimately reduces variability in care and ensures that all patients benefit from the same standard of high-quality care.

For a CCO to take quality and process improvement further as an LHS, it should generate new knowledge about the process and impact from QI interventions. For example, while the (Awakening and Breathing trials, Choice of analgesia and sedation, Delirium monitoring and management, Early mobility [ABCDE]) bundle is recognized as an important quality initiative in critical care, the impact of the bundle on hospital operation and finance is important when advocating for buy-in and resources from the hospital. Taking advantage of standardized protocols across multiple ICUs in a CCO, we leveraged the natural experimental conditions presented by the implementation of early mobilization in one medical ICU and compared it to another similar ICU using a difference-in-difference analysis to understand the clinical, process outcomes, and financial savings that came from the staged implementation of ABCDE. This study demonstrated an improvement in outcomes such as reduction in use of restraints, decubitus ulcer, duration of mechanical ventilation, and length of stay that resulted in a net savings of US$1.9 million for the hospital. However, we also found that institutions may not see an improvement in the duration of mechanical ventilation and reduction in cost with just implementation of ABCD. Implementation of the complete bundle including early mobilization is needed.