Key points

What is health-care quality?

In To Err Is Human , the Institute of Medicine (IOM) defined health-care quality as “the degree to which health services increase the likelihood of desired outcomes” and patient safety (PS) as “freedom from accidental injury because of medical care or medical errors.” These two concepts are fundamentally linked, and many notable authorities have explicitly cited safety as the key dimension of quality, including the IOM, the Leapfrog Group, the Institute for Healthcare Improvement (IHI), the National Quality Forum, and even Hippocrates: primum non nocere. In this article, quality improvement (QI) is used as a broad term that encompasses not only enhanced positive performance but also mitigation or elimination of errors, health care defects, and patient harm.

The IOM’s Crossing the Quality Chasm outlined 6 dimensions of health-care quality. The most fundamental attribute is that care should be safe. Care must also be effective and appropriately dispensed, avoiding underuse and overuse by being provided to all who could benefit and not to those unlikely to benefit. Care should be patient centered; respectful of and responsive to individual patient preferences, needs, and values. Quality care is timely; the right care to the right person at the right time, with waits and delays eliminated or minimized. Care should be efficient, actively seeking to identify and eliminate all forms of waste, be it time, equipment, supplies, or energy. The final dimension of care outlined by the IOM is equitability: the provision of care that does not vary in quality because of personal characteristics such as sex, ethnicity, geographic location, and socioeconomic status. In “Revisiting the Quality Chasm”, Brilli and colleagues reorganized these same IOM dimensions in a patient-centric framework as follows: do not harm me, cure me, treat me with respect, navigate my care, and keep us well. Strategies used to apply evidence to achieve these missions are largely the domain of QI. QI research is the rigorous analysis of what makes QI efforts effective, sustainable, worth the investment, or vice versa.

What is health-care quality?

In To Err Is Human , the Institute of Medicine (IOM) defined health-care quality as “the degree to which health services increase the likelihood of desired outcomes” and patient safety (PS) as “freedom from accidental injury because of medical care or medical errors.” These two concepts are fundamentally linked, and many notable authorities have explicitly cited safety as the key dimension of quality, including the IOM, the Leapfrog Group, the Institute for Healthcare Improvement (IHI), the National Quality Forum, and even Hippocrates: primum non nocere. In this article, quality improvement (QI) is used as a broad term that encompasses not only enhanced positive performance but also mitigation or elimination of errors, health care defects, and patient harm.

The IOM’s Crossing the Quality Chasm outlined 6 dimensions of health-care quality. The most fundamental attribute is that care should be safe. Care must also be effective and appropriately dispensed, avoiding underuse and overuse by being provided to all who could benefit and not to those unlikely to benefit. Care should be patient centered; respectful of and responsive to individual patient preferences, needs, and values. Quality care is timely; the right care to the right person at the right time, with waits and delays eliminated or minimized. Care should be efficient, actively seeking to identify and eliminate all forms of waste, be it time, equipment, supplies, or energy. The final dimension of care outlined by the IOM is equitability: the provision of care that does not vary in quality because of personal characteristics such as sex, ethnicity, geographic location, and socioeconomic status. In “Revisiting the Quality Chasm”, Brilli and colleagues reorganized these same IOM dimensions in a patient-centric framework as follows: do not harm me, cure me, treat me with respect, navigate my care, and keep us well. Strategies used to apply evidence to achieve these missions are largely the domain of QI. QI research is the rigorous analysis of what makes QI efforts effective, sustainable, worth the investment, or vice versa.

Quality improvement, research, and quality improvement research

It is common for classic research and QI interventions to be confused, because they both are seeking to improve patient outcomes in data-driven ways, although typically through different mechanisms. Classic research in early translational stages often defines the best known practices that QI agents strive to implement in later translational work. However, improvements in the performance reliability of baseline clinical practices can reduce noise, improving the sensitivity of analyses designed to detect small incremental improvements in best practice. If children are not afflicted by preventable nosocomial infections, medication errors, or surgical complications, then sensitivity is improved when trying to detect the impact of a novel research intervention on patient outcomes.

This blurred boundary between classic research and QI is well recognized, and derives from the intent of both to result in better outcomes for patients, albeit with a different perspective, through different methodologies, and tolerating different thresholds for risk. This is particularly true for T3 translational research regarding either the operationalization of established best practice or adoption of general principles such as standardization to uniform common practice. Because many peer-review journals require formal institutional review board comment on approval or exemption, many QI/PS projects focused on organizational and operational performance simply go unreported. However, scholarly work in QI/PS has been published with year-on-year increases since the IOM’s 1999 publication of To Err Is Human . Because of the increasing scholarly output in this area of medicine, both in quantity and quality, the expectations for the quality of QI research are also steadily mounting. Calls for scholarly accounts of quality and safety endeavors, along with publication guidelines for proper peer review, have appeared in recent years. Notably, the strengths and weaknesses of the Standards for Quality Improvement Reporting Excellence (SQUIRE) guidelines of 2008 have undergone critical review with updated 2015 guidelines to provide better planning and publication guidance for QI researchers. Many of the methodologies to analyze quality and safety in health care with academic rigor are still in development, young, and under adaptation from other industries and disciplines, and many are neither familiar to practicing clinicians nor embedded in medical education. Such analytical tools from human-factors engineering, psychology, industrial engineering, and manufacturing are increasingly being accepted in the traditional peer-reviewed medical literature.

Perhaps one of the simplest ways to think of QI is as a strategy to close the gap between actual practice and best known practice, be it clinical or operational. The estimated lag time for scientific knowledge generated in randomized clinical trials to be routinely accepted into medical practice is 17 years; a shocking testimony to the size and persistence of the gap between the evidence and actual care, and a provocative invitation to close it. Fig. 1 shows how QI fits in the context of actual, best, and idealized performance. If the graph is taken as a survival or time-to-adverse-event curve from some identified measure of quality or safety, it can be assumed that the ideal outcome is 100% perfect over time. Much conventional research is focused on closing the gap between current best practice and such an idealized practice; that is, taking the best known mousetrap and incrementally making it better. In contrast, much QI is focused on closing the gap between actual practice and best known practice; that is, taking the best-designed mousetrap already known and ensuring that it deploys flawlessly every time it is indicated (and not when it is not) in the specific local context of deployment.

Classic research and QI are synergistic drivers of improvement. Classic prospective interventional research and QI both seek to move clinical care closer to an idealized theoretic best outcome. Classic research often focuses on novel means, such as more sensitive diagnostics or superior therapeutics. QI often focuses on closing the gap between common practice and established best practice, such as through standardization, decision support, and automation. Reducing errors and performance variation in operations through QI can reduce the noise introduced into measures of research interest, thereby improving the sensitivity of research efforts. Thus, clinical research and clinical QI are synergistic.

Health services research (HSR), like classic research, can inform interventional QI efforts, but HSR is typically passive, often taking advantage of natural experiments; it analyzes differences in outcomes based on variations in context or care delivery that already exist instead of implementing practice changes to optimize outcomes.

Unlike clinically or physiologically oriented classic research, HSR studies often incorporate how social factors, organizational structures, delivery processes, financial drivers, and personal behaviors affect health-care access, quality, cost, and outcomes. HSR methodologies, data management, and statistical approaches often draw on the same biostatistical traditions of clinical research. HSR often shows where improvement could or should occur in health care and the degree to which improvement might affect outcomes, and even suggests the factors that may be essential targets for change. HSR may be considered a form of early QI research in terms of informing QI interventions, but it is distinct from the analysis or research performed on interventions undertaken to improve quality. The research methods commonly used to analyze QI interventions are distinct from both traditional research and HSR.

The term translational research encompasses a long sequence of analysis from bench to bedside to practice guidelines. Research guided primarily by physiologic principles occurs in the early stages of translational research (ie, T1 and T2). However, unlike early translational research, late translational research (ie, T3), such as implementation science, performance reliability, or improvement sustainability, is often profoundly influenced by local phenomena (eg, staffing ratios, case mix), individualism (eg, clinicians’ experience base, leadership styles), human factors (eg, psychology, ergonomics), and nonmedical forces (eg, business plans, economics, information technology, industrial engineering). These types of factors often affect the ability to generalize many local or single-site quality and safety interventions, so their inclusion in the analytical plan of QI research is often of utmost importance. As center after center tragically reinvent the wheel locally, most discoveries from T3 translational activity are never published or disseminated in a scholarly manner; a lost opportunity in terms of the knowledge that can be extracted.

In summary, research is the discovery of new knowledge and evidence on which clinicians can base their practices, QI is work designed to deliver the best possible evidence-based care consistently in local contexts, and QI research is the rigorous analysis of the improvement or implementation of the work itself. The clinical or operational outcomes of QI work are a barometer of the effectiveness of the QI intervention, but QI research strives to establish the viability of aims, the appropriateness of scope, the efficacy of change methods used, the interdependent array of results beyond primary outcomes (including process and balancing measures), and the influence of contextual factors either facilitating or impeding improvement efforts. Such clinical QI research produces, at one level, replication studies of primary evidence, but more importantly it creates insights into what contributed to effective or ineffective implementation work in real-world and often diverse settings, allowing others to generalize these insights and either emulate successes or anticipate potential pitfalls when attempting to apply the same evidence.

Models for understanding quality

As experimental statistician George Box observed, “All models are wrong, but some are useful.” Models provide an artificial structure for knowledge that reflects complex phenomena accurately enough to better enable understanding; ideally well enough to enable meaningful interpretation and informed action. Health-care quality is complex, but models can help clinicians to grasp what is essential. Beyond organizing QI work, these models can also serve as the organizational framework for QI research. A few of the more common and useful models for improvement are touched on here.

A common mode for organizing quality and safety analyses relates to the hierarchy of defects in a complex system. Latent system risks, near-misses, and actual harm are points along a continuum. Fig. 2 shows how this continuum matches the QI methods commonly used to remediate such defects. There is a long-standing debate about whether it is more advantageous to measure risk, errors, or harm, but, in truth, each has advantages and disadvantages, and all are widely used.

Hierarchy of defects and methods to remediate.

Detection and elimination of latent defects in a complex system provides the ideal solution to improving quality and safety, because it is the furthest point upstream from harming a patient. Failure modes and effects analysis (FMEA) is a powerful strategy to identify ways in which a complex system can fail because of the known historical performance of constituent parts of a device or process. This strategy allows potential defects to be designed out of the system (or planned countermeasures to be devised) before a design culminates in a product or an active process. FMEA is widely used in manufacturing and engineering industries in which device performance is fairly predictable (as with an intravenous pump or telemetry unit), but it is increasingly applied in the service industry. Limitations of analyzing latent system vulnerabilities include a lack of good historical performance data on which to base the model; the risk of unforeseen perturbations in complex and interdependent systems; unpredictable and dynamic changes in the system; lack of intuitive guidance to the sources of risk; and the theoretic nature of some assumptions and conclusions in the absence of measurable errors or harm.

Another top-tier tool in system improvement is process streamlining through the elimination of waste, be it time, energy, materials, or process complexity. Albert Einstein is quoted as saying, “Make everything as simple as possible, but not simpler.” If a desired outcome can be achieved in fewer steps without loss of fidelity or performance, it is likely to be more reliable, because eliminating unnecessary steps removes some opportunities for errors to creep into a system and simultaneously reduces the number of variables when trying to understand ongoing failures. Furthermore, elimination of waste improves value from a cost-benefit perspective. This kind of streamlining to optimize value-added output is the basis for lean design, originally applied to production lines but increasingly applied to service lines. The contemporary paradigm of lean production and management is based on the Toyota Production System, and lean strategies have been successfully adapted to health care. A full description of lean methodology is beyond the scope of this article, but exhaustive resources are available for interested students.

Progressing along the ladder from latent defect to harm, the next step includes errors. A widely accepted definition for medical error is failure of a planned action to be completed as intended (an error of execution) or the use of a wrong plan to achieve an aim (an error of planning), whether by commission or omission. Thus the drug overdose caused by a decimal error may be considered an error of execution by commission, whereas the treatment of mistaken septic shock instead of the actual adrenal crisis might be viewed as an error of planning in which proper care was omitted. If the error results in no harm, it is commonly labeled a near-miss, whereas if harm occurs it is considered a preventable adverse event. This situation should be considered distinct from nonpreventable adverse events for which ways to avoid the known complication are not established; that is, harm occurring as a consequence of medical care but in the absence of an error (such as with the risk of cardiotoxicity from certain chemotherapeutic agents). Because errors resulting in near-misses are far more common than errors resulting in preventable harm, near-misses provide an attractive target for monitoring and measuring quality and safety on a continuous basis. Analysis of near-misses in an iterative manner can help generate hypotheses for root causes more rapidly than if only harmful events are considered. Focusing on errors can be particularly helpful when related outcome measures are too rare or catastrophic to be acceptable guides (eg, deaths). Such error-based surveillance (eg, compliance with a best practice) is particularly helpful when there is good evidence for key steps or processes firmly established in the medical literature. The ability to identify and monitor compliance with important process measures provides actionable data to an improvement team about how to reduce unnecessary variation and close the gap between actual performance and desired best practice. This ability is the basis of process control, and is often associated with the Six Sigma management strategy used widely in many sectors of industry, including health care.

Although there are advantages to error-based quality-performance analyses, there are notable limitations to acknowledge. Error and near-miss rates vary widely depending on the definitions used for error, the surveillance methods, and even the safety culture of the reporting unit, which confounds the external validity of such quantification. Measures of errors are most helpful if they can be expressed as rates (errors divided by opportunities for that error type), but there is no denominator available for many types of error and harm. Even the numerators can be circumspect because many errors go unreported, and there is an attention bias that favors identification of errors of commission (rather than omission) and errors resulting in harm (rather than near-misses). Data derived from voluntary error reporting are particularly messy. In a survey of pediatric physicians and nurses, half filed incident reports on less than 50% of their own errors, and a third did so less than 20% of the time. It is reasonable to conclude that most practitioners are only aware of the tip of the iceberg when it comes to near-misses, preventable harm, and opportunities for improving health-care quality. Thus, error-based quality assessment may be better applied as a local qualitative and semiquantitative improvement strategy, rather than an as a comparative performance tool.

The measure of actual harm to patients is a final measure of quality and safety in the current hierarchy being discussed. From a high-principled perspective, harm to the patient can be considered a failure to detect and mitigate the latent system defects and combination of conspiring errors or contexts. This situation is often described as the Swiss -cheese model, whereby an error or errors propagate through a system that is designed to intercept such errors but that, because of system dynamism and complexity, does not prevent all harm from reaching a patient. Although risk and near-miss analyses are more proactive, harm analysis is a more reactive process; there is no putting the genie back in the bottle. However, from a pragmatic perspective it is adding insult to injury to witness harm and not try to learn from it. It is worth noting that all errors are not equal; those resulting in harm may be distinct from those that do not, and harmful errors can implicate defects not necessarily apparent in near-misses.

Several organizations have proposed injury-based trigger tools that can be used to provide systematic surveillance measures of harm, such as the Agency for Healthcare Research and Quality’s general and pediatric-specific PS indicators, the IHI’s Global Trigger Tool, and others. Some of these metrics and tools focus on types of harm that are assumed to be preventable or largely so, whereas other tools for measuring harm are inclusive of all readily identifiable harm. One key advantage to measuring all harm is that it provides an opportunity to question the boundary between preventable and unpreventable injuries. If the goal in health care is to eliminate or reduce all harm to patients, then including measures of harm considered unpreventable by traditional medical standards can direct clinicians’ attention toward innovative care or research. It is the medical legacy that types of harm formerly deemed to be a cost of doing business are now considered largely preventable.

Harm-based performance metrics have their limitations too, the most obvious being that the patient is injured in some manner. Because unpreventable adverse events and deaths are, to some extent, expected in hospital settings, such occurrences do not necessarily raise the specter of preventable error. When they are recognized, attribution may not be accurate. Furthermore, the retrospective nature of harm evaluation provokes numerous kinds of bias toward which human perceptions are prone, such as hindsight and outcome bias. If the analysis and attribution of risk, error, or harm are significantly biased, incorrect, or overly simplistic, then the conclusions not only are invalid but also can lead to unnecessary and possibly counterproductive attempts at remediation. Therefore, all risk, error, and harm analysis, as well as planned responses, must be undertaken with such limits and pitfalls in mind. In addition, because all monitoring and corrective strategies have limitations and none are perfectly suited for all applications, it makes sense to use multiple simultaneous approaches for a more robust quality-monitoring and safety-monitoring system. Doing so also helps create cross-validation between sources of perceived risks, error patterns, and actual harm, helping to overcome the weaknesses of each individual approach.