Chapter 3. Improving the Quality of Care in the Intensive Care Unit

• Improving ICU performance demands a systems-oriented approach embodied in total quality management (TQM).
• In most ICUs this requires a major paradigm shift away from the discredited notion that most omissions, errors, and other problems are the fault of individuals, and embraces the idea that meaningful and sustained improvement comes from transforming structures and processes into those that make it hard for people to make mistakes.
• Creating an effective program of TQM is much easier in an ICU having the resources to hire personnel and purchase information systems.
• The most difficult, labor-intensive, and expensive part of TQM is data collection, but there are ways to make this less burdensome.
• Every ICU should implement a system of TQM that is multidisciplinary and inclusive, has the vigorous support of hospital and ICU leadership, and has sufficient personnel and economic support to succeed.

Intensive care units (ICUs) are an important component of modern health care. This importance derives from large costs that are both human and economic. Approximately one third of adult deaths in the U.S. occur in acute care hospitals,1 and as many as half of these occur in ICUs.2,3 An even greater fraction of people spend time in an ICU during the final 6 months of life.3 Among patients admitted to ICUs, 8% to 17% die there, though for some diagnoses the ICU mortality rate is much higher.4–6

Beyond death rates, suffering is common among ICU patients. Half or more of ICU patients endure substantial pain.7–9 Bergbom-Engberg and Haljamae found that 30% of patients surviving mechanical ventilation recall agony, panic, or both.10 Among oncology patients in an ICU, 30% to 75% experienced dyspnea, hunger, anxiety, insomnia, or thirst, with increased suffering during even such innocuous procedures as turning.9 Substantial dissatisfaction among relatives and friends of ICU patients11–13 indicates that suffering in critical illness is not limited to the patients themselves.

To go along with these human costs, the economic costs of ICU care are staggering. Comprising just 8% of acute hospital beds,14,15 ICU care consumes 20% of total inpatient expenditures,15–17 equivalent to 0.9% of the gross domestic product of the United States, or $91 billion in 2001.18 ICUs are a smaller portion of the health care systems in other industrialized countries, but still represent an important segment of medical care and its costs.16,19–23

A necessary starting point for improving ICUs is to recognize that problems of quality are common and serious. A recent Institute of Medicine Roundtable24 stated that:

Serious and widespread quality problems exist throughout American medicine. These … occur in small and large communities alike, in all parts of the country and with approximately equal frequency in managed care and fee-for-service systems of care. Millions of Americans are not reached by proven effective interventions that can save lives and prevent disability. Perhaps an equal number suffer needlessly because they are exposed to the harms of unnecessary health services. Large numbers are injured because preventable complications are not averted. … Quality of care is the problem, not managed care.

The U.S. health care system is by far the most expensive in the world, while ranking thirty-seventh in overall performance, and seventy-second in population health.25 Canadian health care is the tenth most expensive, ranking thirtieth and thirty-fifth on these measures, respectively. This poor performance is not explained by society's violence, poor health habits, or high cholesterol levels.26

There are many factors contributing to this poor performance. Inappropriate use of medical interventions is common throughout the health care system,26–29 with nonadherence to established standards of care being related to poor outcomes.30,31 Medical errors are prevalent, often leading to disability and resulting in 27,000 to 98,000 deaths/year.26,32,33 A study in Veterans Administration hospitals found that 6% to 34% of deaths there may be preventable.34 A study limited to certain subsets of patients with iatrogenic complications in ICUs observed them to occur in 31% of patients, and to be severe in 13%.35 Errors were observed to occur in 1% of all the interventions done each day to patients in an Israeli ICU.36

Another problem is variation in practice and outcomes not explained by patient or illness characteristics. Since wide and widespread variation could not exist if most practitioners practiced optimally, and such variation is evidence that suboptimal care is common. Variation has been found for many outcomes in a broad range of settings, with important differences by geographic region,3,37,38 hospital,39–41 physician,40,42–48 and insurance status or payer system.42,49–53 Such variation occurs in ICUs as well.45,49,51,54–56 Lest one think otherwise, such variation is not just an American phenomenon.48,57–62

Given the primary function of ICUs, it is particularly troubling that they suffer from serious deficiencies regarding palliative and end-of-life care. In addition to the high prevalence of suffering discussed above, caregivers often don't ask about their patients' end-of-life preferences, with consequent discordance between the care desired and that received.7,63 The conversations that do occur are often much delayed.63–66 Communication between caregivers and ICU patients or their surrogates is frequently poor. Azoulay and associates found that 20% to 43% of surrogates did not understand what they had been told about diagnosis, prognosis, and therapeutic options, and that these problems correlated with family-doctor meetings that were too brief.67 In other studies families expressed dissatisfaction with the amount and nature of communication, and many experienced actual conflict with the caregivers.12,68 Even more worrisome was that 24% of patients said they were not aware of the current approach to care, and 47% of those who preferred a palliative approach felt the care they were receiving was inconsistent with that wish.63 In addition, the variation in practice present in other aspects of ICU care is also present in end-of-life and palliative care.3,4,62,64,65,69,70

The data above demonstrate that ICUs are an important, expensive, and problematic component of health care. Therefore, vigorous efforts are needed to critically examine and improve every aspect of how well and how efficiently they perform their functions. Because “quality” is an extremely vague term often used to encompass various, often equally vague, concepts, this chapter will instead refer to the more tangible concept of ICU performance, and use the term performance improvement (PI) in place of the various alternatives.

Since it would require us to believe that ICU care is different than every other area of human endeavor, it is implausible that all ICUs perform equally well, or even that any given ICU is performing optimally. To identify how well a given ICU performs requires quantitation of relevant indices of performance. However, ICUs are complex organizations, and it is difficult to clearly define or measure such indices. In fact, the meaning, scope, and measurement of performance in health care have evolved and broadened over the past two decades. Some of the most widely used approaches to improving performance in health care have proved inadequate. This chapter will discuss both the conceptual basis and practical aspects of a superior method of evaluating and improving ICU performance.

Defining ICU performance is a difficult exercise that embraces diverse elements such as medical knowledge, ethics, economics, systems engineering, sociology, and philosophy. Efforts to measure ICU performance should follow two essential principles: (1) evaluate a variety of measures that span the dimensions of ICU performance; and (2) use performance measures that are directly relevant, or have an unequivocal relationship to measures that possess such relevance.

There are a number of domains within which an ICU should be judged (Table 3-1). While ultimately an ICU exists to serve the medical needs of critically ill patients, it also provides important services for families and friends of patients, health care workers in the ICU, the hospital, and society. Rating ICUs only on narrowly defined measures of health outcomes fails to recognize the larger social value associated with expert care of these patients. In addition, no single metric is adequate to address all of the categories of outcomes listed in Table 3-1.

Assessing performance requires direct measurement of end points that are relevant—to patients, society, and the hospital. Table 3-1 lists relevant measures of ICU performance. Because they are usually easier to measure, surrogate parameters are often used in PI efforts. While assessing surrogate end points is often a component of PI, this should not substitute for measuring and addressing the truly important parameters. For example, errors in drug administration have a proven association with detrimental outcomes,32 and therefore the rate of such errors is a relevant performance measure. Processes instituted within the pharmacy to try and reduce errors, such as altered drug labeling or direct communication with nurses, represent surrogates for the relevant index. While addressing such processes is in fact the necessary means to the end, changing a given process may or may not result in improvement in the rate of such errors, and does not obviate the need to measure it.

The ICU readmission rate is not listed in Table 3-1 because it is at best a questionable indicator of ICU performance. Its potential value derives from observations that readmitted patients have a higher mortality rate and longer length of stay.71 However, for it to be a meaningful surrogate requires that: (1) premature ICU discharge was the cause of a subsequent detrimental outcome due to a problem that was already present in the ICU, and (2) it would not have occurred if the patient had remained longer in the ICU. There are no data that have demonstrated this.71,72 The optimal readmission rate is unknown, and a low one could indicate that on average, patients are inappropriately remaining in the ICU too long, when they could be adequately cared for in less expensive venues.

All of the performance parameters listed in Table 3-1 have limitations. ICU or hospital mortality rates are commonly used measures of ICU performance that are relatively simple to collect. While some data indicate that hospitals with higher death rates have more preventable deaths,30 short-term survival tells us nothing about the things that are much more important to people, long-term survival and quality of life (QOL).73–75 Also, certain attitudes about death and dying can mean that higher short-term mortality represents superior care by virtue of being more concordant with patients' end-of-life wishes.76,77

The laborious nature of collecting data on posthospital survival and QOL undoubtedly contributes to low usage of these important measures.78 It is even more work to combine these into a quality-adjusted measure of long-term survival, such as quality-adjusted life years.79 A variety of questionnaire-based tools have been developed for quantifying patients' QOL.80–83 The most commonly used is SF-36.84 These generally gauge health-related QOL on multiple domains, with heavy emphasis on functioning.80,85 Indeed, ICU survivors often have demonstrated poor QOL.86–88 Assessment of cognitive functioning, depression, and posttraumatic stress disorder also shed light on the long-term consequences of ICU care89,90 (see Chap. 19).

Complication and error rates are often used as measures of ICU performance. These are relevant because of potential causal relationships of such adverse events with increased mortality, morbidity, or costs.32 However, such adverse events do not necessarily lead to clinically relevant consequences, and some studies have found little effect.34,35 Therefore, care must be taken to ensure that an established relationship to relevant outcomes exists for any medical error or complication whose rate is used as a surrogate for ICU performance. Since only a fraction of medical practices have been rigorously proved to be efficacious,91 many deviations from recommended practice may have no such relationship.

Symptom control and end-of-life decision making are important aspects of ICU care. As discussed above, there is much room for improvement in this area. Use of these outcomes as measures of ICU performance has been limited by lack of training and orientation among physicians,92,93 a paucity of pre-existing tools to measure them,94 and other factors.69

Because of the enormous costs of ICU care, resource consumption should be part of every institution's assessment of ICU performance. The best measure that balances simplicity and information content is ICU length of stay. Others that require much effort to acquire include total monetary charges or costs,95 usage of various diagnostic and/or therapeutic procedures, and Therapeutic Intervention Scoring System (TISS) score.96 TISS is a relatively comprehensive attempt to measure resource utilization for ICU patients that works well for cohorts.97,98 However, because spending a lot of money is justifiable if the benefits are commensurately large, while even small expenditures that generate no benefits are wasted, resource use is most relevant in combination with the noneconomic outcomes in Table 3-1. While the formalism associated with this concept, cost effectiveness, is well beyond the scope of local attempts to improve ICU performance,99 it would be a powerful tool to assist society in clarifying the value of ICUs, as well as to assess the performance of individual ICUs. A simpler approach to assessing cost effectiveness that could be adapted within a single ICU depends on case-mix adjustment of short-term mortality and length of stay (see below).100

Effective use of ICU beds is important for a number of reasons. The ICU is an expensive and limited resource.101,102 Moving patients out of the ICU prematurely due to limited availability has serious negative consequences.103 But ICU triage decisions are often inefficient, and can be made more effective without adverse medical consequences.104 Arguing that rationing of critical care is common but often inequitable, Kalb and Miller published a thoughtful framework for consideration of ICU triage in which they propose that the use of critical care be limited to clinical settings in which it has been demonstrated, or at least is presumed, to be cost effective.105 They and others106,107 provide arguments for using medical suitability as the primary determinant of such triage decisions, proposing that a low priority for ICU beds be assigned to those too healthy to require such care and to those too hopelessly ill to benefit. While rates of adherence to written or published admission and discharge standards is a common means of measuring ICU bed utilization, such information is not likely to result in much useful information, as such standards have not been subjected to scientific validation of whether they affect relevant outcomes.108 Possible measures of utility include: (1) the percentage of ICU patients who only receive care that could be provided elsewhere in the hospital, (2) the fraction of patients for whom ICU care represents an exercise in medical futility,109–111 and (3) patients who remain in an ICU for longer than they need its special capabilities.

The importance of satisfaction among patients and their families as measures of ICU performance is highlighted by data documenting that poor communication and dissatisfaction are common.12,13,67,68 Unfortunately, at the local level there is usually more talk than action on this topic. This is partly because unfamiliar survey instruments, questionnaires, or interviews that are time-consuming to administer and analyze, are utilized to gather this data. There are numerous potential dimensions to such surveys in ICU care, including satisfaction with: (1) level of care from physicians, nurses, and other ancillary health care personnel; (2) involvement in decisions regarding care; (3) amount and quality of communications with health care and administrative personnel; (4) outcomes of care; (5) administrative hospital functions such as admissions, discharge, and billing; (6) food; and (7) housekeeping. While there are doubts about the validity of currently used methods,112,113 a variety of tools exist to measure these aspects of satisfaction, including some created specially for use in ICUs.11,13,68,114–117

The satisfaction of all those who work in an ICU should likewise be considered a component of ICU performance. Job dissatisfaction contributes to higher rates of staff turnover,118 which: (1) wastes training time and money;119,120 (2) diminishes the ability of the ICU to perform as an experienced, highly functioning team;121 and (3) further degrades staff morale while increasing stress on managers and remaining staff. The nursing shortage is already substantial, especially in ICUs.122–125 That of respiratory therapists is not far behind,126 and a shortage of ICU physicians is looming.127 Problems related to dissatisfaction and turnover are obviously worse if those who quit cannot be replaced. In fact, dissatisfaction and burnout is common among nurses,128–131 respiratory therapists,132 and ICU physicians.133–136 While staff retention rates are easily obtained from personnel records, data about job satisfaction are collected from questionnaires or interviews. Many survey tools exist to assess job satisfaction, burnout, and related constructs.128,133,136–144

A large class of performance measures quantifies problems in the processes, procedures, and functions occurring within the ICU, or that link the ICU to the rest of the hospital. This group of measures is strongly related to many of the other outcomes in Table 3-1, but possesses independent importance because of its relationship to the quality of interpersonal interactions, and the staff's perceptions of their workplace. An example is inefficiency or conflict in the course of interactions between the ICU and the pharmacy. Even in the absence of identified errors in drug administration resulting from such incidents, they have the potential to lead to errors, waste, or rework, and in any case generate negative feelings among the staff.

The effectiveness of PI efforts depends on the type and quality of performance data collected. If that information is irrelevant, inaccurate, incomplete, or otherwise misleading, the effort expended may be wasted. This is a real problem, in part because in many hospitals PI efforts are primarily an exercise in analyzing individual adverse events (AE), such as complications, errors, or deaths.145 Unfortunately, this inclination has been augmented by recent publicity about medical errors32 and the response of regulatory entities toward such sentinel events.146

Though every performance measure can be phrased in terms of an AE that occurs in relation to individuals, it is fundamental that detection of individual AEs cannot be used to measure performance. This does not mean that analysis of individual AEs has no role in PI, but knowing that any given number of deaths, or even potentially avoidable errors occurred cannot inform us about performance without determining the rate of AEs. The number of individual events must be pooled and expressed as the rate of events by dividing by the aggregate number of at-risk patients, patient-days, or other appropriate denominator. This rate then is compared to some standard. The benchmark for comparison could be some accepted norm, the rate in other ICUs, the rate derived from analyses of databases, or the prior rate in that same ICU. It is this form of cumulative information—rates of death, errors, complications, family dissatisfaction, or others—that truly represents ICU performance.

There are multiple reasons that simply accumulating information about AEs, without their rates, is inadequate and potentially misleading. First, not all AEs are due to poor care. Even with implementation of every known beneficial practice, it is not plausible that AEs will disappear; rather the goal is to reduce their rate to an acceptable level, or below what they were previously. Second, an individual AE does not necessarily cause harm or increase costs (e.g., administering the incorrect dose of a drug often does no harm). Third, even an AE apparently caused by human error that led to harm is not proof that overall performance is poor; for example, it is entirely plausible that a surgeon who once amputated the wrong leg could have a superior overall complication rate. Fourth, the usual methods for identifying AEs, incident reports or chart review, are inefficient, inaccurate, and subject to significant disagreement.34,36,147–150 This last point highlights the necessity of creating reliable methods of systematically collecting accurate data on the performance parameter being surveyed. Both the denominator and numerator must be accurate. This topic is beyond our scope, but a few guidelines are: (1) prospectively collected data are superior to retrospective data; (2) information collected by computerized information systems is superior to that collected by humans, especially if they are specifically programmed to acquire the desired information; and (3) identification of AEs should be based on objective, predefined criteria rather than subjective human judgment.

Measuring ICU performance using variables such as those in Table 3-1 is effected by calculating the rate for binary variables such as survival or complications, and the mean, median, or interquartile range for continuous variables, such as a 0- to 100-point patient satisfaction scale. Whatever the data type, care must be taken to collect a sample size that is large enough to allow reliable statistical comparisons.77,151,152 This point is illustrated by the example of a complication that occurs in approximately 3% of patients, with monthly tracking showing rates that fluctuate between 0% and 6%. If the rate jumps from 1% to 3% from January to February, this cannot be taken as indicative of anything other than statistical variation unless the confidence intervals are narrower than these fluctuations. In this example, there would have to be 1120 patients at risk per month to reduce the 95% confidence interval to ±2%.153 Fewer are needed if the AE rate under consideration is higher, and if wider confidence intervals will suffice to convince oneself that the observed changes are real. A related issue in interpretation of comparative statistical performance data is the phenomenon of regression to the mean.154 This principle tells us that if a performance indicator is either high or low at one time, due to innate statistical fluctuation it will tend on the next measurement to be more intermediate. Thus it is important not to overinterpret short-term changes in performance measurements.

A very important issue is that all these parameters are strongly influenced by patient demographics, comorbidities, and type and severity of acute illness, cumulatively referred to as case-mix. This may be true even for institutional outcomes such as staff satisfaction.155 While raw data are “correct,” they are problematic for comparing ICU performance between different time periods (a primary element of PI) if there have been important changes in case-mix. For example, an increase in ICU mortality over time may not represent worse care if sicker patients were admitted in the second epoch. Even the seasonal differences that occur in some critical illnesses156–158 can create this problem. Case-mix differences are even more important in comparing performance between different ICUs.76,159,160 Thus it is important for PI efforts to collect at least some case-mix data.

Within a single ICU the simplest way to handle this potential problem is to collect and compare case-mix data for the separate time periods under consideration. When appropriate to the performance issue being surveyed, comparisons might be made in the same season of successive years. If the case-mix variables are similar in the different intervals, then it is reasonable to believe that the performance comparisons are valid. The argument against this simple approach is that finding similarity among any given number of case-mix variables cannot exclude the possibility that there are other differences between the cohorts that remain undetected. While the point is valid, this simple approach is reasonable and the only practical one in many institutions having limited resources for PI efforts.

The case-mix variables chosen for assessment should follow the purpose of the PI project. For general purposes it is straightforward to record patient age, gender, race, the presence or absence of important comorbid states, and the organ system most responsible for ICU admission. With more data collection resources one should consider collecting socioeconomic variables such as insurance category (e.g., private or Medicare vs. Medicaid vs. none),161–163 and measures of severity of illness such as the Acute Physiology and Chronic Health Evaluation (APACHE) score, TISS score, or Multiple Organ Dysfunction Score (MODS).96,152,164 Special case-mix variables should be chosen for special purposes. For example, measuring outcomes of mechanically ventilated patients might lead one to record each patient's initial PaO2/FIO2 ratio as a measure of the severity of respiratory failure.

When the case-mix is not similar across time periods, it is best to make adjustments for the observed differences. A common method is to use one of the available ICU risk prediction systems.100,165,166 Some of these, such as APACHE, the Mortality Probability Model (MPM), the Simplified Acute Physiology Score (SAPS), and IMPACT are for general adult ICU patients,164,167 the Pediatric Risk of Mortality (PRISM) score is for general pediatric ICUs,168 and others have been developed for specialized subsets of critically ill patients such as trauma or cardiac surgery patients.169,170 These systems use equations derived from multivariable linear modeling of demographic and clinical data that predict one or more outcome variables for each patient by comparison with the large inception cohort of ICU patients used to create the equations (see Chap. 6). Nonlinear models achieve similar predictive power.171 This approach is, in effect, an application of benchmarking against the outcomes in the inception cohort.146,172 Most of these systems explicitly generate scores reflecting the severity of illness that are used within the equations. These scores should not be confused with the predictions, but are themselves good and convenient tools for evaluating case-mix as described above.

Unfortunately, these systems are very limited, and most of them only predict short-term mortality. This can be used to measure performance using the standardized mortality ratio (SMR). The SMR is calculated as the observed mortality rate of an ICU cohort divided by the average mortality rate predicted by the equations. The SMR is thus a death rate that is case-mix–adjusted by comparison with the original inception group.77,165,166 Though unable to predict individual outcomes with sufficient accuracy to evaluate care for a given patient,173,174 these perform reasonably well for groups of patients.6,165,167,175–177 Other than short-term mortality, the list of performance measures for which validated prediction equations allow case-mix adjustments is short. The proprietary APACHE III system predicts ICU length of stay, days on mechanical ventilation, and the likelihood of receiving “active” intervention, though there has been published validation only for length of stay.54 Similarly, in pediatrics the PRISM investigators have published a prediction of ICU length of stay.55 Validated, ready-made tools to case-mix adjust the other important ICU performance measures do not yet exist.

Even for the few performance parameters included in these systems, there are problems related to calibration,159,160,173 accuracy within some diagnostic subsets,173,178 generalizability to other countries,179,180 aging of the inception data set,166,181 and even their utility in measuring performance.176,182 In addition, none of these systems adjusts for nonphysiologic parameters that may be important determinants of outcomes, such as socioeconomic factors. While the older versions of these systems are in the public domain, some of the most current and usable ones, such as APACHE III and Project IMPACT, are proprietary and expensive.183,184 It should be noted, however, that updating to the latest version often produces only modest increases in predictive power over earlier versions in the public domain.177,185,186

All of the prognostic systems mentioned are prospective. Predictions are generated from data available at the outset of the ICU admission, representing the clinical situation before ICU care was provided. However, use of commercially available, retrospective prognostic systems based on administrative or claims data is growing, in some cases mandated by governmental regulatory agencies. Many of these methodologies factor information about events that occurred during the course of care into the predictions. This represents a fundamentally flawed approach to evaluating performance and should not be used for this purpose. Not only is the accuracy of administrative databases problematic,187–189 but such an approach is not capable of distinguishing whether poor outcomes are due to bad care or to pre-existing severity of illness.76,187,190 Thus an ICU where many patients die due to avoidable complications could fail to be identified as a poor performer by a system that incorporated those complications into its prediction of the risk of death.

A more flexible approach that allows customized case-mix adjustment of any numerical performance parameter is to create your own multivariate regression analysis.152,191 With the performance measure as the dependent variable, the case-mix variables as covariates, and an indicator variable to represent the different time periods, this avoids many of the disadvantages of the ready-made systems. The magnitude and statistical significance of the coefficient for the time period indicator variable then informs whether there has been a change in performance after adjusting for case-mix. A relatively simple implementation might include covariates of age, gender, race, important comorbid states, the organ system most responsible for ICU admission, and a measure of severity of illness. This approach requires some expertise in computer applications and statistical methods, but can be implemented within individual ICUs.192

Special comments must be made discouraging the use of diagnosis-related groups (DRGs) in PI efforts. In most hospitals, the DRG choices are not made by clinicians, and contain significant inaccuracies.193 In addition, the DRG on hospital discharge is generally chosen to maximize reimbursement, and need not bear any relationship to the reason for ICU admission. Indeed, it may reflect a potentially avoidable complication of care.194 Likewise, the admission DRG may not represent the reason for transfer to ICU.

In summary, every ICU should collect cumulative data on a variety of relevant measures of ICU performance. Efforts must be made to ensure the accuracy and completeness of the data collected. The data should be used to calculate summary measures, such as rates or means, for the patient cohort being assessed. Sufficient data must be accumulated to make these rates meaningful. At least some case-mix parameters should be collected as well. If there are substantial differences in case-mix between cohorts being compared, then some adjustment should be made for these differences. However, in the real world, the resources needed to collect and adjust for case-mix may exceed those available, requiring dependence on unadjusted data. In addition, the methodologies used for case-mix adjustment are not perfect. Such limitations should not lead to nihilism and inaction. Though unadjusted data could be misleading, it cannot possibly be more misleading than having no performance data at all. Similarly, adjustments made with the best available, though imperfect, methodologies are better than raw data.195 Since the most common kind of performance comparison done in a single ICU is between two or more time intervals that are contiguous or nearly so, most of the potential difficulties in data interpretation mentioned above (except for the issue of seasonal variation) are probably of little significance.

Improving ICU performance in a given area involves sequential steps of: (1) measuring indices of ICU performance relevant to that area, (2) making one or more interventions aimed at improving those measures, and then (3) remeasuring the indices to document the effect of the intervention.196 Measurement is discussed above. This section will develop a framework for understanding the scope and nature of change.

There are two main domains of change for improving ICU performance. These are the technical components of ICU care, and the organizational features of the ICU. Both medical training and literature almost exclusively emphasize the technical. Thus the attention of clinicians is dominated by choices such as ventilator settings, vasopressors, imaging modalities, and the angle of the bed during tube feeding. While these are important, changes to ICU structures and processes are given little attention, but are at least as significant.194 Indeed, it requires extensive organizational change to ensure effective, uniform application of the technical choices.

A useful framework for addressing technical interventions to improve care in individual ICUs is adapted from Berwick.197 He tells us that we must: (1) know what works, (2) use what works, and (3) do what works adeptly. Knowing what works amounts to reading and understanding the literature. This includes recognizing the limits of current knowledge. With estimates that only 10% to 20% of medical practices are supported by rigorous studies,91 many diagnostic and therapeutic interventions are used in the absence of clearly demonstrated efficacy. Many studies focus narrowly on physiologic efficacy instead of relevant outcomes.198 Also, understanding and correctly interpreting the literature is impeded because many physicians lack the statistical and logical skills needed for such critical evaluation.198,199

As problematic as is our state of knowledge, the single most important deficiency in health care today is ensuring that we use what works, and do it well.3,24,200 Even seminal studies in the most widely read journals often do not change practice.201 To understand this deficiency, it is necessary to recognize that these things do not happen by themselves. Ensuring that all patients uniformly receive every applicable evidence-based “best practice” requires a paradigm shift. The historical, physician-centered paradigm holds individual physicians entirely responsible and free to guide care as they see fit, regardless of whether their choices concur with best practices, and independent of outside oversight except the risk of malpractice suits. Kassirer states that “… when it comes to setting standards of accountability and ethical behavior, our professional organizations and medical institutions have often faltered. In the guise of accountability, their efforts have often yielded lax standards that were intentionally and flagrantly self-serving.”202 Such a system, in which patients are entirely dependent on the knowledge and memory of their physicians, is not rational, and is not in the best interests of the patients or the clinicians. As Wennberg states: “the ‘system’ of care in the United States is not a system at all, but a largely unplanned and irrational sprawl of resources.”107 There are too many important things to know and remember to reasonably expect that any individual physician will always do all the right things. Even simple, universally accepted practices such as giving aspirin and β-adrenergic blockers to patients with acute myocardial infarction are often forgotten, leading to many unnecessary deaths.203 This is not an indictment of physicians—it is unreasonable to harbor such expectations in the modern medical environment. Instead, it is necessary to create structures and implement processes that ensure that every ICU patient, every time, receives all applicable evidence-based practices. Doing so will ensure, for example, that no mechanically ventilated patient receiving tube feedings will fail to have the head of the bed elevated to reduce the risk of nosocomial pneumonia because their physician forgot to write that order among the dozens that were written.204

But ensuring the uniform application of beneficial medical practices is not the only reason we need to re-engineer ICU organization. ICUs are complex systems, and it has been known for many decades in the industry that “the structure of a system significantly determines the performance of the system.”205 By emphasizing that most opportunity for improving performance derives from altering the system itself, Deming, Juran, and others created the most effective method known of improving the performance of complex systems.206,207 This methodology of PI, called total quality management (TQM), applies equally well to service industries such as health care,205,208–210 and tells us that making changes in ICU organization offers opportunities to improve performance that are equal to or greater than the gains from purely technical improvements.211–213 Though the systems approach of TQM is largely alien to health care, it is the recommended method of PI.146

The concept of TQM began with the seminal recognition that only 15% of errors and problems are a result of inadequate performance by individuals. Rather, 85% of the opportunities for performance improvement relate to flaws in institutional systems and processes that hinder the ability of individuals to perform their jobs well.206 As stated by Berwick,209 these original thinkers understood that problems had:

… been built directly into the complex production processes they studied, and that defects in quality could only rarely be attributed to a lack of will, skill, or benign intention among the people involved with the processes. Even when people were at the root of the defects, they learned, the problem was generally not one of motivation or effort, but rather of poor job design, failure of leadership, or unclear purpose. Quality can be improved much more when people are assumed to be trying hard already, and are not accused of sloth.

TQM is about improving the systems that are used to deliver ICU care. It is a continuous and relentless process of examining and improving structures and processes194 for the purpose of improving relevant outcomes. It is driven by systematically collected and appropriately interpreted data on outcomes.208,210,214 In TQM every structure, process, activity, function, and relationship, clinical and nonclinical, that influences the ICU in any way is open to scrutiny and change. Table 3-2 contains an illustrative list of areas for scrutiny.

Examples will aid in making these concepts more concrete. Evans and colleagues developed a computerized antibiotic consultant, available from every hospital workstation, to assist physicians in the selection of appropriate empiric antibiotics.215 They found that this increased the adequacy of the initial antibiotic choice from 77% to 94%, and that the delay in ordering adequate antibiotics was reduced. Leape and coworkers altered the structure of morning work rounds in an ICU to include the participation of a senior pharmacist.216 This simple structural change reduced avoidable drug errors by 65%. Bernstein found that what at first glance seemed like the ultimate in individual human error, surgery on the wrong side of the body, actually represented inadequacy of the procedures that should have been in place to prevent this problem.217

TQM views performance as a moving target that can be improving all the time, even in the absence of identified problems. Benchmarking, most simply against performance in the same ICU at earlier time period, allows for realization of this goal. TQM does not respect existing professional standards; rather it continually demands development of new ones that are more effective and efficient.210 The purpose is to reduce waste, rework, and complexity, while improving relevant outcomes, reducing costs, generating loyalty, and creating a climate of teamwork.208,209

Shifting the focus from identifying and remediating defective workers, to identifying and re-engineering defective structures and processes is necessary to correct “the defects that we have built into our complex medical systems … the waste we encounter in outmoded habits, the rework we pay for when things fail to work correctly the first time, complexity we build into our processes for no good reasons at all, the random and unpredictable variation from our predicted aims.”197 One result of TQM is that it becomes easier for people to do their jobs right, and harder for mistakes to occur.

An important tenet of systems engineering is that organizational performance inevitably deteriorates unless the structures and processes of the organization evolve along with the needs they must serve. The most natural way to cope with a changing environment is to supplement the structures and processes already in existence; as demand grows, the system gets multiplied to scale. However, such growth by accretion inevitably leads to a cumbersome, inefficient system. What is needed instead is evolution of new systems and processes that match the new environment. Obviously, the nature of ICU patients and ICU care, as well as the size of ICUs, has changed dramatically over the years, and projections show that this growth will continue.127 But the organization of many ICUs is not the result of thoughtful planning driven by improving outcomes, but rather reflects their historical origins and subsequent growth by accretion. As a result, their organization is inadequate to support the current demands.

The systems-oriented strategy of TQM is very different from the quality assurance/quality improvement (QA/QI) method that is still used in many ICUs. QA/QI responds to adverse events, errors, or other problems that are found. Obviously, problems are found only after they occur, and so QA/QI is reactive, instead of proactive. Leadership then convenes a group, usually composed of supervisory-level personnel, to do “root cause analysis.” Most commonly, this takes the form of trying to identify the errors made by individuals who allowed the problem to occur. The morbidity and mortality conference held regularly in most institutions is an example of such a flawed approach to improving quality.

In fact, both components of QA/QI, identification of problems and errors, and root cause analysis, are fundamentally deficient. QA/QI relies on occurrence screening to identify problems.145,218 This can take the form of medical record reviews, incident reporting, or use of prospective indicators. An example of an indicator is a document that bedside nurses complete when a mechanically ventilated patient is inadvertently extubated. Retrospective occurrence screening using the medical record is limited by the many inadequacies of that document as a window into actual events.187 Incident reporting is also an inadequate method of identifying adverse events in hospitals. Only a small fraction of adverse drug events are identified by incident reports,147,148 and a study of direct observation in an ICU estimated that only about half of errors were reported.36 So-called audit filters, groups of prospectively defined and collected occurrence screens, are of low sensitivity.219,220 Even prospective occurrence screening can obscure the magnitude of a problem, due to the changes individuals make in their behavior when they know that they are being watched, known as the Hawthorne effect.221 In addition, medical care is so complex and variable that screening for even a large list of specific occurrences is unlikely to have much impact on overall performance.

Root cause analysis is a process in which medical professionals who are knowledgeable in the area under consideration perform a detailed review of an adverse event to identify its causes. As stated by Berwick:222

Even when hospitals find ways to notice the injuries to their patients, their theories of cause often remain scientifically Neanderthal. They cling to unsound but deeply entrenched beliefs. They believe that bad people cause bad errors. … They believe that retrospection will allow them to find a single root cause, even though in fact the very idea of a “root cause” is misleading, since most system failures result from complex interactions between latent failures … and specific actions; conclusions about root causes are often illusions created by hindsight bias.

One reason for the ineffectiveness of root cause analysis is that there is very poor agreement among experts about the causes, and even the occurrence, of adverse events.34,149,150 The poorest agreement is for identifying management errors or negligence as the cause of adverse events.

QA/QI is about meeting minimum standards, detecting poor performance by individuals, and taking punitive actions. It is “at best inefficient, and at worst a formula for failure.”209 Despite prolonged and extensive use, there are almost no published data showing that health care is improved or behavior changed by PI methods that target individuals rather than structures and processes.223 Chassin concludes that there is a “paucity of evidence that, as a whole, previous quality-assurance programs actually did anything to improve outcomes for patients.”224

The ICU is an opportune place to establish a TQM program. ICUs are usually physically localized, well-defined entities that often already possess high degrees of organization. Compared with many other parts of the health care system, there are fewer people working there, making it easier to effect the necessary indoctrination into TQM concepts and to include a substantial fraction of personnel directly in the TQM process. Also, compared to other areas of health care, there are fewer confounding variables, with a relatively high degree of control over them. Methods to adjust for case-mix variation are more highly developed for ICUs than almost any other area of medical care.

TQM is a complex process. Implementing it in a hospital or ICU requires substantial commitments of leadership, resources, and time. It is estimated that 3 to 5 years are needed to adopt and fully institutionalize a new TQM program.210 Outside assistance can be helpful. This can come from a similar institution that has been successful in creating a TQM program, or by contracting with a consulting firm. Specific training is needed for those who will lead the effort. While a complete delineation of its methods is beyond our scope, this section will outline important issues in the implementation of TQM.

Meaningful participation of every type of ICU personnel in TQM is needed because both performance and improvement are recognized to be shared responsibilities. Since the best people to identify and solve problems are those directly involved in them, the basic functional unit of TQM is the multidisciplinary quality team. Unlike most other PI efforts used in health care, TQM is not controlled by clinicians, though obviously they are included in it. Thus for PI purposes, TQM replaces managerial authority with a participatory approach that emphasizes collective rather than individual responsibility.210

Initiating a TQM program within an ICU is difficult, though not impossible, without the existence of a hospital-wide TQM program. In the following five-step outline for implementing TQM in an ICU, it is assumed to be part of such a larger effort.

The first step is that top leadership must take the lead. Those at the top of the administrative and medical hierarchy must make an overt and highly visible commitment that TQM is a clear priority. They demonstrate this by providing the leadership, encouragement, and economic and political support necessary for the institution to evolve into a “TQM culture.” It has been observed that TQM is more readily assimilated in hospitals with corporate cultures that emphasize teamwork and development of human potential, are willing to take risks, and seize on opportunities as they arise.225

Second, investments must be made in education and training. Key administrative and medical managers undergo specialized education and training in the methods of TQM that they then bring back to the institution. Those who will act as facilitators for quality teams also require training. Subsequently the entire ICU staff needs to be familiarized with TQM concepts and the blueprint for improvement. Their involvement at all levels is necessary for TQM to succeed, and it is vital to elicit their support. An element of the human equation is that there must not be fear on the part of personnel that TQM is a smoke screen for downsizing, or that the discovery of problems, even of human error, will lead to retaliation.214,224 The TQM environment must be one of trust. Everyone, from the top to the bottom, must become convinced of the central truth of TQM, that the problems are due to faulty systems, not faulty people. More informed personnel will be more motivated and better participants. Part of the framework for success is that the institutional leadership finds ways to encourage, recognize, and reward individuals who participate effectively.

Third, investments must be made in data gathering and analysis capabilities. As discussed in detail above, TQM requires extensive measurement and statistical analysis of performance parameters. Therefore these functions must be given the necessary institutional support, and often require the acquisition of additional personnel and/or information systems. This is the most costly part of a TQM program, and must be accepted as the price of becoming serious about PI. Some measures of ICU performance should be constantly gathered as surveillance for commonly recognized problems. Examples include rates of major nosocomial infections and overall ICU length of stay. Others are collected for limited times related to a specific PI project. Caution is indicated regarding use of existing clinical personnel to collect data. Because they are in the best positions to accurately collect certain types of data, bedside nurses, respiratory therapists, and physicians can be involved in this part of TQM. However, it is necessary to resist the cost-saving impulse of placing the major burden for data collection on clinical personnel who are already overloaded with work, and who lack special training in this area. Resentment and failure of the program can result if the hospital's top leadership does not provide adequate economic support for TQM.

Fourth, a TQM leadership team for ICU performance is formed. Made up of a multidisciplinary group of administrative, clinical, and nonclinical personnel, this group coordinates TQM activities. It (1) solicits, develops, and prioritizes proposals for areas to be surveyed; (2) selects and supports the individual quality teams created to work on areas of interest; (3) contributes to the choice of primary and surrogate performance measures used to assess ICU performance; and (4) facilitates implementation of proposals for change. Areas given priority often are those that represent high volume, possess high risk, or are perceived to be a problem.

Last, and most importantly, TQM quality teams (QTs) are created and trained. An appropriately constituted QT addresses the area of concern assigned to it by the leadership team. QTs must be of manageable size, usually no more than 8 to 10 individuals representing different departments and different levels. Each QT must be constructed to include all the expertise and experience needed for thorough understanding of the inner workings of the area or topic under consideration. This dictates that the teams be heavily weighted with lower level personnel who actually do the work. No group of ICU workers should be omitted when relevant to the performance issue at hand. For example, housekeepers are an important component of infection control efforts. They, as well as unit clerks and other nonprofessional personnel play important roles that must be addressed within QTs. QTs comprised primarily of managers and physicians are usually doomed to fail because they lack truly intimate knowledge of the systems that make the ICU work.

An ICU may have one or more QTs. A QT may be permanent, or it may be constituted to address a given area, then disband. In smaller ICUs there is often nearly complete overlap between the membership of the leadership team and the quality team.

One person who has undergone TQM training functions as the QT's facilitator. The facilitator guides the group in its central functions: (1) brainstorming to break down the topic of concern into its component structures and processes,208,226–228 and (2) identifying practicable ways to modify or replace those components to improve the relevant outcomes. This procedure typically requires measurement of the relevant outcome parameter of interest, and of surrogate measures related to the structures and/or processes. With institutional support, data are collected, case-mix adjusted if necessary, statistically analyzed, and displayed to illuminate important relationships.228 Following implementation of changes, the performance parameters are remeasured to document improvement.

The focus of a QT can range from big to small. One may be convened to deal with an unusually high rate of death among mechanically ventilated patients. The death rate is the primary, relevant end point, but this team's scope and range will include a large number of structures, processes, and procedures used in the care of such patients. Data collection will be a formidable task for this team. Another team may be asked to find ways to decrease conflict about roles between nurses and respiratory therapists on the night shift.

QTs are potent tools for improvement. In the process of their assigned task they often identify other cumbersome or otherwise deficient processes, even if no obvious error has been related to them. Such proactive problem solving should be encouraged by endowing the QTs with considerable latitude of scope. When embarking on an assigned task, QTs should be encouraged to think expansively. No arbitrary limitations should be placed on what may be altered to improve performance.

In starting their work, QTs should perform a literature review to identify what is already known about the topic under consideration. Practices of proven effectiveness should be incorporated into the structures and processes of every ICU. For many major topics there have been systematic reviews of the literature that can serve as a starting point. However, it is often necessary for one or more members of the QT to review the original studies and summarize the findings for the entire team.

The role in TQM of analyzing individual adverse events deserves mention. As discussed above, such analysis cannot serve as the primary method of PI. However, within an active QT, discussion and analysis of individual events can effectively focus the attention of the group, and assist in identifying the structures and processes that require alteration.229 For this reason, it is still important that ICUs work to identify individual errors and other problems. Morbidity and mortality conference reviewing of each death and major adverse event is recommended, but its limited place within the larger context of TQM must be explicitly understood. There has been much discussion recently of improving error reporting in health care by creating a nonpunitive environment based on the systems and principles of TQM.230 Such approaches have proved successful in the airline industry and elsewhere.231

There are significant challenges to implementing a TQM system in the ICU. These include substantial time commitments, financial investments, and commitment of top leadership in the ICU and hospital. Often the largest impediment to successful implementation of TQM is resistance from physicians. Physicians frequently respond to objective results about performance by insisting that the data are inaccurate, and usually resist proposals perceived as abridging their professional autonomy.200 One contributor to this resistance is that their training and work experience generally do not instill physicians with the concepts of systems engineering. While there are a number of ways to try and change physicians' habits and performance,232 the evidence amply demonstrates that this is difficult.233–235 Nonetheless, gaining physician acceptance and participation in any TQM effort is necessary. Various strategies to achieve this goal have been discussed elsewhere.208,228 One particular experience of note was the effort of the Cleveland Health Quality Choice consortium to gain provider and purchaser confidence for its hospital quality efforts.236 Adapting its approach to the problem of gaining physician confidence in TQM would emphasize: (1) a large-scale effort to incorporate large numbers of physicians in the process from the start, especially those considered opinion leaders; and (2) sensitivity to physicians' concerns about presentation of data in ways that could be perceived as dangerous or threatening to their practices. The role of vigorous physician leadership in this regard cannot be overstated.237

Despite the difficulties, successful examples of implementing TQM in the ICU are there for emulation. Investigators at Intermountain LDS hospital in Utah described TQM implementation in their medical/surgical ICU, with extensive utilization of information systems support.211 They reported improved cost effectiveness, comprised of an 18% reduction in costs, with no change in case-mix adjusted mortality or length of stay. A regional TQM initiative in New England reduced hospital mortality after coronary artery bypass surgery by 24%.213

Improving ICU performance requires making organizational changes. A modest but growing body of research has studied strategies to improve performance via organizational changes in health care systems. These strategies range from simple to complex, and address widely differing aspects of change. They fall loosely into two categories: (1) strategies to increase use of evidence-based best practices, and (2) changes in structures and processes that are not directly related to the technical aspects of medical care. The former include education, clinical practice guidelines, reminders, audit with feedback, ordering policies, and combinations of these individual interventions. The latter are a diverse group of interventions. This section reviews much of that literature as it is applicable to ICUs. Strategies found to be effective in research studies are convenient, ready-made tools that individual ICUs can adopt, and are likely to be effective within their own TQM efforts. Of course, interventions that were effective in one setting may not be so in a different setting; as in all changes made under TQM, it is best if their local efficacy is demonstrated by local data.

Strategies to Increase Use of Evidence-Based Best Practices

Didactic educational interventions are the traditional method of teaching health care practitioners how to optimally care for patients. Continuing medical education (CME) is a requirement for health professionals in most venues. Recently, Davis and coworkers published a systematic analysis of studies assessing the efficacy of traditional didactic CME methods, such as conferences, rounds, meetings, and symposia.238 They found that the quality of these data is poor, that any benefit present was small and apparently limited to CME interventions having interactive elements. Another review of the topic also concluded that passive educational methods did not improve physician performance or patient outcomes.233 One must be careful in reading this literature, because in fact the analyses that seem to show more impressive gains from CME have mixed together didactic education with other modalities such as reminders and audit with feedback.239

Practice guidelines represent attempts to increase use of evidence-based best practices by making available to clinicians a set of recommendations for care. The concept is born of the belief that these practices are omitted because clinicians are not aware of them. They can be simple, or complex and comprehensive. They have been promulgated by professional organizations, governments, consulting firms, HMOs, hospitals, and other groups. Many physicians harbor negative attitudes about practice guidelines.240 In evaluating this body of literature, it is important to distinguish studies that assess clinically relevant outcomes from those that evaluate the surrogate of compliance with recommended patterns of care. This is especially important because many or most published guidelines fall short of the desirable features of practice guidelines.241–244 While some studies of guidelines have demonstrated improvement in outcomes, most have found that the guidelines are of limited effectiveness without concomitant use of other strategies.233,243 Numerous factors have been shown to influence acceptance, use, and effects of practice guidelines. These include their complexity, manner of dissemination, local input into their construction, characteristics of the targeted practitioners, and the ease with which they are accessible at the bedside.233,242–243,249 Nonetheless, it is quite reasonable that practice guidelines be used as tools of ICU TQM, with attention to these acceptance factors, and in conjunction with other strategies.

Because the number of best practices that modern clinicians are expected to remember and implement is prohibitively large, automatic reminders could be a powerful tool to improve ICU performance. The bulk of evidence identifies reminders as one of the most effective methods of improving adherence to best practices.233,242 The major limitation of this concept is effecting reliable and timely delivery of reminder messages to clinicians. Reminders can be delivered to clinicians in a number of ways. The most promising is via computers, as discussed below.

Audit with feedback can be defined as any summary of clinical performance of health care over a specified period of time that is communicated back to practitioners. These may or may not include recommendations for clinical action. Recent systematic reviews have concluded that this body of literature is of poor quality, and that while audit with feedback appears useful, the effects are small.250,251 Isolated studies have, however, observed substantial benefits from audit with feedback. Eagle and associates reported that it led to economically relevant reductions in length of stay for patients admitted with chest pain, congestive heart failure, or syncope in three medical ICUs.252 A review of 26 studies found that audit with feedback was less effective than reminders.233

Another potential method of ensuring best practices is to create groups of orders that have those practices built into them. The ICU is an opportune locale for such an approach by virtue of: (1) the relatively limited subset of diagnoses that gain a patient entry there, and (2) the applicability to many patients of a small set of proven safety practices. Sets of ICU admission orders can easily be created, either on paper or computer, to include by default orders such as prophylaxis against stress gastritis bleeding for patients who are mechanically ventilated.253

A recurrent theme seen in studies to improve use of best practices is that combining multiple strategies is more effective than the use of individual ones.211,213,233,242,254

Changes to Intensive Care Unit Structures and Processes Not Directly Related to Technical Aspects of Care

Table 3-3 summarizes a substantial portion of this literature. The interested reader who uses the listed references for further information will note that evidence-based information is sparse for some of these topics, with conflicting data for others.

The most prominent and best studied of these topics addresses whether the ICU is “closed” or “open.” These terms refer to whether or not all the patients in the ICU at a given time are under the care of a single attending physician. Several groups of investigators have shown that outcomes are superior in closed ICU systems.298–302 However, there are multitudinous organizational differences between different ICUs,303 and the seemingly simple issue of closed vs. open is interrelated to considerations such as: (1) involvement of subspecialists in critical care medicine, (2) the relative roles of the intensivist and the primary attending physician, (3) the presence of residents and critical care fellows, (4) the existence and nature of daily rounds for ICUs having house staff, (5) the role of the ICU medical director, and (6) differences between daytime and nighttime physician coverage. Interpretation of this literature is problematic because: (1) intermixing of these interrelated topics leads to many gradations between open and closed ICU structures; (2) there is substantial variation in study design, outcome parameters assessed, and case-mix adjustments among published studies; (3) many of the individual investigations suffer from weak study design; and (4) the possibility of publication bias. A recent systematic review of these topics avoided restrictive definitions of open vs. closed, and instead analyzed 26 studies according to whether intensivists were involved in a fashion rated as high intensity (defined as a closed ICU staffed by intensivists, or a mandatory consultation by an intensivist) or low intensity (no intensivist involved in ICU care, or elective intensivist consultation).255 High-intensity involvement of intensivists results in lower mortality rates and shorter lengths of stay. After case-mix adjustment, no individual study observed worse outcomes with more intensivist involvement in care. Another study concluded that high-intensity involvement by intensivists in care of all ICU patients in the U.S. would save almost 54,000 lives per year.256 Trying to fill the evidence-based policymaking void present in U.S. health care, a consortium of large businesses, The Leapfrog Group, is attempting to improve hospital outcomes for its employees by making high-intensity involvement of intensivists in ICUs a requirement for any hospital to contract with their health plans.304

Computers and other information technologies are powerful tools for improving ICU performance. Unlike people, computers can keep track of almost limitless amounts of information, never forget, and are essentially flawless in doing their assigned tasks. Table 3-4 lists some ways that computers can advance PI efforts in the ICU. Although there are as yet few studies evaluating whether such technologies improve ICU outcomes,305 their further development and widespread diffusion will likely be necessary for improving health care.200

Computers have facilitated access to clinical data for several decades. In most hospitals clinicians can obtain laboratory and other test results on terminals throughout the hospital. More recently, many hospitals have bought picture archiving communications systems (PACS), making it possible to view radiologic images without leaving the ICU. Although there is conflicting evidence that PACS lead to more timely viewing of images by clinicians,306,307 they do make viewing more convenient and save time.308

Keeping track of patient-related information is an ever-increasing challenge. Data about ICU patients originate from observations and manual measurements made by health care workers, bedside monitors, the laboratories, radiology, devices such as infusion pumps and ventilators, progress notes, clinical flowsheets, the hospital information system, and other sources. In most ICUs, collecting this mass of diverse information requires physicians to read paper records, use several different computer systems, and leave the ICU to obtain information residing elsewhere. This situation makes it difficult and inefficient for clinicians to acquire and integrate all the necessary data, and is obviously prone to oversight and error. Only highly competent computer systems can solve all of these problems.309 Via various interfaces, such systems311,312 can automatically acquire data from electronic connection to bedside devices and other computer systems in the hospital, and manual input via bedside workstations. They can display the data as customizable tables and graphs that people can understand and use at any enabled workstation, including those at remote locations via the internet. Because of these advantages, electronic, paperless medical records should become increasingly common.312–314 The future likely includes use of hand-held devices for wireless remote access to clinical data, and more rapid communication to clinicians of problems such as critical laboratory values.315

Computer technologies have vastly improved access to sources of medical knowledge. Textbooks, databases such as MEDLINE, full text journals, and other authoritative sources are widely available around the clock from internet-ready terminals in ICUs. It no longer requires a time-consuming trip to the hospital library to learn state-of-the-art answers to clinical questions about the patient who just arrived in the ICU. An increasing amount of medical knowledge is even more conveniently accessible within software for hand-held devices.316

Errors in order writing are the most common sort of medical error.32,317 Handwritten orders suffer from illegibility, mistakes in transcription, erroneous dosing, and other ills. Computerized order entry dramatically reduces such errors319,320 and can reduce costs.320 Computerized order entry is already relatively common and increasing in prevalence, driven by safety concerns of influential organizations such as the Leapfrog Group.304

As discussed above, reminders and clinical practice guidelines are strategies to increase use of evidence-based best practices. However, both are limited by problems making the information available to clinicians in a reliable, convenient, and timely way. Information systems can solve these problems. Each time a physician, nurse, or respiratory therapist logs on to the computer for any purpose is an opportunity to deliver a reminder message or offer up a relevant guideline. Having these occur during order entry makes it especially easy for the clinician to execute orders to act on the reminder or initiate the guideline's protocol. Simpler implementations of this concept would be triggered by the order being written.322 For example, a computer-generated reminder could be displayed to order elevation of the head of the bed every time a clinician orders tube feedings.204 Intelligent systems with access to other patient data residing in the electronic medical record are required to allow this concept to achieve its fullest potential. An integrated ICU information system linked to both the ventilator and the electronic medication administration record could identify that a patient is mechanically ventilated and not receiving prophylaxis for stress gastritis,322 and automatically generate a reminder. If the patient was receiving sedatives, the computer could suggest use of the ICU's sedation protocol323,324 and automatically generate the necessary orders for the clinician to implement. Currently, only sparse evidence exists about the effect of such systems. Computerized reminders have demonstrated benefit in most321,325 but not all326 studies. Studies of complex interventions that included computer-aided practice guidelines have observed improvements in care or reduced costs.211,245,327

Extending these uses of information systems is computer-aided decision support (CADS). In CADS, systems having electronic access to patient data provide clinicians with expert medical advice. Take the example of a physician doing computer entry of an order for gentamicin. Combining the latest creatinine value from the laboratory, the patient's weight from the daily nursing notes, and age from the admitting department's computer, the computer system estimates pharmacokinetic parameters and recommends a dosing regimen. When peak and trough levels become available, it would include those in the calculations, and the next time the clinician accessed that patient's electronic medical record, it would suggest revised dosing. These smart systems can help avoid incorrect dosing, and proactively identify drug interactions. Indeed, most studies of CADS have evaluated the effects on choice and dosing of pharmaceuticals. These have found an impressive variety of improvements in clinically relevant outcomes, including better empiric antibiotic choices, better dosing, reduced drug costs, reduction in adverse drug events, and others.215,320,328–332

The potential of computers to assist in management of ICU patients extends beyond providing technical advice. A computer linked to real-time monitoring can be programmed to automatically adjust devices such as ventilators and infusion pumps to maintain optimal parameters in a way that would be impossible for humans to accomplish.333,334 In a recent study of computer control over ventilators in pressure support mode using ordinary closed-loop linear programming, a computer continuously monitoring respiratory rate, tidal volume, and end-tidal PCO2 simultaneously maintained all three within desired limits 93% of the time, while humans did so 66% of the time.333 Use of fuzzy logic controllers may have even wider ICU applications.335 The ability of computers to tirelessly monitor and analyze the entire ICU data stream in real time makes them ideal, and superior to people, for tasks such as early detection of physiologic instability, adverse events,148 and surveillance for care practices outside of established practice guidelines.336

In addition to assisting in care of individual patients, computer applications greatly facilitate implementation of TQM. Even within a single ICU, acquisition, analysis, and presentation of the information necessary for TQM is expensive and labor intensive.194 Setting up the infrastructure for a comprehensive, electronic ICU patient database is an important step that will considerably ease the personnel demands for TQM. Likewise, collection of case-mix variables, and even adjustment for case-mix, can be automated. Manual calculation of scores for APACHE, SAPS, PRISM, or other prognostic systems is cumbersome and time consuming. Computer software linked to hospital information systems can compile, organize, calculate, and store these parameters automatically. Such databases are themselves potent tools for TQM. They can be queried to identify subtle relationships between variables that suggest opportunities for improving performance that are not obvious, even to those intimately involved in day-to-day operations. For example, an evaluation of database information identified laboratory ordering patterns in an ICU that led to policy changes and subsequent cost savings.246 The best system for TQM purposes would be one that also serves as the electronic medical record and clinical information system for the ICU. Such systems do exist.311

Unfortunately, there are currently a number of barriers to widespread use of information technologies to improve ICU performance.309,315,337 These include a paucity of user-friendly commercially available systems, high costs for those that do exist, a lack of interfacing standards leading again to high costs, and opposition from clinicians. Nonetheless, such highly capable systems do exist, and a few forward-looking institutions have invested in them, to their benefit.311,338