In To Err is Human 2 , safety is defined as freedom from accidental injury and error as failure of a planned action to be completed as intended (i.e., error of execution) or use of a wrong plan to achieve a goal (i.e., error of planning). Two types of execution errors exist: errors of commission (unintentionally doing the wrong thing) and errors of omission (unintentionally not doing the right thing). Errors can occur at any step of patient management, including diagnosis, treatment, and prevention.

An error may or may not cause an adverse event. Adverse events are injuries that result from a medical intervention and are responsible for harm to the patient (death, life-threatening illness, disability at the time of discharge, prolongation of the hospital stay, etc.) 2 . A near-miss is an adverse event that either resolves spontaneously or is neutralized by voluntary action before the consequences have time to develop. Adverse events may be due to medical errors, in which case they are preventable, or to factors that are not preventable.

There are two basic approaches to the evaluation and improvement of quality of care. In the room-for-improvement model, problems are identified, plans are then devised to correct the problems, and the effectiveness of the plans is assessed. This approach is known as the Plan-Do-Act Cycle (PDAC) of the Institute for Healthcare Improvement. The second way to measure safety is to use a monitoring system that detects problems and evaluates it periodically using quality indicators. These two approaches are complementary and often are used concomitantly. Thus, the monitoring model can be viewed as a way to seek opportunities for improvement by initiating a PDAC.

Safety measurement requires a self-assessment system for quantifying what we do and how we do it to help us to identify targets for improvement. A surveillance system needs multiple identification methods to detect medical errors and adverse events. These methods are implemented at the national or local level. National governments or agencies have developed reporting systems. At the hospital level, public and private agencies in North America have developed patient safety and improvement programs since 2005, as well as private databases to facilitate adverse-event reporting. In Europe, a safety program called The European Network for Patient Safety (EUNetPAS) was launched in 2008 to develop a culture of patient safety, provide a framework for education and training in patient safety, develop a core European curriculum on patient safety, implement reporting and learning systems, and implement methods to ensure medication safety. At the hospital level, different reporting systems are available to healthcare workers.

• The medical review. Reviews that do not target selected indicators are time-consuming and depend on the information available in the charts. Reviews can focus on selected indicators that can be assessed using the administrative data, discharge summaries, or mortality/morbidity review data. Medical reviews may be conducted manually or electronically using text words or text mining. Factors that may limit the use of the medical review method include absence of electronic medical records, paucity of resources for performing the reviews, variability in the terms used to label adverse events, and spelling mistakes. Failure to standardize the terminology may increase the difficulty of the search and the risk of false-positive results. Moreover, the analysis of documented adverse events requires considerable skill in interpreting the data. A meta-analysis comparing the rate of detection of pharmacists vs. nonpharmacists revealed a high level of adverse-event detection by pharmacists 16 .

• Voluntary reporting is the method most often used to detect medical errors and adverse events. Limitations include underreporting due to time constraints, lack of adequate reporting systems, fear of litigation, a reluctance to report one's own errors, uncertainty of the clinical importance of the events, and the lack of changes after reporting. However, this reporting method is the most useful for inducing behavioral changes, demonstrating the benefits of adverse-event reporting, and allowing us to learn from our errors. The presence of a multidisciplinary safety team might facilitate voluntary reporting.

• Medical errors and adverse events also can be detected by direct observation at the bedside 17 18 . This method is useful for detecting errors by omission. For example, medication errors can occur at any stage of the medication process (prescription, delivery, dispensing, administration, and monitoring), Medication error rates varied in the studies according to the definitions used, the medication process being evaluated, and the method of reporting. A pharmacist at the bedside can collect errors by omissions not detected by voluntary reporting. The medication error rate varied from 7.45/1,000 patient-days with voluntary reporting to 560/1,000 patient-days with daily routine observation of prescriptions 10 12 . Similarly, the presence of a trained clinical research assistant who collected medical errors increased the rate from 2.2/1,000 to 597/1,000 patient-days in the IATROREF studies 14 19 .

• The past several years have seen growing interest in learning from patients' experiences of care safety in all countries 20 , with an older tradition in the United States and the United Kingdom via the CAHPS (Consumer Assessment of Healthcare Providers and Systems) and National Health Service (NHS), respectively. In 2007, the OECD (Organisation for Economic Co-operation and Development) established the patient's experience as a key priority. In the ICU, many patients are too ill to report on their own experience, but information can be obtained from families instead.

Combining these methods to ensure robust reporting of medical errors and adverse events is essential to obtain a global picture of care delivery in the ICU. The above-described surveillance systems require the use of valid indicators. Ideally, each indicator is expressed as a rate with a numerator (number of events, which can be defined easily and accurately) and a denominator (domain of care or population at risk). The surveillance system should include standardized data-collections forms, which should be used by trained staff. Data quality must be checked regularly (via audits and checks of missing data). Event rates may be difficult to determine when the definitions differ across institutions or medical societies or are not accepted by all leaders and when the at-risk population cannot be accurately determined.

According to Avedis Donabedian, three categories of indicators can be used: structure indicators (what we want vs. what we have), process indicators (what we do vs. what we should do), and outcome indicators (what we achieved vs. what we should have achieved). Several societies have published lists of indicators, and Table 1 summarizes the main indicators used in each category. Since 2004, the Outcomerea organization has been working on quality indicators for the ICU. A list was built after searching the electronic MEDLINE database using various combinations of the words "adverse event," "iatrogenic," "intensive care unit," "medical error," and "epidemiology." This list contained 180 reported adverse events. In July 2004, we sent the list of 180 events to 30 experts working in 5 ICU fields (cardiovascular disease, neurology, nephrology, pulmonology, and gastroenterology), who added 415 events, for a total of 575 events. Then, 30 other experts including intensivists and ICU nurses participated in a Delphi process to select indicators exhibiting the following characteristics: precise and simple definition of the event and high incidence of the event, impact on morbidity or mortality, and nonpunitive disclosure. A list of 14 events was chosen as sufficiently long to provide useful data yet not so long as to hinder the feasibility of a multicenter study designed to assess their incidence. To reduce bias in data collection, the steering committee developed detailed definitions for all events, and the definitions were then reviewed and validated by the experts. These indicators are listed in Table 1.

The choice of safety indicators depends on several factors, such as previous quality indicators monitored in the unit, monitoring methods, availability of time to monitor additional indicators and to provide feedback to the team, and whether monitoring of processes is instituted before monitoring of outcomes related to those processes. Improving safety requires time, organization, and resources. The goal is to achieve the best possible quality given our resources. Both process and outcome indicators should probably be selected. The process indicators should be related to robust outcomes and the outcomes should be at least partly preventable. Among nosocomial infections, catheter-related infections exhibit these characteristics 21 22 . Other suitable outcomes are accidental extubation 14 23 , pressure sores 24 25 , falls, rate of readmission within 48 hours 26 27 28 , family satisfaction 29 , and morbidity-mortality conferences 30 .

Comparing the rates of medical errors and adverse events across studies can be challenging due to differences in definitions and to the absence of clear definitions of harms. Even when clear definitions of harms are established before the study, harm rates may be underestimated 14 . Two types of medical errors and adverse events are reported: those related to medications, and those related to procedures or the ICU environment. Administering the right drug to the right patient at the right frequency in the right dose and via the right route represents a challenge for the nursing staff. The Critical Care Safety Study reported an overall rate of 80.5 medication errors associated with harm/1,000 patient-days in medical and coronary-care patients 11 . In the recent worldwide SEE2 study, the rate of parenteral medication errors was 745/1,000 patient-days 10 . With medications given by continuous infusion, the rate was 105/1,000 patient-days 31 . When direct observation at the bedside was used for detection, one medical error was documented for every five doses of medication administered, and among medical errors 23% were errors by omission 17 . Stress ulcer protectors and preventive anticoagulants were among the most often omitted drugs 32 . Vasopressors and catecholamines, insulin, coagulation-altering drugs, antimicrobials, and sedatives were the medications most often involved in medical errors 10 . Insulin and coagulation-altering drugs are associated with numerous errors related to the complexity of dosing and/or monitoring. In recent years, evidence supporting insulin therapy and tight glucose control has led to an increase in the use of insulin in ICU patients 33 34 35 . Clinical trials have demonstrated that this strategy increases the incidence of hypoglycemic episodes 36 37 38 . The IATROREF study found a rate of 757 medical errors/1,000 patient-days and 126 adverse events/1,000 patient-days for insulin administration 14 .

Numerous other medical errors and adverse events related to procedures and equipment have been investigated in the ICU 9 . Mechanical ventilation was associated with at least one incident in 95/137 patients (0.004 per patient and per day of mechanical ventilation) 39 . Pneumothorax, one of the main complications of both barotrauma and catheter insertion, was reported in 1.5% of patients on day 5 after ventilation initiation 40 and was associated with a threefold increase in the risk of death 15 . All tubes, lines, and drains used in the care of ICU patients can be removed accidentally 41 , with an incidence of 22 removals/1,000 patient-days in a French study 42 and 14.5/100 patient-days in a multicenter European study 9 . Maintaining homeostasis is of great importance, and acquired electrolyte disorders can occur as a manifestation of poor quality of care during the ICU stay and can result in increased morbidity or mortality rates 43 44 .

Risk factors for medical errors and adverse events have been extensively studied. They pertain either to the ICU or to the patient. The highly sophisticated treatments, technologies, and diagnostic tools used in the ICU are associated with a high risk of medical errors and adverse events 45 . In the IATROREF study, risk factors for medical errors consisted of mechanical ventilation, insulin use, central catheterization, and unscheduled surgery 14 . A study in a French medical ICU identified age older than 65 years and presence of more than two organ failures as independent risk factors for adverse events 13 . A relationship between severity of illness and adverse events was found in a large multicenter European study in which any organ failure, high or excessive workload, and risk factor exposure time independently predicted adverse events 9 .

The impact of medical errors or adverse events is difficult to assess due to differences in case-mix, confounding factors for mortality, and occurrence of multiple events in the same patients 15 . Sophisticated analysis methods must therefore be used to evaluate relations between medical errors or adverse events and patient outcomes. The IATROREF study identified 1,192 medical errors in 1,369 patients; of these, 183 (15.4%) in 128 (9.3%) patients were adverse events that were followed by one or more clinical consequences (n = 163) or required one or more procedures or treatments (n = 58). After adjustment for risk-factor exposure time, medical errors, even when multiple, had no impact on mortality. In contrast, having more than two adverse events was associated with a threefold increase in the risk of death 14 .

The occurrence of errors is caused by a combination of human factors and system factors 46 . People often make errors, and rates of human error have ranged from 30% to 80% 13 47 48 . What humans do results from interactions between people and the system in which they work. Interventions designed to increase concentration and diligence among healthcare workers are not effective: human errors are unavoidable. Instead, the work conditions should be designed in a way that minimises errors: as stated by Reason, "We cannot change the human condition, but we can change conditions under which humans work" 46 . For example, when two drugs that are very similar in their presentation are stored in the same area, the human-based approach would consist of educating the healthcare workers to pay attention to this similarity to avoid errors. The system-based approach would lead to storage of the two drugs in different places. In the system approach, the key question is not identification of the person responsible for the error but determination of how the error occurred. The mechanism underlying the error is thus identified, without placing blame on the healthcare workers. Then, the organizational flaw can be corrected with the goal of preventing further occurrences of the error.

Since the 1980s, a large amount of work has examined the role for a safety culture in preventing medical errors. Safety culture or safety climate (the two terms are sometimes used interchangeably but "safety culture" is generally seen as a more embracing term than "safety climate") is a concept originally used to describe the safety management inadequacies resulting in major disasters. Thus, the term was first used after the Chernobyl nuclear accident. Now, the concept has evolved to apply to errors at the individual level. The most widely used definition of the safety culture is that developed by the U.K. Health and Safety Commission: the safety culture is "the product of individual and group values, attitudes, perceptions, competencies and patterns of behavior that determine the commitment to, and the style and proficiency of, an organization's health and safety management" 49 . The description of the safety culture concept has been largely empirical. For example, Sexton et al. 50 suggested six dimensions: teamwork climate, job satisfaction, perceptions of management, safety climate, working conditions, and stress recognition; whereas others described a larger number of organizational dimensions 39 40 41 . ICU and hospital organization is a key point in the safety culture concept. The organizational dimension includes human and technological aspects. Concern over the high rate of medication errors has prompted increased interest in using technology to improve safety 51 . New technologies implemented in recent years include electronic health records, clinical decision support with or without a computerized provider order entry system, bar-code medication administration, and smart infusion pumps 51 . Although these technologies decreased the number of errors, there is little evidence for a concomitant decrease in harms 52 53 54 55 . Furthermore, these new technologies have created new errors and harms 56 57 58 . Before considering their implementation, we must define the clinical settings in which they may be effective, and we must address the specific difficulties raised by their use in the ICU. Many errors are related to less-than-ideal human organization. For instance, burnout syndrome occurs in almost half the physicians and one-third of the nurses in French ICUs 59 60 . Burnout syndrome can adversely affect healthcare worker performance, thereby contributing to medical errors and adverse events. Factors that increase the rate of burnout syndrome include high patient volume, high levels of noise and light, long shifts 61 , changes in shift hours, and the occurrence of conflicts 62 . In a study of intern work hours, the traditional intern work schedule involving work shifts longer than 24 hours and a mean of 77 to 81 work hours per week was compared to a schedule designed to decrease sleep deprivation (15-hour shifts at the most, with 60-63 work hours per week) 61 . The traditional schedule was associated with a 22% higher rate of serious errors (193.2 vs. 158.4/1,000 patient-days, p < 0.001), a 20.8% higher rate of serious medication errors (99.7 vs. 85.5/1,000 patient-days, p = 0.03), and a 5.6-fold increase in serious diagnostic errors (18.6 vs. 3.3/1,000 patient-days, p < 0.001). It would be useful to test interventions designed to improve well-being at work and to assess their impact on the rates of medical errors and adverse events 63 .

A number of targets for improvement have been identified 64 65 . The Agency for Healthcare Research and Quality identified five measures that have effects of varying magnitude on physician behavior (academic detailing, audit and feedback, reminder systems, interventions by local opinion leaders, and printed material) 65 . Multifaceted programs or bundles are more effective to improve safety than are isolated measures and have been used in several studies 19 66 67 . In the IATROREF study, we used educational slide shows, printed educational material, and feedback meetings; and we focused on errors administering insulin, errors administering and prescribing anticoagulants, and accidental removal of endotracheal tubes and central venous catheters 19 . Our program was effective in preventing insulin errors and accidental tube/catheter removal. However, significant Hawthorne effects were documented 19 . In a multicenter, cluster-randomized study, a multifaceted program, including feedback meetings, expert-led educational sessions, and dissemination of algorithms, significantly improved process indicators, such as the semirecumbent position for nosocomial pneumonia prevention and measures to prevent central catheter infections 67 . Similarly, a bundle strategy decreased the rate of nosocomial pneumonia 68 . In 2003, the Michigan Keystone ICU Patient Safety Program based on a John Hopkins University model was launched to eliminate catheter-related infections and ventilator-associated pneumonia. The model is based on the four Es: Engage, Educate, Execute, and Evaluate 69 . The Michigan hospitals reported improvements in adherence to guidelines for ventilator-associated pneumonia prevention and decreases in the rates of catheter-related infections 22 70 71 .