Medical errors that affect patient safety have generated huge concern since the publication of "To Err Is Human" 6 years ago [1]. Given the complexity of management in the intensive care unit (ICU) and the nature of human activities, critically ill patients are exposed to adverse events (AEs) induced by medical errors. A large number of studies have focused on AEs and medical errors in ICUs [2-6], one of their main goals being to identify strategies for preventing AEs and thereby improving patient outcomes. Choosing the best AE to serve as an indicator for the risk of medical error is challenging. In 2005, our group conducted a systematic literature review and presented the results to 30 national experts with clinical backgrounds in internal, emergency, and intensive care medicine. Using the Delphi technique, these experts selected 14 AEs that had the following characteristics: high frequency, easy and reproducible definition, association with morbidity and mortality, and ease of reporting without fear of punishment (Iatroref I study) (ref abstract). These AEs were used in a French multicenter study (75 ICUs) for a weeklong incidence evaluation (Iatroref II study) (ref abstract). Preliminary evaluation of the results allowed us to choose the following AEs for the current Iatroref III study: error in insulin administration, error in anticoagulant administration, error in anticoagulant prescription, unplanned extubation, and unplanned removal of central venous catheter. Evidence suggests that guidelines alone without reinforcing strategies may be insufficient to change provider behavior and that the most effective interventions may be multifaceted rather than single-component strategies ([7]). This study will test a composite intervention program. The objectives of the study are to determine whether the introduction of a composite intervention program decreases the predefined AEs.

Study hypothesis: The intervention program will decrease the incidence of the predefined AEs.

Design and methods Definition of adverse events

1. Error in insulin administration Insulin not given exactly as prescribed (including in the written protocol used in the ICU). The error may be a planning error or an implementation error regarding the nature of the insulin, dosage, preparation modalities, or administration modalities. Insulin therapy requires monitoring of capillary blood glucose levels.
2. Error in anticoagulant prescription Non respect of the recommendations about indications, dosages, administration modalities, contraindications, drug-drug interactions, and laboratory monitoring using the VIDAL (French Drug Compendium), learned societies, ICU protocols, and local drug committees. Either unfractionated heparin or low-molecular-weight heparin may be involved
3. Error in anticoagulant administration Anticoagulant not given exactly as prescribed. The error may be a planning error or an implementation error regarding the nature of the anticoagulant, dosage, preparation modalities, administration modalities, dosing times, or dosing intervals. Either unfractionated heparin or low-molecular-weight heparin may be involved.
4. Unplanned catheter removal Unplanned complete removal of a central venous catheter by the patient or by staff while caring for the patient.
5. Unplanned extubation Extubation not planned by the physicians. Unplanned extubations can be categorized as self-extubation by the patient, or accidental extubation for technical reasons (porosity, section of the balloon, poorly secured tube), and during patient care (bathing, taking radiographs, transportation, care).

Randomization procedure We will use a multicenter cluster randomized design to assess the effectiveness of a multifaceted intervention. Our statistician will randomly assign each unit to compliance with ICU guidelines as usual or implementation of the multifaceted intervention to prevent one of the AEs. Randomization will occur at the ICU level. Units that do not use the intervention program will serve as controls. The randomization procedure will be repeated every 2 1/2 months for a total of four 21/2-month periods in 2007-2008. During the month of January, the study will be prepared at each site, where a clinical research assistant will be trained in the use of RheaTM data collection software (http://outcomerea.org/rhea/install).

In the Iatroref II study, anticoagulant prescription or administration errors occurred in 59/1377 (about 6%) patients, insulin administration errors in more than 20% of patients, and unplanned catheter or endotracheal tube removal in 4% of patients. For this study, we collapsed unplanned catheter removal and unplanned extubation into a single group and administration error and prescription error of anticiagulant in another single group. Assuming that more than 2400 patients will be admitted to the four study ICUs during the study period, the study would have greater than 80% power for detecting a 4% absolute decrease in anticoagulant errors, a 3% absolute decrease in unplanned catheter or endotracheal tube removal, and a 10% absolute decrease in insulin treatment errors. We will use the two-sided score test for comparisons (Farrington and Manning). P levels ≤0.05 will be considered statistically significant.

Interventions In each study ICU, a physician trained in clinical research and a a clinical research assistant will prepare the study during the two weeks preceding randomization. The intervention for each AEs will include three components: meeting with the ICU nurses and physicians, distribution of educational materials including a pocket card showing the recommendations about prevention of each AE, and quality improvement sessions twice a month. At each site, the physician trained in clinical research and the clinical research assistant will lead feedback meetings with the ICU staff every 2 weeks 1) to provide feedback about AEs that occurred in the last 2 weeks, 2) to review compliance with recommendations on the pocket card, and 3) to analyze the causes of AEs and suggest practice changes if appropriate. A report on each meeting will be drafted.

Intervention for insulin administration errors

1. Meeting with the ICU staff to discuss the epidemiology of patients admitted with diabetes mellitus outside the ICU, the pathophysiology of insulin resistance in ICU patients, and clinical studies of glycemic control and mortality in wards and ICUs.
2. Pocket card showing how and why to control blood glucose and giving the guidelines for controlling blood glucose levels with insulin infusions according to the protocol in each study ICU.
3. Feedback meeting twice a month on insulin administration errors in the unit, preventability, and appropriate changes in insulin treatment procedures.

Intervention for anticoagulant administration errors

1. Meeting with the ICU staff to discuss the pathophysiology of anticoagulants (heparin and HBPM), indications, AEs, and specific problems in the ICU (e.g., use during dialysis and in patients with renal failure).
2. Pocket card showing the nature of the anticoagulants, AEs, route of administration (IV or enteral), specific problems in the ICU (e.g., use during dialysis and in patients with renal failure).
3. Feedback meeting twice a month on anticoagulant administration errors in the unit, causes of noncompliance with prescriptions, and preventability.

Intervention for anticoagulant prescription errors

1. Meeting with the ICU staff to discuss the pathophysiology of anticoagulants (heparin and HBPM), indications, AEs, and specific problems in the ICU (e.g., use during dialysis and in patients with renal failure).
2. Pocket cards with the nature of the anticoagulants, indications, dosage, route of administration, and Aes, as recommended by the French Society for Anesthesiology in 2005 in a consensus conference.
3. Feedback meeting twice a month on anticoagulant prescription errors in the unit, causes of noncompliance with recommendations, and preventability.

Intervention for unplanned extubation and catheter removal

1. Meeting with the ICU staff to discuss the epidemiology, risk factors, mortality, and prevention of unplanned extubation and catheter
2. Pocket card showing methods for securing catheters and endotracheal tubes, the sedation protocol, and the weaning protocol of each ICU.
3. Feedback meeting twice a month on occurrence of these AEs in the unit, causes, and preventability

Study population All patients admitted to the four study ICUs after January, 2007 will be included. No patients will be excluded. The study has been submitted to the appropriate ethics committee.

Data collection Patients Data are collected in the Outcomerea database. For each patient, the clinical research assistant will complete a case-report form on a computer using data capture software (Rhea) then will import all records to the Outcomerea database. The following information will be recorded prospectively: demographic characteristics (age, sex, and body mass index); underlying diseases using the Knaus classification [8]; admission category (medical, scheduled surgery, or unscheduled surgery), invasive procedures (arterial or venous central catheter, Swan-Ganz catheter, and endotracheal intubation), nosocomial infections (ventilator-associated pneumonia, nosocomial bacteremia, urinary tract infections, surgical site infections, and sinusitis), and decisions to withhold and/or withdraw life-sustaining treatments. The location of the patient prior to ICU admission will be recorded, with transfer from wards being defined as admission to the same hospital or another hospital before ICU admission. Severity of illness will be recorded at admission and once a day thereafter. Day 1 is the interval from admission to 8 A.M. on the next day; all other days are calendar days from 8 A.M. to 8 A.M. The Simplified Acute Physiology Score (SAPS II) [9] at admission and the Logistic Organ Dysfunction (LOD) [10] score are computed using the worst physical and laboratory data during the first 24 hours in the ICU. Duration of stays in the ICU and acute-care hospital and vital status at ICU and hospital discharge will be recorded.

Adverse events All AEs will be collected during all four periods, regardless of the intervention being implemented. For each AE, Rhea software will be used to collect the date and time of occurrence, clinical consequences (from a predefined list), therapeutic consequences (from a predefined list), preventability, severity, and relation with death on a five-item scale (excluded, very less unlikely, less unlikely, unlikely, likely, certainly). An open box will be available for description of the AEs if needed.

Statistical analysis A hierarchical (patients/center) logistic model will be used to evaluate the impact of each intervention. The intervention, as well as interactions between periods and specific interventions, will be taken into account.

Inclusion Criteria:

• patients hospitalised in ICU

Exclusion Criteria:

• age under 18 years