Medication Errors in the Spanish System of Communication of Incidents in Anaesthesia: Ten Year Analysis.

DESIGN AND APPROVAL This observational study was conducted between January 1, 2008, and December 31, 2017, after approval from the institutional review board (Ethics Committee of the La Fe University and Polytechnic Hospital (number) Valencia, using data from the Spanish Anaesthesia and Reanimation Incident Reporting System (SENSAR) to retrieve the medication related patient safety incidents. The methodology followed the recommendations of the Strengthening the Reporting of Observational studies in Epidemiology (STROBE) statement. The Standards for Quality Improvement Reporting Excellence (SQUIRE) were used to write the manuscript.

FRAMEWORK (study setting) All the data analysed in this study was obtained from SENSAR, which is a national, multi centre incident notification system focusing on the perioperative environment. Until December 2017, it included 104 hospitals in Spain and Chile. The incorporation of hospitals into the network was gradual and progressive since its inception in 2008. Voluntarily, anonymously and on a non-punitive basis, anaesthesiologists, resident doctors in training and anaesthesia and resuscitation nurses can communicate any security incident using a web form accessed with a generic key for each assigned centre, ensuring the anonymity of the communicator. In this database, the communication is recorded in a structured form together with a free text for the detailed description of the incident. In each hospital there is a group of professionals in charge of the anonymous and confidential analysis of the incidents based on the London Protocol (*+*), which proposes measures to avoid the repetition of incidents and gives feedback to the notifiers.

The investigators define patient safety incident as any event or circumstance in which there could be, or has been, unnecessary harm caused to a patient (*).

During the analysis process, the incidents reported are classified according to the type of incident through a classification adapted to the perioperative environment. Adverse Events that are not preventable are also included in the researchers classification (annex).

STUDY POPULATION (Subjects) (Data collection) (Observers) The study population consisted of communications and analysis of the SENSAR database between January 1, 2008 and December 31, 2017. The investigators selected incidents classified as medication incidents from the structured information included in the database at the time of notification and subsequent analysis of the incident. In incidents for which there was no classification of type of incident, two authors (ER and YS) individually reviewed each case with the reading of the free texts of the form in search of incidents related to medication. The discrepancies between the two observers were resolved by a third author (DA).

STUDY VARIABLES For the descriptive analysis of the incidents, the investigators used the following categorical variables: type of medication incident, moment in which the incident occurred, drug group involved, and morbidity generated. Regarding the independent variables, the type of medication incident was classified by omission of the administration of the drug, the administration of expired products, the mistake in the selection of drug, recommendation, concentration or points of administration; the moment in which the incident occurred in the prescription, preparation, dispensing, administration or monitoring of the drug, following SENSAR's own classification, adapted to the perioperative environment. For the analysis of the drugs involved, the investigators used their own classification (Table 1).

For the dependent variable, morbidity caused by the incident, the researchers used the WHO taxonomy that orders the morbidity degrees in the following categories: no harm, minor morbidity, intermediate morbidity, major morbidity and death (*).

STATISTICAL METHODS. The researchers use the statistical software, "R", version 3.4.1, to perform the analysis of the sample obtaining, in absolute numbers and percentages, the frequency with which the different types of reported incidents have occurred. To determine the existence of a population-level relationship between the different independent variables used and the damage produced in the patient, the researchers used the Kruskal-Wallis non-parametric test with a level of significance of 5%. The result of this test will be expressed with the obtained contrast statistic and its p-value. To identify the factors that best estimate the risk of producing different degrees of morbidity in the patient, the investigators generated a multiple ordinal logistic regression model. The effects presented in this case will be the estimator, the standard error, the "odds ratio", the confidence intervals at 95% and the p-value for a significance of 5%.