Drug administration errors and their determinants in pediatric in-patients

Abstract

Objective. To quantify the type and frequency of drug administration errors to pediatric in-patients and to identify associated factors.

Design. Prospective direct-observation study of drug administration errors from April 2002 to March 2003.

Setting. Four clinical units in a pediatric teaching hospital.

Study participants. Twelve observers accompanied nurses giving medications and witnessed the preparation and administration of all drugs to all patients on all weekday mornings.

Intervention. None

Main outcome measure. Discrepancies between physicians’ orders and actual drug administration.

Results. During the 1719 observed administrations to 336 patients by 485 nurses, 538 administration errors were detected, involving timing (36%), route (19%), dosage (15%), unordered drug (10%), or form (8% form). These errors occurred for 467 (27%) of the 1719 administrations. Intravenous drugs (OR = 0.28; CI = 0.16–0.49; versus miscellaneous) were associated with fewer errors. Error rates were higher for cardiovascular (OR = 3.38; CI = 1.24–9.27; versus miscellaneous) and central nervous system drugs (OR = 2.65; CI = 1.06–6.59; versus miscellaneous); unspecified dispensing system (OR = 2.06; CI = 1.29–3.29; versus store in the unit); nonintravenous nonoral administration (OR = 4.44; CI = 1.81–10.88; versus oral administration); preparation by the pharmacy (OR = 1.66; CI = 1.10–2.51); and administration by a hospital pool nurse, temporary staffing agency nurse, or nurse intern (OR = 1.67; CI = 1.04–2.68; versus registered full-time nurse). Each additional management procedure in the patient increased the risk of error (OR = 1.22; CI = 1.01–1.48).

Conclusions. The risk factors identified in our study should prove useful for designing preventive strategies, thereby improving the quality of care.

Medication errors have become a major public health concern [1,2]. In-depth studies have been done in adults, most notably in intensive care units (ICUs) [3,4]. Many drugs used in pediatric units are either unlicensed (7–10%) or used off-label (18–64%) [5–7]. The need to prepare dilutions or to open capsules may increase the risk of drug administration errors [8–10]. When computing drug doses, nurses and doctors may make mistakes, which may be life threatening [11–13]. In pediatric units, the physiological immaturity and widely variable body weight of the patients may increase the risk and impact of errors. Kaushal et al. [14] reviewed 10,778 medication orders and identified 616 errors (5.7%). In a retrospective review of medication errors in a pediatric hospital over a 5-year period, Ross et al. showed that nurses were responsible for most of the reported errors (59%) [15].

Of the few direct-observation studies in pediatric in-patients, none investigated the factors specifically associated with drug administration errors. The objective of this study was to identify the type, frequency, potential clinical significance, and determinants of drug administration errors using direct observation in pediatric in-patients.

Materials and methods

Setting

The study was conducted in four pediatric units at a pediatric teaching hospital in Paris, France (440 pediatric and 60 maternity beds): a pediatric intensive care unit (PICU), a neonatal intensive care unit (NICU), a pediatric nephrology unit, and a general pediatric unit. Each of these four units has a fifth-year pharmacy student. Medication orders are written by physicians on a computer using Patient Care System software (PCS®, IBM®, NY, USA). For each patient and each day, a Medication Administration Report (MAR) listing all drugs to be given, with the modalities of preparation and administration, is printed out. If needed, the physician adds instructions to the MAR by hand. The hospital pharmacy dispenses drugs to each unit, on a unit-dose basis. If needed, the nurses take medications from stores available in each unit.

Study design

Undisguised direct observation was used to detect drug administration errors [16]. Nurses and physicians were informed about our objectives. The observers were the 12 fifth-year pharmacy students who rotated in the four study units during the study period and the pharmacy resident. They watched the nurses prepare and administer medications during two consecutive hours in the mornings on the 271 weekdays (weekends excluded) from April 2002 to March 2003. The observers received training by assisting the pharmacy resident in their unit during routine data collection duties, for 8 hours or more. Their performance was then assessed by the pharmacy student and a senior pharmacist, who checked the consistency of the data entered by the observers on the data collection forms. All collected data were reviewed by the pharmacy resident for missing information and inconsistencies. The following data were collected: characteristics of the unit nurse status (unit nurse, pool nurse, or temporary agency nurse); time the nurse had been working in the unit; nurse-to-patient ratio; and nurse workload (number of patients under the care of each nurse, whether a patient was admitted on that day, and number of patients on intravenous infusions for each nurse), characteristics of each patient under the nurse’s care (demographic data, number of daily drug doses, nature of management procedures, and number of infused drugs), and characteristics of the drugs (Anatomical Therapeutical Chemical code, prescription medium, whether licensed for children, dispensing system, preparation by the pharmacy department, route, first administration, pharmaceutical form, and dosage). The observers assigned consecutive numbers to each nurse-observation period (2 hours in the morning) in chronological order. All the nurses in each of the four study units participated in the study. Observers were to intervene only if they detected a potentially life-threatening error. The French National Committee on Informatics Technology and Freedom (Commission Nationale d’Informatique et Liberté) approved this observational study.

Errors

A drug administration error was defined as any discrepancy between printed or handwritten physicians’ orders and drug delivery to the patient, in keeping with the classification developed by the American Society of Hospital Pharmacy [17]. Administration errors were classified into 10 categories: timing errors (greater than 1-hour difference compared with the ordered time), omission, unordered drug, wrong generic drug, wrong dosage, wrong formulation, wrong route, deteriorated drug, technical error in preparation or administration (e.g. wrong infusion flow rate or wrong diluent), and extra dose. A panel composed of two physicians, two pharmacists, one nurse, and one epidemiologist evaluated the seriousness of each error by indicating what their response would have been: no response (decisions normally left to nurses), minor corrective action (discussion with the nurse or telephone call to the pharmacy), additional investigations or monitoring, major treatment modification, or action to eliminate factors contributing to a life-threatening error. When several errors were made during the same administration, error seriousness was assessed based on the most serious error.

Statistical analysis

Qualitative variables were described as frequencies (percentages) and quantitative variables as medians (interquartile range). The frequency of errors (fE) was the number of administrations with one or more errors (nE) divided by the sum of the number of observed drug administrations (nA), and the number of omitted drug administrations (nO) [18].

The sum nA + nO was the total number of opportunities for error [19]. First, we used a bivariate logistic regression model to investigate the relation between error occurrence and drug characteristics. Second, we built a hierarchical logistic regression model with the administration profile at level 1 and the nurse profile at level 2 [20]. Third, we built a second hierarchical logistic regression model with the administration profile at level 1 and the patient profile at level 2. Variables with P < 0.2 were entered into the models and those with P < 0.05 for the deviance comparison were kept in the models. Results are expressed as odds ratios (OR) with the 95% confidence interval (CI). Data analysis was performed using SAS® 8.12 (SAS Institute, Cary, NC, USA) and multilevel regression models were built using MlwiN® 1.1 (Multilevel Models Project, Institute of Education).

Results

During the study period, there were 485 nurse-observation periods, corresponding to 648 patient-days and to 1719 opportunities for error. Table 1 lists the characteristics of the four units and Table 2 those of the 336 patients. At least one management procedure was used in nearly 100% of PICU patients, 54% of NICU patients, 27% of nephrology unit patients, and only 7% of general unit patients. Opportunities for errors (n = 1719) are reported in Table 3. Drug categories were gastrointestinal (n = 615, 36%), anti-infective (n = 339, 20%), blood (n = 250, 14%), central nervous system (n = 158, 9%), hormonal (n = 106, 6%), cardiology (n = 92, 5%), respiratory (n = 81, 5%), immunity (n = 30, 2%), and miscellaneous (n = 48, 3%). One hundred and forty-five (43%) patients experienced at least one error, committed by 190 (39%) nurses. Of the 1719 opportunities for error, 467 led to at least one error (error rate = 27.2%, Table 4). There were 538 administration errors overall (error rate, 31.3%). When timing errors were excluded, the error rate was 302/1719 (17.6%). Errors are described in Table 5. Most route errors consisted in administration through a nasogastric catheter instead of by mouth and were ascribable to absence of ‘nasogastric catheter’ among the options in the PCS® software. Table 4 summarizes the seriousness of observed errors. No potentially life-threatening errors were witnessed. Excluding the 144 errors that required no response eliminated 77% of the timing errors and decreased the error rate to 19%. In the administration-nurse profile model, administration by a nurse intern, temporary staffing agency nurse, or pool nurse increased the risk of error (OR = 1.67; P = 0.03), whereas the number of drugs administered through infusion per nurse was not a significant risk factor (OR = 1.04; P = 0.25) (Table 6). In the administration-patient profile model, the only significant risk factor was the number of management procedures (OR = 1.22; P = 0.04). No random observer effect was found.

Discussion

To our knowledge, this is the largest direct-observation study conducted in pediatric units and including identification of factors associated with drug administration errors. The study lasted nearly 1 year and involved a large number of observers. Although Tissot et al. [4] reported 2009 preparation and administration processes, these corresponded to only 568 administrations [21]. Calabrese et al. [22] performed nearly 6000 observations but confined their study to selected drugs. Other studies [18,19,23–25] included fewer than 600 observations. We chose undisguised observation [4,19,25–28] instead of the disguised technique [18,22,24,28] because observation is more efficient, objective, and reliable than spontaneous reporting [3,29,30] or patient chart review.

The error rates (31.3% overall and 20.5% without timing errors) were similar to those found by Tisdale (24.8 and 8.8%) [24], Tissot (23.2 and 21.7%) [4,21], and Schneider (26.9 and 18.2%) [19]. In a chart review study [26] conducted in the pediatric nephrology unit that participated in our study, the administration error rate was 23.5% overall and 11.7% excluding timing errors. The higher rate in our study is ascribable to the greater efficiency of direct observation in detecting errors, as compared with chart review. In agreement with previous data [18,19,24–26], we found that timing mistakes were the most common type of error.

Interestingly, the only protective factor was intravenous administration, in contradiction to the widely held belief that errors are less likely to occur with the oral route. On the opposite, the transcutaneous, transmucosal, ophthalmic, rectal, nasal, vesical, and auricular routes were associated with a higher likelihood of error compared with the oral route. These routes are often considered less likely to induce toxicity, but on the other hand are less often used than the oral and intravenous routes. Other risk factors for errors were medications found in the patient’s room (and therefore having an unidentified dispensing source and suboptimal storage conditions) and preparation of medications by the pharmacy department to obtain small unit-doses from adult forms. This latter finding adds urgency to pleas that the pharmaceutical industry produce dosage forms suitable for pediatric patients. Errors were less likely to be committed by registered nurses working full-time in the unit than by other categories of nurses: thus, lack of training and limited familiarity with pediatric drug administration probably increased the risk of error. A larger number of management procedures was associated with a higher risk of error. Patients with numerous management procedures were characterized by greater disease severity and a larger number of drug administrations, resulting in more opportunities for errors.

Two previous studies of risk factors for drug administration errors were conducted by van den Bemt et al. [18] and Tissot et al. [25] in adult in-patients. Tissot et al. [25] examined a small number of potential risk factors in a cardiovascular thoracic surgery unit and a geriatric unit: incomplete or illegible prescription was a significant factor in the cardiovascular unit and greater nurse workload in both units. van den Bemt et al. [18] confined their study to ICUs and did not evaluate the potential impact of nurse characteristics; furthermore, they analysed the patient characteristics without considering the hierarchy of the data (drug, patient, nurse). They found more errors for gastrointestinal drugs and fewer errors for blood products and cardiovascular drugs. They also found an increased risk of error with administrations performed on Mondays. We did not include day of the week in our model because this might be a confounding factor and because staffing patterns were similar on all days.

Our study has two potential limitations. First, the nurses were observed only in the morning, which is the busiest time for nurses, and only on weekdays. However, staffing patterns were identical on weekends and weekdays, so that excluding the weekends probably had little effect on our findings, and Donchin et al. [3] and van den Bemt et al. [18] showed that errors were more common during the day than at night. Second, we did not seek to evaluate the impact of errors on clinical outcomes and therefore were unable to use the error severity classification system developed by the National Coordinating Council for Medication Error Reporting and Prevention (NCCMERP) [31] and used by others [4,18,22]. Nevertheless, Dean et al. [32] demonstrated that scoring the seriousness of medication errors without knowledge of patient outcomes is reliable and valid. Importantly, the panel that evaluated error seriousness in our study estimated that one fifth of the errors would have led to major treatment modifications or to additional investigations or monitoring. According to the ‘five rights’, nurses must give the right drug to the right patient in the right dose by the right route at the right time [31] We suggest a number of preventive measures for pediatric wards in Table 7, in accordance with previous guidelines [29,31,33–35].

Conclusion

Our study shows that further evaluation of drug preparation and administration errors is urgently needed. Indeed, using a well-established data collection technique and standardized definitions, we found that errors were common. The risk factors identified in our study should prove useful for designing preventive strategies, thereby improving the quality of care.

Acknowledgements

This work was supported by an educational grant from PHRC RDB/AOM00056, Health Ministry, France. This study was approved by the head nurse Marie-Françoise Houdais. We thank Aurélia Goron, Céline Schwartz, Sandrine Masseron, Morgann Wehrel, Diane Testard, Aurélie Parlier, Rafaël Gandolfi, Nicolas Janus, Manuel Bukudjian, Clément Bouquet, and David Boukobza for collecting the data. Thanks to Antoinette Wolfe for her writing assistance.

References

Author notes

¹Pharmacy, ²Public Health, ³Neonatal Intensive Care Unit, ⁴Nephrology Unit, ⁵Intensive Care Unit, and ⁶General Paediatrics Unit, Hôpital Robert Debré AP-HP, Paris, France