Prevalence and Nature of Medication Administration Errors in Health Care Settings

A Systematic Review of Direct Observational Evidence

Richard N Keers MPharm; Steven D Williams MPhil; Jonathan Cooke PhD; Darren M Ashcroft PhD


The Annals of Pharmacotherapy. 2013;47(2):237-256. 

In This Article

Literature Search Results

Search Process

The electronic database search yielded 20,222 publications for review. Title examination subsequently excluded 19,392 articles (including 2907 duplicates) that were not relevant to preventable medication safety topics. The remaining 830 publications underwent abstract review, and a further 708 were excluded because they did not involve medication administration error research. Full texts for 122 remaining studies were obtained for detailed analysis (Figure 1). Of these, 61 did not meet the inclusion criteria, with the main reasons being review articles (n = 22) and different data collection method (n = 20). Two studies were excluded because only the clinically significant MAE rate could be extracted.[29,30] Examination of the reference lists of relevant review publications and all included studies yielded a further 27 articles for inclusion, resulting in a total of 88 that contained a total of 91 unique studies (4 studies were carried out in multiple countries, giving 8 unique studies[31–34] and 1 study expanded on previously published data and was not considered a unique study).[35,36]

Figure 1.

Summary of systematic process used to apply exclusion/inclusion criteria and select publications for inclusion.

Study Characteristics

Country and Publication Year Twenty-five (27.5%) of the unique studies originated from the US,[17,31,37–59] 22 (24.2%) from the UK,[26,31–34,60–76] 8 (8.8%) from Australia,[35,36,77–83] 6 each (6.6%) from France[33,84–88] and the Netherlands,,[89–94] 5 each (5.5%) from Germany[32,34,95–97] and Spain,[98–102] 4 (4.4%) from Canada,[103–106] and 3 (3.3%) from Malaysia.[107–109] The remaining 7 studies originated from Belgium,[110] Iran,[111] Brazil,[112] Denmark,[113] Malawi,[114] Switzerland,[115] and Turkey.[116] Thirty (33%) studies were published before 2000. A summary of key data extracted from each study is shown in Table 1 .[117,118]

Study Setting Eighty-one (89%) studies were carried out in hospitals (including academic/teaching, tertiary, community, general, army) and 10 (11.0%) in long-term care facilities (including nursing/care/assisted living homes). One study was conducted in a variety of full-time and day care units for individuals with intellectual disability.[91] Of the 81 studies conducted in hospitals, 7 (8.6%) were carried out in pediatric hospitals, 26 (32.1%) in general or unspecified hospitals, 46 (56.8%) in academic/university/teaching/tertiary hospitals, 1 (1.2%) in a mental health institution,[72] and 1 (1.2%) in an army medical center.[51] Five studies considered errors from both academic and nonacademic hospitals;[26,33,43,75,101] the group into which these studies were placed was determined by choosing the type of hospital most commonly studied or within which the most wards were investigated.

Number of Study Settings Of the 91 studies, 26 (28.6%) were carried out in 2 or more institutions. The remaining 65 (71.4%) were conducted in 1 site. Ten studies (11.0%) were conducted at 5 or more sites.[17,42,43,47,50,58,74,75,91,101] Of these, 7 (70.0%) were conducted in long-term care facilities.[17,42,47,50,58,74,91] Two studies stated that multiple sites were used but did not specify the number.[39,41]

Patient Demographics Fifteen studies (16.5% of all studies) were carried out solely in pediatric settings (including 3 studies involving only neonates)[38,51,61,62,71,75,86,93,97,104,105,108,109,115,116] 5 (5.5% of all studies) were carried out in both adult and pediatric units,[26,91,98,100,112] and 31 (34.1%) were conducted in adult units only; 40 (44.0%) studies did not state the age range of the patient populations.

Route of Administration Sixty-one (67.0%) studies considered all routes of administration. Twelve (13.2%) studies investigated MAE associated only with the intravenous route (1 studied parenteral errors),[69] and 1 study[36] (not included in median calculations) reported additional intravenous error data from an earlier paper.[35] The remaining 18 (19.8%) studies evaluated different administration routes such as oral, inhaled, and parenteral, which were often in combination.

Study Design Nurses were observed administering medications in most studies (n = 70, 76.9%). Nurses were observed along with nursing support in 7 (7.7%) cases[42,47,49,58,94,102,110] or medical staff in 6 (6.6%)[26,32,61,107,109,114] cases. Both nurses and parents were observed in 1 study.[97] Pharmacy technicians were observed administering medication in 1 publication.[37] The remaining 6 (6.6%) studies observed a range of nursing support or long-term care staff.[17,50,74,91,92,112]

Study size varied from 100[55] to 7000[42] OE observed. Twenty-four (26.4%) studies were carried out in a before and after format, whereby established systems were evaluated and reevaluated after an intervention. Of these, 8 (33.3%) studies assessed the impact of educational interventions,[46,54,70,93,97,98,108,110] 5 (20.8%) evaluated bar-code administration,[45,48,52,53,57] and the remainder assessed a mixture of electronic system and drug distribution changes. Nine (9.9%) studies tested a traditional system against an alternative concurrently.

Error Detection Pharmacists were the most common data collectors, featuring as the sole data-collecting profession in 36 of 91 (39.6%) studies. Nurses were used in 14 (15.4%) studies as the only type of observer, pharmacy students in 5, pharmacy technicians in 3,[33,92] and medical[108] or nursing students,[116] each in 1 study. In 15 (16.5%) studies, the observer's professional status was not specified. The remaining 16 (17.6%) studies used a combination of professions or agencies/departments.[17,35,38,39,41,43–45,52,53,71,78,94,102,110,112]

Forty-one (45.1%) studies confirmed an error by comparing observations to the physician's order/nurse administration record after the observation period, whereas 29 (31.9%) of the studies involved concurrent error identification. Twenty (22%) studies did not categorically state whether errors were confirmed while observing or after the observation period. One study appeared to describe a process of error determination both during and after the observation period.[77]

Validation of Errors Forty-seven (51.6%) studies reported that more than 1 observer was used to collect data. Of these, 14 (29.8%) reported some form of assessment to determine the consistency of observations recorded between the observers. The assessments included simulated and practice-based examples.[35,43,49,50,53,81,86,90,105,111] Once observation data had been collected, only 17 (18.7%) studies used a process of validation to confirm the presence of a MAE. In the majority of cases (12 of 17), this involved a pharmacist.

Fifty-one (56%) studies attempted to determine the clinical significance of errors that were reported. Three of these[60,81,111] gave only general statements on intervening in cases of harm without explicit assessment criteria and were excluded from the analysis. Two further studies[56,99] contained severity data that could not be attributed to the administration stage of the medication use process and were subsequently excluded from the analysis, leaving 46 (50.5% of all unique studies). Of these, 38 (82.6%) reported severity scales based on methods previously published, and 8[58,62,67,70,77,86,98,110] appeared to use their own criteria. Nine of the 38 (23.7%) studies[41,44,88–92,100,101] reported using the National Coordinating Council for Medication Error Reporting and Prevention (NCC MERP) criteria,[119] 6[26,68,73,74,93,95] based assessment on the method proposed by Dean and Barber,[120] and 5[57,61,84,85,114] based their criteria on the work of Folli et al.[121] Eighteen (18/38, 47.4%) studies used other published criteria for severity assessment.[32,34,35,43,50,59,64,72,80,87,96,102,104,107,109,113] In 34 studies at least 2 assessors were used to determine error severity, 1 assessor was used in 2 studies,[70,84] and in 9 cases, the number of individuals involved in the assessment was not specified.[32,41,61,77,80,101,104,114] Various grades of nurses, physicians, pharmacists, and researchers were used in 32 of the 34 studies reporting 2 or more severity assessors. Common arrangements were use of 2 or more pharmacists (6 studies),[34,89–92] physicians and pharmacists (4 studies),[85,107,110,113] and physicians, nurses, and pharmacists (11 studies).[26,50,58,62,64,68,73,87,93,95,98]

Definition of Medication Error

Fifty-eight studies (63.7%) used 1 or more established MAE definitions/categorizations to either supplement their own or for use in their exact form. The most commonly used of these established definitions were that of Allan and Barker[7] (n = 24) or the ASHP (n = 22).[27,122] The remaining studies either developed their own (n = 28) without referencing published criteria or gave no definition.

Prevalence of Administration Errors

Denominator and Numerator Most studies (n = 82, 90.1%) used a denominator that was considered to meet (or was converted to)[104] the definition of TOE. Of these, 57 of 82 (69.5%) studies reported rates using only the OME denominator, whereas 10 (12.2%) were considered to have used TNE. Fifteen of 82 (18.3%) studies reported sufficient data to allow MAE rate calculations using both numerators. The median error rate (IQR) was 19.6% (8.6–28.3%) of TOE including timing errors (based on 52 studies) and 8.0% (5.1–10.9%) without timing errors (based on 58 studies) using OME. The median error rate was 25.6% (20.8- 41.7%) of TOE including timing errors (based on 22 studies) and 20.7% (9.7–30.3%) without timing errors (based on 23 studies) with TNE. Nine studies reported various denominators that could not be converted to TOE, and 6 used separate denominator values for preparation and administration phases;[32,75,111–113] differences in their definitions precluded comparative grouping. A summary of key results is shown in Table 2.

Country of Origin

OME numerator. The median error rate (IQR) in the US was 18.8% (4.9–23.5%) including WTE (n = 17) and 7.4% (5.2–9.8%) without (n = 18). In the UK this median was 21.7% (6.4–35.9%) including timing errors (n = 6) and 5.5% (3.7–7.7%) without WTE (n = 14), and in Australia the median MAE rate was 17.0% (8.3–18.3%) including timing errors (n = 5) and 8.9% (7.4–10.9%) without (n = 5). In the Netherlands the error rate was 23.3% (19.9–39.8%) (n = 6) and 19.9% (15.1–25.1%) excluding WTE (n = 4); in France this was 23.6% (18.7–27.3%) including timing errors (n = 4) and 11.1% (9.3–14.4%) without (n = 3). More limited data from Canada gave a median of 28.4% (26.1–31.1%) including WTE (n = 3) and 8.4% (5.9–9.8%) without (n = 3); in Spain the error rate was 4.7% (3.7–6.5%) including WTE (n = 4) and 6.9% (5.1–8.8%) without timing errors. In Germany the error rate was 25.8% (14.7–36.9%) excluding WTE (n = 2).

TNE numerator. The median error rate (IQR) in the US was 33.8% (19.3–67.3%) including WTE (n = 4) and 14.7% (7.1- 49.3%) without (n = 4). In the UK this median was 52.3% (25.7–99.6%) including WTE (n = 4) and 26.6% (25.4–57.9%) without WTE (n = 5). In the Netherlands the error rate was 39.4% (25.8–52.4%) (n = 4) and 29.3% (22.5–37.8%) excluding WTE (n = 4); in France the median was 25.9% (23.0–29.3%) including WTE (n = 4) and 19.4% (15.9–20.5%) without WTE (n = 4). In Australia the error rate was 9.5% (6.1–13.0%) including timing errors (n = 2).

Study Setting Studies carried out in hospitals reporting TOE (n = 72) had an MAE rate of 19.1% (10.0–27.8%) including WTE (n = 44) and 7.5% (4.8–10.8%) without errors in timing (n = 49) using OME. With TNE, the MAE rate was 24.2% (19.4–36.3%) including timing errors (n = 20) and 20.1% (9.0–33.9%) without WTE (n = 21). The error rate was 23.3% (7.9–29.8%) including WTE for studies carried out in long-term care institutions (n = 8) and 8.4% (7.9–17.3%) without timing errors (n = 9) using OME. With TNE, these error rates were 36.4% (32.0–40.7%, n = 2) and 22.9% (22.0–23.7%, n = 2) with and without timing errors, respectively.

Patient Demographics Of the 15 studies that involved only pediatric patients, 12 used the TOE denominator. The median error rate was 26.9% (17.4–33.8%) including WTE (n = 11) and 8.4% (5.0–17.6%) without (n = 9) using the OME numerator, whereas the median error rate was 41.2% (26.5–79.0%) including timing errors (n = 4) and 34.8% (17.2–58.2%) without (n = 4) using TNE. Thirty-one studies were carried out with only adult patients; 29 used a denominator mapping onto TOE, giving a median error rate of 19.6% (4.7–27.8%) including WTE (n = 20) and 8.1% (5.5–15.7%) without timing errors (n = 18) with OME, and 25.9% (22.5- 45.1%) including timing errors (n = 9) and 20.9% (16.9–31.8%) without WTE (n = 10) using TNE.

Route of Administration Of the 12 studies that examined only intravenous MAE, 10 used denominators falling within the TOE definition. The median error rate was 85.9% (81.8–89.9%) with WTE (n = 2) and 48.0% (45.0–48.5%) without WTE (n = 3) using OME. The median error rate was 78.6% (51.9–120.6%) including timing errors (n = 3) and 53.3% (26.6–57.9%) without timing errors (n = 5) using TNE. Of the 61 studies observing all routes of administration, 56 used the TOE denominator, giving a median error rate of 19.4% (8.6–27.8%) including WTE (n = 40) and 8.2% (5.5–10.8%) without timing errors (n = 41) when OME was used, and a median error rate of 24.6% (17.4–30.8%) including WTE (n = 18) and 20.1% (9.0–24.6%) without timing errors (n = 17) using TNE.

Types of Administration Error Reported

Timing Errors Definitions of WTE varied among the 69 studies that reported one, with 6 of 69 (8.7%) using ±30 minutes from the prescribing administration time, 37 (53.6%) ±60 minutes, 1 ±75 minutes,[93] 8 (11.5%) determining timing error based on other criteria, and 17 (24.6%) not stating the timing threshold for MAE. Of the number of studies reporting the number of timing errors (n = 64), this was the most common error subtype, being reported in 51 (79.7%) studies as 1 of the 3 most common error subtypes identified. Thirty-five (54.7%) studies that reported WTE reported omission errors and 30 (46.9%) wrong dosage errors as 1 of their 3 most common error subtypes. Of the number of studies not including timing errors (n = 13), omission (n = 10, 76.9%), wrong dosage (n = 9, 69.3%), and unauthorized drug (n = 7, 53.8%) were most often reported as being 1 of the 3 most common subtypes.

Study Setting Seven studies conducted in long-term care institutions contained sufficient data for subcategory analysis. All but 1[17] studied timing errors. Four each reported that omission and wrong dosage followed by 3 each reporting that wrong administration technique (eg, errors in enteral feeding tubes and inappropriate dose form manipulation[91,92] ) and WTE were MAE subtypes among the 3 most common that they identified. Two studies reported wrong preparation as 1 of their 3 most commonly observed subtypes. Seventy studies carried out in hospitals contained data on error subcategories. Of those that studied WTE (n = 58), 48 (82.8%) reported that this error subtype was among the 3 most commonly reported. Thirty-two (55.2%) reported that omission errors and 26 (44.8%) reported that wrong dosage were among the 3 most common. Of publications that did not study timing errors (n = 12), 10 reported omission errors as 1 of their 3 most common error subtypes, followed by 8, which reported wrong dosage and 6 unauthorized drug.

Patient Demographics Thirteen of the 15 studies carried out exclusively in pediatric patients underwent error subcategory analysis. All included WTE and, of these, 12 of 13 (92.3%) reported this as 1 of the 3 most common error subtypes identified. Six each (46.2%) reported wrong preparation and wrong dosage as 1 of the 3 most common error subtypes observed. Four studies each reported wrong administration technique and wrong administration rate as 1 of their 3 most commonly reported error subtypes. Twenty-three of the 31 studies reporting MAE only in adults underwent error subcategory analysis; 22 included timing errors. Wrong-time errors were the most frequently reported among the 3 most common error subtypes, being listed by 68.2% of the studies. Omission (59.1%) and other error (40.9%) were the next most common error subtypes reported, with the latter often involving errors in labeling and documentation.

Route of Administration Twelve studies investigating MAE associated with the intravenous route were analyzed; 6 of these studied WTE, with all reporting this error among their 3 most common error subtypes. All 6 also reported wrong preparation as 1 of the 3 most common. Three studies (50%) reported wrong administration rate as 1 of the 3 most common error subtypes. The remaining 6 studies not considering timing errors reported wrong administration rate and dose omission (n = 5 each, 83.3%) followed by wrong preparation and wrong dose (n = 3 each) as among their 3 most common MAE subtypes. Fifty-two of the 61 studies observing all routes of administration had error subtype data analyzed. The findings reflected the overall results; wrong-time, omission and wrong dosage errors were identified as frequently occurring MAE.

Severity of Administration Errors

Of studies that reported NCC MERP[119] as their criteria to determine harm (n = 9), all but 3[44,88,101] used a modified version. Three studies determined potential harm;[88,90,101] for the remainder, it was unclear whether potential or actual harm was assessed, as no process of patient follow-up was described. Of the comparable NCC MERP-derived data reported from 6 of the 9 studies, severity categories C (error reaches the patient, no harm caused) and D (error reaches the patient, requires monitoring and/or intervention to preclude harm) were most commonly reported, affecting between 45–85% and 2.7–55.1% of errors made, respectively. The remaining categories, B (error does not reach the patient), E (error may have contributed to or resulted in temporary harm requiring intervention), and F (error may have contributed to or resulted in temporary harm requiring initial or prolonged hospitalization) were reported less frequently and found to affect 11.2%,[41] 1.1–9.1%,[41,100] and 1.6%[100] of reported errors, respectively. One study reported no harm.[44]

Two[84,85] of the 5 studies that included criteria defined by Folli et al.[121] used a similar modified version and reported that between 10% and 21% of errors were potentially life-threatening, 26- 42% were potentially clinically significant, and 37–64% were of minor consequence. Data from the remaining 3 studies that did not appear to modify their criteria suggested that 3.3–8.9% of errors made were classified as clinically significant.[57,61] Three studies collected potential severity data based on MAEs that either reached the patient[57,84] or were intercepted and considered clinically significant.[85] Two studies did not specify whether actual or potential harm was reported.[61,114]

Because of differences in application of the severity scale categories and number of assessors the studies utilizing Dean and Barber[120] criteria were not directly comparable. The potential harm associated with intercepted errors that were thought to cause patient harm was determined by 3 studies. Three of the 5 studies reported a mean harm score; this varied from 1.8 to 2.7 depending on the study in question.[68,73,74] Three studies reported a breakdown of the percentages of errors falling into different categories;[26,93,95] potentially severe consequences were recorded in 0.6–6.2% of all errors, moderate for 57.2–60% of all errors, and minor for 33.8- 42.1% of all errors.

Medication Associated with Administration Errors

Ten studies[41,52,54,59,60,77,87,96,104,111] reported how often medications or medication classes were associated with the MAE reported. Two studies[41,111] included a limited list of medications and others[54,96] did not report a complete dataset of which medications were associated with MAEs. The data from these 10 studies were grouped according to therapeutic classification. Five (50.0%) of these studies each listed medications or medication groups belonging to the categories nutrition and blood, gastrointestinal system, and cardiovascular system as being 1 of the 3 most commonly observed medication types associated with MAEs. Four (40.0%) studies each reported medications belonging to the categories infections and central nervous system as 1 of their 3 most commonly observed medication types involved with MAEs.