Preventable Adverse Drug Events Among Inpatients: A Systematic Review

Studies published from 2000 to December 2017 were identified by searching 5 electronic bibliographic databases (Cumulative Index of Nursing and Allied Health Literature, Medline, Scopus, the Cochrane Library, and Embase). The search strategy (Supplemental Tables 4 through 8) included terms for medication adjacent to terms for error found in conjunction with terms for pADEs or harm, with results limited to pediatric hospital inpatients (defined as aged <18 years). Studies had to be in the English language, and researchers had to use chart review or observation to detect medication errors. Because of the recognized underreporting, we excluded studies in which researchers used only voluntary incident reports.^14,–16 Included hospital wards were emergency departments (EDs), ICUs, and general wards (pediatric wards from general hospitals and general wards from pediatric hospitals). Additional exclusion criteria are presented in a Preferred Reporting Items for Systematic Reviews and Meta-Analyses flow diagram (Fig 1).

Two reviewers (P.J.G. and S.A.M.) independently assessed all full-text articles to determine eligibility and rate the quality of published studies (Supplemental Tables 9 through 11). The Newcastle-Ottawa Scale for nonrandomized studies¹⁷ and the Critical Appraisal Skills Program Cohort Study Checklist¹⁸ were employed in addition to the Checklist Assessment of Medication Error Audits, a purpose-built measure that is focused on incident identification methods. Studies identified as “good” through the Newcastle-Ottawa Scale received a point on the overall quality rating, as did those achieving an above average score on the Critical Appraisal Skills Program and the Checklist Assessment of Medication Error Audits (for a possible total quality score of 3 out of 3). Studies were rated as “excellent” quality (3 points), “good” (2 points), “fair” (1 point), and “poor” (no points). Disagreements between reviewers were resolved by discussion.

We defined an ADE as injury or other patient harm resulting from the use of a drug. ADEs were further categorized as being preventable when due to error or not preventable when a drug was used as clinically indicated but the patient suffered an adverse drug reaction (ADR).¹ Because ADRs do not involve actual preventable patient harm, they were excluded from this review but have been reviewed elsewhere.¹⁹ 

Extracted information included medication error and harm definitions, pADE incidence and severity rates, items required for quality assessment, and details of the study sample. In a total of 8 studies, researchers rated pADE severity by using the following scales: National Coordinating Council for Medication Error Reporting and Prevention 9-point rating²⁰ (4 studies^13,21,–23), Overhage and Lukes’ 5-point rating²⁴ (2 studies^12,25), the 5-point Severity Assessment Code²⁶ (1 study²⁷), and the American Society of Hospital Pharmacists 5-point rating²⁸ (1 study²⁹). To allow comparison between studies, these severity scales were collapsed into 3 categories: minor (including harm described to be “insignificant” or “minimal” to “significant” or “temporary” harm); moderate (including “serious” or “permanent” harm); and severe (including “life-threatening” or “fatal” harm).

Study heterogeneity rendered meta-analysis inappropriate. Studies varied by prescribing system (paper chart or CPOE), ward, error type, and denominators (with pADE incidence reported as a proportion of medication error per 100 admissions or patients, per 100 orders, per 1000 patient days, or per 1000 drug administrations).

Twenty-five articles met the inclusion criteria,^{12,13,21,–23,25,27,29,–46} representing 22 unique studies (details shown in Table 1). Eight studies were rated to be of excellent quality,^{12,13,21,38,40,41,44,45} 5 were good quality,^{22,34,36,39,43} 4 were fair,^25,27,30,32 and 5 were poor quality.^{23,29,31,37,42} Most articles (19 of 25) were published after 2006. There was no evidence to suggest that study quality improved over time.

The definitions and range of medication errors varied greatly across studies. Importantly, prescribing errors, particularly dosing errors, were consistently identified to be the most likely medication error found to cause pADEs.^* Despite small variations in wording, our definition of pADE was used across most studies, with the exception of a single study in which the authors assessed “serious medication errors” to include pADEs and nonintercepted errors found not to cause harm.⁴⁵ 

The incidence of pADEs and proportion of medication errors causing pADEs is outlined for individual studies in Table 2 and summarized by hospital ward in Table 3.

A total of 5 US studies,^{12,36,44,–46} 1 Argentinian study,⁴⁰ and 1 New Zealand study³⁸ contained reports of incidences of pADEs across multiple pediatric wards (both general and intensive care wards). Each study was rated to be good to excellent quality. Excluding the Argentinian study, which revealed no actual patient harm,⁴⁰ the incidence was between 1.8 to 74.0 pADEs per 1000 patient days and 0.5 to 3.8 per 100 admissions; in 1 study, researchers reported 0.1 per 100 orders. Severity of pADEs across multiple pediatric wards was reported in only 1 study¹² in which 5 pADEs resulted from 616 medication errors, with 1 rated minor and 4 moderate severity. In the 4 studies that contained reports of the proportion of medication errors causing a pADE, 59 (2.6%) of a combined 2275 errors resulted in a pADE.^35,38,40,44 

Excluding a single study rated to be of poor quality,³¹ the incidence of pADEs in general pediatric wards was reported in 5 studies rated as good to excellent quality.^{13,27,36,40,45} Across 3 studies, there were between 12.2 and 17.0 pADEs per 1000 patient days, 1.5 per 100 orders, and between 3.6 and 7.3 per 100 admissions.^13,27,45 The fourth study included all pADEs divided by the number of medications delivered in the unit and contained a report of 2.1 pADEs per 1000 patient days.³⁶ The remaining study of an Argentinian general ward revealed no actual patient harm.⁴⁰ 

The authors of a single study conducted in the general ward of an Australian children’s hospital rated the severity of harm caused by pADEs in general pediatric wards. The authors identified 13 pADEs over 1 month, with 10 rated to be of minor severity and 3 rated moderate.²⁷ 

In 2 high-quality studies, researchers reported the proportion of medication errors causing pADEs in general pediatric wards.^13,40 The first was undertaken in the pediatric ward of an Ethiopian hospital, and the researchers reported that 8 (1.6%) of 513 errors resulted in a pADE, all of which were rated of minor severity.¹³ The second was conducted in the general ward of a children’s hospital in Argentina, and the researchers identified 201 medication errors but found no pADEs.⁴⁰ 

The incidence of pADEs in PICUs was assessed in 7 studies^{21,23,25,30,34,36,45}; 4 studies were rated to be of excellent quality,^21,36,40,45 and 1 study was rated to be of good quality.³⁴ In the first excellent quality study, the authors reported 29 pADEs per 1000 patient days,⁴⁵ whereas the second included all pADEs divided by the number of medications delivered in the unit and contained a report of 2.0 pADEs per 1000 patient days.³⁶ The 2 remaining excellent quality studies were conducted in Spanish and Argentinian PICUs, and the researchers identified 829 medication errors but found no pADEs.^21,40 The good quality study was of 9 US PICUs, and researchers identified 0.13 pADEs per 100 orders.³⁴ Three additional studies rated to be of poor to fair quality contained reports of 2 and 9 pADEs per 100 orders and 21 pADEs per 1000 patient days.^23,25,30 

Across the 3 studies that contained reports of the proportion of medication errors in PICUs causing a pADE, 30 (3.1%) of a combined 964 errors resulted in a pADE.^21,23,25 In each of these studies, researchers also rated the severity of the pADEs, with a combined total of 22 rated to be of minor severity, 7 moderate, and 1 classified as severe.

A single US study rated to be of good quality contained a report of 3.2 pADEs per 1000 patient days or 0.86 per 1000 medications administered.²² The researchers found that 39 (1.2%) of 3204 medication errors resulted in a pADE. In 2 additional studies rated as fair and excellent quality of Spanish and Argentinian NICUs, researchers identified 855 medication errors but found no pADEs.^33,40 

In 1 study rated to be of poor quality, researchers reported the proportion of medication errors causing a pADE in an Indian NICU and rated their severity.²⁹ The researchers found that 3 of 27 medication errors resulted in a pADE, with 1 rated to be of minor severity and 2 moderate.

Only 2 studies included data on pADEs in EDs in children’s hospitals, and neither provided data on the incidence of pADEs.^29,39 The first study was rated to be of good quality and comprised an audit of 4 rural EDs.³⁹ In that study, researchers identified 90 medication errors; however, no error resulted in a pADE. The remaining study, undertaken in India, was rated to be of poor quality.²⁹ The researchers identified 54 medication errors. Six of those medication errors were found to result in a pADE, with 3 rated to be of minor severity and 3 rated moderate.²⁹ 

In a total of 6 studies, researchers provided data on pADEs in hospitals using health information technologies (HITs)^{22,31,37,41,–43}; in 4 of those studies, researchers used a pre-post design with baseline data already described.^22,31,37,46 In 2 studies, researchers assessed CPOE systems.^41,42 In 1 of those studies, researchers reported the incidence of pADE in 2 Japanese hospitals.⁴¹ That study was rated to be of excellent quality and contained a report stating that 360 (43.6%) of 826 medication errors caused pADEs, with an incidence rate of 2.8 pADEs per 1000 patient days. The remaining study of a CPOE system did not contain a report of the incidence of pADEs but revealed that 40 (1.56%) of 257 dosing errors in a US hospital caused pADEs.⁴² 

Although not a focus of this review, the 4 pre-post studies provide insight into the impact of introducing HITs on the incidence of pADE. In summary, the implementation of CPOE in 2 multiple ward studies rated as good quality resulted in a significant reduction by 1.1 and 1.4 pADEs per 1000 patient days (reducing from 4.5 to 3.4 and 7.9 to 6.5, respectively).^36,37,46 Similarly, after 2 months of use of a barcode medication administration system with dosing alerts in a US NICU, the incidence of pADEs was reduced by 0.5 pADEs per 1000 drug administrations and 3.3 pADEs per 1000 patient days.²² In contrast, a single study rated to be of poor quality revealed that the introduction of CPOE in a UK pediatric ward in a general hospital was associated with an increase in the number of identified prescribing errors from 16 to 31, although no pADEs were found.³¹ 

We identified 22 studies that included information on pADEs among pediatric inpatients. Notably, only 6 studies contained reports of incidence of pADEs across multiple wards,^{12,36,38,40,43,44} 4 of which were conducted in the United States.^12,36,43,44 The incidence of pADEs in pediatric inpatient settings is low, with the highest reported incidence found to be 74 pADEs per 1000 patient days (95% confidence interval [CI]: 64–84). Similarly, the proportion of medication errors estimated to result in a pADEs was low with the highest estimates found to be 15.2% in ICUs and 6.4% across hospital-wide studies. In addition, the severity of pADEs was predominantly minor (temporary harm). Less than 1 in 10 pADEs were rated to be of moderate severity (permanent harm), and life-threatening incidents were extremely rare. In sum, the likelihood of a pediatric patient experiencing a moderate or severe pADE during a hospital stay is fortunately rare.

Few researchers have directly compared the incidence of pADEs between pediatric and adult inpatients. Indirect comparison between 2 highly cited US studies suggests pADEs are less frequent among children. That is, the seminal 2001 study by Kaushal et al,¹² which reported 1.8 pADEs per 1000 patient days among pediatric inpatients, is often compared with another highly cited study of adult inpatients by Bates et al,¹⁰ which reported 3.2 pADEs per 1000 patient days (95% CI: 2.5–4.0). Moreover, direct comparisons made in a single observational study conducted in the ICUs of a Moroccan general hospital revealed the incidence of pADEs among pediatric inpatients to be less than half that of adults (1.5 compared with 4.7 pADEs per 1000 patient days).⁴⁷ However, 2 more recent studies conducted in Japanese hospitals with CPOE revealed a similar incidence in adults and children, with an estimated 2.4 pADEs per 1000 patient days among adults (95% CI: 2.0–2.8)⁴⁸ compared with 2.8 pADEs per 1000 patient days among children (95% CI: 1.9–3.8).⁴¹ In contrast, the risk of medication error is consistently higher among children than adults, primarily because of the increased need for individual weight-based or surface-based dose calculations among pediatric inpatients.⁴⁹ These results suggest that although the risk of medication error in pediatric settings is greater than in adult settings, these errors are less likely to be associated with actual harm to patients.

Only 9 studies were primarily conducted in intensive care and ED settings, and only 4 were rated to be of at least good quality.^21,22,34,39 Nevertheless, there appeared to be a higher incidence of pADEs in these settings compared with general pediatric wards. Across studies, there were up to 29 pADEs per 1000 patient days in intensive care and ED settings compared with up to 17 pADEs per 1000 patient days in general wards. Similarly, the proportion of errors causing pADEs was 1.5% of 5262 medication errors compared with 1.3% of 610 errors. It remains unclear if this difference reflects differences like complexity of care or the greater medication exposure in ICUs.⁵⁰ However, in 1 study, researchers adjusted for medication exposure and found the incidence of preventable events was similar, with rates of 2.0 per 1000 patient days in PICU and 2.1 in a general ward of the same hospital.³⁶ 

Surprisingly few studies have contained assessments of the impact of introducing HITs on pADEs, with 4 studies containing investigations of patient harm,^{21,22,31,37,46} 3 of which included patient harm as a primary outcome.^22,37,46 Notably, the researchers for all 3 of these studies reported a significant overall reduction in the incidence of pADEs after HITs implementation. However, some areas of concern remained. In the first study, researchers reported that underdosing of analgesic medications remained a concern despite available clinical decision support.³⁷ In the second, researchers identified a number of new types of computer-related errors associated with CPOE implementation, such as drop-down menu selection errors.⁴³ These computer-related errors were associated with pADE incidence at 1.6 per 1000 patient days. The third study revealed an increase in medication errors after implementation of a barcoding system, primarily due to an improved ability of the system to detect wrong-timing errors not associated with harm.²² Overall, these findings are consistent with the conclusions drawn from previous reviews that advocate for the introduction of HITs in pediatric hospitals but with a need for further study to improve design and implementation to avoid the inadvertent introduction of new types of errors and harm.^51,–54 

The severity of pADEs was reported in only 7 of the included studies.^{12,13,21,23,25,27,29} A total of 85 pADEs were identified across these studies, of which only 1 was reported to result in a life-threatening outcome. Most preventable adverse events were rated as minor (69 of 85), and researchers for 3 studies found only minor harm.^13,21,29 In comparison, the 1995 study by Bates et al¹⁰ of adult inpatients revealed 14 of 70 pADEs to be life-threatening. Although never directly compared in a single study, pADEs occurring in children may be less severe than pADEs in adults. No studies meeting the inclusion criteria contained assessments of the severity of pADEs in a hospital where the staff use HITs. The absence of studies is notable and deserves future attention because the researchers have investigated the incidence of medication-related harm to children after the introduction of information technologies. More generally, there is a limited number of recent high-quality studies that have gone beyond reporting medication error rates to determine the extent of associated harm in children. This information is central to designing targeted interventions such as electronic decision support, which is focused on error types likely to produce harm.⁵⁵ 

The studies we identified were chiefly of good to excellent quality. Study quality, error detection method, or patient sampling technique did not consistently explain the variation in pADE incidence between studies. However, the included studies had a number of limitations, and caution should be taken in generalizing findings to other settings or contexts. Over half of the studies were conducted in the United States. Secondly, studies were undertaken at single sites, with the most notable exception a study of PICUs from 15 urban teaching hospitals.³⁰ Thirdly, only 12 studies included information on the method used to determine harm associated with medication errors.^† How this assessment was performed in the remaining 10 studies remained unclear, reducing confidence in the findings. The Naranjo et al⁵⁶ algorithm, a tool typically used to assess ADRs, was the most commonly employed technique for this process and was referenced in 4 studies.^12,13,27,45 These findings suggest that the call for a common protocol for the measurement and reporting of patient safety data remains unanswered.^57,58 Finally, the authors of only 2 of the 14 studies that had a prospective chart review addressed the potential impact of the hospital staff’s knowledge that their orders were under review.^21,32 Thus, the remaining prospective studies were not exempt from the “Hawthorne effect.” Similarly, the potential bias of the reviewers’ knowledge of any associated hospital staff or patients was only addressed in 6 of the 22 studies,^{21,29,32,34,39,43} again reducing confidence in the findings.

In sum, we recommend that future researchers take steps to ensure hospital settings are detailed, error identification methods and error definitions are consistent and transparent, and validated tools are used where possible. Determining how pADEs are identified and whether ADEs were preventable and caused by a medication error is a complex process and is often omitted in study methods or oversimplified by stating that this was left to “clinician judgment.” Validated tools are useful in this process, and, in the absence of any recognized gold standards, we recommend the World Health Organization Uppsala Monitoring Centre system for determining harm causation⁵⁹ and the National Coordinating Council for Medication Error Reporting and Prevention index²⁰ for rating harm severity.

Our review had some limitations. Firstly, study heterogeneity prevented formal statistical summary or meta-analysis. A previous review of medication errors in pediatrics included a meta-analysis conducted by combining studies with different error identification techniques, definitions of error, hospital wards, and hospital settings.⁶⁰ Secondly, the strict inclusion criteria limited the number of studies reviewed. However, the decision to exclude studies identifying patient harm by voluntary reporting is supported by several studies, which demonstrated the significant underreporting of incidents compared with chart review or observation.^16,25,38 In addition, the decision to exclude studies undertaken in outpatient and primary care setting was driven by the availability of research in these contexts^61,–63 and our focus on HITs. Finally, we combined study definitions of medication error despite variation in individual study definitions and included error types. As such, studies in which fewer types of medication errors were assessed are likely to identify fewer errors with an associated opportunity to cause a pADE and therefore report a lower incidence of pADEs compared with studies containing assessments of multiple medication error types.

There is limited literature in which the incidence of pADE in children is assessed. Across studies, we found the highest incidence to be 74 pADEs per 1000 patient days, 43 per 100 admissions, 9 per 100 orders, with up to 15.2% of errors resulting in a pADE. Comparisons suggest the risk of any pADE occurring is lower among children than adults. Although few studies contained assessments of pADE severity, the observed harm was predominantly minor. Incidence of pADEs in ICUs was higher than general wards, although this may reflect greater opportunities for error and harm because of higher rates of medication administration and sicker children. The introduction of HITs appears to reduce the incidence of patient harm resulting from medication errors in pediatric settings, but there are surprisingly few studies of the impact of these technologies on medication error outcomes for children.

ADE

adverse drug event

ADR

adverse drug reaction

CI

confidence interval

CPOE

computerized physician order entry

ED

emergency department

HIT

health information technology

pADE

preventable adverse drug event