FormalPara Key Points

Pediatric medication errors are often associated with problems related to prescribing, e.g., dose errors causing under- or overdosing of the medicine.

Computerized physician order entry (CPOE) systems with clinical decision support (CDS) tools have been found to support safe prescribing when customized for the unique needs of pediatric patients.

The successful implementation and maintenance of CPOE–CDS systems in pediatric care settings requires continuous risk management and multidisciplinary teamwork.

1 Introduction

Pediatric patients are prone to adverse drug events, including medication errors [1,2,3,4,5]. Medication errors are defined as “any preventable event that may cause or lead to inappropriate medication use or patient harm while the medication is in the control of the healthcare professional, patient, or consumer” [6]. Especially, prescribing has been identified as a high-risk task within the pediatric medication use process [5, 7]. Pediatric prescribing errors are often associated with dose errors caused by a variety of human or system errors (e.g., a wrong dose resulting from a decimal error or an outdated information of patient’s weight) [8]. However, a significant proportion of pediatric medication errors have been estimated to be preventable with appropriate medication safety measures [9, 10]. Therefore, prevention of dosing errors related to drug prescribing requires the introduction of higher-level systemic barriers, which focus on changes to the system in which individuals operate [11,12,13]. A computerized physician order entry (CPOE) system with a clinical decision support (CDS) integrated into an electronic health record (EHR) represents such a high leverage defense to prevent prescribing-related dose errors [14].

The CPOE system offers an electronic platform to enter the medication order in a structural form [15]. The CPOE system can also include a knowledge-based CDS system, which provides clinical guidance to the physician by combining patient-specific data (e.g., weight, age, or laboratory results) to the general medicines information from medical databases (e.g., interactions) [4, 14, 15]. Knowledge-based CDS systems can be categorized as basic or advanced depending on their ability to utilize patient information [14]. For instance, basic CDS systems may have a simple dose range checking function that ensures that the prescribed medication dose is appropriate for the patient’s weight and age. This type of function is usually based on predetermined minimum and maximum doses, without considering the individual pharmacokinetic and pharmacodynamic characteristics of the patient, while an advanced CDS system may provide, for example, additional dosing suggestions based on a patient’s laboratory results [14, 16]. Alerts are a key function of the CPOE–CDS systems; some CDS systems can produce soft- or hard-stop alerts when the programmed rules (e.g., a dose range) are violated and require an action from the user [14, 17]. Consequently, these systems may prevent medication errors before they reach the patient causing potential harm [18]. CPOE–CDS systems can be homegrown or purchased from a vendor [15]. Some commercial systems can also be customized to meet the needs of the organization or unit, which has been recommend especially for pediatric patients [4, 14, 16, 19,20,21].

Several previous studies have shown that CPOE systems with CDS tools have prevented harmful and even fatal medication errors (e.g., 10- and 100-fold overdoses) [22,23,24,25]. A systematic review by Gates et al. [7] reported the presence of dosage errors in prescriptions based on whether technological systems were in place or not, while Koeck et al. [25] focused on describing all types of pediatric prescribing errors reported in hospitals. Both systematic reviews included few studies considering the effects of CPOE systems with integrated CDS on dosing or prescribing errors in children, yielding both neutral (systems did not show a significant dose error reduction) and positive (CPOE–CDS systems reduced dose errors) results supporting the use of these systems. Moreover, an earlier systematic review by Conroy et al. on interventions to reduce dosing errors within the different stages of the pediatric medication use process found that CPOE systems showed some degree of reduction in medication errors [26]. However, to our knowledge, none of the previous systematic reviews have focused on evaluating and comparing the effects of different combinations of CPOE and CDS interventions to prevent pediatric dosing errors related to prescribing. Consequently, the objective of this systematic review is to examine the effects of CPOE systems with CDS functions on preventing dose errors in pediatric inpatient orders and outpatient prescriptions. The present information is needed to identify and evaluate insights into pediatric dose error prevention and to give recommendations for future research.

2 Materials and Methods

This systematic review follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 criteria and Synthesis Without Meta-Analysis (SWiM) items as an extension to PRISMA criteria (Supplementary Materials, Table S1–S3) [27, 28]. The study protocol was registered in the international prospective register of systematic reviews PROSPERO (CRD42021277413). After literature screening, the protocol was subjected for minor alterations (e.g., defining more specific inclusion criteria for pediatric age group).

2.1 Search Strategy and Databases

The search strategy for this review was formulated around the themes “pediatric,” “clinical decision support,” “computerized physician order entry,” and “dose error” (Table 1). The full search strategy is presented in Supplementary Tables S4–S6. An experienced information specialist from Helsinki University Library was consulted, and the search strategy was piloted several times to ensure inclusion of all relevant publications. The final search was performed in MEDLINE (Ovid), Scopus, Web of Science, and EMB Reviews (including Cochrane Database of Systematic Reviews and Methodology Register, Database of Abstracts of Reviews of Effects and Health Technology Assessment). The search was limited to studies published after January 2006 in English. The chosen time period was considered appropriate as CPOE and CDS systems are evolving technologies and a previous literature review published by Conroy et al. in 2007 found only one study that met these inclusion criteria [26]. The first literature search was conducted on 25 October 2021 with a follow-up on 13 January 2022 and 10 September 2023. In addition, reference lists of included studies and previous systematic reviews were manually searched along with grey literature to supplement the systematic search (Supplementary Appendix 1, Tables S4–S6).

Table 1 Definitions of key terms
Full size table

2.2 Eligibility Criteria

The eligibility criteria were predetermined using a PICOS framework (Supplementary Table S7) [32, 33]. Studies carried out in pediatric settings or in other care settings treating under 18-year-old patients were included (Table 1). The studies could concern both inpatient orders and/or outpatient prescriptions generated by a CPOE system with at least one CDS function. Studies using only a CPOE or a CDS system were excluded, as were studies with a scope that was not on dose error prevention in the prescribing phase.

The included studies had to report the effects of the CPOE system integrated CDS interventions on dose errors as an outcome (Supplementary Table S7). The studies which did not concentrate on measuring the effects of the CPOE–CDS systems on dose error prevention were excluded. Based on several systematic reviews relating the matter from the past decade, it was expected that the included studies would apply varying controlled and noncontrolled study designs to measure the effects of CPOE–CDS systems in preventing dose errors [7, 25, 34,35,36,37]. Classification of the eligible study designs was adopted from the JBI (Joanne Briggs Institution) Manual for Evidence Synthesis as “observational studies” (e.g., analytical cross-sectional studies) and “experimental studies” (e.g., pre–post studies) [38]. When the study had a controlled design (e.g., pre–post studies), three different types of comparators were accepted, including period before implementation of a CPOE system with a CDS, period before the implementation of a CDS system with an already existing CPOE, or period before modification of CDS tool or tools (Supplementary Table S7). These comparators were selected based on their expected ability to demonstrate the effects of CPOE–CDS systems in preventing dose errors. The included publications had to be peer-reviewed scientific original articles or case or short reports. Otherwise, no study method restrictions were applied.

2.3 Study Screening and Selection

For the screening process, the references were imported to Covidence® software platform. Two independent reviewers (HR and EK) conducted screening of the titles and abstracts followed by full text reviews according to the eligibility criteria (Supplementary Table S7). After that, reference lists of included studies and previous systematic reviews were manually searched along with grey literature. In cases of disagreement, a consensus was achieved with a third independent reviewer (SK).

A quality assessment including the risk of bias assessment for the included studies was conducted using the JBI’s critical appraisal tools for quasiexperimental (used for studies with pre–post design) and analytical cross-sectional studies by two independent reviewers (HR and SK) [40]. Any discrepancies were solved with a third reviewer (ARH).

2.4 Data Extraction and Analysis

Data extraction was conducted by two independent reviewers (HR and EK) in Covidence®, where a ready-made extraction template was applied. The template was modified using PICOS criteria and JBI’s mixed methods data extraction form (Supplementary Table S7) [41]. The following information was extracted per included publication: general study information (e.g., authors, publication year, and country), setting, study design and used methods, the number of medication orders and participants, the used CPOE systems and CDS tools and their effects on dose errors, and CDS alert information (if the CPOE–CDS system produced alerts, the types of alerts and how the alerts possibly effected on dose error rates). Disagreements within the extraction templates completed by the two independent reviewers (HR and EK) were resolved through discussion and consensus with a third reviewer (SK) to form the final data.

A meta-analysis could not be conducted due the heterogeneity of the studies. However, vote counting method was used to describe and analyze the quantitative findings of the studies exploring the effects of CPOE–CDS systems and possible CDS alerts on dose errors rates (Supplementary Table S8) [27, 42]. The effect of the CPOE–CDS system on dose error prevention was presented by a visual display of the nonstandardized effects using an effect direction plot [43], a synthesis method suggested in the Cochrane Handbook for Systematic Reviews of Interventions [42]. Quality of evidence of the individual studies was determined by two reviewers (HR and SK) as very low, low, moderate, or high, using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach [44, 45]. Alongside with JBI critical appraisal tools, the GRADE assessment also included an assessment of bias.

3 Results

3.1 Study Characteristics

A total of 17 studies were included from the 4207 identified publications published between 2007 and 2021 (Fig. 1; Table 2). The majority of the publications (65%, n = 11/17) were conducted in the USA, followed by the Middle East (18%, n = 3/17), Europe (12%, n = 2/17), and Asia (6%, n = 1/17). The studies applied pre–post (n = 10) and cross-sectional (n = 7) designs with varying data collecting methods; the most common method (n = 16) was retrospective register data analysis (Supplementary Table S8) [22, 23, 46,47,48,49,50,51,52,53,54,55,56,57,58,59]. In pre–post studies, the effects of implementation or modification of CPOE–CDS system was measured by comparing the rates of dose errors or alert rates before and after intervention. Cross-sectional studies involved data from various time periods to evaluate the impact of CPOS–CDS systems, e.g., on dosing error rates over time. In pre–post studies, a period of time before and after an implementation or modification of the CPOE–CDS system was used to evaluate the effects on dose errors or alert rates, whereas cross-sectional studies involved data from various point of times to evaluate, e.g., the impact on dosing error rates over time. All included studies were conducted in pediatric settings. Pediatric intensive care units (PICU) or other intensive care units were the most common settings (29%, n = 5/17) when the study focused on a specific ward instead of the whole hospital [48, 50, 55, 56, 58].

Fig. 1
figure 1

PRISMA flow diagram [28]

Full size image
Table 2 Description of the included studies (n = 17)
Full size table

The most studies (71%, n = 12/17) included only inpatient orders (Table 2) [46, 48, 50,51,52,53,54,55,56,57,58,59]. Two studies were conducted in outpatient settings, and in one study, outpatient medications were prescribed in a pediatric emergency department [22, 24, 47]. Only 12% (n = 2/17) of the studies included both inpatient orders and outpatient prescriptions [23, 49]. The comprehensive study characteristics are presented in Supplementary Table S8.

3.2 Characteristics of the CPOE–CDS Systems

The studies reported several different CPOE–CDS systems (Table 3 and Supplementary Table S9). In the majority of the studies (94%, n = 16/17), the organization or unit had either a self-developed (homegrown) CDS system or a customized commercial CDS system (Table 4) [22, 23, 46,47,48,49,50,51,52,53,54,55,56,57,58,59]. One study reported a commercial CPOE–CDS system with no additional customizations [24]. In total, 13 different CDS tools were identified, of which 9 were reported as homegrown or customized (Table 3).

Table 3 The identified clinical decision support (CDS) tools and their characteristics assessed in the studies
Full size table
Table 4 The customization of the computerized physician order entry (CPOE) systems with clinical decision support (CDS)
Full size table

The number of CDS tools in one study ranged from 1 to 6 and, when excluding the study that considered multiple different institutions [47], the average number of different CDS tools in one setting was three (mean 3.1 and median 3). The most common CDS tool was dose range check (in 82% of the studies, n = 14/17; Table 3). Other common tools were dosing frequency check and dose calculator (in 47% of studies, n = 8/17). The most customized tools were dose range check (86%, n = 12/14), maximum dose limits (83%, n = 5/6), and frequency check (75%, n = 6/8). In thirteen studies, the used CDS system was basic [22,23,24, 46,47,48,49, 52,53,54, 56, 58, 59], and three were considered as advanced [50, 55, 57]. One study had a basic CDS system with optional advanced tools [51].

3.3 CDS Systems with Alert Functions

In 15 studies, the CDS within the CPOE system comprised soft-stop alerts that can be overridden by the user (Table 1) [22,23,24, 48,49,50,51,52,53,54,55,56,57,58,59]. In four of these studies [48, 55, 56, 58], the CDS also included hard-stop alerts, which cannot be self-overridden (Table 1). When implementing new, or customizing an existing, CPOE–CDS system, usually alert functions were added (n = 9) [23, 24, 48, 50,51,52, 55, 58, 59]. Alternatively, the existing alert systems were modified by either changing the alert logic (n = 2) [54, 57], or modifying other CDS tools (e.g., customizing dose ranges) that had an effect on the alert rate (n = 2) [49, 53]. The prevalence of soft-stop alerts was reported in 59% (n = 10/17) [22, 24, 50,51,52,53,54,55, 57, 58] of the studies, of which two studies also included hard-stop alerts [55, 58]. The number of correct and false alerts were reported in four studies of which all except one [57] reported that the percentual number of false alerts was over 70% [24, 53, 54].

Altogether, ten studies reported the percentage of the overridden alerts, meaning that the orders were either canceled, modified, or executed unchanged [22, 24, 49,50,51,52,53,54,55, 58]. The majority (n = 9/10) of the studies stated that over 50% of the occurred soft-stop alerts were ignored [22, 24, 49,50,51,52,53,54,55]. Order cancelling and modifying were less common; the reported order cancelling ranged from 0% to 21% [51,52,53, 55] and modifying from 0% to 33% [22, 24, 50,51,52,53, 55, 58].

3.4 Effects of CPOE–CDS Systems on Dose Errors

A majority (n = 14) of the studies showed a statistically significant impact on dose error prevention after implementation or modification of a CPOE–CDS system (Table 5). The impacts of CPOE–CDS systems on the effect direction plot was formed based on dose errors in individual studies as well as the statistical significance of the results and the sample sizes of orders in the intervention groups (Supplementary Table S8). In most of the studies (n = 9), the sample sizes were small (< 10,000, or no data reported), followed by medium (10,000–60,000, n = 4 studies), and large (> 60,000, n = 4 studies) sample sizes (Table 5).

Table 5 The effects of a CPOE–CDS system on dose errors with effect direction and the use of clinical decision support alerts. Publications are categorized by the aim and presented according to the effect direction of the outcome measure
Full size table

In 93% (n = 13/14) of the studies reporting benefits in preventing dose errors, the CPOE–CDS systems were homegrown or customized (Supplementary Table S9) [22, 24, 47,48,49,50,51,52,53,54,55,56,57,58]. A statistically significant reduction in dose errors was reported in eight studies, where the most used CDS tools were dose range check (n = 5), dose calculator (n = 4), and dosing frequency check (n = 4) (Table 5) [22, 24, 47, 48, 50, 56,57,58]. Additionally, the majority (n = 7/8) of these studies included a CDS system with an alert function [22, 24, 48, 50, 56,57,58]. The CDS system customization showed a statistically significant increase of appropriate alert rates in four studies [51, 53,54,55]. The CPOE–CDS systems with alert functions were most useful in detecting potential overdoses [23, 51, 52]. In comparison, eight studies reported how alerts contributed, or could contribute to, dose errors (Table 5) [23, 24, 49, 51,52,53,54,55]. Whereas none of the 17 studies reported a statistically significant overall increase of dose errors when using a CPOE–CDS system, the systems could have contributed to some errors, especially underdosing errors, e.g., due the lack of minimum dose ranges, missing indication specificity, and physicians overriding relevant alerts in consequence of alert fatigue [24, 53, 54, 56].

3.5 The Quality of Evidence

All eligible studies met the criteria of the JBI critical appraisal tool assessment (Supplementary Tables S10 and S11). The GRADE approach resulted in low quality in all 17 studies due the observational study designs.

4 Discussion

The present review indicates that a bundle of CDS interventions designed for the pediatric population, instead of using only single CDS intervention, would provide a more comprehensive systems-based approach to prevent dose errors. This is congruent with previous literature, where introducing multiple interventions (e.g., CDS alerts and prescribing rules) has been suggested to be most effective in preventing similar types of medication errors [13, 20, 61,62,63]. We found that especially dose range check, dose calculator, and dosing frequency check functions appear to be the most effective CDS tools in reducing dose errors. Additionally, most studies reporting positive effects of CPOE–CDS systems on preventing dose errors included an alert function, which suggests that CDS alerts are recommended to prevent pediatric dose errors [24, 48, 50,51,52,53,54,55,56,57,58]. Our study demonstrates the growing interest of implementing CDS systems to prevent dose errors over the last 15 years, as the earlier systematic review by Conroy et al. in 2007 found integrated CDS systems rare, while most studies still focused on investigating separately used basic CDS tools or CPOE systems alone [26, 37].

However, not all triggered alerts prevented medication errors, as inappropriate alerts and other warnings were found to hamper the use of a CDS system, which might result in alert fatigue [46, 49, 52,53,54]. This finding was highlighted in certain studies with high numbers of false positives (e.g., Sethuraman et al. [24]), while other studies found higher numbers of actually avoided errors (e.g., Ginzburg et al. [47]). False positive alerts can be caused by several reasons, such as lack of agreement between vendor-supplied dosing rules and clinical practice, variability in the recommended dose between different indications and age groups, and the commonly encountered need for off label use of medicines in pediatric populations [37, 59]. Moreover, the studies showed order modifying and cancellation being rare reactions to alerts. The alert overriding also increased if the user did not understand the meaning of the warnings or, e.g., the dose range limits, or alerting settings were programmed incorrectly, thus causing false alerts [50, 52, 54, 59]. While alert fatigue continues to represent a central issue of CPOE–CDS systems [21, 25, 64,65,66,67], several suggestions have been made for its prevention, e.g., evaluation the suitability of dosing alerts to daily clinical practice and customization of CPOE–CDS systems to pursue reduction in the number of inappropriate warnings [4, 20, 21, 37, 66, 67]. Our findings support this as multiple studies showed a statistically significant increase in appropriate alerts rates after CDS customization [51, 53,54,55]. Additionally, the system defenses should be reinforced by incorporating strong functions that are not dependent on memory, e.g., by changing soft-stop alerts with significant safety risks into hard-stop alerts that cannot be overridden [12, 13, 23, 68].

The majority of the CDS tools targeted for pediatric use were basic, while advanced CDS tools were less common. To the best of our knowledge, there is limited evidence of the reasons why advanced CDS systems are not commonly used. Many of the current CDS functions are designed for adults, and they may not always be compatible with children [57, 69]. For example, when aiming to prevent acute kidney injuries by detecting drug-induced nephrotoxicity, the underlying epidemiological factors in children differ from those in adults [70,71,72]. Therefore, as children are a special patient population and advanced CDS tools (e.g., dose check according to the renal function) may benefit even a smaller population (e.g., children with a renal dysfunction), the cost of an advanced CDS system may be higher per patient when compared to a basic one (e.g., due to additional system customization costs) [73]. Nevertheless, using more advanced CDS systems could improve dose error prevention and prevent alert fatigue as they provide diverse clinical knowledge on medication dosing for individual patients [24, 50, 51, 55, 57, 69]. Similar findings have also been stated by Tolley et al. in 2018 [37]. However, studying the differences between basic and advanced systems in preventing dose errors in pediatric populations would require more rigorous comparative studies.

The present study indicates that when CPOE–CDS systems are being established, they should be critically examined as part of the evidence-based risk management of the user organization. Means for conducting the proactive analysis comprise, e.g., the use of failure modes and effects analysis (FMEA) approach [74]. Furthermore, reactive evaluation of the CPOE–CDS system’s efficiency in preventing dose errors should be examined by observing triggered and overridden alerts and, e.g., by using test patients and orders, as described by Chaparro et al. [14, 20, 75]. It is widely suggested that safeguards should be developed against new types of errors (e.g., technological errors) caused by the CPOE–CDS system [20, 22, 23, 46, 48, 49, 53,54,55,56, 64, 76]. These errors may emerge over time along with frequent system updates due to technology development, changes in current medication information and new medicines entering the market [20, 46, 48, 56, 64]. Therefore, regular monitoring of safety and performance of these systems is necessary by investigating patient safety incident reports, user experiences, and data from CPOE–CDS systems [20, 37, 64, 68]. Furthermore, it is important that healthcare organizations choose competent persons who are responsible for monitoring and making amendments. Customization is recommended to be carried out in a multidisciplinary manner [e.g., a team of physicians, pharmacists, nurses and information, communication technology (ICT) specialists, and possibly representatives of a commercial vendor], based on clinical relevance (e.g., summary of product characteristics of licensed medications or literature regarding off label use), and considering the special characteristic of pediatric patients (e.g., dosing support in NICU settings should include recommendations based on gestational age, postnatal age, or postmenstrual age in combination with weight for certain drugs) [5, 46, 50, 54]. However, the customization process must meet the requirements of local legislation, which may vary between different countries [77]. The progress of customization and engagement of clinicians might be more challenging in countries where the vendors are responsible for the CDS data and local customization is not allowed.

4.1 Limitations and Future Studies

The included studies mainly considered inpatient orders. Therefore, the results may not give a comprehensive view of the effects of CPOE–CDS systems on reducing dose errors in outpatient prescriptions. In relation to the limitations in the review process, the full-text phase interrater reliability between the two reviewers (HR and EK) was moderate (Cohen’s kappa, κ = 0.50) [39]. However, a consensus was achieved by discussion with a third-party reviewer (SK). The included publications had varying observational study designs, and hence, rated as low quality in the GRADE assessment; common limitations were dissimilarities between the studied groups (in pre–post studies) and the lack of a control group (Supplementary Table S8) [47,48,49,50, 55, 56, 58, 59]. However, there is a limited number of publications available on the topic [7, 25], and therefore, the inclusion was not based on the level of evidence quality. Similar findings have been made in previous systematic reviews as well as in a Cochrane review by Maaskant et al. [34] reviewing different interventions to reduce medication errors in pediatrics [25, 35, 36]. Maaskant et al. [34] stated that when the studies concern medication safety changes, e.g., in organizational contexts, experimental designs (such as randomized controlled trials), may not be even fully applicable. Therefore, more rigorous quality assessment criteria for observational studies are urgently needed.

The variation between the studies also contributed to the inability of conducting a meta-analysis or, e.g., a sign test to explore the effects of CPOE–CDS systems statistically. For instance, the sizes of the study populations and the sample sizes (the number of analyzed medication orders, presented in Supplementary Table S8) varied from a few hundred to millions. Moreover, vote counting based on direction of effect used as a synthesis method is less powerful than, e.g., combining p values, and it does not give details on the magnitude of effects [42, 78]. However, we did not observe major differences in the quality of evidence between different included study designs. Consequently, we were able to comprehensively describe the features of different CPOE–CDS systems and their effects on dose error prevention.

The findings of this review are particularly relevant for designing forthcoming studies and the use of CPOE–CDS systems in pediatrics to assess the direction in which limited healthcare resources should be taken. Future studies are needed on which CDS functions are the most effective in preventing dose errors, with an emphasis on outpatient settings and facilities treating both adult and pediatric patients. Additionally, the differences between basic and advanced CDS systems should be further examined in pediatric settings, particularly from the alert fatigue prevention and cost-effectiveness perspective.

5 Conclusions

CPOE systems with CDS functions have a great potential to reduce dose errors and promote pediatric medication safety. System customization for the pediatric population, implementing CDS alerts, and the use of dose range check tool, dose calculator, and dosing frequency check seem to be most advantageous when aiming to prevent dose errors. However, CPOE–CDS systems cannot prevent all dose errors as human errors continue to occur. Moreover, the implementation of CPOE–CDS systems can pose new kinds of at-risk behaviors, such as high rates of overriding of alerts due to alert fatigue. Therefore, systematic actions are needed to optimize the safe use of CPOE–CDS systems in pediatrics. More studies are needed, particularly on the effect on dose error prevention when comparing basic and advanced CDS tools and the effects of different individual CDS functions and settings.