The inclusion criteria were as follows:

• Population: This review included papers on children and adult patients of any age, race, ethnicity, or gender who were admitted to inpatient or emergency services for any condition.
• Interventions: The review examined any type of IT, such as Electronic Health Records (EHR); Computerised Physician Order Entry (CPOE); decision support systems and barcoding; and any type of educational intervention of frontline staff, patients, and families aimed at reducing the patient identification errors rate.
• Comparisons: Any comparator employed within the included studies.
• Outcomes: The primary outcome reported patient identification errors rate during a hospital stay. Moreover, surrogate outcome measures related to the patient identification process included patient Identification Bands (ID bands) errors and patient identity checks error rates. The secondary outcomes were the causes of patient identification errors and wrong-patient electronic orders rate.
• Types of studies: This review considered both experimental and quasi-experimental research designs, including non-randomised and Randomised Controlled Trials (RCTs), pre-post studies, and interrupted time-series studies. Other research designs of quantitative nature, such as observational, analytical and descriptive studies were also considered for inclusion.

The search strategy aimed to locate both published and unpublished studies, and this review employed a three-step procedure. An initial limited search of Medline (NICE Healthcare Databases Advanced Search) and CINAHL (EBSCO) was undertaken, followed by an analysis of the text words in the title, abstract, and index terms used to describe the article.

A second search using all identified keywords and index terms was performed across all databases: the Cochrane Central Register of Controlled Trials, Medline, CINAHL (EBSCO), EMBASE, SCOPUS, Latin American and Caribbean Health Sciences (LILACS) and Medes. The search for unpublished studies included: ProQuest Dissertations and Theses, Google Scholar, National Health Service Improvement, Dart-Europe, System for Information on Grey Literature in Europe (Open Grey), Banco de Teses – CAPES, and Theses Canada.

Subsequently, the reference lists of all studies selected for critical appraisal were screened for additional research. All databases were searched from inception until March 2020. Moreover, studies published in the English, Portuguese, and Spanish languages were considered for inclusion to avoid language bias and missing key studies with significant findings related to the aim of this review.

A systematic approach was applied based on the research question [23], and a Medline search strategy was developed (Table 1). Individual search strategies were formulated for each database and are available in Supplementary file 1.

The search results were uploaded to EndNote ^TM online (Clarivate Analytics, PA, USA). Following the removal of duplicates, the relevant studies were retrieved and their citation details were imported into the JBI System for the Unified Management, Assessment and Review of Information - JBI SUMARI (JBI, Adelaide, Australia), and selected according to the eligibility criteria using a two-step process:

• The title and abstract screenings were examined by two independent reviewers for assessment of the review’s inclusion criteria. Disagreements were resolved through discussion.
• Full-text screening studies were retrieved and assessed in detail by two independent reviewers based on the inclusion criteria. Any disagreements between them were resolved through discussion or by a third reviewer.

To ensure the quality of the analysed evidence, a cut-off point for inclusion was applied with at least 60% of the questions with affirmative answers in the JBI's critical appraisal tools [21]. Eligible studies were assessed by 2 independent reviewers, and any disagreements were resolved through discussion or by a third reviewer.

Data were extracted using the standardised JBI extraction tool [21] and included specific details regarding the interventions, populations, study methods, and outcomes. Discrepancies were resolved through discussion among the reviewers. When required, the authors of primary quantitative studies were contacted for missing information or clarification of unclear data.

A meta-analysis was unfeasible because of the heterogeneity of the population, study design, interventions, and outcomes, as well as the insufficiency of data for statistical pooling. A summary of findings was created using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) for rating the quality of evidence [24].

Patient identification errors, indicated in the protocol of this systematic review [23], were scarce and were only addressed in two studies. Therefore, the ID band errors and patient identity check error rates were analysed as substitute outcomes. Research has suggested that surrogate outcomes should be considered in the absence of relevant evidence for healthcare providers and the population [24-26]. Additionally, the wrong patient electronic orders rate was considered as the secondary outcome.

These deviations from the protocol were because these outcomes were a fundamental part of the patient identification process in the hospital setting; consequently, they were highly relevant to answer the review question and may provide decision support to improve patient safety.

The search identified 2,943 potentially relevant studies (three articles were obtained through the reference lists of all identified reports). Of these, 1,634 were duplicate records. The remaining 1,309 records were screened by title and abstract. After examination, 1,230 studies failed to meet the inclusion criteria. The full texts of the remaining 79 records were retrieved. Subsequently, two reviewers independently reviewed each article; of these, 61 studies were excluded because they did not address a minimum of one inclusion criterion, and one article was unavailable (the list of full-text articles excluded with reasons is available in Supplementary File 2). Overall, 18 studies were appraised for methodological quality, and six were omitted. Finally, the remaining 12 records were included in the review. The flow chart presented in Fig. (1) displays the details of the search results as well as the selection process and reasons for exclusion.

Fig. (1). PRISMA flow diagram of study selection.

The reviewers independently assessed the methodological quality of the 18 studies. When necessary, for additional data or clarification, the authors of the papers were contacted; however, not all of them provided further information. Furthermore, 6 studies failed to reach the cut-off point for inclusion [27-32]. The list of citations and reasons for the excluded research is provided in Supplementary File 3.

Of the 12 studies included in this review, the RCT [33] scored nine out of 13 (69.2% quality score) for quality assessment with a moderate risk of bias (Table 2). The allocation of the intervention groups was not concealed, thus including potential performance bias. Due to the nature of the educational intervention, it was not possible to blind the participants and auditors. However, there was no blinding of the researchers, which may have induced a performance bias risk.

All 11 pre-post studies met at least six of nine criteria (60% quality score) with a moderate risk of bias (Table 3). Only two studies [34, 35] included a control group. However, the comparisons were made ‘within the groups only’ and reduced the possibility of exploring the consequences on the groups of exposure to different treatments in addition to the intervention of interest. In seven studies [35-41], it was unclear whether the participants were included in similar comparisons, contributing to the risk of selection bias. Additionally, in three studies [42-44], it was uncertain whether the results were derived from various measures before and after the intervention. Therefore, it was not possible to explore the changes in the ‘in time’ effect in each group and compare them between groups. Furthermore, one study [34] presented inconclusive data regarding loss of follow-up, limiting its internal validity. There was a lack of clarity with respect to whether the results were measured in a reliable way for two studies [34, 42] that may have threatened the validity of the statistical inference and its relationship between the intervention and the outcome. Lastly, one study [43] did not present conclusive details of the statistical analysis that may have caused an error in statistical inference.

The 12 studies were conducted between 2002 and 2017 and included one RCT [33], two controlled pre-post-test studies [34, 35] and nine single group pre-post-test studies [36-44]. A table describing the included studies is provided in the appendix.

Overall, 11 studies were published in English [33-35, 37-44], while only one in Spanish [36]. Furthermore, eight were single-centre studies conducted in multiple countries, including the USA [34, 41-43], Spain [36], Australia [44], the United Kingdom [37] and Switzerland [38]. Moreover, four multi-centre studies were conducted in five emergency services in New York City, USA [35], one in Canada, Australia and the USA [39], one in England, the USA, Canada, Norway, Brazil, and Australia [39], and one in six hospitals across the USA [40].

Among the included studies, sample types varied between patients, ID bands, electronic orders, and medication rounds; they ranged from 282 [41] to 11,377 [40] patients. The studies conducted in adult services were as follows: two studies in inpatient wards [37, 39], two in Accident and Emergency Services - A&E [36, 44], one in Theatres [38] and one in both intensive care unit (ICU) and ward [39]. Furthermore, five studies were carried out in neonatal and paediatric settings. Of these, three were developed in both neonatal and paediatric ICUs [34, 40, 41], one in a paediatric ward [43] and one in a single paediatric ICU [42]. In addition, one study was conducted in both adult and paediatric A&Es [35].

The follow-up period varied significantly between studies and ranged from less than six months [ 40, 44] to three and half years [34]. The interventions of interest examined in the studies were categorised into four major groups as described below.

Four studies included educational staff interventions [ 33, 38, 39, 41]. The RCT [33] compared a tag on blood bags educating the staff to verify the patient’s wristband and identity before transfusion against control (no addition of the tag on blood bags). One study [38] implemented a protocol to confirm patient identity with periodic audits and ‘just in time’ staff feedback after an error had been identified, discussing inadequate identification verification issues. Furthermore, supplementary training was provided at staff meetings. Two studies [39, 41] standardised ID bands, followed by audits and consequent staff feedback after each error had been identified to raise awareness about the importance of correct patient identification.

Only one study [40] combined staff education regarding correct patient identification with a partnership with families and patients through education and audited wristbands providing prompt feedback to them about the importance of identification for safe care. A complementary poster was developed to be hung in the patient’s room, reminding that the ID bands should remain on them constantly.

Five studies applied IT interventions. Franklin et al. [37] used a closed-loop system incorporating electronic prescribing, ward-based automated dispensing, and barcode patient identification to enhance patient identity check prior to medication administration. Hyman et al. [43] modified the EHR provider interface, including a verification screen (with the patient’s photograph) that requested the employees to confirm patient identity when ordering. Three studies [34, 35, 42] employed the CPOE ID re-entry function to verify patients’ identity; additionally, two of these [34, 42] also incorporated a new name convention for newborns.

Furthermore, two studies combined IT resources with an educational staff strategy. Cuadrado et al. [36] implemented the CPOE in addition to staff education, emphasising the different measures of improvement in relation to patient misidentification. Spain et al. [44] implemented two interventions: staff education alone (regarding the technology, patient identity checks, and the risks of misidentification) and combined with an armband barcoding scan.

The measures used for primary outcomes varied across studies and were classified as follows:

• Patient identification errors rate was addressed by two studies measuring them in the analytical request process [36] and the total number of reported patient identification errors [43].
• The ID band errors rate was emphasised by three studies reporting the percentage of wristband errors rate [39-41].
• Patient identity check errors rate in four studies was addressed by measuring the number [33] and the percentage of incorrect patient identity checks [37, 38, 44].

The secondary outcome causes of patient identification errors were not measured in any of the studies included in this review. The wrong-patient electronic orders rate was addressed in three studies [34, 35, 42].

Two studies measured the ID band errors rate outcome. Walley et al. [41] reported a statistically significant decrease in the rate by 4% compared to baseline (95% CI [2.5–5.5] p<.001). Similarly, Howanitz et al. [39] indicated a statistically significant decrease in the overall mean ID wristband error rate by 4.3% compared to baseline (p=.001). Despite the overall positive results, there was considerable heterogeneity among studies regarding population, sample size, and intervention delivery.

The patient identity check errors rate outcome had inconsistent results. Garnerin et al. [38] reported an increase in the full compliance of the patient identity check from 9.7% to 58.7% compared to baseline (p<.001). The results from the RCT [33] showed that the label on the blood bag failed to improve bedside patient identity check when comparing the intervention with the control group on immediate (pooled OR 1.09; 95% CI: 0.54-2.17) and long-term effects (pooled OR 0.49; 95% CI: 0.22-1.06). Meta-analysis could not be conducted due to the different types of interventions employed in the included studies.

Only one study [40] delivered this combined intervention and had a statistically significant reduction in ID band errors rate by 77% after the interventions were compared to baseline (p=.001). Nonetheless, this study was conducted at a multi-centre site where each hospital had its own standard process to confirm patient identity, and the authors did not ensure interrater reliability.

Franklin et al. [37] demonstrated a statistically significant reduction of 63% in the patient identity check errors rate during medication administration compared to baseline. The authors noted that full compliance in examining patient identity was not achieved because of incorrect practices, such as the bar code’s affixation on patients’ furniture. Moreover, the study had unclear information about the inclusion of participants in any similar comparisons.

Hyman et al. [43] reported a 25% decrease (51 versus 37) in the patient identification errors rate after the intervention compared to baseline. However, the research did not provide further statistical analysis and researchers relied on voluntary reporting of recognised errors.

The following 3 studies assessed the secondary outcome of the wrong-patient electronic orders rate. Adelman et al. [42] found a statistically significant reduction of 36.3% in this rate as compared to baseline (OR 0.64; 95% CI: 0.42-0.97). Similarly, Green et al. [35] reported a statistically significant 30% and 24.8% decrease in the rate after the intervention in the short-term impact (RR=0.70; 95% CI: 0.63-0.77) and 2 years after the intervention, respectively (RR=0.76; 95% CI: 0.69-0.83). Adelman et al. [34] also indicated a statistically significant reduction in wrong-patient electronic orders. In the neonatal ICU, intervention 1 (CPOE ID re-entry function) reduced the frequency of wrong-patient orders by 48.7% (p<.001), and the combined intervention (with new name convention) yielded a 61.1% decrease from baseline (p<.001). In the non-neonatal ICU, the former and the latter had 34.1% (p<.001) and 28.4% (p<.001) reductions from baseline, respectively.

Despite Adelman et al. [34] and Green et al. [35] using control groups, the comparisons were made only ‘within groups’ resulting in the absence of a direct comparison between the control and experimental groups. Moreover, the heterogeneity presented within the studies in this group precluded meta-analysis.

Cuadrado et al. [36] found a patient identification errors rate of 6.18 per 10,000 requests in the analytical process. The authors reported a statistically significant decrease of the patient identification errors rate, in the analytical process (when compared within the intervention periods only) from 17.4 (after combined intervention) to 6.37 (end of study) per 10,000 requests (p<.0001). No statistically significant reduction in patient identification errors was found when comparing the baseline with the intervention period, concluding that the estimates of the effect may be imprecise.

Spain et al. [44] found a statistically significant reduction in the patient identity check errors rate before blood collection compared to baseline with staff education alone: when the patient was not asked to state their name 33.2% (p<.001) and their date of birth 35.1% (p<.001); this was further augmented by combining the armband scanner: 67.4% (p<.001) and 71.4% (p<.001) reductions in blood collection when the patient’s name and the date of birth were not asked, respectively. However, the cost of education alone appeared to be low compared to the armband scanner technology. Despite the positive effects, the study had a small sample size and short follow-up time.

The results of this review demonstrated that educational interventions, using audit and feedback to improve accountability and raise awareness about the importance of correct patient identification, may lead to a small, but significant, decrease in the ID band errors rate [39, 41] and a larger reduction of the patient identity check errors rate [38]. However, both outcomes had limited results due to the small number of high-quality studies. The use of a reminder to educate the staff about the importance of patient identification before blood transfusion was ineffective in reducing the rate of patient identity check errors [33]. This finding is similar to another review [11], which reported that the reminders could simply increase the amount of information during the delivery of care where there is a lacking on account of professionals.

Table 4. GRADE Summary of Findings.

Outcomes	Impact	Number of Studies	Certainty of Evidence (GRADE)	Comments
Educational staff interventions
Patient ID band errors rate	Significant impact reported in two studies	2 pre-post-test studies	VERY LOW	Lack of control group in one study. Heterogeneity: different populations and interventions
Patient identity check errors rate	Significant impact reported in one study	1 RCT and 1 pre-post-test study	VERY LOW	RCT with true randomization, lack of allocation concealment and blinding and small sample size. Lack of control group in one study. Heterogeneity: different interventions
Educational staff intervention combined with a partnership with families and patients through education
Patient ID band errors rate	Significant impact reported	1 pre-post-test study	VERY LOW	Lack of control group, unclear inclusion of participants in any similar comparisons
IT interventions
Patient identity check errors rate	Significant impact reported	1 pre-post-test study	VERY LOW	Lack of control group, unclear inclusion of participants in any similar comparisons
Patient identification errors rate	Evidence is insufficient to determine impact	1 pre-post-test study	VERY LOW	Lack of control group, estimates of effects are not precise
Wrong patient electronic orders rate	Significant impact reported	3 pre-post-test studies	VERY LOW	Lack of control group in one study. Two studies with control group, however the comparisons were made within groups only. Heterogeneity: different populations
IT interventions combined with an educational staff strategy
Patient identification errors rate	Some significant impact reported	1 pre-post-test study	VERY LOW	Lack of control group, selective outcome reporting
Patient identity check errors rate	Significant impact reported	1 pre-post-test study	VERY LOW	Lack of control group, small sample size and short follow-up time
Causes of patient identification errors	-	-	-	-

GRADE Working Group grades of evidence
High certainty: We are very confident that the true effect lies close to that of the estimate of the effect.
Moderate certainty: We are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.
Low certainty: Our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect.
Very low certainty: We have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.

The ID band errors rate had a significantly higher reduction when audit and feedback interventions were combined with a partnership with families and patients through education. However, this evidence was limited to only a single group pre-post study [40]. Despite the increasing awareness of the demand for patient-centered strategies to improve the safety of healthcare delivery, additional studies are required to develop a complete picture of the impact of service users’ engagement on the patient identification process. Moreover, healthcare leaders need to facilitate a collaborative approach between the recipients of care and staff to enable relevant and effective patient participation [46-48].

As highlighted in a previous report [1], IT is an essential resource for improving quality. However, in this review, the evidence was insufficient to support decision-making regarding the use of IT interventions, and its effectiveness was based on the results of five pre-post studies. Overall, these studies reported that the CPOE framework might lead to a significant reduction in the wrong-patient electronic orders rate [34, 35, 42]. However, patients’ photographs in the EHR had insufficient evidence to determine the effect on the patient identification errors rate [43]. Nevertheless, it may be considered a good tool to measure near misses and to assist in preventing adverse events. Electronic prescribing, automated dispensing, and barcode technology can reduce the rate of patient identity check errors [37]. These findings support other reviews that suggest these technologies can have positive benefits on patient safety outcomes [49-51]; however, there is limited evidence regarding this [52].

The effectiveness of health-related IT interventions combined with an educational staff strategy has been described in two pre-post studies. Overall, they reported a significant reduction in the patient identification errors rate [36] and a significant reduction in the patient identity check errors rate (44). These results are also in accord with other reviews in which educational strategies have been found to enhance nurses’ adherence to patient safety attitudes and assist in the proper use of IT in hospitals [53, 54]. It is worth noting that education alone was more cost-effective than combined with an armband scanner to reduce the rate of patient identity check errors. Hospital managers should weigh the potential benefits of these technologies against possible challenges with their implementation and costs [55].

This review provided some evidence for the use of staff and patient education and IT interventions in reducing patient identification errors as well as the rates of ID band errors, patient identity check errors, and wrong-patient electronic orders in the hospital setting. However, it is also worth considering that all the included studies had methodological limitations and delivered many interventions to different populations and contexts. Consequently, the findings denote a lack of a valid effect, making it difficult to ascertain which interventions were the most effective.