A Systematic Review of Methods for Medical Record Analysis to Detect Adverse Events in Hospitalized Patients.

OBJECTIVE: In this systematic review, we evaluate 2 of the most used trigger tools according to the criteria of the World Health Organization for evaluating methods.
METHODS: We searched Embase, PubMed, and Cochrane databases for studies (2000-2017). Studies were included if medical record review (MRR) was performed with either the Global Trigger Tool or the Harvard Medical Practice Study in a hospital population. Quality assessment was performed in duplicate. Fifty studies were included, and results were reported for every criterion separately.
RESULTS: Medical record review reveals more adverse events (AEs) than any other method. However, at the same time, it detects different AEs. The costs of an AE were on average €4296. Considerable efforts have been made worldwide in health care to improve safety and to reduce errors. These have resulted in some positive effects. The literature showed that MRR is focused on several domains of quality of care and seems suitable for both small and large cohorts. Furthermore, we found a moderate to substantial agreement for the presence of a trigger and a moderate to good agreement for the presence of an AE.
CONCLUSIONS: Medical record review with a trigger tool is a reasonably well-researched method for the evaluation of the medical records for AEs. However, looking at the World Health Organization criteria, much research is still lacking or of moderate quality. Especially for the cost of detecting AEs, valuable information is missing. Moreover, knowledge of how MRR changes quality and safety of care should be evaluated.

Several studies have shown significant rates of adverse events (AEs) that cause harm to patients during their stay in the hospital.[1-3] Therefore, interest in the implementation of quality and safety programs that prevent these harmful events has grown. According to the report To Err Is Human (1992) of the Institute of Medicine, at least 44,000 people (and possible even 98,000) died of AEs each year in U.S. hospitals.[3] An update, 15 years after this first report, showed little improvement and stresses the need for further improving patient safety.[4] In the Netherlands, a report by the Dutch Institute for Research in Healthcare evaluated care-related harm in Dutch hospitals. It was estimated that yearly approximately 1700 patients (4.1% of the total number of inpatient deaths) die of unintentional preventable harm. Follow-up after 6 years showed slight improvement, but still 2.6% of the inpatient deaths seemed preventable. In other countries, an incidence between 2.5% and 13.5% has been found.[5-10]

To diminish care-related harm, there are several instruments to detect AEs: direct observation, incident reporting systems, autopsy reports, and mortality and morbidity conferences are some examples.[11-14] For medical record review (MRR), trigger tools are often used to prevent time- and cost-consuming investigation of all records. The most well-known trigger systems are the Harvard Medical Practice Study (HMPS) trigger system[15] with 18 triggers[15,16] and the Global Trigger Tool (GTT), developed by the Institute for Healthcare Improvement (IHI) with 54 triggers.[17]

Information on sensitivity, specificity, and positive and negative predictive values of these screening tools for detecting AEs is important in striving for effective and affordable methods to detect AEs. Also, MRR, independently of the method used, is a costly process because it takes time of both experienced physicians and nurses; hence, it is of importance that the process is indeed improving quality and safety.

However, adequate detection of AEs is only a small part of the review process. The total procedure also involves feedback to the medical departments, adjustments in the delivery of care, and hence improved outcome for patients resulting in less AEs. Other recent reviews highlighted MRR on the reliability perspective[18] or describe the use and potential value of the GTT.[19,20] Our aim was to describe the 2 most used trigger tools (GTT and HMPS) with the whole process in mind, starting with the detection of the triggers to the feedback to the departments and the effect on the patient safety. We were interested in all stages of this screening method. The World Health Organization (WHO) criteria for evaluating methods offer, in our opinion, a suitable framework to analyze all stages of this screening process. Therefore, we investigated the total screening process based on MRR and if this is evidence based according to the WHO criteria for evaluating methods.[21] With these criteria in mind, we searched the literature for evidence about the use of this specific method to improve patient safety.

METHODS

Search Strategy and Information Sources

We identified potentially eligible studies by searching PubMed, Embase, and the Cochrane Library for every criterion (and the explanation) described in Box 1. Our search was restricted to studies in English or Dutch published between 2000 and 2017 because we assumed that older results might not be applicable because of rapidly changing health care. The search strategy and corresponding search terms are shown in Table S1 (Supplemental Digital Content 1, http://links.lww.com/JPS/A263). The flowchart for every criterion separately is shown in Figure S2 (Supplemental Digital Content 2, http://links.lww.com/JPS/A264).

Explanation of the WHO criteria (Box 1)

1. Effectiveness in capturing the extent of harm (in different environments)

Comparative studies, presenting the numbers of AEs found with MRR and another method

2. Availability of reliable data

Interrater reliability of MRR

3. Suitability for large-scale or small, repeated studies

Acceptability and feasibility of MRR by institutions and professionals

4. Costs (financial, human resources, time, and burden on system)

5. Effectiveness in influencing policy

Effect of (the results of) MRR on national, regional, or local policy

6. Effectiveness in influencing hospital and local safety procedures and outcomes

Intervention studies that evaluate the effect of MRR on daily practice and the AEs

7. Synergy with other domains of quality of care

The IHI defined 6 domains of quality of care (safe, effective, patient-centered, efficient, timely, and equitable).[22] In this criterion was described whether MRR can be of influence on several of these domains.

Selection Criteria and Process

Studies were included if MRR was performed with either the GTT or the HMPS in a hospital population with a wide variety of patient groups. Suitable study designs were observational studies. Reviews, posters, comments, studies solely focusing on adverse drug events, and patient younger than 18 years were also excluded. Furthermore, we excluded studies that used computer detection for finding triggers and/or AEs. Duplicate references were removed using the software program EndNote (EndNote X8; Thomson Reuters, New York, New York). Afterward, all retrieved citations based on the titles and abstracts were screened. Study selection was performed in duplicate by 2 independent reviewers (D.O.K. and R.J.M.W.R.) according to the aforementioned inclusion and exclusion criteria. Study eligibility was thereafter assessed by reading the full text. Disagreements were resolved by discussion and consensus in which both reviewers had an equal vote.

Quality Assessment

The quality of the included studies was evaluated using the Gilbert criteria complemented with criteria composed by Worster and Badcock. These criteria have been developed to assess the quality of MRR studies and include, for example, the following: assessment on training of abstractors, presence of abstraction forms, and evaluation of the interrater agreement.[23-25] Quality assessment was performed in duplicate by D.O.K. and R.J.M.W.R. Rating categories used in the assessment were “present” or “missing.” These were transformed into 1 and 0 and, after scoring the numbers, were added together, resulting in a score between 1 and 15. Scores between 0 and 5 were seen as weak; between 6 and 10, reasonable; and between 11 and 15, good. The results are shown in Table S3 (Supplemental Digital Content 3, http://links.lww.com/JPS/A266).

Data Extraction and Analysis

Data extraction was performed by D.O.K. General data extracted from full-text included the following: author, year of publication, country, study aim, study design, instrument(s), sample characteristics, number of screening criteria, number of reviewers, and number of records analyzed for the measurement of the reliability. Extracted outcome data included the following: prevalence of AEs, κ agreement for AEs, percentage agreement, summary of main results, and costs of AEs (Table S4, Supplemental Digital Content 4, http://links.lww.com/JPS/A265) Using basic descriptive statistics (mean with confidence intervals [CIs] if available), we summarized these variables. Costs were not corrected for inflation. Different currencies were transformed to euros to make comparison easier. We used the exchange of August 2018 (1€ = 1.13 U.S.$). The results of this systematic review are presented according to the Preferred Reporting Items for Systematic Reviews and Meta-analyses guidelines.[26] (Supplemental Digital Content 5, http://links.lww.com/JPS/A267).

RESULTS

Our search resulted in a total of 832 citations (steps 1–4), and after title and abstract screening, 752 were discarded, leaving 80 full-text articles to assess their eligibility. Another 63 were excluded after full-text evaluation, leaving 17 for inclusion. After reading the references in these studies, we found 76 additional studies. The included studies originated from 24 countries, and sample size varied from 96[27] to 210 million[28] patients. Some studies were relevant for more than 1 criterion. Hereinafter, we report the results for every criterion separately and according to the trigger tool (GTT or HMPS). The average quality of the included studies was reasonable (score 8.3).

Criterion 1. Effectiveness in Capturing the Extent of Harm (in Different Environments)

Sixteen studies of reasonable quality were identified, which compare the use of the MRR with the use of another method to detect AEs in a hospital setting. A wide diversity of methods was compared with the GTT, such as reporting notification systems,[29,30] a hospital survey on patient safety,[31] patient safety indicators (PSIs),[29,32] and complaints and claims by patients and their relative and incident reports.[33,34] For the HMPS, the outcomes were primarily compared with incident reports, as well as patient-reported AEs[35] and patient complaints.[36]

GTT

Kurutkan et al[30] found that the GTT was 19 times more sensitive compared with internal voluntary reporting for the detection of AEs. Another study by Farup[31] detected an inverse association between the patient safety culture survey and AEs. Kennerly et al[29] found that voluntary reports and PSIs captured less than 5% of the total AEs, and this was also found by other studies.[32,37] Rutberg et al[34] found that 6.3% of the AEs detected by the GTT were reported with voluntary reporting. Mull et al[33] compared 3 methods (quality improvement programs, PSIs, and voluntary incident reporting) with GTT and concluded that 12% of the AEs were also detected by one of the other methods.

Harvard Medical Practice Study

Christiaans-Dingelhoff et al[38] linked 4 reporting systems with MRR: informal and formal complaints by patients or relatives, medicolegal claims by patients or relatives, and incident reports by health care professionals. Less than 4% of the AEs identified by record review were found in at least 1 of the 4 reporting systems.

Michel et al[14] compared 3 methods: cross-sectional, prospective, and retrospective review of records. The prospective and retrospective methods identified similar numbers of AEs (70% versus 66% of the total), but the prospective method detected more preventable AEs (64% versus 40%). The cross-sectional method showed a large number of false positives and identified none of the most serious AEs.

Several studies compared the number of AEs found when using incident reports compared with the use of HMPS. Blais et al[39] showed that in 15.5% of the cases with an AE, an incident report was present. Also, Sari et al[40] detected underreporting by voluntary incident reporting; only 24% of all patient safety incidents and only 5% of those resulting in patient harm were detected with the HMPS method. de Feijter et al[36] compared both patient complaints, MRR and incident reporting systems; they found that the type of AE found was dependent on the method used. Therefore, they recommend using a combination of methods when assessing patient safety in a hospital. The findings by Macharia et al[41] were in line with the other studies.

Weissman et al[35] found that 23% of the patients had at least 1 patient-reported AE and 11% according to MRR. The agreement between these 2 were poor for occurrence of any type of AE and slightly better for life-threatening or serious AEs. Bjertnaes et al[42] showed a significant correlation between the 2 measurement methods.

Criterion 2. Availability of Reliable Data

Twenty-seven suitable studies of reasonable quality regarding the availability of reliable data in studies that used a trigger tool method for MRR were found.

Trigger Tool

The IHI method showed a positive predictive value of 30.4% (95% CI, 13.3%–47.6%).[30,32,43-48] There was only one study that calculated the negative predictive value, which was 99%.[32] The agreement on the presence of a trigger had on average a moderate agreement (κ = 0.48; 95% CI, 0.18–0.77).[2,44-46,48,49]

In studies using the HMPS, the positive predictive value was on average 33.4% (95% CI, 21.1%–45.8%).[5-7,9,40,50-56] The agreement between the nurses on the presence of a trigger showed on average a substantial agreement (κ = 0.63; 95% CI, 0.55–0.71).[5,40,51,52,54,55]

The AE Assessment Strategy

Within the studies using the IHI trigger tool,[17] the κ on the presence of an AE was on average 0.67 (good agreement; 95% CI, 0.56–0.82).[1,2,30,37,43-46,49,57-60] In 3 studies, the agreement on the severity of the AE was investigated, which showed an average κ of 0.40 (fair agreement; 95% CI, 0.06–0.73).[1,43,58,59]

In HMPS studies, there was a moderate agreement between medical doctors on the presence of an AE in the medical record, with a κ value on average of 0.58 (95% CI, 0.33–0.82).[5,9,40,48,52-54,61]

Criterion 3. Suitability for Large-Scale or Small, Repeated Studies

Forty-three suitable studies of reasonable quality regarding the suitability of MRR for large-scale or small, repeated studies were found.

In the last decades, several large-scale studies have been executed to assess the prevalence of AEs in hospitals on a national level or the financial impact of AEs. The smaller studies were used for the training of the reviewers,[49] for comparison with the detection rate of other methods,[29,32,37,38,62] or for assessing the interrater reliability[45] of the review.

Size

The size of the studies varied from 15[49] (training) records to 40,851,[63] and the number of hospitals investigated also varied from 1 to 25 hospitals.

Cross-Country Comparison

Deilkas et al[64] compared the AE rate between Norway and Sweden. Norway had significantly higher AE rates of surgical complications. Swedish hospitals had significantly higher rates of pressure ulcers, falls, and “other” AEs. No significant difference between overall AE rates was found between the 2 countries.[64]

Criterion 4. Costs (Financial, Human Resources, Time, and Burden on System)

Of the 12 studies found concerning the costs of AEs, 6 reported these results based on MRR; the quality of these 6 studies was reasonable. Others reported the costs based on a hospital claim database,[28] a national medical and drugs claim database, and a hospital cost accounting system.[65] Furthermore, some studies used consensus-based methodology, diagnostic coding error,[66] and estimation of social cost.[67] The costs of an AE were on average €4296 (range, €2600–€6436).[6,55,60,68-70]

Next to the cost of an AE itself (the clinical consequences and the patient harm), the cost of finding an AE using MRR is also important. With this we mean the cost of nurses for trigger analysis, and physicians for the AE detection in records and their administration. Keep in mind that for detecting one AE, many records without AEs have to be analyzed. No studies on the cost of detecting an AE were available. Based on our own data (unpublished), the cost of detecting an AE with HMPS was €150.000 on a yearly basis, and that for a single potentially preventable AE was approximately €1800.

Criterion 5. Effectiveness in Influencing Policy

We found no trials or studies but only reports of projects concerning this issue.

Short Overview of Studies Mentioned in the Context of the European Union Network for Patient Safety and Quality of Care

In the Netherlands, the reports by the Dutch Institute for Research in Healthcare have been published in the context of a research program named patient safety in the Netherlands. The first study was first performed to gain insight into AEs in Dutch hospitals.[54] The 2 follow-up studies were executed to evaluate whether the safety programs had a positive influence on these AEs.[71,72]

Based on experience in Norway and Sweden, MRR by the GTT method gives a valuable overview of the kind and incidence of AEs affecting patients and a good starting point for intensified patient safety improvement work.[63,73]

Criterion 6. Effectiveness in Influencing Hospital and Local Safety Procedures and Outcomes

Our search revealed 5 studies investigating changes in AE rates during the study period.

Kennerly et al[46] showed a 7% reduction in AEs in 2 years (on average, 3.5% per year). Suarez et al[74] found during a 6-year study period a decrease of 2.5% (on average, 0.4% per year). Deilkas et al[63] showed that AE rates decreased from 16.1% to 13.0% in 2 years (1.55% per year).

However, Rutberg et al[34] found no improvement during the 4-year study period in which the GTT was used, despite several initiatives for improving the quality in the hospital, similar to Landrigan et al[1] and Mortaro et al.[75] Three national studies in the Netherlands (2004–2012) showed no changes in overall AE but did show a decrease of 45% regarding preventable AEs.[71] However, the latest update in 2017 did not show a further decrease of the preventable AEs and preventable deaths but did show a 2% decrease in overall AE.[76] Landrigan et al[1] also found slight improvements in a 5-year period in the United States.

Criterion 7. Synergy With Other Domains of Quality of Care

We found no trials or studies but only reports of projects concerning this issue. The IHI has defined 6 domains of quality of care.[22] Medical record review has common ground with a few of these domains; safe, effective, timely, and equitable. During MRR, a committee assesses whether AEs have occurred. The goal is to improve care, making it safer. Furthermore, it is evaluated whether the specific treatment for a particular patient was correct, right on time (effective and timely), and independent of personal characteristics (equitable). The project “Deepening our Understanding of Quality Improvement in Europe” investigated the relation between quality systems and patient-related outcomes. Almost 200 hospitals in 8 European countries participated in the Deepening our Understanding of Quality Improvement in Europe project. Beside questionnaires, MRR and data registries were analyzed. One of the conclusions of this project was that the presence of quality systems has a positive effect on the safety culture in a hospital.[77]

DISCUSSION

Our study clearly shows that there is abundant literature concerning MRR in hospitals. However, almost 75% of these articles were of rather moderate quality (steps 1–4). The first 4 WHO criteria relate to the characteristics of MRR concerning validity, reliability, and costs. They could effortlessly be extracted from the existing literature. The last 3 criteria relate to the ability of MRR to generate improvements in safety procedures and the quality of safety programs. Data concerning criteria 5 and 7 were indirectly described or concealed in reports, and therefore, we were unable to evaluate the quality according to the quality checklist.

The literature we found in relation to the first criterion showed that MRR reveals more AEs than any other method. However, at the same time, MRR detects different AEs compared with other methods. Furthermore, we found a moderate agreement for the presence of a trigger and a good agreement for the presence of an AE for the GTT. For the HMPS, we found a substantial agreement for the presence of a trigger and a moderate agreement for the presence of an AE. Also, MRR seems suitable for both small and large cohorts as shown in several studies with different sample sizes.

The costs concerning AEs can be the cost of the event itself (which is usually the topic of the literature we found), but also the cost of MRR. It is striking that most studies investigating costs of AEs only evaluated costs related to the event. The only study we found evaluating also the costs of the detection method was published by Bates et al,[78] but was not included in the current study. The costs for the detection of a single AE in 1995 was 103€, and that for a preventable AE was 241€ (€11.10 for every admission). Translating this to the current situation means a considerable amount of money for the detection instrument, let alone the costs of the AEs themselves. Because there is no agreement on which costs exactly should be taken into account concerning AEs, comparison between studies is difficult. For future research, it is important to have a complete overview of all costs involved. It should contain not only the cost of the detection instrument and the direct cause of the AE itself, but also loss of working days, up to the implementation of other protocols and their costs to prevent AEs. Only with these total costs we will be able to estimate the costs per quality-adjusted life-year to see if this is acceptable. The aforementioned is also underlined in a report by Øvretveit[79] and a more recent report by the Organization for Economic Co-operation and Development.

The results of the fifth criterion show that MRR can have an effect on health care policy. In Europe, the Network for Patient Safety and Quality of Care has been active for 4 years. This network was cofunded and supported by the European Commission within the Public Health Program. Its focus was “to improve Patient Safety and Quality of Care in Europe by supporting the implementation of good organizational practices and safe clinical practices in health care organizations and through sharing of information and experiences.” The Network for Patient Safety and Quality of Care builds on the experience of the European Union Network for Patient Safety project (2008–2010), which established patient safety platforms in several European member states. The main outcome will be the consolidation of the permanent network for patient safety.

Considerable efforts have been made worldwide in the health care systems to improve safety and to reduce errors in the treatment for patients. As is shown in criterion 6, these efforts have translated into only slight improvements in the overall safety of patients or in better quality of care. Finally, the last criterion shows that MRR has focused on several domains of quality of care.

Michel[80] has composed an overview of strength and weaknesses of available methods for assessing AEs. Medical record review is one of these methods. Since then (2003), no update has been performed to investigate how well the trigger tools comply according to the WHO criteria. Instead of giving an overview of the available methods, we decided to focus on the 2 most used trigger tools. Also, in this overview, Michel did not give an insight into the quality of the included studies, rather compared the methods with each other. Furthermore, he evaluated the evidence-based rating of the methods for estimating AEs, for each criterion.

Our systematic review had several strengths and some limitations. We used extensive search criteria (Table S1, Supplemental Digital Content 1, http://links.lww.com/JPS/A263) in several databases and also screened the references of the finally selected articles, which, to our opinion, minimizes the risk of missing important literature. Moreover, we used an accepted WHO strategy to evaluate this screening tool. Furthermore, we combined several quality checklists for the assessment of the quality of the included studies.

An important limitation for drawing aggregate conclusions was the different methods studies deployed. Although the same trigger tool method was used, almost every study adapted the triggers or the review process slightly. Besides that, different definitions and scales were used for both AEs and their preventability. Moreover, some studies used external reviewers, some internal reviewers, or experienced reviewers. Although the intention was to evaluate the quality of all included studies, this was only possible for 5 of 7 WHO criteria. The reason was that the others were not suitable for evaluation according to our 15 quality requirements. These were actually more reports than studies. For other studies, we found that important detailed information was not reported (e.g., information on positive and negative agreement, and reproducibility regarding the individual triggers). We only searched for studies from the year 2000 and onward. There is a chance that important older studies are therefore missed. However, we doubt that these findings would still be generalizable to today’s care because of rapidly developing health care and quality and safety improvements. Also, we were forced to write a systematic narrative review; because of the number of different studies, we were not able to combine the numbers and create a meta-analysis.

Between 2000 and 2017, many hospitals have switched from paper medical records to electronic medical records. In this period, many hospitals switched from paper health records to electronic health records. Electronic health records have several advantages for MRR because the records are accessible, readable, and easier to process.[81-83] A next step will be automatic detection of triggers and AEs, but at the moment, this is not optimal.[84-88]

In conclusion, MRR with one of the trigger tools is a well-researched method for the evaluation of medical records for AEs. However, looking at the WHO criteria for the evaluation of methods, much research is still lacking or of moderate quality. More information concerning costs of the detection method and the improvements in care and patient outcome is needed. Only with this information that MRR could be evaluated on its cost-effectiveness. Moreover, more insight into how MRR changes quality and safety of care is needed. We found no studies analyzing the whole string that starts with the application of the triggers and ends with quality and safety improvement for individual patients at acceptable costs.