Near-miss events detected using the Emergency Department Trigger Tool

Abstract

Objectives

Near-misses include conditions with potential for harm, intercepted medical errors, and events requiring monitoring or intervention to prevent harm. Little is reported on near misses or their importance for quality and safety in the emergency department (ED).

Methods

This is a secondary evaluation of data from a retrospective study of the ED Trigger Tool (EDTT) at an urban, academic ED (data from 10/1/2014 - 10/31/2015; 92,859 eligible visits). All patients aged ≥18 completing a visit were eligible. We ran the EDTT, a computerized query for triggers on 13 months of ED visit data, reviewing 5,582 selected records using a two-tiered approach. Events were categorized by occurrence (ED vs present on arrival [POA]), severity, omission/commission, and type, using a taxonomy with Categories, Subcategories and cross-cutting modifiers.

Results

We identified 1,458 ED near-misses in 1269 (22.7%) of 5582 records and 80 near misses that were POA. Patient care events represented the majority of ED near misses, including delays in diagnosis, treatment, and failure to monitor, primarily driven by ED boarding and crowding. Medication events were second most common (17%), including 80 medication administration errors. Of 80 POA events, 42% were related to over-anticoagulation. We estimate that 19.3% of all ED visits include a near miss.

Conclusion

Near miss events are relatively common (22.7% of our sample, 19.3% in the population) and are associated with an increased risk for an AE. Most events were patient care-related (77%) involving delays due to crowding and ED boarding followed by medication administration errors. The EDTT is a high-yield approach for detecting important near misses and latent system deficiencies that impact patient safety.

INTRODUCTION

Near-misses include conditions that have the potential to cause harm, intercepted medical errors and events impacting patients that require monitoring or intervention to prevent harm from occurring. It has long been recognized in high-risk and high-reliability industries such as aviation, that reporting of near-miss events is important for identifying and addressing latent safety issues to prevent serious adverse events. This observation has important lessons for healthcare, where near misses often go undetected.^1,2 Near miss reporting in health care is included as a specific recommendation in the landmark IOM report To Err is Human, yet it is not part of routine quality and safety surveillance in medicine.² It has been observed that errors are commonplace but rarely result in AEs, that AEs are uncommon and are infrequently due to error, and that patient safety efforts should focus on preventable and ameliorable harm, rather than error per se. This is understandable. However, it is also noted that near misses represent “an unsafe situation that is indistinguishable from a preventable adverse event except for the outcome. A patient is exposed to a hazardous situation but does not experience harm either through luck or early detection.”³ This argues for attention to near misses and their value as targets for prevention efforts.

The emergency department (ED) is recognized as having high potential for near misses yet detection and reporting of these is typically very poor.^4–8 Voluntary incident report systems are not felt to be adequate surveillance systems for adverse events; their performance for capture of near-misses is less clear.^9–12 In this manuscript we present near-miss events detected using the ED Trigger Tool (EDTT) and discuss possible implications for quality improvement.

METHODS

Study Design

This is a secondary analysis using data from our retrospective observational study to derive, validate and optimize the ED Trigger Tool (EDTT).^13,14

Study Setting and Participants

This study was conducted at an urban, adult, academic medical center using data from 92,859 visits by 58,497 unique patients (10/1/2014 – 10/31/2015). All ED patients aged ≥18 were eligible for inclusion. Visits in which patients left without being seen were excluded. This study was approved by our university institutional review board.

Patient and Public Involvement

Neither patients nor the public were involved in the design, conduct, reporting or dissemination of this research.

Record Selection

We identified a broad set of candidate triggers in a prior study, using an expert consensus process. In the present study, we mapped these candidate triggers to our electronic medical record (EMR). We applied the automated trigger query to 13 months of ED visit data, validating the trigger query against manual review and assessing for associations of triggers with AEs. This initial step required that 25% of records were deliberately selected to ensure that reviewers saw each trigger enough times to perform analyses; the remaining 75% were randomly selected from those with one or more triggers in addition to a small number of records with no triggers for comparison. Following the derivation phase, independent validation records were randomly selected for review based on presence of at least one of the triggers found to be associated with AEs in the derivation phase. In total, we reviewed 5,582 records (6.0% of all visits) by 5,187 unique patients (8.9% of patients): 5,439 records with ≥1 trigger and 143 with no triggers. Record selection by phase was as follows: (1) derivation (N=1,726 records with 1+ triggers and 60 records without triggers);and (2) validation (N=3,713 records with 1+ triggers and 83 without).

Record Reviews

Three first level (L1) nurse reviewers worked through lists of visits, which were assigned to pairs of L1 reviewers for dual independent review. Reported events were randomly assigned to one of two second level (L2) reviewers (RTG, RMS) who could agree with or modify the classifications of the event, disagree & decline, and add missed events.

Event Classification

Reviewers indicated whether an event occurred prior to arrival (POA) or in the ED, was an act of omission or commission, and rated severity using the National Coordinating Council’s Medication Event Reporting and Prevention (MERP) Index, ¹⁵ modified to use the term ‘event’ rather than ‘error’ as the events described do not necessarily represent errors. Here, we present only near miss or ‘non-harm’ events, with MERP levels: (A) conditions were present with the capacity to cause harm; (B) an error occurred but was intercepted and did not reach the patient; (C) an event that reached the patient but did not cause harm; and (D) an event occurred requiring monitoring to confirm no harm resulted or requiring some intervention to prevent harm from resulting. ¹⁶ Events were categorized using a taxonomy of ED AEs that focuses on the nature of an event experienced by patients¹⁷ including common Categories (e.g., Medication-related; Patient Care-related) as well as several subcategories within each Category (e.g., bleeding event, allergic reaction) and inclusion of up to three (of 46) cross-cutting modifiers (i.e., can be used across any category or subcategory) to further describe events. Events where one led to the next or sharing a common root were considered Cascading and were counted as a single event, which is a common feature of trigger tools.^18,19

Outcomes

We present near miss events, stratified by those occurring in the ED vs. present on arrival (POA) along with the associated Category/ Subcategory (across all levels of severity). We then present near misses by cross-cutting themes, rolling up event descriptors regardless of whether common terms were used as a subcategory or as a modifier in our taxonomy. This allows us to look at all cases irrespective of the main Category or proximate cause and provides a broader picture of their incidence. Cross-cutting terms include: 1) a subcategory that is used within two different Categories (e.g., bleeding event); 2) any modifier used across Category/Subcategories (e.g., opioids); and 3) terms that could be used in the taxonomy as a Modifier or a Subcategory (e.g., glycemic event). We provide the overall incidence of these cross-cutting themes and present common associations between pairs of descriptors.

Analyses

This descriptive analysis presents counts and proportions and 95% confidence intervals. We report frequencies of category/subcategory and cross-cutting themes, the proportion of overall events the subcategories represent and an assessment of event type severity by NCC MERP. The discussion is limited to instances where a sufficient number of events were observed (generally, 5 or more). Additional details within Categories/Subcategory were derived from review of the event narratives (not quantified). We used the Apriori algorithm to identify association rules between pairs of themes and report those with a minimum support (# times observed) of five and minimum confidence of 25%. We eliminated association rules which were not statistically significant after correcting for multiple comparisons (Fisher’s exact test and the Benjamini-Hochberg procedure for a 1% false discovery rate). Because the focus of our study was on events occurring in the ED, we analyze and discuss POA events separately.

These frequencies are specific to the set of events in visits triggered using the EDTT (i.e., in a selected sample). We then use population trigger frequencies and direct standardization to estimate prevalence in our patient population (using, e.g., the frequency of the procedural sedation trigger in the entire population of 92,859 visits, and the estimate of the risk of an event given presence/absence of the trigger from the 5,582 records we reviewed). This provides some background denominator estimate of cases when seeking to determine how often AEs such as hypoxia or airway events occur in the context of procedural sedations.

All analyses were conducted with SAS 9.4. Model development used the Python sklearn package.²⁰

RESULTS

Sociodemographic and clinical descriptors are presented in Table 1. There were no differences in age, sex or race compared to those with no events found. Patients with near misses tended to have lower acuity level but had a higher admission rate. Overall, we identified 1,458 ED near misses in 22.7% of the 5,582 visits reviewed. We also identified 80 POA near misses, discussed separately. The large majority of ED near misses were MERP A (87.0%, conditions present with the capacity to cause harm). This was largely driven by the volume of patients with either the trigger C52:”ESI [Emergency Severity Index] 1-3 with boarding time >2 hours” or C30: “ED Boarding >6 hours” for admitted patients.. The remainder were almost all MERP C (6.0%) and D (6.5%), where an event reached a patient but either did not cause harm or required monitoring or intervention to prevent harm. Intercepted errors without further intervention (MERP B) comprised less than 1% of all near misses.

Table 1.

A. Patient Characteristics
		Population	Study	No Near Misses	1+ ED Near Miss
		N=58,497	N=5,187	N=4,079	N=1,236
Age	Median	42.7	51.5	51.8	50.9
Sex	Female	53.7%	53.8%	53.7%	54.2%
	Male	46.3%	46.2%	46.3%	45.8%
Race	Black	55.7%	56.5%	56.2%	58.6%
	White	41.8%	41.5%	41.7%	39.5%
	Other	2.5%	2.1%	2.1%	1.9%
B. Visit Characteristics
		Population	Study	No Near Misses	1+ ED Near Miss
		N=92,859	N=5,582	N=4,313	N=1,269
ESI	A- Resuscitation	2.2%	7.2%	7.3%	6.7%
	B- Emergent	30.0%	40.7%	41.7%	37.5%
	C- Urgent	51.8%	45.8%	43.4%	53.7%
	D- Semi-Urgent	15.1%	6.1%	7.3%	2.1%
	E- Non-urgent	0.8%	0.2%	0.3%	---
Disposition	Discharged	63.7%	44.4%	47.1%	35.3%
	Admit	34.2%	52.0%	49.3%	61.0%
	Expired	0.3%	1.2%	1.5%	0.5%
	Transfer	1.0%	1.6%	1.2%	2.8%
	AMA	0.7%	0.8%	0.9%	0.4%

The figures for patients with or without a near miss (4079 + 1236) sum to greater than the number of unique patients (5187) since a patient may have made a visit with a near miss and another with no near miss.

Patient care-related events were the most common (n=1,125, 77.2%; Table 2), followed by Medication-related events (n=248, 17.0%). Table 3 gives, respectively, the subcategory frequencies for near misses, by main categories; and dyad (Category-Subcategory) as percentages of all near misses. Delays dominate the Patient care and Ancillary services categories (90.1% and 50%, respectively). Medication near-misses mostly fall into the diffuse “other” category (60.5%), with Wrong Dose/ Medication/ Route/ Order/ Patient as the next most common (32.3%). Malposition was most common for Surgical/Procedural near misses (66.7%). There are too few Device-related near misses (n=8) captured to present. Among category/subcategory dyads, the most common was patient care/delayed treatment, accounting for 68.7% of all near misses. This was driven primarily by long waiting room delays, delays in consultant services including procedures and inpatient boarding in the ED. Table 4 provides a sample of illustrative vignette cases that provide additional event detail. Table 5 provides frequency of cross-cutting terms (whether used as a Subcategory or Modifier, regardless of Category).

Table 2.

	Frequency			MERP score distribution
	N	%		A	B	C	D
Patient Care	1,125	77.2%		95.8%	0.09	2.0%	2.1%
Medication	248	17.0%		67.0%	2.42	15.3%	15.3%
Ancillary Services	44	3.0%		29.6%	4. 6%	40.9%	25.0%
Care Coordination	18	1.2%		50.0%	---	39.0%	11.1%
Surgery/ Proc.	15	1.0%		6.7%	---	13.3%	80.0%
Device	8	0.5%		12.5%	---	---	87.5%
Total	1,458			87.0%	0.6%	6.0%	6.5%

MERP = Medication Error Reporting and Prevention Index

Table 3.

	N	As % of category	As % of all near misses
Patient Care
Delayed Treatment	1,002	89.1%	68.7%
Delayed Diagnosis	11	1.0%	0.8%
Other	90	8.0%	6.2%
Failure To Monitor	15	1.3%	1.0%
Hypotension	3	0.3%	0.2%
Transfusion Reaction	3	0.3%	0.2%
Fall	1	0.1%	0.1%
Medication
Other	150	60.5%	10.3%
Wrong Dose/ Medication/ Route/ Order/ Patient	80	32.3%	5.5%
Allergic Reaction	7	2.8%	0.5%
Hypoxia	4	1.6%	0.3%
Delirium, Confusion	2	0.8%	0.1%
Hypotension	2	0.8%	0.1%
Bleeding	1	0.4%	0.1%
Nausea And Vomiting	1	0.4%	0.1%
Over-anticoagulation	1	0.4%	0.1%
Ancillary Services
Delay	22	50.0%	1.5%
Wrong Exam/ Patient/ Result/ Test	12	27.3%	0.8%
Specimen-related issue	8	18.2%	0.5%
Other	2	4.5%	0.1%
Care Coordination
Inter-/ Intra-dept Communication Issue	11	61.1%	0.8%
Documentation Issue	3	16.7%	0.2%
Sending Institution Communication Failure	2	11.1%	0.1%
Other	1	5.6%	0.1%
Outpatient Coordination / Communication Issue	1	5.6%	0.1%
Surgery/ Procedural
Malposition	10	66.7%	0.7%
Bleeding	3	20.0%	0.2%
Other	2	13.3%	0.1%
Device			0.0%
Malfunction/ Migration/ Occlusion	4	50.0%	0.3%
Infiltration/Extravasation	2	25.0%	0.1%
Other	2	25.0%	0.1%

Table 4.

Example Vignettes

Category	Subcategory	Modifiers	MERP	Site*	Event Narrative
Ancillary Services	Wrong Exam/Patient/Result/Test	Lab Services	B	ED	Point of care INR test result was high, prompting ordering of reversal agents. These were then cancelled when whole blood INR resulted withing normal range.
	Wrong Exam/Patient/Result/Test	Radiology	D	ED	Initial ED CT scan interpretation of a peri-colonic abscess prompted admission for operative management. The following day, re-review of the CT determined there was no fluid collection or abscess and the patient was discharged.
	Specimen Related Issue	Blood Bank	B	ED	Type & screen sample for ED patient waiting to go to the OR was lost. New sample had to be sent and procedure was delayed.
Care Coord.	Inter-/Intra-departmental Communication Issue	(None)	C	ED	Patient taken urgently to OR from the ED without chart, allergy band, or name band.
	Documentation Issue	Wrong Patient/ Site/ Procedure	A	ED	MD note documented 2 EKG interpretations; one was for a different patient.
	Discharge Instructions Issue		A	ED	Patient sent home with wrong discharge instructions.
Device	Malfunction/ Migration/ Occlusion	Pulmonary; Procedural	D	ED	During procedural sedation team observed “loss of capnography likely due to equipment malfunction.” Patient had BVM ventilatory assistance; no decrease in oxygen saturation.
	Infiltration/Extravasation	Vascular	D	ED	Patient’s iv access infiltrated and was removed. No further intervention.
Medication	Over-anticoagulation		D	POA	Patient on coumadin with INR 6.5 in the ED (and no bleeding). IV vitamin K ordered, administered, and patient was monitored.
	Wrong Dose/ Medication/ Route/ Order/ Patient	(None)	A	ED	Insulin ordered on wrong patient, but not administered.
	Other	Electrolyte abnormality	D	POA	Incidental finding of hypokalemia thought due to patient’s home diuretic medications. Asymptomatic (so no AE) and treated with po replacement.
	Allergic Reaction	Opioids	D	ED	4mg morphine ordered, stopped at 2mg infused due to itching and redness at iv site. No intervention required.
Patient Care	Failure to Monitor	Fall/Fall Risk	C	ED	Patient with altered mental status and designated as a fall risk found ambulating throughout the ED.
	Delayed Treatment	ED Boarding/Crowding	A	ED	Delay in initial and scheduled antibiotic administration in patient with prolonged ED waiting room and boarding times.
	Delayed Treatment	ED Boarding/Crowding	A	ED	Patient with chief complaint of abdominal pain (ESI 3) had a 6hr waiting room LOS before placement in ED treatment room.
	Transfusion Reaction	(None)	D	ED	Intubated patient developed hives after being given multiple units of blood products. Monitored without other intervention or changes in care.
	Other	Language Barrier/ Interpreter Services	A	ED	Translator phone service broken. Patient’s husband had to call their son to translate.
Surgery/ Procedural	Malposition	GI	D	ED	Nasogastric tube placed for medication administration was curled in esophagus on xray requiring advancement and repositioning
	Bleeding	GU	D	ED	Blood noted after urinary catheter placement. Monitored, without further interventions required.

Table 5. – Cross cutting themes.

Frequency = Number of near misses with term / total number of near misses (=1,458)

item	N	Frequency
Delayed treatment	1,012	69%
ED boarding/crowding	997	68%
Other	247	17%
Documentation issue	238	16%
Medication Administration Error (wrong dose/ medication/ route/ order/ patient)	80	5%
Psychiatric	32	2%
Delay	22	2%
Failure to monitor	21	1%
Delayed diagnosis	20	1%
Radiology	19	1%
Opioids	18	1%
Lab services	13	1%
Wrong exam/ patient/ result/ test	12	1%
Fall/ fall risk	11	1%
Inter-/ intra-departmental communication issue	11	1%
Malposition	10	1%
Blood bank	9	1%
Gastrointestinal	9	1%
Specimen-related issue	8	1%
Allergic reaction	7	0%
Cardiac	7	0%
Pharmacy	6	0%
Benzodiazepines	5	0%
Hypotension	5	0%
Intubation/ airway	5	0%
Language barrier/interpreter services	5	0%
Neurological	5	0%
Procedural	5	0%
Test/lab result	5	0%
Vascular	5	0%
Bleeding	4	0%
Genitourinary	4	0%
Hypoxia	4	0%
Malfunction/ migration/ occlusion	4	0%
Orthopedic	4	0%
Propofol	4	0%
Wrong patient/site/procedure	4	0%
Self injury	3	0%
Transfusion reaction	3	0%
Cascading event	2	0%
Delirium, confusion, oversedation, mental status	2	0%
Electrolyte abnormality	2	0%
Glycemic event	2	0%
Infiltration/extravasation	2	0%
Iv contrast	2	0%
Pulmonary	2	0%
Sending institution failure to communicate	2	0%
Angioedema	1	0%
Confusion	1	0%
Delay in ordering (consultation, lab, imaging)	1	0%
Dysrhythmia	1	0%
Otolaryngology	1	0%
Nausea and vomiting	1	0%
Outpatient coordination or communication issue	1	0%
Over-anticoagulation	1	0%
Sepsis	1	0%
Skilled Nursing Facility/ Home Health / Nursing Home	1	0%

Prevalence of near misses

There were two difficulties when modeling occurrence of near misses from the trigger data alone: the preponderance of delays (>70% of events, overall) and the heterogeneity in the types of the remaining events. Despite this, a predictive model (logistic LASSO) for presence/absence of near misses achieves an AUC of 0.85 in both training and test samples (random 80-20 split). A more focused model with presence of delay-related near misses as the outcome (rather than any event) performs better (AUC ~0.90 in both test/train samples). Expectedly, much of the information comes from triggers C30 (Boarded in the ED >6hrs), C52 (Arrival to room time of >2hrs if ESI 12 or 3) and C6 (ED length of stay >6hours). Extrapolating to the entire population using direct standardization, we estimate that 19.3% of all visits had an ED near miss (95%CI: 18.2-20.5), and that 15.1% (13.8% - 16.3%) had a delay-related near miss.

Multiple events

We found that 262 visits had multiple events (which may be AEs or near misses). In our selected sample, which is enriched for adverse events, the presence of a near miss marginally increases the likelihood of an AE (8.9% records with a near miss event also had an AE, compared to 7.2% for those without, p = 0.04).

POA Near Misses

We also detected 80 near miss events that were present on arrival (POA). POA near misses thus accounted for a much lower percentage (5.2%) of all near misses compared to the much higher proportion of detected AEs that were POA (61%).¹³ Significant differences between POA and ED near misses were noted, in terms of distribution of severity scores and the main category of event types (p < 0.0001). Compared to ED events, POA near-misses were more commonly medication-related (POA: 60% vs ED: 17%) and less often related to patient care (18.8% vs 77.2%). Of the 80 POA near miss events, 34 (42.5%) were attributed to over-anticoagulation (compared to 1/1,458 in ED near misses). POA events were also more likely to have MERP Index levels of C and D rather than A and B (86% versus 12.4%; p < 0.0001). This is largely driven by over-anticoagulation events, 94% (32/34) of which had a MERP score of D. Apart from over-anticoagulation, delays are mentioned 13 times in POA near misses (16.3% compared to 72% in ED near misses) and communication/coordination issues 8 times (8.8% compared to 17.4% in ED).

DISCUSSION

In the course of derivation and validation of the computerized ED Trigger Tool, we applied a query of electronically specified triggers to our EMR data of 95,000 visits occurring in a 13-month period. About half of these visits contained one of 95 candidate triggers - a smaller proportion were triggered when using a subset of 29 validated triggers (Supplemental Table) to identify AEs in the ED (not near misses, not POA). Reviewing 5,582 selected records revealed a 23% near miss event rate. This selected sample is not intended to present an incidence rate, nor to be representative of those in other EDs, but rather speaks to the ability of the trigger tool to identify areas for quality improvement and performance over time. Because we have the trigger distribution for this entire population and know the associations between triggers and near miss events, we are able to estimate that 19.3% of ED visits to our facility over 13 months included a near miss event.

Many studies have focused on specific near miss event types in the ED and discrete interventions for improvement, for example, medication-related near miss events, their relation to aspects of the electronic order entry^21,22 and their interception by embedding of pharmacists in the ED.²³ However, broad, systematic explorations of near misses in emergency care and surveillance methods are very limited. One ED-based study focusing on near misses and unsafe conditions is from a pediatric ED network using a voluntary reporting system. Across 18 pediatric EDs it found 487 near misses and unsafe conditions in 16.8% of the incident reports filed with variability in reporting and in types of events. The authors emphasized the importance of these events for quality improvement purposes: “sites with culture of reporting near-misses and unsafe conditions have potential of gaining better understanding of latent safety issues…the top four reported areas likely reflect real importance to the safety of ED patients—medication issues, laboratory and radiology potential errors and the perception that many providers do not follow approved policy and procedures.”

It has been observed that event reporting systems miss the large majority of AEs and physician participation in reporting is low. Various approaches have been used to enhance capture of near miss events. Some describe use of “good catch” systems, rewarding nurses for event reporting.^24,25 Others have created ED-specific tools for physician reporting.²⁶ Patient surveys and interviews have been used to enhance capture of near misses and AEs, but with a low response rates and yields.^27,28 Though the EDTT requires standard record review, this approach identifies near misses with relatively robust yield, without the need for voluntary reporting systems or surveys. Conducted for the purposes of a research study, our record reviews were a bit more rigorous than what would be required for routine quality and safety surveillance. Yet, the average time for our first level reviewers remained low, at 7.4 minutes on average per chart.¹⁴

Reports on near misses often focus only on medical errors, which are a subset of near miss or non-harm events. Researchers may also use different definitions of near misses and adverse events. The IHI definition of an AE requires that there be physical patient harm and some intervention in response to the harm. Unsafe conditions, intercepted events, errors reaching the patient but not causing harm or events that require intervention to prevent harm are all considered near misses (MERP Index levels A-D respectively). The definition of harm used by the Agency for Healthcare Research and Quality (AHRQ),³ for example, does not require occurrence of an intervention. Because of this, many events we consider near misses would be characterized as adverse events in other studies, which underscores the importance of these events and has import for the reporting of AE and near miss rates.²⁹ For example, by the definitions we used, an allergic reaction treated with antihistamines would be an AE while a rash that is just monitored would be a near miss. A 10-fold dosing error that hits the patient but does not result in harm (due to preventative intervention) might not be reported or detected in a surveillance system that focuses solely on AEs, (where the definition of an AE requires intervention).

Whereas many AEs that occur in the ED are present on arrival, (estimated by us to account for 9% of annual ED visits), the near misses we identified generally occurred in the ED. This makes sense, as patients in a nursing facility, for example, would not be sent to the ED for a near miss but would be sent for evaluation after harm. Near miss events detected were largely Patient Care-related rather than Medication-related which reflects a different pattern from AEs. As a broad category, Patient Care includes many more activities than medication administration and associated events, with more opportunities for errors or other problems that do not translate to definitive harm.

Delays in diagnosis and treatment were often related to prolonged waiting room and ED boarding times, accounting for a large proportion of events detected (nearly 70%). These events were frequently detected with triggers that identified visits with waiting room stays for patients of over 2 hours for patients with an ESI 1-3 or admitted patients with boarding times exceeding 6 hours. It is likely that many patients whose waiting or boarding times fell below these thresholds also had delays, so this approach may underestimate their occurrence. When we were able to confirm or reliably infer harm from delayed diagnosis and/or treatment (e.g. delay in time-sensitive care for MI after a long period in the waiting room), we considered such events to be AEs. However, events for which we were unable to confirm or infer harm were considered near misses. The latter category might include delayed antibiotics for infections, delays in other medications for various conditions being treated in the ED, delays in joint reductions, etc. Because ED boarding and crowding are currently ubiquitous in the US, they risk being perceived as normative, but in fact represent a significant risk to patient safety. Numerous studies describe the impacts of crowding and ED boarding, and their associations with increased morbidity and mortality.^8,30–36

Our study has some limitations. Our findings are based on a single-center retrospective record review design. Though this approach has inherent limitations, it is nevertheless considered the gold standard in adverse event studies. Events detected are partly driven by the triggers used in selecting records for review. However, the trigger base used included a broad, consensus-based set of candidate triggers in the derivation phase of the parent study and presence of one of 30 of these triggers found empirically to be associated with AEs for the validation phase. We also systematically reviewed ~140 records without triggers. Though the data are from 2014-15, (reviews performed between 2016-2019). this does not lessen the significance or our findings, the intent being to demonstrate the nature of near misses detected using the EDTT and the level to which these can be described. Changes in patterns and frequencies of near-misses over time and across sites are expected, and the EDTT provides an approach to detect such changes. The AE frequencies we report are thus not intended to generalize to other EDs and other time periods. We do not expect the nature of near misses themselves (e.g. 10-fold dosing errors, crowding, treatments leading to values outside therapeutic windows) to fundamentally change, except in their frequencies. Similarly, we do not expect the relationship between triggers and events to significantly change over time.

CONCLUSIONS

In this single center study, near miss events were relatively common (22.7% of our sample, 19.3% estimate in the population) and were associated with an increased risk for an AE. Most near misses were patient care-related (77%) involving delays related to crowding and ED boarding. Medication administration errors were the second most common events. Over-anticoagulation comprised most POA events. Though directed primarily at detecting AEs, the ED Trigger Tool, is a high yield approach for detecting important near miss events and latent system deficiencies that impact patient safety.