Emergency Department Volume and Delayed Diagnosis of Pediatric Appendicitis: A Retrospective Cohort Study

Abstract

Objective:

To determine the association of emergency department (ED) volume of children and delayed diagnosis of appendicitis.

Summary Background Data:

Delayed diagnosis of appendicitis is common in children. The association between ED volume and delayed diagnosis is uncertain, but diagnosis-specific experience might improve diagnostic timeliness.

Methods:

Using Healthcare Cost and Utilization Project 8-state data from 2014–2019, we studied all children with appendicitis <18 years old in all EDs. The main outcome was probable delayed diagnosis: >75% likelihood that a delay occurred based on a previously validated measure. Hierarchical models tested associations between ED volumes and delay, adjusting for age, sex, and chronic conditions. We compared complication rates by delayed diagnosis occurrence.

Results:

Among 93,136 children with appendicitis, 3,293 (3.5%) had delayed diagnosis. Each twofold increase in ED volume was associated with an 6.9% (95% confidence interval [CI] 2.2, 11.3) decreased odds of delayed diagnosis. Each twofold increase in appendicitis volume was associated with a 24.1% (95% CI 21.0, 27.0) decreased odds of delay. Those with delayed diagnosis were more likely to receive intensive care (odds ratio [OR] 1.81, 95% CI 1.48, 2.21), have perforated appendicitis (OR 2.81, 95% CI 2.62, 3.02), undergo abdominal abscess drainage (OR 2.49, 95% CI 2.16, 2.88), have multiple abdominal surgeries (OR 2.56, 95% CI 2.13, 3.07), or develop sepsis (OR 2.02, 95% CI 1.61, 2.54).

Conclusions:

Higher ED volumes were associated with a lower risk of delayed diagnosis of pediatric appendicitis. Delay was associated with complications.

In this retrospective study of 93,136 children visiting 997 EDs in 8 states, every 2-fold increase in pediatric appendicitis volume was associated with a 24.1% decrease in the risk of delayed diagnosis. Delayed diagnosis was associated with intensive care utilization, perforated appendicitis, abdominal abscess drainage, multiple surgeries, and sepsis.

INTRODUCTION

Appendicitis is the most common surgical emergency in children.^1–3 Delayed diagnosis of appendicitis can result in adverse outcomes including perforated appendicitis, leading to lengthy hospitalizations, repeated operations, long-term intravenous antibiotics, bowel necrosis, or sepsis.^4–6 It is also the second-most common condition involved in malpractice claims in pediatrics, chiefly due to diagnostic errors.^7,8 Diagnosis in the emergency department (ED) is a particular challenge because children may be developmentally unable to articulate symptoms, and conditions with overlapping presentations are common.

Diagnosis is a complex process, so institutional experience may be a key predictor of timely diagnosis.⁹ Higher condition-specific volume is associated with improved mortality and surgical quality in several conditions, but the relationship between volume and diagnostic quality is less clear.^10–12 Lower institutional volume has been associated with false positive diagnoses of pediatric appendicitis, but whether institutional volume is associated with delayed diagnosis is unknown.¹³ ED volume of pediatric patients is a proxy for an ED’s pediatric experience, and pediatric appendicitis volume is a proxy for diagnosis-specific expertise.¹⁴ The objective of this study was to evaluate the association of volume with delayed diagnosis of pediatric appendicitis; we hypothesized that appendicitis volume would be more strongly associated.

METHODS

Design, Setting, and Participants

We conducted a retrospective cohort study of children under 18 years old diagnosed with appendicitis who visited any ED in Arkansas, Florida, Georgia, Iowa, Maryland, Nebraska, New York, or Wisconsin between 2014–2019. The states included 20.9% of all US children in 2019. Data were obtained from the Healthcare Cost and Utilization Project State Inpatient and ED Databases, and states were chosen based on availability of longitudinal identifiers.¹⁵

Children with appendicitis were identified using International Classification of Diseases, 9th and 10th Editions, Clinical Modification (ICD-9-CM and ICD-10-CM) codes (ICD-9-CM 540–542; ICD-10-CM K35–37). Each child’s first diagnosis in the database was used for case identification. We excluded children transferred who had no record from the receiving ED (as might happen after crossing state lines), those hospitalized within a week prior to appendicitis diagnosis (as we were interested in ED-attributable delays), and children with no longitudinal identifier. All excluded children were counted toward EDs’ volumes, but not analyzed as participants.

Outcomes

The outcome was a probable delayed diagnosis of appendicitis using a previously validated algorithm that accurately identifies delays using administrative data and was separately validated in both pediatric and community EDs.^16,17 All potentially delayed diagnoses had two ED encounters within 7 days, with the second encounter resulting in a diagnosis of appendicitis. The algorithm assigns a probability of delay for each patient based on the interval between encounters, initial encounter diagnoses, presence of perforation at diagnosis, and other factors. The algorithm was derived and validated in distinct cohorts of children and the predicted probabilities of delay were shown to have acceptable accuracy.¹⁶

We defined a probable delay as a likelihood of delay ≥ 75%. Children without a previous ED encounter within 7 days of the appendicitis diagnosis or those with a delay likelihood < 75% were considered not to have a delay. The earliest opportunity to diagnose appendicitis was termed the “index encounter,” which was defined as the initial encounter for patients with delayed diagnosis and the diagnosis encounter for those with timely diagnosis.

Secondary outcomes were complications of appendicitis and included hospital length of stay, intensive care unit hospitalization, perforated appendicitis, abdominal abscess drainage procedure, bowel resection, multiple (>1) abdominal surgeries, and sepsis (each defined in the eMethods).⁴ All secondary outcomes were ascertained from the diagnosis encounter.

Variables

There were two exposure variables of interest: annual total ED volume of pediatric patients (<18 years old) and annual ED volume of pediatric patients with appendicitis, each calculated as the ED-level annual mean over the study period. Volumes were log₂-transformed so that estimates of effect were per doubling of volume, similar to other studies of volume effects.¹²

Covariates included patient- and ED-level features. Patient features were those that have previously been associated with delayed diagnosis of appendicitis, including age (<5 years, 5–10 years, or >10 years), sex, and the presence of a complex chronic condition (CCC).¹⁸ Hospital features that we hypothesized might mediate a volume-delayed diagnosis association included teaching status, presence of a pediatric-specific ED, log₂-transformed number of pediatric surgical operations per year, presence of ultrasound services, and presence of computed tomography (CT) services (each defined in the Online Methods, Supplemental Digital Content 1). Hospital features were obtained from the American Hospital Association (AHA) Annual Survey. Since the main data source did not have linkages to AHA data for Georgia hospitals, models containing hospital features were conducted excluding patients in Georgia.

Models

Our modeling strategy adhered to principles for high-quality volume-outcomes association studies including the use of volume as a continuous variable, reporting model fit, the use of hierarchical modeling, and reporting on the relative importance of variables.^19,20

Hierarchical logistic regression models with the outcome of probable delayed diagnosis were constructed. An ED-only model assessed variability in delayed diagnosis rates between hospitals and included only a random intercept for ED. We then created two volume-adjusted models for ED volume and appendicitis volume, separately adding each to the ED-only model as fixed effects. Patient-adjusted models were created by adding patient features to the volume-adjusted models. Finally, we created full models by adding hospital features to the patient-adjusted models. A complete description of model specifications is in the Online Methods, Supplemental Digital Content 1.

Analysis

We summarized demographic information. Primary diagnoses given at the initial ED encounters among patients with delayed diagnosis were grouped using Clinical Classifications Software.²¹

The association of volume and delay was reported as the fixed effect of volume, with and without adjustment for patient features. We reported predictions of the probability of delayed diagnosis by volume, calculated as marginal means and 95% confidence intervals (CIs) at the means of each covariate.

Model fit

We used three metrics to evaluate model fit (the closeness of model predictions to actual data) and explained variance (how much a model explains differences in delayed diagnosis frequency).^11,19 First, to assess overall model fit, we used the Akaike information criterion (AIC). Second, likelihood ratio tests (LRTs) evaluated the improvement in fit of a nested model over its parent model (for example, by comparing a model adding patient features to a volume-only model). Third, we compared changes in random intercept variance estimates between models to assess how much between-ED variation in delay was explained by the addition of variables.¹¹ For example, if adding volume to a random intercept-only model resulted in a decrease in between-ED variance, it would indicate that differences in volume accounted for variation in delay rates between EDs.

Mediators

Evaluation of hypothesized mediators of the volume-delayed diagnosis association was performed by comparing the full models to patient-adjusted models. We examined the importance of each hospital feature in the full models by evaluating the change in model fit introduced by removing each hospital feature individually.¹¹ The degree of mediation was measured as the change in coefficient for volume when adding hospital feature to the model, with 10% defined as the cutoff for mediation.²² Georgia was excluded from the mediator analysis because the state’s data source lacked hospital characteristics.

Complications

Delayed diagnosis may predispose patients to a worse course. We therefore compared illness outcomes between patients with and without delayed diagnosis. Complication rates were compared by presence of delayed diagnosis using bivariable statistics including rank sum test for length of stay and univariable logistic regressions for categorical variables.

Sensitivity and Exploratory Analyses

The robustness of the volume-delay relationship was tested using several sensitivity analyses. We recreated volume-adjusted models for several subsets of the whole cohort: excluding patients transferred during the index encounter (to eliminate any lack of clarity in attribution of delayed diagnosis to a given ED), and excluding patients in Arkansas and Florida hospitals (because of the number of excluded encounters due to missing longitudinal identifiers). We also repeated the models stratified on each of several variables: sex (given the significant differences in the differential diagnosis of abdominal pain), insurance (as insurance may create differential access to care that could affect the time to presentation and the associated likelihood to identify appendicitis as the condition progresses). Finally, we reconstructed volume-adjusted models at different thresholds of delayed diagnosis likelihood: >0%, ≥50%, and ≥90%.

In a post-hoc analysis, we explored the relationship of hospital payer mix with delayed diagnosis. For each hospital we determined the proportion of all ED patients (children and adults) with Medicaid or no insurance (i.e. self pay or free care). We constructed payer models by adding the payer mix variable to the volume-adjusted models.

Statistical significance was defined as a two-sided p < 0.05. All analyses were conducted using R version 4.2.0 (R Foundation, Vienna, Austria).

RESULTS

There were 120093 children with appendicitis evaluated in 997 EDs. We analyzed 93136 (77.6%) after excluding 26619 (22.2%) for a missing longitudinal identifier (of whom 22091 were in Florida), 306 (0.3%) for hospitalization within a week before the diagnosis of appendicitis, and 32 (0.0%) for having an index encounter transfer without receiving encounter information. Demographic characteristics are shown in Table 1.

TABLE 1:

Characteristics of patients and hospitals

Patient characteristics (n=93136)	n (%)	Missing, n (%)
Age (years)		0
<5	4299 (4.6)
5–10	39155 (42.0)
>10	49682 (53.3)
Female	37064 (39.8)	0
State		0
Arkansas	2411 (2.6)
Florida	17151 (18.4)
Georgia	16122 (17.3)
Iowa	6123 (6.6)
Maryland	7549 (8.1)
Nebraska	3430 (3.7)
New York	30782 (33.1)
Wisconsin	9568 (10.3)
Insurance		114 (0.1)
Public	45648 (49.0)
Private	41547 (44.6)
Other	5827 (6.3)
Complex chronic condition	8597 (9.2)	0
Transferred during index encounter	10001 (10.7)	0
Abdominal procedure during diagnosis encounter	76286 (81.9)	0
Length of stay during diagnosis encounter, median (IQR) days	1 (1–2)	56 (0.1)
Hospital characteristics (n=997)
ED volume of children per year, median (IQR)	3472 (1296, 8021)	0
Pediatric appendicitis volume per year, median (IQR)	7 (3–18)	0
Pediatric surgical procedures per year, median (IQR)	7 (1–27)	0
Teaching hospital	147 (17.1)	138 (13.8)*
Pediatric ED	324 (37.7)	138 (13.8)*
Computed tomography services	841 (97.9)	138 (13.8)*
Ultrasound services	813 (94.6)	138 (13.8)*

^*132/138 (96%) missing observations were due to unavailability of linkages between Georgia data and American Hospital Association Survey data

Probable delayed diagnosis of appendicitis occurred in 3293 (3.5%) patients. The interval between encounters was a median of 1 day (interquartile range [IQR] 0–1 days). Among these patients, 1733 (52.6%) changed EDs between their initial and diagnosis encounters. When patients did change EDs, the annual pediatric volume was a median 6.0 times higher (IQR 2.3–12.7) at the diagnosis ED compared with the initial ED. The most common initial encounter diagnoses were abdominal pain (1761, 53.5%), nausea and vomiting (323, 9.8%), other gastrointestinal disorders (300, 6.4%), non-infectious gastroenteritis (212, 6.4%), and urinary tract infection (158, 4.8%) (Table, Supplemental Digital Content 2).

A twofold increase in pediatric ED volume was associated with a 6.9% (95% CI 2.2, 11.3) decreased odds of delayed diagnosis without adjustment for patient features (Figure 1). With adjustment for patient features the decrease was 8.3% (95% CI 3.4, 12.8). Every twofold increase in pediatric appendicitis volume was associated with a 24.1% (95% CI 21.0, 27.0) decreased odds of delayed diagnosis without adjustment for patient features and 26.0% (95% CI 23.0, 29.0) with adjustment. ED volumes ranged from 45 to 199358 children/year and predicted delayed diagnosis frequencies were 6.7% (95% CI 4.7, 9.5) and 2.8% (95% CI 2.1, 3.7) at the two volumes respectively, without adjusting for patient features. Appendicitis volumes ranged from 1 to 1139 children/year and predicted delayed diagnosis frequencies were 11.0% (9.5, 12.7) and 0.7% (0.6, 1.0) respectively, without adjusting for patient features.

Figure 1: Association of ED volume and delayed diagnosis.

Estimated frequencies and 95% confidence intervals of delayed diagnosis depending on annual ED volume of children (left) and annual ED volume of children with appendicitis (right). Estimates are based on marginal predicted probabilities with (green) and without (blue) adjustment for patient features taken at the means of other model covariates.

Model fit

Model fit was improved by adding pediatric ED or appendicitis volume (Table 2). ED volume or appendicitis volume accounted for 5.1% and 36.6% of between-ED variability in delayed diagnosis frequencies, respectively (Table 2). Patient features improved model fit but did not account for changes in between-ED variability.

TABLE 2: Association of doubling ED or appendicitis volume and delayed diagnosis of appendicitis.

Model fit is shown using the Akaike information criterion (AIC). A null model with fixed intercept only had an AIC of 24139. Likelihood ratio tests (LRTs) compared each model to its parent; the parent of the volume models was the random intercept only model; the parents of the volume/patient models were the volume models. Model standard deviations (σ) describe the amount of between-ED variability in delayed diagnosis rates.

Model	Volume Effect, aOR (95% CI)	AIC	LRT p	Between-ED variability, σ (95% CI)	Between-ED variance explained, %
ED random intercept only		26460	<0.001	1.02 (0.95–1.11)	0
ED volume	0.93 (0.89, 0.98)	26454	0.005	1.00 (0.92–1.08)	5.1
ED volume and patient features	0.92 (0.87, 0.97)	25865	<0.001	1.05 (0.97–1.14)	−5.5
Appendicitis volume	0.76 (0.73, 0.79)	26310	<0.001	0.82 (0.75–0.89)	36.6
Appendicitis volume and patient features	0.74 (0.71, 0.77)	25709	<0.001	0.85 (0.78–0.93)	31.0

Mediators

The only hospital characteristic that mediated the association of ED volume and delayed diagnosis was pediatric surgical volume, which fully accounted for the ED volume-delay association (Table, Supplemental Digital Content 3). Teaching hospital status (aOR 1.67, 95% CI 1.32, 2.11) and pediatric surgical volume (aOR 0.76 per twofold increase, 95% CI 0.73, 0.79) were each associated with delayed diagnosis, and each improved model fit. The association of appendicitis volume and delayed diagnosis was modestly mediated by pediatric surgical volume, which decreased the volume coefficient toward the null by 15.6%. Model fit was improved only by adding teaching hospital status (aOR 1.47, 95% CI 1.17, 1.85).

Complications

Length of stay was longer in patients with delayed diagnosis of appendicitis (median 2 days [IQR 1–4] versus 1 day [IQR 1–2]). Patients with delayed diagnosis were more likely than those with timely diagnosis to be hospitalized in the intensive care unit (OR 1.81, 95% CI 1.48, 2.21), have perforated appendicitis (OR 2.81, 95% CI 2.62, 3.02), undergo abdominal abscess drainage (OR 2.49, 95% CI 2.16, 2.88), have multiple abdominal surgeries (OR 2.56, 95% CI 2.13, 3.07), or develop sepsis (OR 2.02, 95% CI 1.61, 2.54) (Table 3). There was no significant difference in the frequency of bowel resection.

TABLE 3:

Outcomes of patients with timely or delayed diagnosis of appendicitis.

Outcome	Timely N=61174 n (%)	Delayed N=2240 n (%)	OR (95% CI)	p
Hospitalization, median days (IQR)	1 (1–2)	2 (1–4)		<0.001
Intensive care unit hospitalization	1608 (1.8)	105 (3.2)	1.81 (1.48, 2.21)	<0.001
Perforated appendicitis	16761 (18.7)	1291 (39.2)	2.81 (2.62, 3.02)	<0.001
Abdominal abscess drainage	2452 (2.7)	215 (6.5)	2.49 (2.16, 2.88)	<0.001
Bowel resection	334 (0.4)	15 (0.5)	1.23 (0.73, 2.06)	0.39
>1 abdominal surgery	1410 (1.6)	129 (3.9)	2.56 (2.13, 3.07)	<0.001
Sepsis	1094 (1.2)	80 (2.4)	2.02 (1.61, 2.54)	<0.001

CI: confidence interval, IQR: interquartile range, OR: odds ratio

Sensitivity and Exploratory Analyses

Delayed diagnosis frequencies varied from 1.4–5.7% depending on the threshold of delay likelihood used to define delayed diagnosis. With one exception, the association of ED volume and delayed diagnosis was not significantly different than in the main analysis (Table 4). In the private payer subgroup, the association of ED volume and delay was more pronounced: −13.9% per doubling of ED volume (95% CI −18.8, −8.5). In all sensitivity analyses, the association of appendicitis volume and delayed diagnosis was not significantly different from the main analysis.

TABLE 4: Sensitivity analyses.

Each included a random intercept for ED and a volume term. Analyses were not adjusted for patient features. The effect of volume is reported as the change in delayed diagnosis frequency per twofold increase in volume.

			Annual ED volume of children		Annual volume of pediatric appendicitis
Cohort	Patients	Delayed diagnosis, n (%)	Effect of volume, % (95% CI)	SD of ED random intercept	Effect of volume, % (95% CI)	SD of ED random intercept
Main analysis	93136	3293 (3.5)	−6.9 (−11.3, −2.2)	1.00 (0.92, 1.08)	−24.1 (−27.0, −21.0)	0.82 (0.75, 0.89)
Index transfers excluded	83135	3250 (3.9)	−5.7 (−10.1, −0.9)	0.96 (0.89, 1.04)	−22.4 (−25.3, −19.3)	0.80 (0.73, 0.87)
Age ≥ 5 years	89390	3012 (3.4)	−7.9 (−12.3, −3.1)	1.00 (0.92, 1.08)	−24.6 (−27.5, −21.5)	0.82 (0.75, 0.89)
Females only	37064	1542 (4.2)	−10.0 (−14.9, −4.7)	0.94 (0.85, 1.05)	−25.0 (−28.2, −21.6)	0.75 (0.66, 0.84)
Males only	56072	1751 (3.1)	−5.8 (−11.1, 0.1)	1.06 (0.97, 1.17)	−23.6 (−27.0, −19.8)	0.88 (0.79, 0.97)
Public payer only	45648	1891 (4.1)	−8.0 (−13.1, −2.3)	0.99 (0.90, 1.09)	−24.1 (−27.4, −20.6)	0.80 (0.72, 0.89)
Private payer only	41547	1138 (2.7)	−13.9 (−18.8, −8.5)	0.98 (0.86, 1.10)	−26.0 (−29.6, −22.3)	0.78 (0.67, 0.89)
AR/FL excluded	73574	2552 (3.5)	−5.8 (−11.1, −0.0)	1.03 (0.94, 1.13)	−24.2 (−27.8, −20.4)	0.86 (0.78, 0.94)
Delay > 0% likely	93136	5298 (5.7)	−4.0 (−8.0, 0.3)	0.92 (0.86, 0.99)	−22.2 (−24.8, −19.5)	0.74 (0.68, 0.80)
Delay ≥ 50% likely	93136	3865 (4.1)	−5.1 (−9.5, −0.4)	0.99 (0.91, 1.06)	−23.2 (−26.0, −20.2)	0.81 (0.74, 0.88)
Delay ≥ 90% likely	93136	1340 (1.4)	−3.3 (−9.1, 3.2)	1.06 (0.95, 1.17)	−21.3 (−25.1, −17.2)	0.88 (0.79, 0.99)

CI: confidence interval, ED: emergency department, SD: standard deviation

Payer mix was associated with delayed diagnosis: as the hospital proportion of patients with Medicaid or no insurance increased, the likelihood of delayed diagnosis increased (Table, Supplemental Digital Content 4). In the hospital volume model, the odds of delayed diagnosis was 2.4-fold (95% CI 1.2, 4.8) higher among patients visiting hospitals with ≥75% Medicaid or no insurance compared with <25%. In the overall pediatric appendicitis volume model, the increase was 3.0-fold (95% CI 1.7, 5.3).

DISCUSSION

In eight states representing one fifth of US children, probable delayed diagnosis of pediatric appendicitis occurred in 3.5% of children. EDs exhibited substantial variation in the frequency of delayed diagnosis of pediatric appendicitis. Delayed diagnosis was most likely in EDs with lower volumes of children generally and children with appendicitis specifically. More than one third of the variability in EDs’ delay frequencies was explained by pediatric appendicitis volume. Delays were associated with serious complications of appendicitis including longer hospitalizations, perforated appendicitis, abscess drainage procedures, multiple surgical procedures, and sepsis. Thus, children visiting EDs with lower volumes are at greater risk of delay-associated complications.

The association of appendicitis volume with delayed diagnosis was substantial, while overall ED volume was only modestly associated with delay. Diagnosis-specific experience may be more important than pediatric experience generally in the diagnostic evaluation of children with possible appendicitis. Indeed, pediatric surgical volume fully mediated the association of ED volume and delayed diagnosis, but only modestly mediated the effect of appendicitis volume. It is possible that specialty- and diagnosis-specific resources are relevant to reducing the risk of delayed diagnosis. This would not be surprising, since the children most likely to experience a delayed diagnosis are those with more subtle, atypical presentations, which are more difficult to discern from more benign, common conditions.²³

The problem of delayed diagnosis is probably more widespread than our data indicate. We used a conservative definition in this study, to limit false positive assignments of delayed diagnosis. In sensitivity analyses, relaxing the definition of delayed diagnosis increased the frequency of delayed diagnosis to as high as 5.7%, concordant with other studies.²⁴ The diagnoses at the preceding encounters of patients with delay appeared related to the ultimate diagnosis of appendicitis, with abdominal conditions accounting for more than 90%. Regardless of the threshold of certainty used to define delayed diagnosis, the volume-delay association was consistent, and similar to prior estimates.²⁵

What can be done to improve diagnostic timeliness for pediatric appendicitis? First, evidence-based abdominal pain protocols for children should be available and applied in all emergency care settings. Protocols relying on the pediatric appendicitis score or pediatric appendicitis risk calculator can provide estimates of the likelihood of appendicitis.^26,27 Appendicitis imaging is cost-effective even at low probabilities of appendicitis, although imaging must be interpreted in the context of individual patients.^28,29 Protocol recommendations are often not followed in cases of missed appendicitis, even among highly experienced clinicians.²³ Such protocols have the potential to narrow disparities in appendicitis care and could also reduce unnecessary imaging in patients with a very low probability of appendicitis.³⁰ Second, pediatric emergency medicine and pediatric surgical expertise should be made more available to community EDs. Pediatric emergency medicine specialists exist almost exclusively in urban areas, and 87% of US counties have no such specialists.³¹ Teleconsultation is a promising means of bringing pediatric expertise to lower-volume community EDs.³² Regional pediatric referral centers and community EDs have partnered successfully to make decisions about management and disposition across a range of complaints.³³ Finally, graduate and continuing medical education offerings in emergency medicine should focus on improving physician accuracy in evaluating the likelihood of conditions like pediatric appendicitis.³⁴

Rates and predictors of delayed diagnosis have previously been challenging to investigate for two reasons: first, studies that can specifically identify cases with delay have generally required expert case review, which is laborious and costly.³⁵ This approach is also prone to selection bias, since such studies are typically conducted in academic medical centers, yet most children are seen in community EDs.³⁶ Second, large, population-based studies are insufficiently specific because they often rely on an over-inclusive definition of delay, such as the existence of any preceding encounter before diagnosis.^24,37,38 In this study, we overcame these limitations through the use of a large sample size, investigation of all types of EDs, and the use of a specific method of delay identification.

This study had two main limitations. First, we did not evaluate clinical records of patients assigned as having delayed diagnosis, limiting our certainty about the presence of delay. However, the approach used to identify patients with delay was previously shown to have high accuracy.¹⁶ Additionally, the interval between ED encounters among delayed diagnosis patients, as well as their first-visit diagnoses have face validity for patients likely to have experienced delay. Second, there may be residual confounding in the diagnosability of patients between EDs; patients visiting higher-volume EDs may have had more clinically apparent appendicitis. Given the effect size of the appendicitis volume estimate, confounding would have to be very severe to eliminate the association, which is unlikely. In addition, our estimates were consistent across a range of sensitivity analyses, suggesting that the population of patients was irrelevant to the size of the volume effect.

In conclusion, higher pediatric volume and pediatric appendicitis volumes in EDs were associated with a lower risk of delayed diagnosis of appendicitis. Delayed diagnosis was associated with longer hospitalizations and a greater likelihood of intensive care unit hospitalization, perforated appendicitis, abdominal abscess drainage, multiple abdominal surgeries, and sepsis. Making resources from higher-volume EDs available to lower-volume EDs may improve diagnostic timeliness and outcomes.

Data Availability Statement:

The authors cannot share the data due to agreements with the data source, however they are publicly available for purchase. The statistical code to generate the results are available from Dr. Michelson upon reasonable request.