Headache is a common complaint among children presenting to the emergency department (ED) and can be due to serious neurologic and nonneurologic diagnoses (SNNDs). We sought to characterize the children discharged from the ED with headache found to have SNNDs at revisits.
We performed a multicenter retrospective cohort study using data from 45 pediatric hospitals from October 1, 2015, to March 31, 2019. We included pediatric patients (≤18 years) discharged from the ED with a principal diagnosis of headache, excluding patients with concurrent or previous SNNDs or neurosurgeries. We identified rates and types of SNNDs diagnosed within 30 days of initial visit and compared these rates with those of control groups defined as patients with discharge diagnoses of cough, chest pain, abdominal pain, and soft tissue complaints.
Of 121 621 included patients (57% female, median age 12.4 years, interquartile range: 8.8–15.4), 608 (0.5%, 95% confidence interval: 0.5%–0.5%) were diagnosed with SNNDs within 30 days. Most were diagnosed at the first revisit (80.8%); 37.5% were diagnosed within 7 days. The most common SNNDs were benign intracranial hypertension, cerebral edema and compression, and seizures. A greater proportion of patients with SNNDs underwent neuroimaging, blood, and cerebrospinal fluid testing compared with those without SNNDs (P < .001 for each). The proportion of SNNDs among patients diagnosed with headache (0.5%) was higher than for control cohorts (0.0%–0.1%) (P < .001 for each).
A total 0.5% of pediatric patients discharged from the ED with headache were diagnosed with an SNND within 30 days. Further efforts to identify at-risk patients remain a challenge.
Headache is a common complaint in pediatric emergency departments (EDs) and is usually due to benign etiologies. However, some are due to serious diagnoses. We have limited understanding of children discharged with headache and later found to have serious disease.
We identified children discharged from pediatric EDs with headache and later found to have serious diagnoses. The rate of serious diagnoses is low (0.5%); these patients underwent more testing at the index ED visit than patients without serious diagnoses.
Headache is prevalent in the pediatric population, with up to 80% of children diagnosed with headache in one year1,2 and up to 40% experiencing headache in one week.3 Although there are outpatient resources for headache management,4 it is one of the most common neurologic reasons for admissions.5–7 Similarly, headache is a common chief complaint in pediatric emergency departments (EDs).8,9 Although the majority of headaches are benign, researchers in single-center prospective10 and retrospective9,11–15 studies estimate that 1% to 15% of patients with headache evaluated in the ED have serious etiologies, including vascular malformations, neoplasms, and bacterial meningitis. The identification of such cases in the ED requires careful evaluation, often aided by diagnostic testing. Neuroimaging, including computed tomography (CT) and MRI, is an important diagnostic tool to evaluate many serious neurologic conditions. However, such testing is not always indicated. Current guidelines from the American Academy of Neurology recommend neuroimaging in children with headache only if they have an abnormal neurologic examination or symptoms to suggest neurologic dysfunction.16 Although over-testing exposes children to the risks of radiation, sedation, incidental findings, and increased cost and length of stay,17–19 under-testing can lead to missed serious neurologic and nonneurologic diagnoses (SNNDs).20–22
To date, there has been limited investigation into the frequency of missed SNNDs among children discharged from the ED. Previous studies in children have been limited to single centers and small cohorts of a few hundred patients.9–15,23 Therefore, our objective was to evaluate the rates of SNNDs among pediatric patients discharged from the ED with a diagnosis of headache by using a large, multicenter cohort.
We performed a multicenter, retrospective cohort study using administrative data provided by the Pediatric Health Information System (PHIS), a database that contains billing data from ED, inpatient, ambulatory surgery, and observation units from geographically diverse children’s hospitals in the United States affiliated with the Children’s Hospital Association (Overland Park, KS). Contributing hospitals are located in 26 states and the District of Columbia. Encounters are deidentified but assigned unique patient identifiers to allow for tracking across multiple encounters. The Children’s Hospital Association and participating hospitals jointly ensure the data quality and integrity. We included data from 45 hospitals in PHIS for this study. The Ann and Robert H. Lurie Children’s Hospital Institutional Review Board deemed this study exempt from the requirement for informed consent.
We included patients ≤18 years of age with an International Classification of Diseases, 10th Revision, Clinical Modification (ICD-10-CM), principal diagnosis code of headache (R51.X [headache], G43.X [migraine], and G44.X [other headache syndromes]) who were billed for ED encounters and discharged between October 1, 2015, and March 31, 2019.
For patients with multiple ED encounters within the assessed time frame, we considered the first encounter as the index visit and subsequent encounters as revisits. We excluded ED encounters that were transfers from another hospital or resulted in admissions. As in previous studies,15,23,24 we excluded encounters associated with trauma or poisoning (any associated ICD-10-CM diagnosis code of S00-T88). We excluded patients who had an SNND made during or previous to the index visit. We also excluded those with previous history of neurosurgeries, defined as the presence of any previously billed procedure performed by a physician with a specialty of neurologic surgery. We applied these exclusions by considering all encounters available in PHIS for each patient, including inpatient and ambulatory surgery encounters. To determine which patients met the exclusion criteria of a previously diagnosed SNND before October 2015, when ICD-10-CM was universally implemented, we used generalized equivalence mapping, a resource provided by the Centers for Medicare and Medicaid services,25 to perform backward mapping of the ICD-10-CM codes to International Classification of Diseases, Ninth Revision codes.
To evaluate if our outcomes were confounded by a baseline incidence of SNNDs, we performed analyses of SNND rates for four other index visit diagnoses: cough (R05), chest pain (R07), abdominal pain (R10), and soft tissue complaints (M79), consistent with previous methods.24 We applied the same exclusions for these cohorts as for the primary cohort.
We evaluated the following: demographics (age, sex, race, ethnicity, primary payer, and geographic region), chronic conditions, and diagnostic testing. We categorized age into subgroups of infants and toddlers (0–2 years), young children (3–5 years), older children (6–12 years), and adolescents (13–18 years). We defined chronic conditions using the complex chronic conditions (CCCs) flag in PHIS, which is based on encounter-level diagnosis codes.26 Diagnostic tests at the index visit were defined as neuroimaging (ie, head CT or brain MRI), EEG, blood, and cerebrospinal fluid (CSF) testing.
Because secondary headaches of significant etiology can be of neurologic or nonneurologic origin, our outcomes of interest were patients with SNNDs, billed as either a primary or associated diagnosis on revisits within 30 days of the index visit. A priori, we determined 30 days as the follow-up period for diagnosis on the basis of a similar previous study in the adult ED population.17 Furthermore, 30 days is a time frame during which many acute etiologies can re-present.14,27,28 If a patient was diagnosed with multiple SNNDs during the 30-day period, the first revisit where one or more SNNDs were diagnosed was considered the revisit by which time to diagnosis was calculated.
Two investigators (A.Z.Z. and J.R.M.) developed the list of SNNDs on the basis of previous literature defining serious causes of secondary headache.14,15,24,29,30 The list consists of >100 diagnoses, each representing a distinct ICD-10-CM billable code. These were consolidated into 22 mutually exclusive categories based on the diagnoses (Table 1). For example, malignant neoplasm of the pituitary gland (C75.1) was combined with malignant neoplasm of cerebral meninges (C70.0, C70.9) and other diagnoses of central nervous system malignancies to make the category of malignant neoplasms.
For all included hospitals, we examined the length of time available for previous assessments of SNNDs and neurosurgery using data provided by the Children’s Hospital Association. We assessed the rate of return visits to the ED over the 30-day period and the proportion of patients having the outcome of interest during this time. Because patients may have had >1 SNND identified during the 30-day period, we identified the time to the first SNND diagnosis with respect to the number of days (0–7 days and 8–30 days) and number of ED revisits until diagnosis (1, 2, 3, 4, and ≥5). We reported demographics, presence of complex conditions, and rates of diagnostic testing. For control cohorts, we identified rates of SNNDs within 30 days of their index visit and compared overall rates of SNNDs with the primary cohort. Analyses were conducted with R version 3.6.1 (R Foundation for Statistical Computing, Vienna, Austria; https://www.R-project.org/).
Sensitivity and Post Hoc Analyses
First, to assess the impact of only including those with a principal discharge diagnosis of headache, we performed a sensitivity analysis expanding the inclusion criteria to include patients discharged from the ED with headache coded in any of the discharge diagnosis positions. Second, to evaluate for differences in diagnostic testing at the index visit between patients diagnosed with SNND at the index visit versus those diagnosed at revisits, we performed a post hoc analysis. In this analysis, we expanded the inclusion criteria as follows: (1) to include patients with any diagnosis of headache as in the aforementioned sensitivity analysis and (2) to include patients who had an SNND diagnosed at their index visit, regardless of whether they were discharged or admitted. We compared the rates of diagnostic testing among patients who had an SNND diagnosed on their index visit with those with an SNND diagnosed within 30 days of the index visit using χ2 testing. The P values <.05 were considered statistically significant.
A total of 181 423 encounters were identified in the initial search query. After applying exclusions, 121 621 patients were retained in the study cohort (Fig 1). The median patient age was 12.4 years (interquartile range: 8.8–15.4 years), and 57.1% were female. The 45 included hospitals contributed data to PHIS for a median duration of 14.8 years (interquartile range: 7.6–15.8 years) by the end of the 30-day follow-up period on April 30, 2019.
Patients With SNNDs
A total of 608 patients (0.5%; 95% confidence interval: 0.5%–0.5%) had ≥1 SNND identified within 30 days of the index visit. Of these, 37.5% had their first SNND diagnosed within 7 days of the index visit (Fig 2), and 80.8% had their first SNND diagnosed at the first ED revisit (Fig 3). A total of 13 189 patients (10.8%) had ≥1 ED revisit within 30 days of their index visit. Of patients with revisits, 82.7% had 1, 13.6% had 2, 2.8% had 3, 0.7% had 4, and 0.3% had ≥5 revisits. Most diagnoses of headache occurred in children ≥3 years of age. There was little variation in the rate of SNNDs by sex, race, ethnicity, insurance type, and geographic region. Patients with CCCs were more likely to have an SNND diagnosed at ED revisits (1.5%) compared with those without CCCs (0.5%) (Supplemental Table 5).
Seven hundred ninety SNNDs were identified among 608 patients meeting criteria for the primary outcome (Table 2). The most common SNNDs were benign intracranial hypertension (24.1%), cerebral edema and compression (15.2%), and seizures (11.5%). Patients with SNNDs had higher rates of neuroimaging (MRI and CT), blood, and CSF studies compared with patients without SNNDs (P < .001 for each, Table 3). For example, 25.0% of patients with SNNDs had neuroimaging at the index visit, compared with 16.0% of patients without SNNDs.
We analyzed rates of SNND on revisits among four control cohorts with principal discharge diagnoses of cough, chest pain, abdominal pain, and soft tissue complaints. The incidence of SNND in control cohorts ranged from 0.0% to 0.1% (Table 4). These were significantly lower than the 0.5% incidence of SNNDs identified in the primary headache cohort (P < .001 for each control cohort comparison).
To explore whether our primary outcome is similar for patients with any associated diagnosis of headache and not only a principal diagnosis of headache, we used an expanded inclusion criteria of patients with any (primary or associated) diagnosis of headache. A total of 205 927 patients were included (Supplemental Fig 4). The rate of SNND in this cohort was 0.4% (95% confidence interval: 0.4%–0.4%). The demographics, types of SNNDs, and extent of diagnostic testing were similar to the primary cohort (Supplemental Tables 6 and 7).
Post Hoc Analysis
We performed an additional analysis to determine if patients diagnosed with SNNDs at revisits had different rates of testing at the index visit compared with patients diagnosed with SNNDs at the index visit (Supplemental Fig 5). Patients diagnosed at the index visit underwent significantly more neuroimaging, blood, CSF, and EEG studies than patients diagnosed at revisits (P < .001 for each, Supplemental Table 8).
In this multicenter study of pediatric patients discharged from the ED with headache, the rate of SNNDs diagnosed within 30 days was 0.5%. Although low, this figure was significantly higher than rates among control cohort patients with other diagnoses. Nearly 40% of patients with SNNDs were identified within a week of the index ED visit, and 80% of patients were identified at the first ED revisit. Patients diagnosed with an SNND at ED revisits had higher rates of diagnostic testing at index visits than those without SNNDs. Benign intracranial hypertension, cerebral edema and compression, and seizures were the most common SNNDs.
Our finding of an SNND rate of 0.5% among pediatric patients discharged from the ED with headache is lower than that found in other studies, which have reported rates between 1% and 15%. One single-center study reported SNNDs in 4% of the 526 patients discharged from the ED with headache.5 Another single-center study of 295 patients with headache found that 4% were subsequently diagnosed with SNNDs within 5 days of the index ED visit.14 There are several potential reasons for the differences in our rate. First, our larger sample may have resulted in regression toward the mean. Second, we excluded patients with previous neurosurgery and SNND and those with trauma and SNND at their index visit. The rate of serious disease is likely higher in these excluded patients. Third, we limited our outcome to ED revisits, which does not account for other visits and diagnoses in the outpatient setting.
Our SNND rate of 0.5% is similar to the findings in a recent study of adult patients in which researchers used similar inclusion criteria and a 30-day follow-up period.24 The most common SNNDs in that study were cerebral artery occlusion, subarachnoid hemorrhage, and transient cerebral ischemia. These etiologies, which carry high mortality or morbidity, constituted 40% of the total SNNDs in that study. In contrast, such etiologies represented a minority of pediatric patients in our study (<6%). The most common SNNDs identified in our study generally carried lower morbidity and mortality. An examination of the 86 diagnoses of hypertension (Table 2) revealed 3 patients with an ICD-10-CM diagnosis of hypertensive encephalopathy (diagnosis code I67.4). The other 83 patients were diagnosed with essential hypertension only (diagnosis code I10), which encompasses malignant, benign, and unspecified hypertension. Hypertension is typically diagnosed in the outpatient setting, unless the patient has hypertensive emergency or urgency or signs of end-organ damage.31–33 Therefore, it is possible that some of the 83 patients did not meet criteria for the diagnosis but rather were erroneously labeled as hypertensive because of elevated blood pressures at the ED revisit. Therefore, the true rate of serious diagnoses is likely lower than 0.5%.
In our study, SNNDs requiring acute interventions, such as ventricular shunt malfunction and malignant neoplasms, occurred at low rates. Cerebral edema and compression was the second-most common SNND identified, representing 15.2% of all patients with SNNDs and <0.1% of the overall patient cohort discharged from the index ED visit with headache. On review of the 120 patients with this condition, 51 patients (43%) had ≥1 other SNNDs, including, most commonly, ventricular shunt malfunction (n = 21), malignant neoplasms (n = 13), and intracranial hemorrhage (n = 11). This suggests that many patients were diagnosed with cerebral edema and compression secondary to another SNND.
A larger proportion of patients diagnosed with SNNDs at revisits had diagnostic testing at index visits compared with those not diagnosed with SNNDs. These findings suggest that patients ultimately diagnosed with SNNDs were subject to a higher degree of scrutiny at their index visit. Therefore, additional efforts to identify SNNDs at the index visit would be limited. In such patients, timely diagnosis may be facilitated through close outpatient follow-up. For example, one center referred a subset of children for neurology clinic appointments within one week of their ED visit for headache, thereby offering a safety net for patients with signs and symptoms concerning for serious secondary etiologies.15 Another study revealed that children who established headache center follow-up after the index visit were less likely to revisit the ED with headache.14 Furthermore, some SNNDs, such as hypertension, require longitudinal follow-up to definitively establish the diagnosis.31–33 Our results also highlight the need for additional research to identify which patients require neuroimaging at the index visits.16 Given the rarity of abnormal findings, a case-control strategy may best identify factors associated with pediatric SNNDs.
The findings from this study are subject to the limitations of a retrospective analysis using billing data, including errors with coding accuracy and data availability.34 Notably, the PHIS data set does not contain clinical data, including vital signs and neurologic examination findings. In addition, this study only included patients at specialty pediatric hospitals, which may limit generalizability because an estimated 85% of children in the United States seek care in general EDs.35–37 Hospitals contributed data to PHIS over varying time periods, and it is possible some patients included in the cohort were not identified as having previous SNNDs or neurosurgeries. This misclassification would result in an overestimation of the SNND rate. A repeat analysis limited to 33 hospitals that contributed data for ≥5 years before the study inclusion period revealed a similar rate of SNNDs at 30-day follow-up (0.5%), suggesting that this was not a substantial limitation (results not shown). We did not have data from outpatient visits that may have occurred, because PHIS does not include these data. Similarly, we were unable to capture subsequent visits to a different ED than the index ED. However, previous studies have indicated that only a minority of revisits are to a different ED. Researchers in one study evaluated return visits among nearly 700 000 pediatric patients at 22 EDs and found that only 0.3% of patients had a revisit to a different ED within one year of the index visit.38 In another study of pediatric patients in general and pediatric EDs, 97.7% of patients were occasional ED users, of which only 7.7% visited ≥2 hospitals in a one-year period.39 Therefore, it is unlikely that a substantial number of SNNDs were missed because of revisits to a different ED.
Among patients discharged from the ED with a diagnosis of headache, the rate of SNNDs identified within 30 days of the initial visit was 0.5%. Overall, there was a low incidence of high-acuity disease. Patients found to have SNNDs underwent more diagnostic testing at the index visits compared with patients without SNNDs, suggesting high initial clinical suspicion. For the subset of patients with clinical suspicion for serious etiologies but none identified at the index visit, this represents an opportunity for close outpatient follow-up.
Dr Zhou designed the study, analyzed and interpreted the data, drafted the initial manuscript, and approved the final manuscript as submitted; Dr Marin designed the study, supervised the data analysis and interpretation, reviewed and revised the manuscript, and approved the final manuscript as drafted; Dr Hickey conceptualized the study, supervised the data analysis and interpretation, reviewed and revised the manuscript, and approved the final manuscript as drafted; Dr Ramgopal conceptualized and designed the study, analyzed and interpreted the data, reviewed and revised the manuscript, and approved the final manuscript as drafted; and all authors approved the final manuscript as submitted and agree to be accountable for all aspects of the work.
FUNDING: No external funding.
POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no potential conflicts of interest to disclose.
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.