Nonspecific Diagnoses and Return Visits Among Children Discharged From the Emergency Department

We performed a retrospective study from 40 US hospitals which contribute data to the Pediatric Health Information System (PHIS), a database containing administrative data from US children’s hospitals affiliated with the Children’s Hospital Association (Lenexa, KS). This study was approved by our institutional review board.

We included children (aged <18 years) discharged from the ED between July 1, 2021, to June 30th, 2022. We identified the first ED encounter per each unique patient during the study period, defined as the “index encounter.” We excluded patients with in-ED mortality and patients with a disposition status as left without being seen. We also excluded patients who had only billing discharge diagnosis codes (as ICD-10-CM codes) starting with V to Y and Z. Diagnosis codes of V to Y indicate external causes of morbidity (such as cause of injury or location of injury). Codes starting with Z describe factors influencing health status and contact with health services, such as do not resuscitate status or family history. We excluded these classes of diagnoses and neither can be considered as either specific or nonspecific diagnoses.

Our primary and secondary outcomes of interest were ED return visits within 7 and 30 days of the index encounter, respectively.⁹  Our predictor of interest was diagnosis, categorized as nonspecific or specific. Nonspecific diagnoses were defined as having all (primary and any secondary) ICD-10-CM codes starting with R (nondisease-based general symptoms and signs; (eg, cough). We further subcategorized these patients using the principal ICD-10-CM code by section. The section for “general symptoms and signs” (R50–R69) was further divided by individual codes (eg, R50 for fever). The ICD-10-CM section for cardiorespiratory abnormalities was divided into subcategories for cardiac and respiratory abnormalities. Subcategories occurring in <0.1% were grouped into an “other” category (Supplemental Table 4). Patients with specific diagnoses had at least 1 specific diagnosis (Fig 1). We similarly classified return visits as having a specific or nonspecific diagnosis.

We acquired variables known to be associated with return visits to develop adjusted models,^10–12  including payer status (public, private, other, or unknown)¹¹ ; race (non-Hispanic white, non-Hispanic Black, Hispanic/Latino, and others/>1); presence of a chronic complex condition (CCC, using criteria by Feudtner et al)¹³ ; Child Opportunity Index (COI; v2.0)¹⁴ ; and age (infant [aged <1 year], toddler [aged 1–3 years], child [aged 4–11 years], and adolescent [aged ≥12 years]). We included the variables of race and ethnicity on the basis of previous research demonstrating an association between these social constructs and return visits.^9,15,16 

We described our study sample overall and by each outcome. We performed time-to-event analysis, considering encounters after the outcome window as right-censored. We calculated the median time to return visit among children who had a nonspecific diagnosis and a specific diagnosis on their index visit and compared these proportions with the log-rank test. We compared the proportion hospitalized on their return visit using the χ² test. We identified the proportion children with a nonspecific diagnosis on their index visit who had a nonspecific diagnosis on their return visit. We evaluated the most common pairs of diagnoses on index and return visits among children with a nonspecific diagnosis on their index visit. To evaluate the association between nonspecific diagnoses and return visits, we developed a Cox proportional hazard model for each outcome. These models were developed by (1) considering nonspecific diagnoses as a single category, and (2) dividing nonspecific diagnoses into subcategories. The presence of a specific diagnosis was used as the reference group. We expressed our results as adjusted hazard ratios (aHRs) with 95% CI. Analyses were performed using R, v4.1.2 (R Foundation for Statistical Computing, Vienna, Austria).

First, we used generalized linear mixed models to evaluate the association between nonspecific diagnosis and return visits, using hospital as a random effect. Second, we repeated our analysis when defining nonspecific diagnosis as the presence of a nonspecific diagnosis only on the basis of their primary diagnosis (disregarding secondary diagnoses).⁷  Third, to classify diagnoses into broader groupings for a sensitivity analysis, we classified index and return visit primary diagnoses using the Diagnosis and Grouping System, a consensus-derived classification scheme which classifies diagnoses into 21 major groups.¹⁷  We repeated modeling using an outcome defined as a return visit in which the diagnosis group of the index and return visit matched. Fourth, because the application of diagnosis codes among children with medical complexity may differ from those without these conditions, we performed our analysis after removing patients with CCCs from our sample.

We identified 2 580 354 ED encounters from 1 914 182 unique patients during the study period. After subsequent exclusions, we identified 1 870 100 children discharged from the ED as index visits (Fig 2). A total of 73 956 (4.0%) had a 7-day return visit and 151 410 (8.1%) had a 30-day return visit. A nonspecific diagnosis was present among 294 701 (15.8%) index visits. Demographics are provided in Table 1.

Seven-day return visits were slightly higher among those with a nonspecific primary diagnosis on their index visit (4.2% vs 3.9%; P < .01 by log-rank test). The median time to return visit was the same (2 days, interquartile range [IQR] 1–4 days) among patients with a nonspecific and a specific diagnosis on their index visit. For children with a 7-day return visit, a similar proportion of infants were hospitalized among those with a nonspecific diagnosis (22.2%) and with a pecific diagnosis (22.0%) on their index visit (P = .58). A nonspecific diagnosis on the return visit occurred in 85.7% and 14.3% of children who had a nonspecific diagnosis and a specific diagnosis at the time of their index visit, respectively. The most common pairings of diagnoses with a specific diagnosis on return visits were those with fever or vomiting. When 7-day return visits resulted in another nonspecific diagnosis, the same principal diagnosis was frequently used (Table 2).

The aHR of having a 7-day return visit was higher among those with a nonspecific diagnosis on their index encounter (aHR 1.08, 95% CI 1.06–1.10). In a model which included subcategories, nonspecific diagnoses with the highest aHR of a return visit were fever (1.36, 95% CI 1.31–1.41), convulsions (1.27, 95% CI 1.17–1.38), digestive system and abdomen (1.17, 95% CI 1.13–1.20), and headache (1.17, 95% CI 1.05–1.29). Respiratory disorders (0.84, 95% CI 0.80–0.88), emotional state and behavior (0.84, 95% CI 0.76–0.92), and other general symptoms and signs (0.69, 95% CI 0.60–0.80) had lower aHRs of return visits (Table 3).

Children with a nonspecific diagnosis had a similar proportion of 30-day return visits compared with those with a specific diagnosis (8.6% vs 7.9%; P = .06 by log-rank test). The median time to return visit was 7 days (IQR 2–17) among patients with a nonspecific diagnosis and 8 days (IQR 3–18) among patients with a specific diagnosis on their index visit. A similar proportion of children were admitted to the hospital among those with a nonspecific diagnosis (17.4%) compared with those with a specific diagnosis (16.9%) at the time of their first return visit (P = .09). A nonspecific diagnosis occurred more frequently upon revisit among children with a nonspecific diagnosis on their index visit (86.1%, compared with 13.9% among patients with a specific diagnosis on their index visit). Diagnosis pairings between index and return visits were similar when using a 30-day return visit window as observed for a 7-day window (Supplemental Table 5).

The presence of a nonspecific diagnosis was associated with a slightly higher aHR of a 30-day return visit in a multivariable model (aHR 1.01, 95% CI 1.01–1.03). In models using subcategories of nonspecific diagnoses, many had a higher adjusted odds of a 30-day return visit relative to those with a specific diagnosis. The highest aHRs were among convulsions (1.22, 95% CI 1.15–1.29), enlarged lymph nodes (1.22, 95% CI 1.03–1.45), and headache (1.10, 95% CI 1.02–1.19). Syncope (aHR 0.69, 95% CI 0.62–0.76) and respiratory disease (aHR 0.94, 95% CI 0.92–0.97) had a lower aHR of a 30-day return visit.

Older children had a lower aHR of return visits compared with infants. Commercially insured patients had a lower aHR of return visit compared with publicly insured patients. Children from lower COI neighborhoods had a higher aHR of return visit compared with those from very high COI areas. Black non-Hispanic children had a lower aHR of 7-day return visit compared with white non-Hispanic infants. Black non-Hispanic and Hispanic children had a higher aHR for 30-day return visits compared with white non-Hispanic children.

When using a generalized linear mixed model, the adjusted odds ratio for a child with a nonspecific diagnosis on their index visit was 1.10 (95% CI 1.07–1.12) for a 7-day return visit and 1.03 (95% CI 1.02–1.05) for a 30-day return visit. Models using subcategories resembled the proportional hazard models (Supplemental Table 6).

When nonspecific diagnoses were determined using the primary discharge diagnosis, a higher proportion (n = 458 995, 24.5%) were classified as having a nonspecific diagnosis. The aHR of a nonspecific diagnosis for 7- and 30-day return visits was higher (aHR 1.15, 95% CI 1.13–1.17 and aHR 1.08, 95% CI 1.07–1.10, respectively). Findings in subcategories were similar to the primary analysis (Supplemental Table 7).

Return visits occurring within the same category occurred in 37 729 (2.0%) within 7 days and 65 174 (3.5%) within 30 days. The aHR of a nonspecific diagnosis leading to a 7-day and 30-day return visit was 1.16 (95% CI 1.13–1.19) and 1.12 (95% CI 1.10–1.15), respectively. The evaluation within subcategories resembled the primary analysis (Supplemental Table 8).

When excluding children with CCCs, the aHR for a nonspecific diagnosis to result in a 7-day return visits was 1.09 (95% CI 1.06–1.11) and to a 30-day return visit was 1.02 (95% CI 1.01–1.03). Models performed using subcategories again resembled the primary analysis (Supplemental Table 9).

We evaluated the association of a nonspecific diagnosis with pediatric ED return visits. Approximately 15% of children discharged from the ED had a nonspecific diagnosis. Overall, children with nonspecific diagnoses were 8% more likely to have a 7-day return visit.

Our finding of an 8% increased likelihood of a 7-day return visit is small but remains present in adjusted models. This effect varied on the basis of diagnosis and was greater in sensitivity analyses. Marshall et al demonstrated an association among diagnostic uncertainty and a nonspecific primary diagnosis code among admissions.⁷  Extending these findings to the current study, nonspecific diagnosis coding may be indicative of diagnostic uncertainty by the treating physician. However, this comparison has not been performed in an ED-based population, where priorities of clinicians (such as the establishment of a safe patient disposition, with or without diagnostic certainty) differ from those caring for hospitalized children.⁷ 

The higher proportion of return visits among children with nonspecific diagnoses is likely because of several causes. Disease progression may be among the most important, given the high number of children with fever who were later diagnosed with viral syndromes, or children with vomiting later diagnosed with gastroenteritis demonstrated in our pairwise analysis. For many of these pairings, progression occurred along expected lines, suggesting that establishing expectations with patients and discharge teaching may play a role in preventing some unnecessary return visits. Although some patients may have had a missed clinically important diagnosis on their initial visit, this effect is likely minimal based on previous work, suggesting that missed diagnoses accounted for <4% of pediatric ED return visits.¹⁸  A qualitative study performed among pediatric emergency medicine and hospitalist providers identified important themes within the diagnostic process, including an appreciation for the “test of time,” prioritizing disposition over diagnosis, and effective communication with families.³  Some potentially unnecessary return visits, suggested in our pairwise analysis, may be mitigated through discussions with primary care providers¹⁹  and through multimodal and engaged discharge processes that ensure families have clear expectations about what to expect and when to return to care.²⁰ 

Certain diagnoses are more clearly linked to a higher aHR of diagnostic uncertainty compared with others. The overall association when evaluating our findings by subcategory suggest that some nonspecific diagnoses (such as fever) have a higher risk of return visit than others (respiratory disorders). Patients with fever, for example, more clearly differentiate into a specific diagnosis over time. One study noted that febrile children who made a return visit had higher temperature elevations and longer durations of fever compared with those who did not have an ED return visit.²¹  Similarly, a patient with vomiting alone may more fully meet criteria for gastroenteritis after disease progression.

Other findings corroborate previously reported research about ED return visits in children. We found, for example, an association between lower COI areas and return visits, which has been previously reported.¹⁰  Our univariable associations with race and ethnicity demonstrated associations with this variable and return visits. However, their effect sizes were lower in adjusted models, suggesting that other factors (such as payer status and neighborhood opportunity) limit the role of race and ethnicity on return visits. Some of these findings may be because of difficulties in outpatient care access reported among Black and Hispanic children.^22,23  The association with CCCs and return visits is consistent with the higher degree of hospital-based care required for these patients.^12,24 

Our findings suggest of important areas for future investigation. Further work is required to establish a connection more definitively between diagnostic uncertainty and diagnosis coding in the ED. The study by Marshall et al used a case-control study with electronic medical record data to ascertain diagnostic uncertainty.⁷  Granular, and ideally, multisite chart reviews, such as have been performed for pneumonia,²⁵  fever,²⁶  or urinary tract infection,²⁷  will allow for the identification of diagnosis codes most strongly associated with diagnostic uncertainty and provide measures of sensitivity, specificity, and predictive value. In addition, qualitative research will be needed to evaluate how these codes are used in practice by physicians (whether truly for diagnostic uncertainty, expediency, defensive practices, or for other reasons) and to evaluate how a nonspecific diagnosis and diagnostic uncertainty are perceived by patients and families.

Our findings are subject to limitations. Some ED return visits may occur for unrelated reasons from the index visit (e.g., cough, followed by a return visit for a motor vehicle collision). However, previous work has suggested that there is substantial overlap between diagnoses among index and return visits for 3- and 7-day return visits when using Diagnosis and Grouping System groupings, though this decreases more for 30-day return visits.⁹  Studies which have evaluated individual cases suggest that a variable proportion of ED return visits are for unrelated reasons, ranging between 4% and 30%.^18,28,29  We were unable to evaluate return visits to other (non-PHIS) hospitals, which likely underestimates the overall ED return visit proportion.³⁰  Our findings may not generalize to nonpediatric EDs.³¹  PHIS may have errors related to data abstraction and coding, and does not contain clinical information needed to delineate disease progression versus missed diagnoses. Finally, different hospitals and coders may apply primary and secondary diagnosis codes differently. Despite these limitations, our study provides important information about the potential role of diagnostic uncertainty on pediatric ED return visits.

We used an administrative data set to identify the association between nonspecific diagnoses and ED return visits. Children discharged from the ED with nonspecific diagnoses had more return visits in adjusted models compared with those with specific diagnoses. These findings highlight the need for further research and multimodal approaches to better identify diagnostic uncertainty within medical record data and to ultimately improve care delivery during these complex situations.