Delayed diagnosis of new onset pediatric diabetes leading to diabetic ketoacidosis: a retrospective cohort study

Abstract

Objectives

Patients with a delayed diagnosis of diabetes are more likely to present in diabetic ketoacidosis (DKA). The objective of this study was to assess the prevalence, risk factors, and consequences of missed pediatric diabetes diagnoses in emergency departments (EDs) potentially leading to DKA.

Methods

Cases of children under 19 years old with a first-time diagnosis of diabetes mellitus presenting to EDs in DKA were drawn from the Healthcare Cost and Utilization Project database. A total of 11,716 cases were included. A delayed diagnosis of diabetes leading to DKA was defined by an ED discharge in the 14 days prior to the DKA diagnosis. The delayed diagnosis cases were analyzed using multivariate analysis to identify risk factors associated with delay, with the primary exposure being child opportunity index (COI) and secondary exposure being race/ethnicity. Rates of complications were compared across groups.

Results

Delayed diagnosis of new onset diabetes leading to DKA occurred in 2.9 %. Delayed diagnosis was associated with COI, with 4.5 , 3.5, 1.9, and 1.5 % occurring by increasing COI quartile (p<0.001). Delays were also associated with younger age and non-Hispanic Black race. Patients with a delayed diagnosis were more likely to experience complications (4.4 vs. 2.2 %, p=0.01) including mechanical ventilation, as well as more frequent intensive care unit admissions and longer length of stays.

Conclusions

Among children with new-onset DKA, 2.9 % had a delayed diagnosis. Delays were associated with complications. Children living in areas with lower child opportunity and non-Hispanic Black children were at higher risk of delays.

Introduction

Diabetic ketoacidosis (DKA) is a life-threatening condition that is present at the time of diagnosis in approximately 28–35 % of children with type 1 diabetes mellitus [1, 2]. The presence of DKA at the time of diagnosis is associated with an increased risk of mortality, longer hospital stays, and suboptimal glycemia over time [3], [4], [5]. Given substantial increases in the prevalence of diabetes worldwide, ensuring a timely diagnosis is critically important [6]. Unfortunately, the diagnosis of new-onset diabetes mellitus is delayed in 14–38 % of pediatric patients, which increases the risk for initially being diagnosed while in DKA [7], [8], [9], [10], [11]. Previous studies have shown that younger children, particularly those younger than 2 years old, and children identifying as ethnic minorities are more likely to have delays in diagnosis compared to older patients [12]. However, apart from these two features, it is unclear what factors influence missed diabetes diagnoses which may predispose to DKA and subsequently place patients at higher risk for serious complications.

Missed diagnoses in the emergency department (ED) are common in general, occurring in approximately 5 % of adult and pediatric visits across the United States [13]. Contributors span from systems-level factors including overcrowded EDs and long wait times, clinician-specific factors such as increased cognitive burdens, and patient-specific factors like presenting symptoms and health literacy. In recent years, the number of children presenting to EDs has been rising without a concomitant rise in number of primary care physicians, further stressing EDs nationwide and negatively affecting clinical outcomes [14].

The aims of this study were to identify rates, predictors, and complications of ED-attributable delayed diagnosis of new-onset diabetes leading to DKA across a multistate population.

Methods

This was a retrospective cohort study of children under 19 years old with a first-time diagnosis of diabetes mellitus presenting in DKA. Cases were drawn from the Healthcare Cost and Utilization Project (HCUP) database. Episodes of DKA occurred between January 1, 2015, and December 31, 2019, across all EDs in eight states: Arkansas, Florida, Georgia, Iowa, Maryland, Nebraska, New York, and Wisconsin. We evaluated records for patients through 2014 to allow a 12-month wash-in to maximize capture of only new-onset cases. This study was approved by the Lurie Children’s Hospital’s Independent Review Board (2023–5967).

The cohort was identified using International Classification of Diseases 9th and 10th Editions, Clinical Modification codes (ICD-9: 250.1 (diabetes with ketoacidosis), ICD-10: E08.1 (diabetes mellitus due to underlying condition with ketoacidosis), E09.1 (drug or chemical induced diabetes mellitus with ketoacidosis), E10.1 (type 1 diabetes mellitus with ketoacidosis), E11.1 (type 2 diabetes mellitus with ketoacidosis)). Patients were excluded if they were unable to be followed between multiple ED visits because of a missing longitudinal identifier, if a transferred patient had no record from a receiving hospital which would not allow for outcome determination (e.g., if the patient was transferred out of state), if the patient was hospitalized within two weeks prior to the episode, or if the patient left against medical advice on the first presentation.

The primary outcome was delayed diagnosis of diabetes presenting in DKA, defined by an ED discharge in the 14 days prior to the DKA diagnosis [15]. 14 days was chosen based on evidence that estimates hyperglycemia-related symptoms appear 12 ± 8 days before a new onset diabetes mellitus diagnosis, increasing the likelihood that hyperglycemia was present at the index visit [16]. After the rate of delayed DKA diagnosis was calculated, the cases of delayed diagnoses were analyzed to identify risk factors associated with delay. Secondary outcomes were complications of DKA including presence of cerebral edema, coma, mechanical ventilation, and mortality. Cerebral edema was defined using diagnosis codes for cerebral edema (ICD-9: 348.4–348.5, ICD-10: G93.5–G93.6, S06.1X0A, S06.1X1A, S06.1X2A, S06.1X3A, S06.1X4A, S06.1X5A, S06.1X6A, or S06.1X7A–S061X9A), or a current procedural terminology code for hypertonic saline or mannitol (J7130, J7131, J2150). We determined alternative diagnoses from the preceding ED visits among patients with delayed diagnosis of DKA, which were categorized using Clinical Classifications Software (Healthcare Cost and Utilization Project (HCUP), 2019) based on the primary diagnosis. We also measured utilization outcomes including intensive care unit (ICU) admission and hospital length of stay (LOS).

The primary exposure was the Child Opportunity Index (COI). COI is a composite measure of neighborhood features that promote healthy childhood development and was applied based on patient zip code. The score incorporates 29 indicators across three primary domains of education, health/environment, and social/economic [17]. The secondary exposure was race/ethnicity categorized as Asian or Pacific Islander (API), non-Hispanic (NH) Black, Hispanic, Native American, NH-White, or Other. Covariates analyzed included age group (4 years or younger, 5–9 years, 10–14 years, and 15–18 years) [18], sex, payer (Medicaid, private, uninsured, or other), and presence of a complex chronic condition (CCC) [19] based on ICD codes.

Analysis

We first reported the proportion of patients with delayed diagnosis, the time between ED visits among those with a delay, and the number of ED visits needed to reach a diagnosis. Proportions of children with each alternate diagnosis were reported among those with delay (Supplementary Table 1). Trends in delayed diagnosis were summarized as a relative change per year. The change was determined using a logistic regression model with the dependent variable of delayed diagnosis and independent variable of year-quarter in which each patient was discharged.

To evaluate the association of delayed diagnosis with COI, we first determined the raw proportions of delay occurring by COI quartile, evaluating the association using a chi-square test. To address potential confounding of the relationship between COI and delay, we conducted multivariable logistic regression. All potential confounders were included as independent variables in the model, which were all covariates except for payer. We considered payer to be a mediator because it is a component of COI and lower income is a requirement for Medicaid eligibility. Type of diabetes (type 1, type 2, other/unspecified) was reported in initial patient characteristics, however, given uncertainty of diabetes type at the time of initial diagnosis, it was not included in the multivariable model. Cases with missing race or COI variables were excluded from the model.

We reported raw complication rates as proportions and evaluated the association of delay with complications using an odds ratio with 95 % CI. We also assessed the association of delay with utilization using a chi-square test for ICU admission and a rank sum test for LOS.

As a sensitivity analysis, the definition of delay was adjusted to a 7-day revisit period and the multivariable analysis was repeated. Statistical significance was defined as p<0.05. Analysis was performed using SPSS (Version 28.0, Armonk, NY) for statistical testing and R 4.2.0 (R Foundation, Vienna, Austria) for data cleaning.

Results

We included 22,017 unique patients with DKA, of whom 4,445 (20.2 %) were excluded due to prior diagnoses of diabetes mellitus in 2014, 5,845 (26.5 %) for a missing longitudinal identifier, and 11 (0.04 %) for prior admission within 14 days. We therefore analyzed 11,716 (53.2 %) patients. Demographics according to COI quartile are displayed in Table 1.

Table 1:

Demographics by COI quartile.

	COI 1 n=2,756	COI 2 n=2,527	COI 3 n=2,982	COI 4 n=2,544	p-Value
Age, years					0.06
0–4	256 (9.3 %)	255 (10.1 %)	320 (10.7 %)	272 (10.7 %)
5–9	591 (21.4 %)	545 (21.6 %)	589 (19.8 %)	536 (21.1 %)
10–14	1,199 (43.5 %)	1,072 (42.4 %)	1,354 (45.4 %)	1,151 (45.2 %)
15–18	710 (25.8 %)	655 (25.9 %)	719 (24.1 %)	585 (23.0 %)
CCC	105 (4.0 %)	100 (4.0 %)	121 (4.1 %)	98 (3.9 %)	0.97
Male sex	1,425 (51.7 %)	1,361 (53.9 %)	1,576 (52.9 %)	1,342 (52.8 %)	0.48
Race					<0.001
API	35 (1.3 %)	43 (1.7 %)	35 (1.2 %)	38 (1.5 %)
Hispanic	454 (16.5 %)	301 (11.9 %)	251 (8.4 %)	129 (5.1 %)
Native American	13 (0.5 %)	3 (0.1 %)	7 (0.2 %)	5 (0.2 %)
NH-Black	1,026 (37.2 %)	422 (16.7 %)	321 (10.8 %)	138 (5.4 %)
NH-White	686 (24.9 %)	1,210 (47.9 %)	1738 (58.2 %)	1,612 (63.4 %)
Other	211 (7.7 %)	148 (5.8 %)	127 (4.3 %)	116 (4.6 %)
Payer					<0.001
Private	700 (25.4 %)	945 (37.4 %)	1,608 (54.0 %)	1781 (70.0 %)
Medicaid	1835 (66.6 %)	1,368 (54.1 %)	1,149 (38.5 %)	597 (23.5 %)
Uninsured	115 (4.2 %)	101 (4.0 %)	100 (3.4 %)	71 (2.8 %)
Other	104 (3.8 %)	109 (4.3 %)	122 (4.1 %)	86 (3.4 %)
Diabetes type					<0.001
Type 1	2,235 (81.1 %)	2,123 (84.0 %)	2,625 (88.0 %)	2,296 (90.3 %)
Type 2	265 (9.6 %)	210 (8.3 %)	182 (6.1 %)	124 (4.9 %)
Other/unspecified	256 (9.3 %)	194 (7.7 %)	175 (5.9 %)	124 (4.9 %)

API, Asian/Pacific islander; COI, child opportunity index; CCC, complex chronic condition. NH, non-Hispanic. Only patients with non-missing COI were included in this table.

There were 342 (2.9 %) delayed diagnoses of new-onset diabetes presenting in DKA. The median interval between the preceding visit and diagnosis visits was 2 days (interquartile range [IQR] 1, 6), and 78 % of those were within 7 days. Of the 342 cases of delayed diagnosis, 322 (94.2 %) had one preceding visit within 14 days, 18 (5.3 %) had 2 prior visits, and 2 (0.6 %) had three prior visits. Among children with delayed diagnosis, the most common primary diagnoses at the preceding ED visit were other upper respiratory infections, nausea and vomiting, other gastrointestinal disorders, and abdominal pain (Supplementary Table 1). The likelihood of delayed diagnosis increased by 7.9 % per year over the study period (95 % CI 4.1–11.7).

Delayed diagnosis was associated with COI (chi-square p<0.001), with delay occurring in 4.5 , 3.5, 1.9, and 1.5 % of cases by increasing COI quartile (Figure 1).

Figure 1:

Percentage of delayed diagnoses by COI quartile.

In the adjusted analysis, children in the lowest COI quartile (OR 2.2, 95 % CI 1.4–3.5) and second quartile (2.2, 95 % CI 1.4–3.4) had a higher risk of delayed diagnosis compared with the highest quartile. Younger age and NH-Black race were each independently associated with delayed diagnosis (Table 2).

Table 2:

Risk factors for delayed diagnosis of new onset diabetes presenting in DKA. Raw delay rates are reported. Adjusted odds ratios were generated from a logistic regression model including all covariates shown in the table.

Risk factor	Delay rate	Adjusted odds ratio (95 % CI)
COI quartiles
1st COI quartile	4.5 %	2.2 (1.4–3.5)
2nd COI quartile	3.5 %	2.2 (1.4–3.4)
3rd COI quartile	1.9 %	1.3 (0.9–1.9)
4th COI quartile	1.5 %	Reference
Age
0–4 years	7.1 %	4.4 (2.9–6.5)
5–9 years	2.7 %	1.6 (1.1–2.5)
10–14 years	2.5 %	1.3 (0.9–1.9)
15–18 years	2.2 %	Reference
Race
API	4.4 %	2.1 (0.9–4.6)
Hispanic	3.2 %	1.3 (0.9–2.0)
Native American	6.7 %	3.7 (0.8–16.1)
NH-Black	5.5 %	2.6 (1.9–3.5)
NH-White	1.8 %	Reference
Other	3.2 %	1.3 (0.8–2.2)
CCC
Yes	3.5 %	1.4 (0.8–2.5)
No	2.9 %	Reference

API, Asian/Pacific islander; COI, child opportunity index; DM, diabetes mellitus; CCC, complex chronic condition; NH, non-Hispanic. Due to missing data, 9,079/11,716 (77.5 %) observations were included in the model.

Complications occurred in 260 (2.2 %) patients; these complications included cerebral edema (1.2 %), coma (1.1 %), mechanical ventilation (0.9 %), or death (0.2 %) (Table 3). Patients with a delayed diagnosis of new-onset diabetes in DKA were more likely to experience a complication than those without delay (OR 2.1, CI 1.2–3.6), including higher likelihoods of mechanical ventilation and ICU admission, and longer LOS (rank sum p<0.001).

Table 3:

Outcomes according to delayed diagnosis vs. no delay in diagnosis.

	Delayed diagnosis	No delay	Odds ratio	p-Value
Any complication	15/342 (4.4 %)	245/11,374 (2.2 %)	2.1 (1.2–3.6)	0.01
Cerebral edema	8/342 (2.3 %)	138/11,374 (1.2 %)	1.9 (1.0–3.9)	0.06
Coma	7/342 (2.0 %)	118/11,374 (1.0 %)	2.0 (0.9–4.2)	0.07
MV	7/342 (2.0 %)	99/11,374 (0.9 %)	2.4 (1.1–5.0)	0.02
Died	1/340 (0.3 %)	18/11,325 (0.2 %)	1.9 (0.2–13.8)	0.54
ICU admissiona	192/342 (56.1 %)	5,376/11,374 (47.3 %)	1.4 (1.2–1.8)	0.001
LOS (median [IQR])	2 [1, 4]	2 [1, 3]	n/a	<0.001

ICU, intensive care unit; LOS, length of stay; IQR, interquartile range; MV, mechanical ventilation. ^aICU admission may be affected by hospital policy.

We repeated the main analysis defining delayed diagnosis as having a previous ED discharge within 7 days instead of 14. In this sensitivity analysis, there were 270 delayed diagnoses (2.3 %). The rate of delay in the lowest COI quartile was 3.4 % and the highest COI quartile was 1.3 % (p<0.001). In multivariate logistic regression, the significance and direction of associations were consistent with the main analysis (Supplementary Table 2).

Discussion

Across eight states, 2.9 % of children with new-onset diabetes presenting in DKA had a delayed diagnosis. Children living in the lowest-opportunity areas have more than twice the risk of delayed diagnosis as those in the highest-opportunity areas. NH-Black children have an additional doubling of risk of delay. Children with delayed diagnosis were more likely to incur serious complications such as mechanical ventilation. Taken together, our findings indicate that disparities exist in diagnostic delays in DKA, and that such delays are associated with serious complications of illness. Beyond that, our findings indicate the importance of work to understand and improve diagnostic quality, an area of growing national importance as we recognize how common diagnostic errors are [20].

There are several potential factors driving the association between COI and delayed diagnosis. COI is derived from multiple elements across three primary domains: education, health/environment, and social/economic, any of which could play a role in a delayed DKA diagnosis. COI reflects healthcare measures including health insurance coverage, which has been shown to impact accessibility to care. Low-resource intensity ED visits, defined as visits with no laboratory work, imaging, procedures, or admission, are more likely among children from very low COI neighborhoods [21]. While a higher rate of low-resource intensity ED visits may reflect a lack of accessibility to primary care among children from low COI neighborhoods, it may also be an indicator of healthcare system and individual biases. Even when controlling for covariates including COI, we found that NH-Black race was independently associated with an increased risk of delayed diagnosis. Our results are congruent with a large cross-sectional study which showed that NH-Black and Hispanic children in pediatric EDs were significantly less likely to have imaging studies obtained during their visits than NH-White children [22]. Taken together, these findings reflect a serious disparity in the way that pediatric emergency care is being delivered to marginalized children across the US. Not only does a delayed diagnosis of DKA lead to a higher likelihood of short-term complications, but a prior study noted that NH-Black and Hispanic children with type 1 DM were more likely to have increased hemoglobin A1c over time [1], highlighting the importance of timely diagnosis and initiation of multidisciplinary care for all.

The prevalence of delayed diabetes diagnoses across all care sites ranges from 14-38 % based on the definition of delayed diagnosis and the methodology [7], [8], [9], [10]. This study is the first to our knowledge to quantify delayed diagnoses that are specifically ED-attributable, explaining the lower rate of delayed diagnoses in this cohort. Our study is concordant with others in showing the special risk of delayed diagnosis conferred by young age [10]. This is unsurprising given the relatively low prevalence of type 1 diabetes in young children and the high prevalence of viral illnesses, making misdiagnoses more common [23]. We found that the rate of delayed DKA diagnosis increased slightly over the four-year period studied, consistent with previous findings of increasing rates of delayed diagnosis of type 1 diabetes both with and without ketoacidosis and increasing rates of DKA at diagnosis [7, 24].

This study has significant implications for the approach to the undifferentiated patient in the ED. The knowledge that patients from neighborhoods with low COI, young patients, and NH-Black patients are more likely to have a delayed diagnosis of diabetes possibly leading to DKA should prompt ED systems, from initial triage through the provider encounter, to include DKA in the differential diagnosis. Given the high percentage of patients who received alternative diagnoses related to gastrointestinal symptoms at the index visit, we would propose as a next step to investigate the cost effectiveness of universal point-of-care glucose screening for patients with nausea or vomiting of unclear etiology. While our findings highlight the effects that socioeconomic disparities and racial biases have on healthcare outcomes, further research is needed to investigate the specific elements of COI that can be targeted to alleviate these inequalities, as well as interventions that can reduce individual biases such as implicit bias trainings [25].

Our study had several limitations, the first of which was the lack of clinical data available. While clinical outcomes were obtained from ICD and CPT codes, we were unable to evaluate severity of DKA using laboratory data or follow long-term clinical consequences of delayed diagnoses. Similarly, the data source cannot account for hospital policies around ICU admission or outpatient diabetes management that potentially could impact LOS. Second, although we had a one-year lookback for prior diabetes diagnoses, patients with known well-controlled diabetes or who moved in state would not have been excluded. Finally, the race variable was derived from HCUP, which imports race data from the electronic medical record. Race is often assigned to patients at registration rather than self-identified, which is not always congruent with self-identified race [26].

Conclusions

Overall, among children with new-onset diabetes presenting in DKA, 2.9 % had a delayed diagnosis. Delays were associated with complications. Children living in areas with lower child opportunity and NH-Black children were at higher risk of delays. This information is essential to reduce diagnostic disparities and subsequent complications in vulnerable patients.

Supplementary Material

Supplementary Material

Supplementary Material

This article contains supplementary material (https://doi.org/10.1515/dx-2024-0024).