Rewers A, Chase HP, Mackenzie T, Walravens P, Roback M, Rewers M, Hamman RF, Klingensmith G. Predictors of Acute Complications in Children With Type 1 Diabetes. JAMA. 2002;287(19):2511-2518. doi:10.1001/jama.287.19.2511
Author Affiliations: Department of Pediatrics, Pediatric Emergency Medicine (Drs A. Rewers and Roback), the Barbara Davis Center for Childhood Diabetes (Drs Chase, Walravens, M. Rewers, and Klingensmith), and Department of Preventive Medicine and Biometrics (Drs Mackenzie, M. Rewers, and Hamman), University of Colorado School of Medicine, Denver.
Context Diabetic ketoacidosis and severe hypoglycemia are acute complications
of type 1 diabetes that are related, respectively, to insufficient or excessive
insulin treatment. However, little is known about additional modifiable risk
Objective To examine the incidence of ketoacidosis and severe hypoglycemia in
children with diabetes and to determine the factors that predict these complications.
Design, Setting, and Participants A cohort of 1243 children from infancy to age 19 years with type 1 diabetes
who resided in the Denver, Colo, metropolitan area were followed up prospectively
for 3994 person-years from January 1, 1996, through December 31, 2000.
Main Outcome Measures Incidence of ketoacidosis leading to hospital admission or emergency
department visit and severe hypoglycemia (loss of consciousness, seizure,
or hospital admission or emergency department visit).
Results The incidence of ketoacidosis was 8 per 100 person-years and increased
with age in girls (4 per 100 person-years in <7; 8 in 7-12; and 12 in ≥13
years; P<.001 for trend). In multivariate analyses,
sex-adjusted and stratified by age (<13 vs ≥13 years), the risk of ketoacidosis
in younger children increased with higher hemoglobin A1c (HbA1c) (relative risk [RR], 1.68 per 1% increase; 95% confidence interval
[CI], 1.45-1.94) and higher reported insulin dose (RR, 1.40 per 0.2 U/kg per
day; 95% CI, 1.20-1.64). In older children, the risk of ketoacidosis increased
with higher HbA1c (RR, 1.43; 95% CI, 1.30-1.58), higher reported
insulin dose (RR, 1.13; 95% CI, 1.02-1.25), underinsurance (RR, 2.18; 95%
CI, 1.65-2.95), and presence of psychiatric disorders (for boys, RR, 1.59;
95% CI, 0.96-2.65; for girls, RR, 3.22; 95% CI, 2.25-4.61). The incidence
of severe hypoglycemia was 19 per 100 person-years (P<.001
for trend) and decreased with age in girls (24 per 100 patient-years in <7,
19 in 7-12, and 14 in ≥13 years). In younger children, the risk of severe
hypoglycemia increased with diabetes duration (RR, 1.39 per 5 years; 95% CI,
1.16-1.69) and underinsurance (RR, 1.33; 95% CI, 1.08-1.65). In older children,
the risk of severe hypoglycemia increased with duration (RR, 1.34; 95% CI,
1.25-1.51), underinsurance (RR, 1.42; 95% CI, 1.11-1.81), lower HbA1c (RR, 1.22; 95% CI, 1.12-1.32), and presence of psychiatric disorders
(RR, 1.56; 95% CI, 1.23-1.98). Eighty percent of episodes occurred among the
20% of children who had recurrent events.
Conclusions Some children with diabetes remain at high risk for ketoacidosis and
severe hypoglycemia. Age- and sex-specific incidence patterns suggest that
ketoacidosis is a challenge in adolescent girls while severe hypoglycemia
continues to affect disproportionally the youngest patients and boys of all
ages. The pattern of modifiable risk factors indicates that underinsured children
and those with psychiatric disorders or at the extremes of the HbA1c distribution should be targeted for specific interventions.
Diabetic ketoacidosis (DKA) and severe hypoglycemia are the major life-threatening
complications of type 1 diabetes in children. Both are theoretically preventable.
The incidence rates of DKA and severe hypoglycemia have been reported largely
by studies completed before 1993.1- 5
However, less is known about the incidence and predictors for DKA and severe
hypoglycemia in the US population in the era after the Diabetes Control and
Complications Trial (DCCT).6- 8
Few of the previous studies examined predictors for DKA and severe hypoglycemia
in children6,8- 11
and even fewer used a prospective design.6,8,12,13
Diabetic ketoacidosis often leads to an emergency department (ED) visit
and hospital admission and contributes to the high costs of care for children
with type 1 diabetes.14,15 Cerebral
edema, a devastating complication of DKA, is one of the leading causes of
mortality among children with type 1 diabetes.16- 18
The prevalence of DKA has been estimated to be up to 30% among the patients
newly diagnosed with diabetes,19,20
but the incidence of DKA later in the clinical course of the disease is less
documented. The incidence of DKA in adolescent patients enrolled in the DCCT
was 2.8 per 100 patient-years in the intensive treatment group (n = 92) vs
4.7 per 100 patient-years in the conventional therapy group (n = 103).21
The true incidence of severe hypoglycemia is also unknown. Previous
studies, based on different definitions of severe hypoglycemia, provided rates
ranging from 3.1 to 85.7 episodes per 100 patients per year.6,7,12,13,22
The DCCT found that intensive insulin treatment in adolescent patients increased
the frequency of severe hypoglycemia (26.7 per 100 patient-years) almost 3
times more than that observed in those receiving conventional therapy (9.7
per 100 patient-years).21,23 The
incidence of hypoglycemia continues to be high even in those with poor glycemic
control and hemoglobin A1c (HbA1c) levels higher than
Poor glycemic control, family and school problems, low socioeconomic
status, ethnicity, sex, and lack of adequate health insurance have been reported
to increase risk of acute complications in children with type 1 diabetes.9,19,24 However, these predictors
have not been sufficiently characterized to allow for effective preventive
The purpose of this study was to determine the incidence and the predictors
of DKA and severe hypoglycemia in a large contemporary cohort of children
with type 1 diabetes. We hypothesized that the incidence of DKA and severe
hypoglycemia in children with type 1 diabetes remains high, despite improvements
in diabetes care. In addition, we hypothesized that underinsurance (a surrogate
for low socioeconomic status) and the presence of psychiatric disorders predict
DKA and severe hypoglycemia, independently of demographic factors, duration
of diabetes, and the level of glycemic control. If confirmed, these modifiable
risk factors could be targeted in intervention programs aimed at preventing
acute diabetic complications.
In 1993, a comprehensive computerized patient record system was established
at the Barbara Davis Center for Childhood Diabetes, Denver, Colo, to evaluate
prospectively selected diabetes-related outcomes in relation to risk factors,
treatment patterns, and comorbidity. This system was expanded in 1995. For
this study, we selected a cohort of 1243 children (583 girls, 660 boys) who
met the following inclusion criteria: (1) diagnosis of diabetes (not secondary
to other conditions) and continuous insulin treatment, (2) infancy to age
19 years at the last visit, (3) residence in the 6-county Denver metropolitan
area, and (4) at least 1 outpatient visit at the Barbara Davis Center during
the study period January 1, 1996, through December 31, 2000. This study was
approved by the Colorado multiple institutional review board.
The median age of participants was 13.0 years (interquartile range [IQR],
IQR, 9.0-16.0 years). The median age at diagnosis was 7.0 years (IQR = 4.0-11.0
years) and the median duration of diabetes at the half point of the follow-up
period was 3.3 years (IQR, 1.2-7.0 years). An average patient had 12 diabetes
clinic visits and a median follow-up period of 3.5 years (IQR, 1.7-4.9 years).
The information regarding acute events, HbA1c levels, insulin dose,
insurance status, body mass index (BMI), and comorbidities was collected at
Prior to study completion, 13.8 % of the participants were lost to follow-up
(1.3% in 1996, 1.5% in 1997, 1.9% in 1998, 5.1% in 1999, and 4.0% in 2000),
mostly because their families had moved out of state.
Acute events were identified by interview at the time of each visit
and recorded in a computerized record system.
Diabetic ketoacidosis was defined as an episode
of hyperglycemia and ketoacidosis leading to an ED visit and/or hospital admission.
The episodes of DKA at the onset of diabetes were excluded. Computerized patient
record data were verified retrospectively by reviewing medical records at
The Children's Hospital in Denver, where most of the admissions occurred.
We reviewed charts of 100 children, randomly chosen from among 274 children
hospitalized at the children's hospital with an International
Classification of Diseases, Ninth Revision, code 250.1 hospital admission
between 1996 and 2000. Detailed review of these charts identified 148 potential
DKA episodes. Of the 42 episodes at diabetes onset, 12 who lived outside the
study area and 17 who were not Barbara Davis Center patients were excluded.
The remaining 77 admissions were matched with the patient clinic records.
The study-patient records were found to be 92% (71/77) complete in recording
a possible DKA episode. Next, we applied the DCCT26
definition of DKA to the 77 patients admitted to the children's hospital,
including (1) blood glucose level higher than 250 mg/dL (>13 mmol/L); (2)
presence of large or moderate ketone level in urine or serum; and (3) arterial
blood pH level lower than 7.30, venous blood pH level lower than 7.25, or
serum bicarbonate lower than 15 mEq/L. Of the 71 episodes found in records
at both institutions, 67 (94%) met the definition. Of the 6 records found
only in the children's hospital, 5 met the definition. In summary, we found
documentation of DKA in the electronic patient records to be 92% complete
and 94% accurate.
Severe hypoglycemia was defined as a hypoglycemic
episode leading to loss of consciousness or seizure or resulting in an ED
visit or hospital admission. A validation study of the completeness of hypoglycemic
events documentation in the patient record has been previously published.27 Briefly, families of 884 patients treated at the
Barbara Davis Center completed a questionnaire designed to ascertain all episodes
of hypoglycemia that occurred during the years 1993 through 1998. The DCCT21,23 definitions of severe and milder
episodes of hypoglycemia were used to develop the questionnaire. After reviewing
records from 1993 through 1998, we found that the 11 221 computerized
clinical records of the 884 patients were 98% complete for severe and 62%
complete for milder hypoglycemic events. Because the DCCT definition of milder
episodes includes "episodes requiring assistance" are difficult to standardize
in young children, we included only severe hypoglycemic events in our study.
Underinsurance was defined as a lack of insurance,
receiving Medicaid, participating in the Colorado Resident Discount Program,
or having Child Health Plan Plus insurance. We used underinsurance as a main
indicator of low socioeconomic status because it is easier to ascertain and
standardize insurance coverage over time than it is to determine a family's
income or educational level, level of dysfunction, or other resources.
Psychiatric disorder was defined as having
1 or more of the Diagnostic and Statistical Manual of Mental
Disorders, Fourth Edition Axis I and Axis II diagnoses, including major
depression, bipolar, anxiety, or panic disorder and/or use of psychotropic
medication at any time during the study period. Attention deficit hyperactivity
disorder and eating disorders were not included in this definition. Major
psychiatric disorders were selected because of previous clinical observations
that they may reflect systematically missed insulin injections, missed meals,
infrequent blood glucose testing, suicidal attempts, chaotic lifestyle, and
lack of compliance.4
The mean insulin dose (per kilogram of body weight) and the mean HbA1c level (measured in percentage) during the study period were calculated
for each patient after excluding data from the initial year after diagnosis.
Body mass index (BMI) was calculated as weight in kilograms divided by height
in meters squared.
Statistical analyses of the data were performed using SAS software version
8.2 (SAS Institute Inc, Cary, NC). Incidence rates of DKA and severe hypoglycemia
were calculated by summing the number of events that occurred during the follow-up
and dividing by total person-years of follow-up and presented per 100 person-years.
Medians and IQRs were calculated for the demographic and laboratory patient
characteristics and are reported in Table
1 and Table 2.
We identified predictors of severe hypoglycemia, using univariate and
multivariate Poisson regression of the counts of severe hypoglycemia. This
was done using PROC GENMOD function in the SAS software program. In these
Poisson regressions, variable follow-up times were accounted for by using
offset equal to the logarithm of the follow-up times. The same methods were
carried out to determine predictors of DKA. Regression models were sex adjusted
and stratified by age (<13 vs ≥13 years) because of significant interactions
between age and HbA1c level, mean insulin dose per kilogram, and
Backwards stepwise logistic regression analysis was used to identify
predictors of the recurrence of DKA or severe hypoglycemia after adjusting
for the length of the follow-up. The Hosmer-Lemeshow goodness-of-fit test
indicated very good fit for the models reported. There was little evidence
of collinearity in inspection of tolerance warnings, SEs, and correlation
matrix. Regression models were stratified by age (<13 vs ≥13 years)
because of significant interactions of some independent variables with age.
Generally, interaction terms were considered significant at P<.01. For DKA significant interactions between age and HbA1c was P = .06; psychiatric disorders, P = .01; insulin dose, P = .009;
and race, P = .03. For severe hypoglycemia significant
interactions between age and sex were P = .08; HbA1c, P<.001; and psychiatric disorders, P = .009.
There were 320 hospital admissions and/or ED visits due to DKA in the
study cohort during 3994 person-years of follow-up. The overall incidence
of DKA was 8 per 100 patient-years and increased significantly (P<.001 for trend) with age in girls (4 per 100 person-years in those
aged <7, 8 in those 7-12, and 12 in those ≥13 years; P<.001 for trend), but not in boys (7 per 100 person-years in those
aged <7, 5 in those 7-12, and 8 in those ≥13 years; Figure 1). Demographic and laboratory characteristics of patients
with and without DKA are compared in Table
1. Using univariate analysis, female sex, longer duration of diabetes,
higher mean HbA1c level, higher reported insulin dose, the presence
of psychiatric disorders, and underinsurance predicted DKA.
The Poisson multivariate regression analysis to identify potentially
modifiable predictors of DKA was age stratified and sex adjusted. Figure 2 summarizes factors found to be significantly
and independently predictive of DKA. In younger children, the relative risk
of DKA increased with higher HbA1c level (relative risk [RR], 1.68
per 1% increase; 95% confidence interval [CI], 1.45-1.94; P<.001) and with higher reported insulin dose (RR, 1.40 per 0.2
U/kg per day; 95% CI, 1.20-1.64; P<.001). In older
children, the RR increased with higher HbA1c level (RR, 1.43 per
1% increase; 95% CI, 1.30-1.58; P<.001), higher
reported insulin dose (RR, 1.13 per 0.2 U/kg per day; 95% CI, 1.02-1.25; P<.02), underinsurance (RR, 2.18; 95% CI, 1.65-2.95; P<.001), and the presence of psychiatric disorders in
boys (RR, 1.59; 95% CI, 0.96-2.65) and especially in girls (RR, 3.22; 95%
CI, 2.25-4.61; P<.01 for interaction).
There were 768 severe hypoglycemia events in the study cohort during
3994 person-years. The overall incidence of severe hypoglycemia was 19 per
100 patient-years and decreased significantly with age in girls (24 per 100
patient-years in those aged <7, 19 in those 7-12, and 14 in those ≥13
years; P<.001 for trend) but not in boys (23 per
100 patient-years in those aged <7, 22 in those 7-12, and 20 in those ≥13
years) (Figure 3). Demographic and
laboratory characteristics of patients with and without severe hypoglycemia
are compared in Table 2. Using
univariate analysis, non-Hispanic white ethnicity, longer duration of diabetes,
the presence of psychiatric disorders, and underinsurance predicted severe
The Poisson multivariate regression analysis to identify potentially
modifiable predictors of severe hypoglycemia was age-stratified (<13 vs ≥13
and sex adjusted). Figure 4 summarizes
factors found to be significantly and independently predictive of severe hypoglycemia
risk, which for younger children increased with diabetes duration (RR, 1.39
per 5-year increment; 95% CI, 1.16-1.69; P = .001)
and underinsurance (RR, 1.33; 95% CI, 1.08-1.65; P<.008).
In older children, the risk increased with lower HbA1c level (RR,
1.22 per 1% decrease; 95% CI, 1.12-1.32; P<.001),
duration (RR, 1.34 per 5-year increment; 95% CI, 1.25-1.51; P<.001), underinsurance (RR, 1.42; 95% CI, 1.11-1.81; P<.005), and the presence of psychiatric disorders (RR, 1.56; 95%
Nearly 60% of DKA episodes occurred in 5% of children with recurrent
events (≥2 episodes) while 79% of severe hypoglycemia episodes occurred
among 14% of children who had recurrent events. A detailed characterization
of these patients is presented in Table
1 and Table 2. Eighty
percent of episodes of either event occurred among just 20% of children in
the cohort. Factors predictive of recurrent DKA and severe hypoglycemia were
evaluated in age-stratified multiple-logistic regression models, adjusted
for sex and length of the follow-up.
In younger children, the recurrence of DKA was associated with higher
levels of HbA1c (OR, 1.81 per 1% increase; 95% CI, 1.26-2.60; P<.001) and with higher reported insulin dose (OR, 1.62
per 0.2 U/kg per day; 95% CI, 1.10-2.38; P<.02).
In older children, recurrence of DKA was associated with higher HbA1c (OR, 1.64 per 1% increase; 95% CI, 1.26-2.14; P<.001),
higher reported insulin dose (OR, 1.63 per 0.2 U/kg per day; 95%, 1.18-2.24; P<.003), underinsurance (OR, 3.39; 95% CI, 1.57-7.34; P<.002), and the presence of psychiatric disorders (OR,
4.39; 95% CI, 2.05-9.40; P<.001).
Recurrence of severe hypoglycemia in younger children was not associated
with any specific predictors. In older children, the recurrence of severe
hypoglycemia was associated with diabetes duration (RR, 1.54 per 5 year increments;
95% CI, 1.13-2.10; P<.006), and underinsurance
(RR, 1.93; 95% CI, 1.10-3.38; P<.03).
The advantages of our study include prospective design, large sample
size (2-4 times larger than that of any previous prospective studies), and
broad representation of patients of different age, ethnicity, socioeconomic
status, and diabetes duration. However, the study is not strictly population-based,
representing an estimated 80% of all children with type 1 diabetes living
in the Denver metropolitan area. Future studies will have to address potential
urban and rural differences in the risk for acute diabetic complications related
to distance from tertiary pediatric diabetes centers. Although the design
of our study minimized underreporting or misclassification of events, it has
not eliminated these limitations completely due to multiple providers involved
in data collection.
Few previous studies have validated completeness of event ascertainment
using independent sources of information. In our study, the electronic patient
record proved to be a complete and accurate source. On the other hand, our
data collection system, designed to capture in real time only major diabetes
care outcomes, lacks detail concerning severity of acute events, related complications,
and costs of care.
Accrual into the study continued throughout the observation period,
and there has been a 14% loss to follow up. However, these factors were accounted
for at the level of analyses by calculation of exact person-year denominators
for participants affected by left or right censoring.
This is one of few studies to evaluate systematically potential predictors
for DKA and severe hypoglycemia in children. In this study, children with
type 1 diabetes, while managed by a state-of-the-art multidisciplinary team,
still experienced significant numbers of severe, potentially preventable complications.
The overall incidence of DKA, 8 per 100 patient-years, in this study population
was higher than that reported in the highly selective DCCT population, but
comparable observational studies before and after DCCT1,3
are summarized in Table 3.
Adolescent girls were at the highest risk for DKA, with the incidence
rate of 12 per 100 patient-years consistent with previous reports.3,19 This could be related to issues of
body image because adolescent girls with diabetes often omit insulin injections
to lose weight.30 Girls with recurrent ketoacidosis
have also been shown to exhibit more behavioral problems, lower social competence,
and higher levels of family conflict.31 We
found somewhat different predictors for DKA in prepubertal vs older children:
poor glycemic control reflected by higher HbA1c and reported higher
insulin dose predicted DKA independent of age. Higher reported insulin dose
may represent lower endogenous insulin secretion with longer diabetes duration
or insulin resistance due to puberty or obesity. In patients who miss insulin
injections, higher recommended insulin dose may reflect futile efforts on
the part of the clinician to control hyperglycemia. Among older children,
in addition, underinsurance and the presence of psychiatric disorders were
significant predictors of DKA. These factors can lead to poor compliance with
diabetes treatment and to omissions of insulin injections, the most frequent
cause of ketoacidosis.19
Our study may have underestimated the effect of mental disorders, because
eating disorders were excluded from the definition of psychiatric disorders.
Eating disorders are frequent in children with diabetes and significantly
affect patient care30 but may be difficult
The annual incidence of severe hypoglycemia in this study was 19 per
100 patient-years, similar to or higher than previously reported (Table 4). Some of the discrepancies can
be explained by different definitions of severe hypoglycemia that investigators
have used in the previous studies. A Joslin Clinic study6
used a definition similar to ours, but in a cohort of older children, aged
7 to 16 years, and found a lower rate of 8 per 100 patient-years. Consistent
with previous reports,10,12 the
highest incidence of hypoglycemia in our cohort was in the youngest children.
In older children, it was still higher than that reported from the Joslin
Clinic.6 However, our cohort included children
with documented psychiatric disorders and unstable living conditions who are
at higher risk for severe hypoglycemia but were excluded from the Joslin Clinic
study. Male teenagers in our cohort had a severe hypoglycemia incidence rate
(21 per 100 patient-years) similar to that in adolescents participating in
the DCCT intensive treatment group (26.7 per 100 patient-years) while the
rate in female teenagers (15 per 100 patient-years) was closer to that in
the DCCT conventional treatment group (10 per 100 patient-years).21
Somewhat different predictors anticipated severe hypoglycemia in prepubertal
vs older children. Duration of diabetes and underinsurance predicted severe
hypoglycemia in both age groups. The increased risk of hypoglycemia with longer
duration of diabetes is a well-known phenomenon.11,33
Underinsurance was our proxy for low socioeconomic status and unstable living
conditions. Those factors could have a significant influence on a glycemic
control and the rate of acute complications, as previously reported.23 On the other hand, studies from countries with nationalized
health care systems have shown that socioeconomic status did not appear to
play a major role in the risk of hypoglycemia.10,11
Among older children, lower HbA1c levels and the presence
of psychiatric disorders were additional significant predictors of severe
hypoglycemia. Beyond the first year of diabetes, HbA1c levels below
8% usually mark intensive insulin therapy and portend an increased risk for
severe hypoglycemia.26 Psychiatric disorders,
on the other hand, lead to severe hypoglycemia through less frequent blood
glucose testing, and more chaotic diet and sleep and wake patterns.
In this study, we have confirmed that DKA and severe hypoglycemia are
a recurrent problem in some children with diabetes. Recurrence appears to
be related to age, duration of diabetes, blood glucose control, presence of
psychiatric problems, and socioeconomic factors, which is in agreement with
previous studies.5,31 The presence
of previous episodes may help to identify patients at greater risk for developing
Acute complications in children with type 1 diabetes increase directly
and indirectly25,35 the costs
of care. Direct medical care charges associated with DKA episodes represent
28% of the direct medical care charges for all patients, and 56% for those
with recurrent DKAs.36 In 1997, the average
charge per DKA episode was $6444. Assuming that the incidence rate of DKA
in Denver applies to all estimated 150 000 prevalent cases of type 1
diabetes patients younger than 20 years in the United States,37
the direct medical cost of DKA in this age group exceeds $77 million, annually.
Using the incidence rate of severe hypoglycemia in Denver and the average
annual cost of severe hypoglycemia estimated at $174 per person,25
the direct medical cost for severe hypoglycemia in US children is $26 million.
Thus the combined direct medical care charges for DKA and severe hypoglycemia
in US children with diabetes probably exceeded $100 million per year during
the late 1990s. Further studies are needed to update these figures and to
estimate, in addition, indirect cost (eg, lost productivity and diminished
quality of life).
Although death or cerebral edema appear to be uncommon outcomes of acute
diabetic complications in our population with easy access to specialized care
(we found no death and only 2 cases of documented cerebral edema in the study
cohort), these are much more likely among patients with geographic or financial
barriers to care.37
Intervention programs that improve glycemic control have been shown
to reduce the risk for DKA and severe hypoglycemia and medical charges for
these complications.38 However, there is a
paucity of efficacy data concerning additional medical and psychosocial support
interventions to improve outcomes in children at risk. Our data suggest that
aggressive treatment of psychiatric disorders and improved socioeconomic status
(including better insurance coverage for and access to diabetes care) should
be strongly considered.
In conclusion, US children with type 1 diabetes are at high risk for
DKA (8 per 100 patient-years) and severe hypoglycemia (19 per 100 patient-years).
Those underinsured, with psychiatric disorders, or at the extremes of HbA1c distribution should be targeted for interventions aimed at these
modifiable risk factors.