Abstract
Introduction
Hyperglycemia is a common toxicity affecting between 16% and 24% of children undergoing treatment of acute lymphoblastic leukemia (ALL) during the induction, or initial, phase of treatment (Gatzioura et al., 2016; Tsai et al., 2015; Zhang et al., 2014). Hyperglycemia is associated with diabetic ketoacidosis and increased risk for life-threatening infections in children receiving ALL treatment (Dare et al., 2013; McCormick et al., 2020; Roberson et al., 2008). Children who develop hyperglycemia during treatment may be at increased risk for type 2 diabetes mellitus as survivors (Williams et al., 2019).
The predominant cause of hyperglycemia is the use of two medications in combination during the induction phase, corticosteroids and asparaginase (Howard & Pui, 2002). Exposure to corticosteroids causes increased hepatic gluconeogenesis. Corticosteroids also increase insulin resistance leaving the body's cells less able to use glucose (Pui et al., 1981). Asparaginase causes a reduction in insulin from the pancreatic ß cells in response to glucose (Howard & Pui, 2002; Pui et al., 1981). When corticosteroids and asparaginase are given together the effect is synergistic. That is, the increased demand for insulin caused by exposure to the corticosteroids in combination with a reduced insulin response, consequent to the administration of asparaginase, increases the likelihood that hyperglycemia will result (Bostrom et al., 2013).
Amiel et al. (1986) found that insulin resistance is increased during puberty in both diabetic and non-diabetic youth. It is likely this hormonal influence underlies the increased incidence of hyperglycemia in children ages 10 years and older (Grimes et al., 2019; Koltin et al., 2012). Excessive adipose tissue is associated with insulin resistance (Esbenshade et al., 2013; Meenan et al., 2019).
Associations between insulin resistance, race, and various measurements of economic status have been reported (Chen et al., 2019; Iguacel et al., 2018; Lee et al., 2019; van Den Berg et al., 2012). Families exposed to poverty or racial discrimination experience high levels of stress. The effect is repeated and prolonged activation of the body's stress responses leading to the release of cortisol causing a decrease in hepatic and extrahepatic sensitivity to insulin, thus contributing to insulin resistance (Condon, 2018). Additionally, exposure to conditions associated with poverty can exacerbate the development of insulin resistance in children through health threats, including poor regulation of diet, altered sleeping patterns, and less than optimal physical activity (Fuller-Rowell et al., 2019). Despite the known relationships between insulin resistance and race and socioeconomic status, little attention has been given to the role of these factors as predictors of hyperglycemia during ALL treatment.
Although it is often transient in nature, resolving when corticosteroids and asparaginase are stopped, hyperglycemia during ALL induction therapy is associated with increased mortality (Sonabend et al., 2009; Weiser et al., 2004; Zhang et al., 2014). Insulin resistance due to corticosteroid exposure causes an increased demand for insulin, which then binds to insulin-like growth factor receptors on the leukemic cells (Wang et al., 2019). It is hypothesized that this drives proliferation of these cells leading to lower rates of survival in both adults and children who develop treatment-related hyperglycemia (Wang et al., 2019; Weiser et al., 2004; Zhang et al., 2014). In a cohort of children diagnosed with ALL between 1999 and 2002, Sonabend et al. (2009) found that 5-year overall survival rates differed by the degree of severity of hyperglycemia. Children who maintained normal glucose levels had higher rates of survival (98.5%) compared to those who developed mild hyperglycemia (92.3%) and those with the highest elevations in serum glucose levels (73.7%; p < .001; Sonabend et al., 2009). Similarly, in a study of children diagnosed in China between 2008 and 2012, survival was 83.8% in those who developed hyperglycemia compared to 94.9% in those who did not (p = .014; Zhang et al., 2014). It should be noted that a retrospective analysis by Roberson et al. (2009) did not find an association between hyperglycemia and 5-year overall survival between those who developed hyperglycemia (81.9%) and those who did not (89.1%, p = .28; Roberson et al., 2009).
Hispanic children, Black children, and those living in areas with concentrated poverty have lower rates of overall survival following the diagnosis of ALL (Abrahao et al., 2015; Acharya et al., 2016; Bona et al., 2016; Goggins & Lo, 2012; Schraw et al., 2020; Walsh et al., 2017). Because of the association between hyperglycemia and increased mortality in children with ALL, we sought to determine if there racial and economic disparities in its incidence (Sonabend et al., 2009; Weiser et al., 2004; Zhang et al., 2014). Our study analyzed a large and diverse cohort of children hospitalized with ALL in the United States over the course of 1 year. We took advantage of the large sample size to examine race and socioeconomic status as predictors in the development of hyperglycemia.
Methods
Sample Selection
We conducted a secondary analysis of the 2016 Healthcare Cost and Utilization Project Kids' Inpatient Database (HCUP KID; Agency for Healthcare Research and Quality, 2018). The HCUP KID contains demographic and administrative information on individual hospital stays representing 80% of the pediatric hospital discharges that took place in the United States during 2016. Records in the HCUP KID were included in our sample if any of the discharge diagnoses contained the International Classification of Diseases, 10th Revision, Clinical Modification (ICD-10-CM) codes for ALL (C91.00, C91.01, or C91.02). Figure 1 depicts the process that was used to delimit the HCUP KID file to the sample of interest.
Process for the delimitation of the 2016 Healthcare Cost and Utilization Project Kids' Inpatient Database (HCUP KID) file to the sample under investigation.
Study Variables
Hyperglycemia, the outcome variable, was identified through a search of the 30 discharge variables in each record for two ICD-10-CM codes: R73.x (elevated glucose level excluding diabetes mellitus) and E09.x (drug or chemical induced diabetes mellitus). These two codes were used to best capture hyperglycemia that occurred as a result of treatment and not due to preexisting diabetes.
The HCUP KID does not contain a variable designating Hispanic and non-Hispanic ethnicity. Cases in which the hospital has reported both race and Hispanic/non-Hispanic ethnicity are recoded as Hispanic race in the HCUP KID race variable. Thus, this study will refer only to Hispanic race, and not race plus Hispanic/non-Hispanic ethnicity (Agency for Healthcare Research and Quality, 2018). Household financial status was defined using the variable provided in the HCUP KID, median annual household income for the patient's zip code. This variable varies by year. In 2016, the four levels were defined as ≤$42,999, $43,000–$53,999, $54,000–$70,999, and $71,000 or above (Agency for Healthcare Research and Quality, 2018).
Previous research has established that older children and adolescents are at greater risk for hyperglycemia during ALL treatment so age was converted to a dichotomous variable: 9 years and younger and 10–20 years (Grimes et al., 2019; Koltin et al., 2012). Obesity was included as a predictor because it is linked to insulin resistance and has been identified as a risk factor for hyperglycemia in children with ALL (Esbenshade et al., 2013; Grimes et al., 2019; Meenan et al., 2019). Documentation of the ICD-10-CM code for obesity (E66.x) determined the presence of obesity. Sex, as recorded at each hospital is coded as male and female in the HCUP KID database.
Statistical Analysis
Clinical, demographic, and social variables were analyzed to control for their possible influence on the outcome of hyperglycemia in hospitalized children diagnosed with ALL. Since all the variables were categorical, frequencies and percentages were calculated to describe the sample. Determination of the inclusion of each predictor variable in the final analyses was done through bivariate analyses using the χ2 test, α set at 0.10. Multilevel models were built with hospital-specific random intercepts to accommodate the clustering effect of children nested within the same hospital. To determine the independent effects of each predictor unadjusted multilevel logistic regression models were created for each variable that met the inclusion threshold. The relationship between race, annual median household income, and the risk of hyperglycemia was assessed in an adjusted multilevel logistic regression model using 95% confidence intervals, α set at 0.05. The odds ratios resulting from these analyses represent the odds that hyperglycemia will occur in the presence of the predictor variables, in comparison with their absence. Analyses were performed using R package LME4 (Bates et al., 2015).
Ethical Considerations
The Institutional Review Board of Rutgers University, the State University of New Jersey, determined this study was exempt from review.
Results
A description of the sample and a comparison of discharge records with and without the diagnosis of hyperglycemia, as well as the results of the bivariate analyses, are found in Table 1. Hyperglycemia was documented in 955 (5.3%) of the 18,077 hospital discharge records. When obesity was present, the rate of hyperglycemia was 19.5% in comparison with 4.6% when there was no evidence of obesity (p < .001). The rate of hyperglycemia was higher in children between 10 and 20 years of age (8.2%) than it was when the age was 9 years and younger (3%; p < .001). Black children (6.7%) and Hispanic children (6.2%) were more likely to have documentation of hyperglycemia compared to White children (4.5%; p < .001). The rate of hyperglycemia was 6% in the lowest income quartile and 4% in the highest (p < .001). There was a difference in the rate of hyperglycemia between female and male children, 6.2% and 4.6%, respectively (p < .001).
Characteristics of the Sample (N = 18,077 Hospitalizations of Children With ALL).
The p value was determined through analyses using χ2 test with Yate's continuity correction.
Between group significance p = <.001.
Annual median household income for the child's zip code area in the year 2016 in the United States.
Note. ALL = acute lymphoblastic leukemia.
The results of the multilevel binary logistic regression models testing the relationship between predictors and the odds of developing hyperglycemia can be found in Table 2. Compared to White children, Black children were found to be at increased risk of developing hyperglycemia while hospitalized with ALL (OR 1.57, 95% CI [1.25, 1.98], p < .001). This risk remained evident when the model was adjusted to include household income, age, sex, and obesity (OR 1.37, 95% CI [1.09, 1.74], p = .008). Hispanic children were at greater risk for hyperglycemia than White children (OR 1.47, 95% CI [1.26, 1.71], p < .001). However, this increased risk did not persist when the model was adjusted (OR 1.16, 95% CI [0.99, 1.37], p = .065). In comparison with children residing in areas where the median annual income was ≥$71,000, the odds of hyperglycemia was greater in the second-lowest income areas ($43,000–$53,999; OR 1.38, 95% CI [1.11, 1.71], p = .004) and the lowest income areas (≤$42,999; OR 1.33, 95% CI [1.07, 1.66], p = .009) after the model was adjusted for race, age, sex, and obesity. The risk for children living in the second-highest income areas ($54,000–$70,999) did not differ (OR 1.18, 95% CI [0.94, 1.47], p = .150).
Odds Ratio for Developing Hyperglycemia During Hospitalization With ALL a .
Hyperglycemia was documented in 955 of the total 18,077 hospital discharge records.
Annual median household income for patient's zip code.
Note. ALL = acute lymphoblastic leukemia; OR=odd ratio; CI = confidence interval.
Children and adolescents ages 10 years and older had greater odds of developing hyperglycemia than those 9 years and younger (OR 2.58, 95% CI [2.24, 2.99], p < .001). Females were at higher odds than males (OR 1.47, 95% CI [1.29, 1.68], p < .001). Children diagnosed with obesity were at higher odds than those without this diagnosis (OR 3.61, 95% CI [2.92, 4.45], p < .001).
Discussion
The development of hyperglycemia during ALL therapy is associated with higher rates of mortality (Sonabend et al., 2009; Weiser et al., 2004; Zhang et al., 2014). Equitable outcomes are reliant upon the illumination of all possible pathways between race, economic status, and inferior survival. Thus, we sought to investigate potential disparities in the incidence of hyperglycemia in hospitalized children with ALL, as this may represent one modifiable factor contributing to inferior survival rates for non-White children and those living in low-income households.
In this large and diverse sample, Black children hospitalized with ALL were 37% more likely to be diagnosed with hyperglycemia than White children when socioeconomic status, age, sex, and obesity were controlled. Hispanic children were found to be 47% more likely to develop hyperglycemia than white children. However, the association for Hispanic children was attenuated with the addition of other contributing factors. Previous studies that included race as a predictor did not find differences in their samples (McCormick et al., 2020; Pui et al., 1981; Roberson et al., 2009; Sonabend et al., 2009; Williams et al., 2019).
Children residing in areas with a median household income in the lower two quartiles faced greater odds of developing hyperglycemia compared to those living in the highest quartile. These findings differ from the secondary analysis of administrative data conducted by McCormick et al. (2020) which did not show an association between annual household income and hyperglycemia during ALL treatment. However, in that study, there was no adjustment for the co-morbidity of obesity, a strong predictor of hyperglycemia (McCormick et al., 2020).
We found older children and adolescents between 10 and 20 years to be at greater risk for the development of hyperglycemia than those under the age of 10 years. This finding is consistent with previous research of children with ALL and is likely due to the natural rise in insulin resistance that occurs during puberty (Amiel, et al., 1986; Gatzioura et al., 2016; Koltin et al., 2012; McCormick et al., 2020; Tsai et al., 2015; Zhang et al., 2014). It should be noted that the incidence of ALL is highest among children between the ages of 2 and 5 years, therefore the mean age of 8.8 years potentially reflects an overrepresentation of older children and adolescents in our sample (Hunger & Mullighan, 2015). Our finding that females hospitalized with ALL are at increased risk for hyperglycemia was consistent with the findings of McCormick et al. (2020). But, the majority of studies did not find this association (Gatzioura et al., 2016; Koltin et al., 2012; Lowas et al., 2009). In this sample, children who had documentation of obesity in the hospital discharge record were also at risk of having hyperglycemia. Obesity induces insulin resistance and has been found to be a consistent predictor of hyperglycemia in children undergoing ALL treatment (Esbenshade et al., 2013).
We discovered in this diverse nationally representative sample that Black children and those who lived in areas with lower household income were at disparate risk for hyperglycemia while hospitalized with ALL compared to White children and those living in areas with the highest median household income level. It is hypothesized that hyperglycemia is associated with increased mortality through the proliferation of leukemic cells in its presence (Wang et al., 2019; Weiser et al., 2004; Zhang et al., 2014). Despite overall gains in the survival of children following the diagnosis of ALL, non-White minority groups and those living in neighborhoods with concentrated poverty lag behind and persistently have higher rates of mortality (Abrahao et al., 2015; Acharya et al., 2016; Bona et al., 2016; Goggins & Lo, 2012; Schraw et al., 2020; Walsh et al., 2017). Our findings of racial and economic gradients in the incidence of hyperglycemia reveal a potential pathway contributing to these disparate mortality rates worthy of further exploration.
Study Limitations
A limitation of the HCUP KID database is the possibility of inaccurate administrative coding of diagnoses by the hospitals submitting information. We were reliant upon the use of diagnostic codes to identify hyperglycemia in the sample and did not have access to laboratory data for validation. This limitation may have also contributed to the low percentage of children with obesity in our sample since our study relied solely upon the documentation of obesity as a medical diagnosis and not on the measurement of height and weight. Obesity was documented in 4.3% of our sample and this is in contrast to the estimated 18.5% of children between the ages of 2 and 19 years in 2015 to 2016 who were obese (Hales et al., 2017).
Another limitation to the analysis of the HCUP KID was the unavailability of disease and treatment-specific details, including family history of diabetes, higher baseline fasting serum glucose level, and evidence of leukemia in the central nervous system (Koltin et al., 2012; Meenan et al., 2019; Tsai et al., 2015; Zhang et al., 2014). Unavailable treatment information, including the timing and dosages of corticosteroids and asparaginase, was also a limitation of our study.
A further limitation to our study stems from the way in which race is coded in the HCUP KID file. When a hospital submits information about a child's Hispanic or non-Hispanic ethnicity, in addition to race, that case is coded as Hispanic and the race that was recorded at the hospital level is disregarded (Agency for Healthcare Research and Quality, 2018). This has the potential to undercount children who identified not only as part of the Hispanic ethnic group, but also as belonging to the White and Black racial groups. It must also be noted that race in the HCUP KID file is recorded at the individual hospitals and differs by institution leading to the possibility of inconsistent categorization of children in racial and ethnic groups (Agency for Healthcare Research and Quality, 2018).
The measure of household economic status available in the HCUP KID and used in this study, zip code-level median household income, may fail to capture the more direct ways in which exposure to the stress of poverty contributes to an underlying predisposition to hyperglycemia in children exposed to ALL therapy. Measurements of household material hardship (food and housing insecurity, unmet medical needs, and suspension of utility services), limited parental education, poor access to outdoor space for physical activity, and household chaos have been shown to provide better estimations of economic disadvantage in families that lead to adverse health outcomes (Neckerman et al., 2016; Schreier et al., 2014).
Study Implications
Our analysis of this large and diverse sample determined that Black children and those living in areas of economic disadvantage are more vulnerable to developing hyperglycemia while hospitalized with ALL. This complication has been implicated as a possible contributor to inferior rates of survival in ALL patients (Sonabend et al., 2009; Zhang et al., 2014). Despite gains in survivorship, Black children and those living in areas with increased levels of poverty continue to lag behind (Abrahao et al., 2015; Acharya et al., 2016; Bona et al., 2016; Goggins & Lo, 2012; Schraw et al., 2020; Walsh et al., 2017). It is important to conduct research to identify potential contributing factors to the disparate outcomes that persist despite improved survival for children following the diagnosis of ALL. Prospective studies are needed to overcome the limitations of our study attributed to the use of previously collected administrative data. This research should include detailed clinical and treatment factors, as well as more representative measures of household material hardship. This study lays the groundwork for intervention research with the goal of conservatively managing hyperglycemia in children at the greatest risk. Finally, it is incumbent upon researchers to ensure that there is diverse racial and socioeconomic representation in these studies.
Conclusions
Our secondary analysis of the 2016 HCUP KID file was limited by the availability of clinical and treatment details, applicability of the measurement of socioeconomic status, and the system used for the coding of racial and ethnic identification. Despite these limitations, the role of race and economic security in the development of hyperglycemia during ALL treatment draws attention to a possible contributor to the disparities that exist in overall survival that face Black children and those living in poverty.
Footnotes
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
