Association of Area Deprivation and Diabetic Ketoacidosis Readmissions: Comparative Risk Analysis of Adults vs Children With Type 1 Diabetes

Abstract

Diabetic ketoacidosis (DKA) is a serious complication in adults and children with type 1 diabetes (T1D). Recurrent admissions for DKA have been associated with increased mortality, and patients with four or more readmissions have a sixfold increased risk of death (1). Studies evaluating the association of socioeconomic factors and DKA readmissions in adults are limited, so we must rely on inferences from the pediatric population in whom this has been more extensively studied. Although these inferences do provide insight, there are inherent differences between these two populations that may make extrapolation of findings unreliable. We are not aware of comparative studies evaluating whether predictors of DKA differ between adults and children with T1D. Historically, studies exploring the role of socioeconomic status (SES) and DKA have relied on SES surrogates, including household income, education, insurance status, and occupation (2). Although single measures of SES can be a good proxy for SES, composite measures are more likely to fully capture latent SES (3).

Area deprivation index (ADI) is a geographically based measure of SES that takes into account several measures, including income, employment, education, and housing quality. ADI has been associated with several T1D outcomes, including glycemic control, microvascular disease, and neuropathy (4–9), but there are no studies evaluating the association of ADI and DKA readmissions in adults or children in the United States. In Scotland, Govan et al. (10) evaluated the role of area deprivation in DKA admissions nationally over a 2-year period and found that adults with T1D from the most deprived areas had 4.5 times higher odds of DKA readmission than those who were least deprived.

The objective of this study was to determine whether area deprivation is associated with DKA readmission and evaluate whether this association differs among pediatric and adult patients in the United States. We hypothesized that DKA readmissions would be associated with higher degrees of area deprivation. Our secondary objective was to explore whether other patient factors, some of which may act as social determinants of health (e.g., health insurance), predict DKA readmissions and whether such predictors differ among pediatric and adult populations. To our knowledge, no previous study has assessed the link between area deprivation and DKA readmissions or examined differences in predictors of this outcome between adults and children with T1D living in the United States.

Research Design and Methods

Study population

This study used the Maryland Health Services Cost Review Commission’s (HSCRC’s) database, which collects data from all acute care hospitals in the state. Using International Classification of Diseases (ICD) diagnostic codes, we selected all patients with an admission diagnosis of T1D who had at least one admission for DKA between 1 January 2012 and 31 December 2017. Patients were identified by their primary admission ICD-9 diagnosis code. We used four ICD-9 codes (250.11, 250.13, 250.21, and 250.23) that code for either ketoacidosis or hyperosmolarity in a patient with T1D that is either well controlled or poorly controlled. Hyperosmolarity ICD-9 codes 250.21 and 250.23 accounted for the primary diagnosis for <2% of admissions. These admissions were included, given that ∼60% of participants who had an admission with a primary diagnosis of hyperosmolality had either a secondary diagnosis of DKA or other admissions for which the primary diagnosis was DKA. Furthermore, given the prevalence of mixed DKA with hyperosmolality in T1D (11–15) and the lack of convention for coding these patients (11), the very low number of patients (∼1%) with hyperosmolality codes only were considered to have negligible influence on the generalizability of the findings to patients with T1D and DKA.

Patients were categorized into two groups, adult and pediatric patients, on the basis of their age at the index admission, which we defined as the first DKA admission in the study period. In this study, pediatric was defined as 2 to 19 years of age and adult as 20 years of age or older according to the classification of the HSCRC and consistent with the common practice of following patients in a pediatric setting beyond the legal age of adulthood.

Given the focus of this study, patients not residing in Maryland or its bordering states (Virginia, West Virginia, Pennsylvania, and Delaware) or the District of Columbia were excluded because they were less likely to be readmitted to a Maryland hospital.

Predictor variables

The HSCRC database reports data by admission and includes hospital and patient-level data. Because the outcome of interest was DKA readmission, the covariates analyzed in this study represent characteristics from the index admission. Our primary variable of interest was ADI, which considers several neighborhood-level factors including, but not limited to, income, employment, education, and housing. We used the 2013 ADI developed by the Health Resources and Services Administration and refined by Kind et al. (16) at the University of Wisconsin-Madison. In this database, ADI is reported for each individual state in deciles from 1 to 10, where 1 represents the least deprived group and 10 the most deprived. ADI values are usually reported at the block level (zip code +4); however, the HSCRC reports only five-digit zip codes. Therefore, to calculate the ADI for a patient’s five-digit zip code, we averaged the ADIs for the blocks within that zip code. In a review of a neighborhood atlas for the state of Maryland by ADI (17), there appeared to be three geographically distinct patterns of ADI, with clustering of ADI deciles of 1 to 3 in the Washington, DC, and Baltimore suburbs, 4 to 7 in central Maryland and the mid-Eastern shore, and 8 to 10 in Western Maryland and the southeastern portion of the state. Thus, we classified ADI into three categories defined by tertile: least deprived, moderately deprived, and most deprived.

Other covariates in this study included age, sex, race, ethnicity, marital status, severity of illness, and discharge against medical advice (AMA). Age was reported in the HSCRC as 19 grouped categories: Group 1 included toddlers aged 2 to 4 years; all other categories were in increments of 5 years until group 18, which included patients ≥85 years of age. We classified races in three groups: white, black, and other (Asian, Pacific Islander, American Indian, biracial, other, or unknown). Ethnicity was categorized as Hispanic or non-Hispanic. Payer groups included private insurance, Medicare, Medicaid, self-pay or charity, and other (workmen’s compensation, Maternal and Child Health Title V Program, Veterans Affairs, and other government programs). Severity of illness categories (minor, moderate, major, and extreme) were provided by the HSCRC according to All Patient Refined Diagnosis Related Groups.

Outcome variable

The primary outcome was readmission for DKA occurring in the study period. We evaluated DKA readmission as a categorical outcome with three levels: no readmission, one to three readmissions, and four or more readmissions. The rationale for selecting these categories was based on a previous study that demonstrated a substantial increase in mortality for patients with five or more DKA admissions (i.e., four or more readmissions) (1).

Statistical analysis

Descriptive statistics were used to characterize the two study populations (adult and pediatric) by outcome. All variables were categorical variables, and the χ² test was used to evaluate overall differences in proportions among adults compared with pediatric patients. Within each group, we also analyzed differences between patients with one to three and four or more DKA readmissions compared with those with no readmissions.

Multivariable logistic regression was used to evaluate the association of DKA readmission with predictor variables separately for adults and pediatric groups. Two fully adjusted regression models were developed for each DKA outcome: one to three readmissions (model 1) and four or more readmissions (model 2), with the reference group for each model being no readmissions. Because all exposure variables were thought to be potentially related to readmissions, the fully adjusted models included all predictor variables, whether or not they were statistically significant on univariate analysis. In regression analyses, age category and severity of illness scores were treated as ordinal variables, for which change was a one-unit increase within a category. To determine whether there were statistically significant differences between the pediatric and adult groups, models with interaction terms were evaluated for each variable with adjustment for the other covariates.

Patients with missing data were not included in the regression. The only variable with missing data was ADI; ∼2.5% of patients had zip codes that were not associated with an ADI score. An ADI was not available for zip codes in three scenarios: (i) those associated with a post office box and not geographically representative, (ii) unique zip codes often assigned to a business or large-footprint entities that have a large volume of mail delivery, or (iii) coastal areas or offshore zip codes that were block groups.

All analyses were performed using Stata Statistical Software, version 14.2 (StataCorp). A two-sided P value ≤0.05 was considered statistically significant. A two-sided P value for interaction (P_int) ≤0.20 was considered statistically significant per convention owing to multiple comparisons (18).

Results

During the study period, there were 4037 patients with at least one DKA admission, of whom 732 (18%) were pediatric patients and 3305 (82%) were adults. Maryland was the state of residence for 92% of patients. Readmission rates were relatively similar between groups: Among pediatric and adult patients, ∼20% had one to three readmissions and 6% had 4 four or more readmissions.

The characteristics of pediatric patients admitted with DKA differed statistically from those of adults for all variables except ethnicity. Overall, the pediatric group was predominantly non-Hispanic (96%), white (52%), female (54%), and adolescent (77%) with private insurance (50%) and from the most deprived areas (44%). Unlike in the pediatric group, among adults, females were in the minority (48.3%). The proportion of patients living in moderately deprived areas was higher among adults than pediatric patients (47% vs 41%, respectively), and the proportion of patients with private insurance was lower among adults than pediatric patients (37% vs 50%, respectively). Similarity to the pediatric group, the predominant racial/ethnic group was non-Hispanic white in the adult group, but the two groups differed as, there was a higher prevalence of blacks (45% vs 37%). Severity of illness was higher among adults than children, with 46% vs 13%, respectively, having severity of illness scores in the major/extreme categories. Adults were more likely to be discharged AMA (4% vs 2%). Baseline characteristics for the pediatric and adult populations by readmission outcome are reported in Table 1.

The results of the univariate analyses of predictors and DKA readmissions, both of which evaluated groups by readmission outcomes compared with the reference of no readmission, are reported for pediatric and adult patients in an online repository (17). For pediatric patients, there were statistical differences across age, race, payer, and area deprivation in those with one to three DKA readmissions. In addition to these variables, female sex, discharge AMA were also significant in the group with four or more readmissions. For adults, all predictor variables were significant except for female sex and severity of illness in the one to three DKA readmissions group. All variables were significant for the four or more readmissions outcome.

Table 2 shows the results of the fully adjusted logistic regression models for one to three and four or more readmissions for pediatric and adult populations. There were several notable differences in the strength of association of socioeconomic and other predictors among pediatric and adult populations. For the outcome of one to three DKA readmissions, Medicaid insurance was predictive for both adult and pediatric patients. Additional predictors for children only included increasing age category and living in moderately deprived and most deprived areas. Variables that were positively associated with one to three DKA readmissions in adults only were Medicare status and discharge AMA, whereas black race and Hispanic ethnicity were inversely associated. For the outcome of four or more DKA readmissions, variables that were positively associated in both adults and children were female sex, Medicaid insurance, and living in the most deprived area. For this outcome, increasing age category and living in a moderately deprived area were also positively associated variables in the pediatrics group only. Among adults only, increasing age category and being in the “other” race category were inversely associated with the outcome of four or more DKA readmissions, whereas discharge AMA was positively associated.

In those who were from moderately and deprived areas, area deprivation differed among the pediatric and adult populations for both outcomes of one to three readmissions (P_int < 0.01 and P_int = 0.010, respectively) and four or more readmissions (P_int = 0.10 and 0.20, respectively). In the pediatric group, those who were moderately deprived had an OR of 4.26-for one to three readmissions compared with patients who were least deprived (95% CI, 1.79- to 10.12) and an OR of 7.87-for four or more readmissions (95% CI, 1.02 to 60.81). In those who were most deprived, the OR increased to 4.04 for one to three readmissions (95% CI, 1.67 to 9.73), which increased to 7.76 for four or more readmissions (95% CI, 1.04 to 58.11). Among adults, in those who were moderately deprived, there was no statistical difference in odds of one to three or four or more readmissions. In those who were most deprived, no association was observed with one to three DKA readmissions; however, the OR significantly increased to 2.23 for four or more readmissions (95% CI, 1.16 to 4.25).

With respect to insurance status, nonprivate insurance was generally associated with an increased risk of readmission among both pediatric and adult patients. Compared with private insurance, Medicaid insurance increased the odds of one to three readmissions by 1.85 (95% CI, 1.17 to 2.90) in pediatric patients and 1.39 (95% CI, 1.11 to 1.75) in adult patients (P_int = 0.06); the corresponding ORs for four or more readmissions were 2.81 (95% CI, 1.35 to 5.87) and 1.99 (95% CI, 1.32 to 3.01), respectively (P_int = 0.64). In the one to three readmissions outcome, self-pay/charity increased the odds of readmission in adults but not in pediatric patients (P_int = 0.41).

Although increasing age category was associated with DKA readmission in pediatric patients—with odds of four or more DKA readmissions as high as 11.66 (95% CI, 3.74 to 36.37)—increasing age was inversely associated with DKA readmission among adults, with a 29% decrease in odds of four or more readmissions (OR, 0.71; 95% CI, 0.66 to 0.76) for each one-unit increase in age category (P_int < 0.01). Females had increased odds of readmissions among both pediatric and adult groups, although the effect size was larger in the pediatric population than in the adult population for the outcome of four or more DKA readmissions (OR, 2.31 vs 1.51, respectively; P_int = 0.20). Severity of illness was not associated with either DKA readmission outcome among children, but it was associated with four or more DKA readmissions among adults (OR, 1.79; 95% CI, 1.42 to 2.25). Similarly, discharge AMA was not associated with readmissions among pediatric patients, but it was associated with both DKA readmission outcomes among adults, with an OR of 1.87 (95% CI, 1.06 to 3.31) for four or more DKA readmissions (P_int = 0.19).

Discussion

In this study, we found that lower SES was associated with DKA readmissions among adult and pediatric patients with T1D however, there were several important differences with respect to individual socioeconomic factors in these two populations. Most importantly, area deprivation showed a much stronger association among pediatric patients than among adults, as evidenced by (i) greater effect sizes for both DKA readmission outcome measures compared with adults, (ii) greater magnitude of change in effect size when the severity of the outcome (one to three vs four or more DKA readmissions) was compared in children and adults; and (iii) statistically increased odds for both DKA readmission outcomes observed among children at less severe degrees of deprivation (i.e., for both moderate and most deprived groups), whereas only the most deprived adults were observed to have only the more severe outcome (four or more DKA readmissions). A similar pattern was observed for Medicaid insurance, whereby the pediatric group had a larger effect size than adults, as well as a greater change in the OR between the one to three and four or more DKA readmission outcomes. These findings suggest that pediatric patients with T1D may be more susceptible to the negative outcomes associated with lower SES.

The association of area deprivation and DKA readmissions has not been evaluated in children. One study in Scotland evaluated the association of area deprivation with the incidence of DKA hospitalization in adult outpatients with T1D (10). The study found that increasing deprivation was associated with increased odds of DKA admission. We are not aware of any studies, even in the social literature, that have compared the influence of SES on acute health outcomes in pediatric patients with that in adult populations. It is thought that adult SES is closely linked to SES in early life (19), and the cumulative risk model argues that the effect of SES is compounded over the life course; our study conflicts with this theory, which suggests that adult patients of lower deprivation—who were likely deprived throughout their life course—should have higher odds of negative health outcomes (19). Other theories argue that childhood SES and adult SES are autonomously important and that the latter may mitigate effects of early life (19).

In this study, we did not measure the association of SES across the life course but rather as a predictor in these two different populations with T1D. There are several possible explanations for the stronger association of deprivation and DKA in pediatric patients, including dependency and shorter duration of diabetes. The care of a child with T1D is dependent on a parent or guardian. Parents of lower SES are known to have more life stressors, competing responsibilities, and more diabetes-related family stress (20); therefore, it may be difficult to prioritize the child’s diabetes management. Although adults of low SES may have similar life stressors, their longer duration of diabetes and greater experience in diabetes self-management may counteract the negative effects of deprivation.

Other notable variables associated with DKA readmissions were race, ethnicity, age, and sex. Our study found no significant difference in odds for DKA readmissions among different racial and ethnic groups in pediatric patients. In adults, we observed that blacks, other races, and Hispanics generally had decreased odds of DKA readmission compared with whites and non-Hispanics for some outcomes. Among children, Malik et al. (21) found no difference in the frequency of DKA readmission among races compared with whites, except for non-Hispanic blacks, who had an OR of 2.4-for having a higher number of DKA admissions. In the adult literature, there are few data evaluating race and recurrent DKA readmissions, but the T1D Exchange did study predictors of at least one DKA hospitalization in the prior 12-month period in clinic patients with T1D (22). Although this population differs from ours, no significant differences were observed among race and DKA hospitalization (22). The reason for the racial/ethnic differences observed in our study is unclear; we postulate that the lower odds of DKA readmission in blacks and Hispanics could potentially have resulted from misclassification of ketosis-prone type 2 diabetes, a condition that is more prevalent in black and Hispanic populations (23) and theoretically would pose a lower risk of recurrent DKA than T1D.

Our study found a positive association with DKA readmission and age category in pediatric patients but an inverse relationship in adults. These findings differ from a Children’s Hospital System study that observed an association between DKA readmission and increasing age, with a peak in the 10- to 11-year-old age group (21). Our findings in adults are similar to those in the published literature (10). Overall, the age and DKA readmission pattern we observed parallels the age and glycemic control trend reported by Miller et al. (24): worsening of hemoglobin A1C in the teenage years, which rapidly peaked in the late teens and improved significantly in the fourth decade of life and onward. The increased odds for readmission in late childhood and early adulthood are associated with the need for independence and self-determination, which interferes with diabetes self-management and increases the likelihood of high-risk behaviors (25). Moreover, at the pediatric to adult transition, many patients are often lost to follow-up and/or disengage in care (26).

We found that females in both pediatric and adult populations had a higher OR of DKA readmission, but the effect size was larger in pediatric patients than in adults (2.31 vs 1.51 for four or more readmissions; P_int = 0.20). Females have had worse glycemic control and outcomes in T1D (27). It has been found that girls and women have higher perceived disease burden, lower quality of life, and increased anxiety compared with males (25, 27). Males have more active coping, less avoidant behavior, less need for social support, and less depressive symptoms than females (28). Eating disorders are also more prevalent among girls and women (29). The higher odds of DKA readmission among girls compared with women may be explained by several factors. Adolescent girls are known to struggle with body image, which may result in eating disorders or intentional omission of insulin injections for weight loss (29). Omission of insulin in this setting is an established risk factor for recurrent DKA admissions (30), poor glycemic control, and mortality (31). Females admitted with recurrent DKA often exhibit more behavioral problems, lower social competence, and higher levels of family conflict than males (32). As these females grow into adulthood, some of these behaviors may dissipate (33).

This study has several strengths. We used data from a large cohort of patients over a 5-year study period, with nearly complete ascertainment of predictors and outcomes. Because patient-level data were available, we were able to follow a patient through multiple admissions over several years, and we have a representative sample of patients from infancy to end of life. We used ADI, which is a validated measure of SES, and we attempted to adjust for relevant confounders in the outcome measure, such as severity of illness.

Limitations of the study include the lack of laboratory confirmation of T1D or the diagnosis of DKA, as we relied upon ICD-9 coding provided by the HSCRC for classification. The HSCRC also captures only admissions occurring in the state of Maryland; we may have underestimated readmissions because we were unable to capture out-of-state DKA admissions. We averaged area deprivation by zip code rather than by neighborhood, which some have argued may not be appropriate because deprivation can vary within a given five-digit zip code; nonetheless, we do not believe this assumption interferes with our ability to make inferences because of the relatively homogeneous geographical distribution of deprivation in the state of Maryland by zip code (17). Our findings may not be generalizable to all areas of the United States. Finally, we are limited in our ability to elucidate the cause(s) of recurrent DKA admissions (i.e., behavioral vs health system issues), as we did not have information on outpatient visits, medications, or providers’ assessment of the direct cause of the DKA readmission (e.g., omission of insulin, cost barriers, acute illness).

Conclusion

Lower SES as measured by area deprivation is associated with recurrent DKA admissions among pediatric and adult patients with T1D with more pronounced influence in children. Other demographic factors, including trends with age, sex, race, ethnicity, and insurance type confer different risks in these two populations. Further studies are needed to understand the mechanisms behind these associations and to inform development of targeted strategies to address disparities specific to each population.

Acknowledgments

Financial Support: E.E. was supported by the Clinical Research and Epidemiology in Diabetes and Endocrinology Training Grant of the National Institute of Diabetes and Digestive and Kidney Diseases through grant No. T32 DK062707.

Disclosure Summary: The authors of this publication have nothing to disclose.

Abbreviations:

Abbreviations:
 
• ADI

  area deprivation index

 
• AMA

  against medical advice

 
• DKA

  diabetic ketoacidosis

 
• HSCRC

  Health Services Cost Review Commission

 
• ICD

  International Classification of Diseases

 
• Int

  interaction

 
• P_int

  P value for interaction

 
• SES

  socioeconomic status

 
• T1D

  type 1 diabetes

References and Notes