BACKGROUND: Malnutrition is known to be associated with higher morbidity and a risk factor of readmissions in the adult population. In this study, we explore the effect of malnutrition in pediatrics because it may differ from the adult population.
METHODS: Data for all inpatient encounters at a tertiary children’s hospital within a 2-year period corresponding to 19 702 visits were obtained. The data included demographics, socioeconomic status, registered dietitian diagnosis of malnutrition, and variables of the LACE readmission model. We excluded all neonates and patients older than 21 years. A multivariable logistic model was obtained by implementing best subset regression on these variables, controlling for demographics and socioeconomic status, and considering all possible 2-way statistical interactions between malnutrition and the variables for demographics and socioeconomic status.
RESULTS: We discovered a statistical interaction effect between a patient’s age and malnutrition status (P value = .002) with respect to odds of unplanned 7-day readmission. It is indicated in this interaction term that patients who were malnourished had higher odds of readmission than patients who were not malnourished. Furthermore, younger patients who were malnourished were at increased odds of readmission than their older peers, whereas among patients who were not malnourished, younger patients were at reduced odds of readmission.
CONCLUSIONS: The statistical interaction effect revealed that a patient’s risk of readmission is jointly modified by the patient’s age and malnutrition status. This finding advances our understanding of the complex picture of the simultaneous risk factor of unplanned 7-day readmissions in pediatrics.
Malnutrition has been shown to be common among adult patients who are hospitalized1 and to be associated with increased complications, length of stay (LOS), cost, and increased risk for readmission.2–4There is a scarcity of recent studies in regards to the prevalence of malnourished pediatric patients who are hospitalized and potential complications in the United States.5 The prevalence of illness-related malnutrition in hospitalized children has been estimated at 6% to 51%.6 However, the lack of a uniform definition for the severity of pediatric malnutrition (before 2013) is likely responsible, in part, for the lack of prevalence and impact data in pediatrics.6 Worldwide, the prevalence of pediatric malnutrition of hospitalized patients in developed countries has been estimated at 5% to 27%, depending on the criteria used.7–10
On the other hand, there has been an abundance of reports over the last several years in regard to pediatric readmissions, their prevalence, causes, metrics, and preventability.11–15 Malnutrition has not been specifically addressed as a factor in any of the reports. In contrast to published adult studies on readmission,16 pediatric rates of readmission are much lower and the contributing factors different. There has been an increased focus on malnutrition since the 2013 Academy of Nutrition and Dietetics and the American Society for Parenteral and Enteral Nutrition (ASPEN)17 consensus on malnutrition characteristics. The momentum for the pediatric studies on readmissions has been driven predominantly by the concern regarding the loss of reimbursement for “preventable” readmissions as is happening with adult hospitals and the use of readmission rates as a measure of quality of care in most pediatric organizations.18
The purpose of our study was to assess the relationship between malnutrition in the general pediatric inpatient population and unplanned 7-day readmissions in a retrospective study with a large sample at a children’s hospital. Several medical conditions may covary with malnutrition. In this study, we aimed to find the aggregate effect of malnutrition averaged over all such scenarios. The Ohio Children’s Hospitals’ Solutions for Patient Safety recognizes a 7-day (instead of a 30-day) metric for quality reporting because the 7-day metric may be more indicative of preventable readmission than a 30-day metric in pediatrics.19,20
This study was approved by the Internal Review Board of the Children’s Hospital of Orange County (institutional review board #160992).
Data from all inpatient encounters within a 2-year period at Children’s Hospital of Orange County were obtained. This included demographic variables (sex, age, and race and ethnicity), type of health insurance as payer (taken to be an approximation of or proxy for socioeconomic status), acute and/or emergent admission, number of previous emergency department (ED) visits within the last 6 months, LOS, preregistration status, and registered dietitian (RD) diagnosis of malnutrition. We excluded all chemotherapy encounters as well as encounters that ended in patient death. Furthermore, we excluded all neonates (patients <4 weeks of age) and patients older than 21 years of age. For each qualifying encounter, we determined the readmission status using the preregistration status to differentiate between planned and unplanned readmissions. Therefore, our outcome variable of interest was an indicator for unplanned readmissions within 7 days of a previous discharge from the hospital. We followed the Ohio Children’s Hospitals’ Solutions for Patient Safety recommendations on exclusion criteria and provided a flowchart of how patient encounters were excluded in Fig 1.
LACE Readmission Model Variables
We considered the LACE readmission model variables, a priori, as putative factors for inclusion in the final model. The acronym, LACE, is derived from the variables included in the model: LOS, acute and/or emergent admission, the Charlson’s Comorbidity Index Score, and number of ED visits in the previous 6 months (excluding visits that resulted in hospitalization).21,22 It was developed and validated in the adult population, but since its validation, studies have revealed that the constituent variables are independent risk factors for readmission in pediatrics as well.
All patients admitted to the hospital receive a nutritional screening by registered nurses in accordance with the Joint Commission standard. The result of this nutrition screening is used by the RD in determining who gets an assessment for malnutrition. The RD classifies patients that they diagnose with malnutrition into mild, moderate, or severe malnutrition using the 2013 definitions described by an interdisciplinary work group of ASPEN.6 The process for diagnosing malnutrition (shown in Table 1) depends on whether the child has only a single data point for use as criterion (weight to height ratio, BMI, mid-upper arm circumference, and length to height ratio) or when a child has historical medication information and 2 or more data points available for use as criteria as determined by ASPEN.6 We used the resulting diagnosis from the RD’s note to determine patients diagnosed with malnutrition and dichotomized the corresponding variable for malnutrition as an indicator of the absence of a diagnosis for malnutrition or the presence of any degree of malnutrition.
We had complete data for all variables except for payer in which there was 0.28% missing due to clerical (data entry) error. We proceeded with imputation23 of the missing payer data via a logistic regression imputation model and under a missing completely at random assumption for the missing data mechanism. Missing completely at random implies that the probability of missing data on a variable is not related to some observed data in the analysis or the value of the missing data itself.23,24
Summary statistics were calculated and a bivariate analysis of the variables under consideration was conducted. In the development of a multivariable model, we implemented a best subset logistic regression selection algorithm to find the optimal model for the odds of unplanned 7-day readmission. Best subset regression is a method for selecting the best regression model through an exhaustive search over all possible models25 that included main effects and 2-way interactions. The statistical tool for model comparison and selection that we used was the Bayesian Information Criteria, a tool that favors the most parsimonious model that explains the most variation in the data.26,27 The inclusion of 2-way interaction terms determines if there are any significant complex joint effects between malnutrition and the sex, race and ethnicity, age, or payer of the patient. In other words, these interactions reflect the changing effect of a variable over a range of values of another variable. In our case, the interactions answer the following question: “Does the role (if any) that malnutrition plays in modifying one’s risk for being readmitted differ by payer, age, race, or sex?” We used the δ method28–30 in cases in which we had to estimate the SE of the transformation of the regression coefficients to appropriately calculate corresponding confidence intervals (CIs). All statistical analyses were conducted using the R statistical programming language, version 18.104.22.168
The sample consisted of 19 702 encounters with 759 unplanned readmissions corresponding to a readmission rate of 3.85%. The average age of the patients was 7.5 years with a SD of 6.0 years, 45.9% were female, 48.8% were Hispanic, and 61.5% were low-payer. We defined patients with private or nongovernmental insurance plans as high-payer, and patients on Medicaid or other low-income plans as low-payer. Average values of the putative risk factors of readmission (LOS, acute and/or emergent admissions, comorbidity scores, and ED visits) for all encounters, patients who were malnourished, and patients who were not malnourished are given in Table 2.
We conducted a summary and bivariate analysis to assess the unadjusted effect of each variable on the odds of readmission as shown in Table 3. We found that all variables except sex attained unadjusted statistical significance at an α level of .05.
We developed a multivariable model using best subset logistic regression selection approach. Our results reveal that the following variables were significant predictors of readmission (α level of .05): LOS (P value < .001), acute and/or emergent admission (P value = .002), comorbidities (P value < .001), number of ED visits (P value < .001), the main effect term for age (P value < .001), the main effect term for malnutrition (P value = .007), and an interaction between malnutrition and age (P value = .002). Patients with a longer LOS were more likely to be readmitted than patients with a shorter LOS. A patient with acute and/or emergent admission had higher odds of readmission compared with patients with routine admission, and every unit increase in comorbidity score and ED visits were associated with increased odds of readmission, with an odds ratio (OR) of 1.17 and 1.16, respectively.
The main finding in our study was the 2-way interaction term between malnutrition and age that was found to be a significant predictor of unplanned 7-day readmission (P value = .002). The presence of this interaction term modifies the main effect terms of age and malnutrition (hence the absence of ORs for age, malnutrition, and their interaction in Table 4). In Fig 2, we plot the predicted probability of readmission by malnutrition status, keeping sex, race, and socioeconomic status constant while considering the difference in average values of the LACE variables across both groups of patients. Because we have an interaction between a dichotomous and a continuous variable, separate regression lines with different slopes and different intercepts are required to model the effect of age within the 2 groups of patients. In other words, it is indicated in the interaction that age has effects with opposite signs on the 2 groups of patients. Within the group of patients who were malnourished, the odds of readmission are higher for younger children. On the other hand, within the group of patients who were not malnourished, the odds of readmission are highest in older patients. The odds decrease with increasing age in the first group but increases with age in the second. Furthermore, the odds of readmission among the patients who were malnourished tend to be higher than those who were not malnourished.
In addition to plotting the predicted probability of readmission, it is important to estimate OR and corresponding 95% CIs for different combinations of age and malnutrition status. In Table 5, we provide ORs to throw more light onto the varying effect of age on odds for readmission by malnutrition status. We can see that all scenarios selected had ORs implying significance at the .05 α level.
The main finding in our study was the identification of a statistically significant interaction between age and malnutrition status regarding unplanned 7-day readmissions in pediatric patients. The interaction is such that the odds of readmission among patients who were malnourished were higher than the odds for patients who were not malnourished. Among patients who were malnourished, the odds of readmission were higher in younger patients and decreased with age. And finally, the odds of readmission increased from younger to older patients among those who were not malnourished. Why the odds of readmission among patients who were malnourished would be higher in younger patients was not part of our study and was somewhat unexpected. This may have to do with a period of rapid growth seen in patients <2 years old when malnutrition may be more detrimental.5,7 The model also identified known risk factors for readmission, such as LOS, acute and/or emergent admission, comorbidities, and ED visits, with effect sizes comparable to those found in other studies.32
The prevalence of malnutrition in 4.7% of the sample of patients in this study is lower than most reported estimates,8–10 but such comparisons are difficult because of differences in how malnutrition may be assessed across different institutions. In our study, we used the recommendations of ASPEN, 2013 for the definition and diagnosis of malnutrition.6 In earlier studies, Waterloo Criteria was used and obesity was included,5 which we did not include in the present study.
Our findings of 3.9% of unexplained readmissions at <7 days and 10.5% at <30 days (in the sample of patients in this study) are comparable to other large studies in which multiple children’s hospitals were involved.13,15 Although malnutrition has rarely been studied previously in the context of pediatric readmissions, it has been researched in several adult studies.3,4 In a 2011 retrospective study, Allaudeen et al3 found that in a cohort of >6000 adult patients, weight loss was a statistically significant risk factor for readmission. Mudge et al,4 also in 2011, in a prospective study of 142 patients who were hospitalized and >50 years of age with 2 or more hospitalizations in the past 6 months, found that being underweight was, again, a significant predictor of readmission. This association is not surprising in that malnutrition leads to abnormal immune function, increased infectious complications, delayed wound healing, muscle weakness, loss of lean body mass, and developmental delay, all of which are potential contributors to severity of illness.6
This study had several limitations, including the retrospective design involving a single children’s hospital. This may limit the applicability to community hospitals that also care for children. Another limitation is the absence of data on possible readmissions of the patients in the study to other hospitals. Lastly, association, not causation, is indicated in our findings.
In our study, we clarify and indicate the presence of a complex joint effect of age and malnutrition status on the odds of readmission. One important clinical implication may be the identification of a subset of malnourished infants and children that can be targeted to reduce unplanned readmissions. Future researchers might incorporate multiple institutions and focus on a prospective observational or interventional study.
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
FUNDING: No external funding.
POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no potential conflicts of interest to disclose.
- Shaughnessy EE,
- Kirkland LL
- Pileggi VN,
- Monteiro JP,
- Margutti AV,
- Camelo JS Jr.
- Hain PD,
- Gay JC,
- Berutti TW,
- Whitney GM,
- Wang W,
- Saville BR
- Wallace SS,
- Keller SL,
- Falco CN,
- et al
- Auger KA,
- Teufel RJ II.,
- Harris JM II.,
- et al
- Toomey SL,
- Peltz A,
- Loren S,
- et al
- House SA,
- Coon ER,
- Schroeder AR,
- Ralston SL
- ↵Children’s Hospitals’ Solutions for Patient Safety. Our goals. 2017. Available at: www.solutionsforpatientsafety.org/about-us/our-goals/. Accessed January 1, 2017
- van Walraven C,
- Dhalla IA,
- Bell C,
- et al
- Enders C
- James G,
- Witten D,
- Hastie T,
- Tibshirani R
- Neath AA,
- Cavanaugh JE
- Cox C
- Jackson CH
- ↵R: A Language and Environment for Statistical Computing [computer program]. Version 3.4.0. Vienna, Austria: R Development Core Team; 2017.
- Copyright © 2018 by the American Academy of Pediatrics