Disparities in the Diagnosis and Management of Infants Hospitalized With Inadequate Weight Gain Free

We conducted a single-center retrospective cohort study at a large, free-standing children’s hospital in the Midwest. Infants aged 2 weeks to 12 months hospitalized on the Hospital Medicine service between May 2016 and June 2021 with a primary International Classification of Diseases Ninth Revision or Tenth Revision (ICD-9 or ICD-10) discharge diagnosis of failure to thrive or inadequate weight gain (Table 1) were eligible.¹⁸  Children were excluded if they did not meet the established diagnosis of inadequate weight gain by using an adaptation of recently published diagnostic criteria⁸  in which a patient must meet 1 or more of the following criteria: (1) weight-for-age less than the fifth percentile, (2) length-for-age less than the fifth percentile, (3) weight deceleration across >2 major percentiles since birth, (4) conditional weight gain in lowest fifth percentile using World Health Organization (WHO) growth velocity standards, and (5) not returned to birth weight by 2 weeks old. Because we sought to include only index admissions, readmissions and hospital transfers were excluded because were those with an alternative primary reason for admission. A structured electronic health record query was followed by a chart review to determine exclusions (Fig 1) and collect additional data. This study was deemed exempt by the institutional review board.

We chose to examine race as the primary exposure to investigate potential inequities experienced by children with inadequate weight gain that may result from systemic, institutionalized, and structural racism, interpersonal and personally mediated racism, and implicit biases.¹⁹ 

At our institution, registration staff who undergo standardized training record race and ethnicity per parental report. Racial categories in our system include American Indian and Alaska Native, Asian, Black or African American, Middle Eastern, Native Hawaiian and other Pacific Islander, white, and unknown. Multiple categories can be selected for each patient. To honor specific racial and ethnic categories over collective terms, analyses were limited to patients who identified as exclusively non-Hispanic Black or exclusively non-Hispanic white. Because of the limited racial and ethnic diversity of our study population, all other patients were excluded from analyses (Fig 1).

The documented etiology of inadequate weight gain was the primary outcome. Etiologic categories were designated as (1) objective organic, defined as a medical condition with objective diagnostic criteria (eg, genetic disorder, congenital heart disease, oral dysphagia resulting in aspiration or a need for tube feeds), (2) subjective organic, defined as GER and CMPI, and (3) nonorganic, defined as the identification of child-centered feeding difficulties (eg, oral aversion, mild dysphagia without aspiration, insufficient transfer of breastmilk), parent-centered feeding difficulties (eg, improper formula mixing, parent unaware of necessary feeding schedule or amount required, insufficient breastmilk supply), or neglect and abuse without comorbid organic etiology. Children with an unknown etiology but a significant diagnostic workup pending at discharge for a suspected objective organic disease were excluded from the analysis. Although the inadequate weight gain etiology is often multifactorial, patients were assigned to 1 prioritized etiologic category. A priori, we defined objective organic as taking the highest priority and nonorganic as taking the lowest priority.

The documented etiology of inadequate weight gain was determined via progress notes and discharge documentation review. Before independent review, all reviewers (CS, SP, EC, SR, CH, and RM) were trained on a subset of data to enhance interrater agreement. Two separate reviewers reviewed each record independently. Disagreements were resolved by discussion and third-reviewer adjudication.

Dichotomous secondary outcomes included a social work (SW) consult during admission, child protective services (CPS) involvement, fortification of feeds (caloric content of recommended feeds at discharge >20 kcal per ounce regardless of whether the fortification happened outpatient or as a result of prematurity), nasogastric (NG) tube placement, gastroenterology (GI) consult, and speech therapy (ST) consult.

Covariates included patient age, admission weight z-score, birth weight z-score, the presence of medical complexity, and the census tract-level “deprivation index.” Admission weight z-score was determined by using the WHO growth chart by sex for infants born at >37 weeks’ gestation. For infants born at <37 weeks’ gestation, the Fenton growth curve was used until a gestational age of 48 weeks, after which the corrected age on the WHO curve was used. Birth weight z-score was similarly calculated. The presence of medical complexity on admission was defined by using the Children’s Hospital Association definition of children with medical complexities.²⁰  The deprivation index ranged from 0 to 1, with higher values indicating greater socioeconomic deprivation. It is calculated from 6 census tract-level variables from the 2018 5-year US Census American Community Survey, including (1) fraction of households below the federal poverty level, (2) median annual household income, (3) fraction of adults with less than high school education, (4) fraction of population without health insurance, (5) fraction of households receiving public assistance, and (6) fraction of vacant housing units.^21,22  Each patient’s home address, reported by the caregiver at the time of admission, was geocoded and spatially joined to census tracts and the deprivation index by using 2015 TIGER/Line address range files with custom, offline geocoding software.²³  Children were excluded if they were in the custody of their home county (ie, foster care) because their home census tract could not be determined.

Descriptive statistics are presented as a mean with an SD or a number with a percentage. The differences in etiologies and covariates across races were tested by using Wilcoxon rank and χ-square tests. To further characterize the relationship between race and the etiology of inadequate weight gain, we used a Bayesian difference-in-means approach. An uninformative β prior $(π∼B(1,1))$ was used, and the posterior of Black patients was subtracted from the posterior of white patients. The difference in means was reported as a percentage with a 95% credible interval (CI), which can be interpreted as containing the true parameter with 95% probability.²⁴ 

We used Bayesian multinomial logistic regression to estimate the independent association between race and etiology (subjective organic was the reference group). Covariates, defined a priori, included age, admission weight z-score, birth weight z-score, presence of medical complexity, and deprivation index.

Bayesian difference-in-means and multivariable logistic regression were used for the evaluation of secondary outcomes. With these models, we exclusively examined patients who did not have objective organic etiology. Subjective organic etiology was the primary exposure, and a logistic regression model was implemented for each management outcome. A priori, age, admission weight z-score, birth weight z-score, medical complexity, and deprivation index were chosen as covariates. Additional details can be found in the Supplemental Information section.

Statistical analyses were performed by using R, v.4.0.3 (R Foundation for Statistical Computing) using the packages gtsummary and MCMCpack. Frequentist P values were 2-tailed. A threshold of P < .05 determined statistical significance. Difference-in-means was considered statistically significant if the 95% CI did not cross 0. Odds ratios and adjusted odds ratios calculated from logistic regression models were considered statistically significant if the 95% CI did not cross 1.

Of 380 infants, the majority were white and had a nonorganic etiology of inadequate weight gain (Table 2). White infants were younger at admission, had higher birth weight z-scores, higher admission weight z-scores, and lower deprivation indexes than Black infants (Table 2). A difference in posteriors (white–Black) revealed that the rates of the objective organic etiology were similar by race (−1.5%, 95% CI [−8.4 to 4.8]); however, Black patients were more likely to have nonorganic etiology (−9.1%, 95% CI [−17.4 to −0.5]), and white patients were more likely to have subjective organic etiology (10%, 95% CI [3.1 to 16.5]; Supplemental Fig 3, Supplemental Table 4).

In multinomial logistic regression using subjective organic as the referent group, Black patients had 2.3 times higher unadjusted odds (95% CI 1.17 to 4.76) of the nonorganic etiology of inadequate weight gain compared with white infants (Supplemental Table 5). When accounting for confounding, there was no association between race and etiology (adjusted odds ratio 0.8, 95% CI [0.4 to 2.08]); rather, deprivation index was the primary predictor for named etiology of inadequate weight gain. For every 0.1-point increase in deprivation index, the odds were 1.8 times greater (95% CI 1.37 to 2.41) for the nonorganic etiology relative to subjective organic. Figure 2 reveals the probability of each etiology by race as the deprivation index value changes and all other covariates are held at the mean value by race. Also reflected in this figure are calculated probabilities for the etiology of inadequate weight gain. The calculated probability of the mean Black patient having a nonorganic etiology was 73% compared with 54% for the mean white patient, the calculated probability of the mean Black patient having subjective organic etiology was 16% compared with 37% for the mean white patient, and the calculated probability of the mean Black patient having objective organic etiology was 11% compared with 9% for the mean white patient.

Infants with GER or CMPI, compared with those with nonorganic etiology, were significantly less likely to have an SW consult and CPS involvement and significantly more likely to have an NG tube placed, a GI consult, and an ST consult (Table 3, Supplemental Fig 4, Supplemental Table 6). Specifically, infants with GER or CMPI had 0.16 times the odds of CPS involvement and 2.4 times the odds of receiving an ST consult than infants with nonorganic etiology. There was no difference between the 2 groups in the rate of the fortification of feeds in the adjusted models (Supplemental Table 7).

We found inadequate weight gain etiology to be associated with race in bivariate analyses. In adjusted analyses, race was not significant; however, infants from more socioeconomically deprived neighborhoods were more likely to be diagnosed with nonorganic etiologies and less likely to be diagnosed with subjective organic etiologies of inadequate weight gain. Given the legacy of structural racism, socioeconomic deprivation in our sample disproportionately affected Black infants. As such, the average Black infant, and infants living in more deprived neighborhoods, were less likely to be diagnosed with GER or CMPI and more likely to be diagnosed with improper feeding techniques, neglect, or breastfeeding failure. Such differences proved influential given that the named etiology of inadequate weight gain was associated with different management practices. Infants with subjective diagnoses of GER or CMPI were more likely to receive medical interventions, including NG tube placement and ST consults. In contrast, infants diagnosed with nonorganic etiologies were more likely to have SW and CPS involvement. The recognition of such inequities can help us develop strategies to mitigate biases and ramifications of structural racism in clinical care.

When neighborhood socioeconomic deprivation was included in a multivariable model, there was no association between race and the named etiology of poor weight gain in our study. We challenge the frequent assertion that race is a reflection of shared genetic composition and, instead, endorse that race is a reflection of social, cultural, and environmental factors.²⁵  In our cohort, there were significant differences between racial groups in many covariates (deprivation index, birth weight z-score, admission weight z-score, and age). Black infants in our cohort had significantly lower z-scores at the time of admission and were older. These differences between cohorts likely follow upstream disparities in prenatal and neonatal care that are present before admission for inadequate weight gain and suggest that Black infants may have presented at higher severity and later in their course of poor weight gain. This may reflect limited or delayed access to outpatient care or follow-up. As discussed in a recent JAMA Pediatrics editorial, “the inclusion of characteristics as confounding variables in models assessing the impact of race on health can imply questionable assumptions about where in the causal pathway racism begins to exert an effect.”²⁶  While attempting to select clinically relevant confounders for our analysis, we chose covariates that in part are impacted by racism, creating a collider bias. For example, historic racism contributes to Black patients living in poorer neighborhoods (higher deprivation index), having inequitable obstetric care (potential link to lower birth weight, lower z-score at admission),²⁷  and having decreased access to care (potential link to lower z-score and older admission age). Thus, we risk unintentionally disregarding the mechanisms by which racism contributes to inequitable health outcomes by considering only adjusted models.

Our findings add to an existing body of literature evaluating the association between socioeconomic status (SES) and inadequate weight gain. Data from 2 Japanese birth cohorts revealed that infants in the lowest quartile of household income were 1.3 times more likely to experience inadequate weight gain than infants in the highest quartile.²⁸  However, large studies in the United Kingdom revealed no association between SES or educational status and inadequate weight gain.^29,30  Authors hypothesized that more robust modern welfare systems in the United Kingdom protected children of lower SES from experiencing inadequate weight gain due to poverty.^28,31  In support of the effectiveness of such welfare programs, a multisite cross-sectional survey study in the United States revealed that infants who received Special Supplemental Nutrition Program for Women, Infants, and Children assistance were less likely to be underweight than those eligible for the Special Supplemental Nutrition Program for Women, Infants, and Children but who did not receive it because of difficulties accessing the assistance.¹⁰  Our study builds on these findings by identifying disparities across different levels of neighborhood deprivation in subjective areas of the diagnostic process and subsequent management.

We posit that the disparities we identified are influenced by social drivers of health, such as food security, health literacy, and in-home support. Such drivers likely play important roles in trajectories for infants with poor weight gain caused by nonorganic etiologies, especially in more socioeconomically deprived neighborhoods. As pediatricians, we must recognize these social drivers of health, optimizing the access to and quality of outpatient care after birth to limit the need for inadequate weight gain hospitalizations. This includes providing caregiver education on optimizing feeding practices, completing effective health-related social risks and needs screening, and pursuing high-fidelity connections to appropriate community resources.¹²  It also requires us to inquire without judgment about and educate ourselves on the conditions in which children grow and develop as we pursue diagnosis and management plans.

We also consider that GER and CMPI may be underdiagnosed in infants from neighborhoods with greater socioeconomic deprivation and overdiagnosed in infants from neighborhoods with less socioeconomic deprivation. Two constructs may illuminate how biases encroach on the diagnostic process and perpetuate either under- or overdiagnosis. First, CMPI is a commercially influenced diagnosis because research and guideline development have been heavily influenced and marketed by formula companies.¹³  Caregivers who live in neighborhoods with less socioeconomic deprivation are more likely to have higher incomes, more knowledge of this diagnosis, and a more powerful voice with their pediatrician; hence, they may be more likely to receive this diagnosis.³²  Second, we must consider the role of illness scripts rooted in problematic biases. Consider 2 different hospitalized infants: (1) one who receives primary care in a federally qualified health center with a social history significant for food insecurity and an open CPS case on a sibling whose mother is absent from the bedside, and (2) one from the suburbs followed by a private practice pediatrician with a mother at the bedside describing frequent spit-ups and fussiness with feeds. Pattern recognition in these cases may lead to anchor bias and diagnostic momentum affecting subsequent reasoning, problem-solving, and management.³³ 

In addition to finding that the named etiology is associated with SW and CPS involvement, an examination of covariates in the model revealed that deprivation index was associated with SW and CPS (Supplemental Table 4). The illness scripts above, as well as this disparity in CPS involvement, may contribute to a phenomenon of “parental blame,” in which a parent is blamed for the poor health of their child. Black women disproportionately feel blamed for infants’ illnesses in the NICU compared with white women,³⁴  and research suggests that mothers of infants with inadequate weight gain experience similar feelings of blame.³⁵  Parental blame has been shown to limit engagement with the health care team in the NICU;³⁶  this likely carries over to other health care contexts and may shape later interactions with the health care system, spurring mistrust that produces and amplifies racial gaps across health outcomes.²⁵  This complements literature revealing poverty and Black race as risk factors for CPS and maltreatment diagnostic codes.³⁷ 

One potential mitigation tactic to reduce the disparities in CPS involvement may be to standardize SW consults in infants admitted with inadequate weight gain. This may lessen the risk of overlooking neglect cases, especially in patients of white race or from higher socioeconomic status. It may also help the health care team make meaningful connections between families in need and community resources. However, many infants do not have the psychosocial etiologies of inadequate weight gain, and SW consults are not always necessary or high-value. Additionally, we must acknowledge that although standardizing SW consultation may mitigate the inherent bias that providers experience in deciding when and for whom to consult SW, it also runs the risk of shifting the bias from one health care provider (the managing physician) to another (the social worker). Reducing disparities requires more than referral standardization and self-reflection. It will require the acknowledgment and mitigation of systemic and structural forces and biases inherent to our approach to care delivery.³⁸ 

Our study has several limitations. Although this is a single-center study, our hospital system is unique in that it is the only hospital that admits children in our regional catchment area, which includes urban, suburban, and rural geographic locations. We also have >60 different hospitalists who manage our patient population across the 2 sites of care included in this study. The primary county for our patient population has poor neonatal health outcomes, as evidenced by an infant mortality rate worse than the national average throughout the study period.³⁹  For this reason, some of the disparities identified in our study may be more significant than in other geographic locations. Our population was defined by ICD-9 and ICD-10 codes, so patients had to be coded correctly to be included in our study. To honor the granularity of racial and ethnic categories, we chose not to reduce or combine populations, limiting our sample size to patients who identified as exclusively non-Hispanic Black or white.⁴⁰  This contributes to limited generalizability. Additionally, although our institution’s practice is to ask a caregiver about their child’s race, there are imperfections in this self-reporting process. Our study was also limited to what occurs during hospitalization for inadequate weight gain, a diagnosis predominantly managed in the outpatient setting. We chose to highlight the hospital medicine inpatient setting (rather than the subspecialty inpatient setting) to capture those with undifferentiated inadequate weight gain. It is unclear how the inclusion of patients admitted to subspecialty services (eg, GI or Cardiology) would impact our findings. We also did not include patients managed in an outpatient setting for inadequate weight gain because focusing on inpatient cases allowed us to highlight more severe cases and was the most reliable way to collect data on a broad range of patients. We are the only hospital that admits children in our region, and we do not have access to most primary care practice records. Despite our concentration on the inpatient setting, we expect that our findings would be relevant across the inpatient-to-outpatient continuum. We acknowledge that additional research into the intrusion of bias in outpatient care for infants with inadequate weight gain is needed. Finally, the disparities that we highlighted in this study have numerous upstream contributors (eg, prenatal growth and pregnancy complications) that we were unable to examine because of limitations in our dataset. More research is needed to determine the impact of inequitable prenatal and neonatal outcomes on hospitalizations for inadequate weight gain.

In a health care system in which trust differs by race because of demonstrated racism in medicine, it is critical that communities that have been historically marginalized, mistreated by health care systems in the past, and continue to experience inequities today feel heard and cared for when seeking medical care. The under- and overdiagnosis of subjective organic inadequate weight gain etiologies and implications of caregiver blame are modifiable variables that should be carefully considered when caring for these infants. The next steps include surveying providers about drivers of diagnostic decisions, obtaining feedback on practice patterns, and pursuing quality improvement initiatives that promote equitable care delivery.⁴¹