Introduction

The increasing survival rates following pediatric critical illness have led to a shift in focus towards the long-term outcomes of critically ill children [1, 2]. Unfortunately, many survivors are at risk of experiencing later neurocognitive, emotional, or behavioral impairments [3,4,5,6,7,8,9,10,11,12]. These neuropsychological impairments have a multifactorial cause, which complicates the accurate risk prediction during and after critical illness [3, 4]. Currently, identifying at-risk patients relies on neuropsychological assessment, but higher-order cognitive functions, such as memory and executive functions, can only be reliably assessed from school-age onwards. As a result, recognizing at-risk patients for academic difficulties often occurs after the manifestation of neuropsychological deficits [13,14,15,16]. Implementing preventive interventions earlier in life could positively impact neuropsychological development, leading to a growing interest in identifying early predictors measurable during infancy.

Growth and nutritional outcomes could possibly serve as early predictors, considering that critical illness negatively affects these factors in children during hospitalization and in the post-hospitalization period, with neonates and infants being particularly vulnerable [17,18,19,20,21,22]. In neonates and infants, poor growth and inadequate nutritional status at the time of admission to the pediatric intensive care unit (PICU) have been linked to unfavorable short-term health outcomes, including prolonged length of PICU stay, increased risk of infection, and increased mortality [23, 24]. However, the association between growth and nutritional status and long-term neuropsychological outcomes, such as neurocognitive functions or emotional and behavioral problems, remains unknown. Unraveling this association in critically ill neonates and infants is crucial since the first year of life is critical for brain development [25,26,27,28]. During this time period, the brain is particularly vulnerable for external factors that can influence its development, such as hypoxia, stress, and exposure to anesthetics [15, 29,30,31]. Given the increased risk of weight deterioration and poor nutritional status in neonates and infants surviving critical illness during this critical time window, further investigation is needed to understand the possible association between growth measures during initial hospital admission and neuropsychological outcomes later in life in this population.

The current study is a secondary analysis of the multicenter, randomized controlled trial (RCT): the pediatric early versus late parenteral nutrition in critical illness (PEPaNIC) trial [32].

This RCT demonstrated that withholding from supplementary parenteral nutrition (PN) during the initial week of admission to the pediatric intensive care unit (PICU) yielded superior short-term outcomes. These benefits encompassed a reduced occurrence of new infections, a shorter duration of stay in the PICU, and diminished direct healthcare expenditures, as opposed to the administration of parenteral nutrition on the day of PICU admission [33]. In the long term, withholding PN for 1 week did not exert adverse effects on survival, anthropometric measurements, health status, or neurocognitive development [7, 8]. Furthermore, it was observed that, at 2 years and 4 years follow-up, children in the delayed-PN group had better scores on parent-reported executive functioning and/or emotional and behavioral problems and/or improved visual-motor integration, in comparison to the children in the early-PN group. Critically ill children aged 29 days to 11 months at time of exposure were identified as most vulnerable to developmental harm evoked by early-PN [34]. Therefore, in the current secondary analysis, we first focus on investigating the weight trajectories of critically ill neonates and infants admitted to the PICU from admission to 4-year follow-up. Second, we aim to identify possible predictors of these weight trajectories. Lastly, we examine the association between weight growth during hospital stay and growth and neuropsychological outcomes at 4-years follow-up.

Materials and methods

Study design

This is a secondary analysis of the multicentre pediatric early versus late parenteral nutrition in critical illness (PEPaNIC) trial randomized controlled trial in a subgroup of children aged < 1 year at randomization. The PEPaNIC trial was conducted at the University Hospitals Leuven, Belgium; Erasmus MC – Sophia Children’s Hospital, Rotterdam, the Netherlands; and Stollery Children’s Hospital, Edmonton, AB, Canada. The study included 1440 critically ill infants and children admitted to the pediatric intensive care units of the participating centers between 2012 and 2015. For additional information regarding the trial protocol, see Appendix S1 (Supplementary files). The full study protocol has been published [32]. Additional to the inclusion criteria for the PEPaNIC RCT, inclusion criteria for the current analysis were (1) the availability of anthropometric measurements on admission and on the last day in hospital and (2) the availability of neuropsychological outcomes at 4-year follow-up. Individuals with a syndrome diagnosis (for the applied definition of a syndrome, see Appendix S2 (Supplementary files)) or a non-physiological growth per week more than two times the age-appropriate norm [35, 36] were excluded. As anthropometry at PICU and hospital discharge were only measured in the Erasmus MC-Sophia Children’s Hospital, only children included at this site were eligible for inclusion in this secondary analysis. This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Ethics Committee of Erasmus MC (NL49708.078). Written informed consent was obtained from the parents.

Weight trajectories from PICU admission to 4-year follow-up

Weight was measured to the nearest 0.01 kg using calibrated scales [24], upon PICU admission, at PICU discharge, at hospital discharge, and at 2- and 4-year follow-up. Early growth was defined as change in weight Z-score from admission to the last day in PICU and from admission to the last day in hospital. Weight Z-score was defined as weight-for-age Z-score in children < 1 year old and BMI-for-age Z-score in children ≥ 1 year old [37]. For neonates, birthweight-for-gestational age Z-scores [38] were used until the age of seven days, to account for physiological weight loss during the first week of life. Change in weight Z-score was categorized as decline (change in weight Z-score < 0, rounded to two decimals), no change (change in weight Z-score = 0), and incline (change in weight Z-score > 0).

To investigate whether clinical parameters predict growth in weight outcomes, weight trajectories were plotted for different subgroups: diagnostic group (surgical cardiac, surgical other, and medical), neonates (age < 1 month) at admission yes/no, malnourished at admission (weight-for-age Z-score < −2) yes/no, and malnourished at hospital discharge yes/no.

4-year follow-up of neuropsychological outcomes

Outcome was assessed by pediatric psychologists, who were not involved in the management of the patients during their stay in the pediatric intensive care unit. They were strictly masked to treatment allocation. Parents and caregivers were not masked to treatment allocation and were not actively informed about the initial PEPaNIC study results or the 2-year outcome results [7].

For all participants determined as being able to undergo neurocognitive testing (Appendix S3 (Supplementary files)), performance for a broad range of neurocognitive functions and emotional and behavioral problems was scored using age-appropriate, validated, internationally recognized parent-reported questionnaires, and clinical tests with adequate normative data [7]. At the follow-up visit, body weight and height were measured. A clinical neurological examination was done to assess gross neurological abnormalities, and findings were used to determine cognitive testability. Testability was determined by screening the medical file or on clinical judgment before the start of the cognitive assessment by the physician or psychologist and confirmed by the parents or caregivers.

The test battery at the 4-year follow-up assessment included parent-reported questionnaires, i.e., the Behavior Rating Inventory of Executive Function [39] (BRIEF, for executive functioning, expressed in T scores, with mean 50 and standard deviation (SD) 10) and the Child Behavior Checklist [40] (CBCL, for emotional and behavioral problems, expressed in T scores, with mean 50 and SD 10). On both questionnaires, higher scores indicate more problems.

Additionally, clinical tests consisted of the age-appropriate version of the Wechsler Intelligence Quotient Scale [41, 42] (WPPSI-III-NL, expressed in standard scores, with mean 100 and SD 15), the Beery Developmental Test of Visual-Motor Integration [43] (VMI, expressed in a scaled score, with mean 10 and SD 3), and tasks of the Amsterdam Neuropsychological Task Battery (ANT) [44] (ANT-Baseline Speed (alertness and reaction time expressed in z-scores with mean 0 and SD 1) and ANT-Tapping (motor coordination as number of taps)). For the clinical tests, a higher score indicated better functioning, with the exception of ANT-Baseline Speed.

Statistical analysis

Growth in weight parameters

In case of multiple recorded weight measurements at the same day, the first measurement was used for all measurements registered at that day. Growth per week during hospital stay was calculated by subtracting weight in kg at admission from weight in kg at hospital discharge and dividing the difference by the length of hospital stay in weeks. Plausibility of calculated growth in weight rates was checked, and all individuals with a calculated growth per week of more than two times the age-appropriate average growth per week [35, 36] were excluded.

Descriptive statistics

Variables are reported as proportions, mean (SD), or median (interquartile range (IQR)) as appropriate. Descriptive statistics were compared between (sub)groups using (paired) two-sample t-tests, Kruskal–Wallis tests, chi-squared test, or ANOVA analysis of variance.

Correlation and association analyses

Correlation between growth from hospital admission to 4-year follow-up and outcome measures at 4-year follow-up were assessed by computing a correlation matrix of the univariate association between all complete pairwise observations, using Spearman’s rho.

Multivariable linear regression analyses were done on the pooled imputed dataset (see Appendix S3 (Supplementary files)) with the β-estimates reported to investigate the association between growth in weight z-scores during hospitalization and outcome. All multivariable analyses were adjusted for predefined covariates, being sex, diagnostic group (cardiac surgery, other surgery, and medical), age group at admission (neonate versus non-neonate), severity of illness (PIM3 score), parental smoking behavior before admission to the PICU (in-house smoking or no in-house smoking), and occupational level of the parents or caregivers (see Appendix S5 (Supplementary files)). Analyses were not corrected for the treatment strategy, as previous analyses showed no association between treatment strategy and growth during hospital stay [24].

Sensitivity analyses

Sensitivity analyses were done by repeating linear regression analyses with the change in growth per week in kilograms instead of weight Z-score during hospital stay, as the weight Z-score takes into account sex- and age-specific growth norms. Finally, analyses were also done with both weight for (gestational) age Z-scores at PICU admission as well as at hospital discharge, to assess the interplay between static anthropometric measurements at the beginning and end of a period of critical illness and neurodevelopmental outcome.

Statistical software

Z-scores were calculated with use of Growth Analyzer Research Calculation Tool version 4 and Fenton 2013 Preterm Growth Chart version 6. Statistical analyses were done using RStudio (version 4.1.2) [37, 45]. Two-sided p values of 0.05 or less were considered statistically significant and were corrected for multiple comparisons by controlling the false discovery rate (FDR).

Results

Of the 309 Dutch PEPaNIC participants aged < 1 year, 121 children were eligible to be included in the analyses; see Fig. 1 for the participants flowchart.

Fig. 1
figure 1

Flowchart of study participants

Full size image

Baseline characteristics

Baseline characteristics are shown in Table 1. Characteristics of individuals included and not included in the analyses can be found in Table S1 (Supplementary files). Those not included only differed from those included in terms of median duration of hospital stay (included: 13 (8–29) versus not included: 12 (6–24), p = 0.048). Patients were evenly distributed among the randomization strategies. Median age at admission was 21 days, 53.7% were neonates, and median age (IQR) at 4-year follow-up was 4.37 (4.22–4.52) years. More detailed information regarding the diagnostic categories of the participants can be found in Table S2 (Supplementary files).

Table 1 Demographic and clinical characteristics of the study population
Full size table

Early growth in weight parameters

The mean weight Z-scores of critically ill neonates and infants during PICU admission, at PICU discharge, and at hospital discharge differed significantly from age- and sex-specific population means, see Table S3 (Supplementary files). At 4-year follow-up, weight Z-scores did not differ from age and sex-specific population means. Twenty-four percent of the infants had acute malnutrition (weight for age Z-score ≤ −2) on admission; at PICU and hospital discharge, this was 25.6% and 28.1%, respectively. During follow-up, the percentage of malnutrition decreased to 3.8% at 2-year follow-up and 8.3% at 4-year follow-up.

Mean weight Z-scores declined during hospital stay (−0.46 (0.82), p < 0.0001). The decline in Z-score was greatest in the PICU (−0.29 (0.64), p < 0.0001); the mean Z-score also declined from discharge from the PICU up to discharge from the hospital (−0.17 (0.72), p = 0.012) (Fig. 2). Median weight gain per week in kilograms was mostly below age-appropriate norms for the different age groups and phases of hospital admission (Table S4 (Supplementary files)).

Fig. 2
figure 2

Mean and standard deviation weight Z-scores

Full size image

Follow-up measurements at 2- and 4-year post-PICU admission showed a significant increase in weight Z-score compared to all measurements during hospital admission. Weight Z-score decreased from 2- to 4-year follow-up (−0.20 (0.94), p = 0.032) (Table S5 (Supplementary files).

Predictors for change in weight Z-score from PICU admission to follow-up

Diagnosis category upon admission was not significantly associated with weight Z-scores during hospital admission nor at follow-up (Fig. 3A and Table S6 (Supplementary files)). Neonates were admitted and discharged from PICU and hospital with a higher mean weight Z-score than non-neonates (Fig. 3B and Table S6 (Supplementary files)). Patients diagnosed with malnutrition (weight for age Z-score ≤ −2) at PICU admission also had a lower weight for age Z-score during hospitalization than those with no malnutrition at PICU admission (Fig. 3C and Table S6 (Supplementary files)). Those diagnosed with malnutrition at hospital discharge already had a lower mean weight for age Z-score at PICU admission compared to patients with no malnutrition at hospital discharge. Mean weight Z-score increased from hospital discharge to 2-year follow-up in both groups (mean increase of 2.67 (1.38), p < 0.0001 and 1.03 (1.25), p < 0.0001). At 2-year follow-up, patients with malnutrition at hospital discharge still had a lower mean weight Z-score compared to those without malnutrition at hospital discharge (Fig. 3D and Table S6 (Supplementary files)). This difference had disappeared by 4-year follow-up.

Fig. 3
figure 3

Weight Z-score trajectories in different subgroups. A Weight Z-score trajectory per diagnostic group, B Weight Z-score trajectory neonates versus non-neonates, C Weight Z-score trajectory by nutritional status at admission, D Weight Z-score trajectory by nutritional status at hospital discharge

Full size image

Associations between early growth during hospitalization and anthropometric and neuropsychological outcomes at 4-year follow-up

Change in weight Z-score and change in weight gain in kilograms per week during hospitalization were not significantly associated with growth and neuropsychological outcome at 4-year follow-up (Tables 2 and S7). Weight for (gestational) age Z-score at PICU admission was associated with height (β 0.22, 95%CI 0.10–0.35, p = 0.014) and weight Z-scores (β 0.20, 95%CI 0.08–0.33, p = 0.022) at 4-year follow-up (Table S8). Weight for (gestational) age Z-score at hospital discharge was associated with weight Z-score at 4-year follow-up (β 0.24, 95%CI 0.11–0.38, p = 0.014) (Table S9). Anthropometric and neuropsychological outcomes assessed at 4-year follow-up were not correlated (Fig. S2).

Table 2 Associations between change in weight Z-score during hospital admission and anthropometrics and neuropsychological outcomes at 4-year follow-up
Full size table

Discussion

In this secondary analysis of the PEPaNIC randomized control trial, we found that both change in weight-for-age Z-score and growth in kilograms per week were not associated with anthropometric or neuropsychological outcomes at 4-year follow-up. As to their weight trajectories during hospitalization, weight gain was markedly less than age- and sex-appropriate norms, expressed in a decrease in weight Z-score and a mean growth per week in kilograms below the age-specific norms. Moreover, we found a high, persistent prevalence of malnutrition during hospitalization, with the proportion of neonates and infants fulfilling the criteria for acute malnutrition (weight-for-age Z-score < −2) increasing from 24.0% at PICU admission, to 25.6% at PICU discharge, and 28.1% at hospital discharge.

This high prevalence of malnutrition is in accordance with previous studies in which a prevalence of malnutrition op to up to 24% has been found [46, 47]. However, in the years following critical illness, critically ill neonates and infants appear to be able to catch up, also those who were malnourished upon admission. In this study, most participants with malnutrition on PICU admission were also malnourished at hospital discharge (22 out of 27, 81.5%). As to what might explain this, LOS can be of interest. We did not investigate a possible association between LOS and change in nutritional status during hospitalization. However, no difference in LOS between those who were malnourished on PICU admission and those who were not, nor between those who were malnourished at hospital discharge and those who were not. With a median LOS of 13 days, this should be long enough to expect sufficient weight gain in most study participants. Regarding the difference in weight Z-score between those who were malnourished at hospital discharge and those who were not persevering to 2-year follow-up, but disappearing at 4-year follow-up, this may indicate the amount of time needed for these young children to overcome the disadvantage they started off with due to their critical illness.

The association between anthropometric measures and short term, medical outcome has been studied frequently. Poor nutritional status had been associated with longer duration of mechanical ventilation [48,49,50,51], longer duration of PICU stay [23, 47, 49, 51,52,53], and mortality [23, 51,52,53,54]. In a cohort of children with congenital heart disease, impaired weight gain was predictive of late post-surgical mortality [19]. A secondary analysis of the PEPaNIC RCT has shown that less decrease in weight Z-score during PICU admission was associated with a lower risk of new infection and a higher likelihood of earlier alive discharge from PICU [24].

On the other hand, the association between weight gain or other anthropometric measures during hospital stay and long-term, neuropsychological outcome has, to our knowledge, only been studied in children undergoing surgery for congenital heart disease [55,56,57,58]. In a study of 107 infants undergoing open-heart surgery for congenital heart disease, neurodevelopmental outcome at one year was not associated with growth failure. It was shown that impaired weight gain before surgery was followed by catch-up growth after surgery [57]. Also, in another study of 143 infants, impaired intellectual abilities at 6 years were not associated with growth in weight [58]. Only in a study of 72 infants with congenital heart disease 3-month weight-for-age Z-score was associated with lower mental developmental index and psychomotor developmental index scores at 12 months [56].

Within the PEPaNIC RCT that included the total PICU population, it has been demonstrated that withholding supplementary parenteral nutrition during the first week of a PICU admission did not harm physical and neurocognitive development assessed 2 or 4 years after critical illness, compared to early-parenteral nutrition [7, 8]. So, even though macronutrient intake was less in the late PN-group, there was no effect of PN strategy on weight Z-trajectories [24, 33]. Moreover, analyses evaluating the role of age at PICU admission on the effect of early PN showed that children aged 29 days to 11 months were most vulnerable for the observed developmental harm of early PN [34]. These previous PEPaNIC findings together with the lack of association between growth during hospital stay and 4-year neuropsychological outcomes of our current study suggest that inability to achieve intake targets and even subsequent impaired growth in weight during critical illness, especially in infants, may not be detrimental for longer term outcome of PICU survivors, opposed to what has been proposed before [55, 56, 59, 60].

Question remains what factors are associated with the observed worse outcome of young PICU survivors in some neuropsychological domains. In previous studies in infants with congenital heart disease, impaired height trajectories were associated with worse neurodevelopmental outcomes [55, 56]. However, head circumference at birth, postnatal factors, genetic comorbidity, and epigenetic changes are important factors which must be taken into account concerning impaired height trajectories and neuropsychological outcome [61]. Also, when evaluating outcome of children after PICU discharge, it is important to include baseline (health) status and family factors in multifactorial predictive models [62, 63]. The use of such predictive models to identify those critically ill children most at risk already during PICU stay is clinically relevant, as it enables the possibility for tailored long-term follow-up, which is essential to prevent further deviation of normal development as much as possible. Relevance of early identification of predictive factors is further underlined by an analysis of the change in neurodevelopmental outcome of PEPaNIC participants from 2- to 4-year post-PICU admission [64]. This analysis showed that developmental impairments remained prominent in the general PICU population group. Within the investigated time-window, impaired growth in height, impairment in executive functioning, and in intelligence aggravated. Based on these findings, it was determined that additional investigation is necessary to understand the long-term impact of pediatric critical illness on development into adulthood [64].

Limitations

The major strength of this study is the longitudinal growth assessment of a large group of former PICU neonates and infants and their neuropsychological functioning at 4-year follow-up.

The study has several limitations despite its strengths. First, since only a subset of the total trial population was analyzed, there may be selection bias. However, this bias is unlikely as participation was based mainly on the center rather than clinical variables that could influence the association being investigated. Additionally, a comparison of baseline and demographic characteristics between infants from Rotterdam included and not included in this showed no differences. As this analysis composes of only a subgroup of the PEPaNIC trial, not all covariates corrected for in the original PEPaNIC analyses [7, 8] could be added in the models of this analysis. Therefore, the overall effect of the association of weight gain and neuropsychological outcome may be diminished.

Second, the study relied on weight measurements to evaluate growth, which may be affected by factors like edema and fluid retention during illness. To mitigate this issue, growth parameters were examined throughout the entire hospital admission, not just during PICU stay. The weight measurements followed a consistent protocol for all participants, reducing the likelihood of unreliable, and potentially biased measurements. Unfortunately, assessing body composition through methods like air plethysmography, dual X-ray absorptiometry, or bioelectrical impedance was not feasible for critically ill children. Alternatively, mid-upper arm or calf circumference could have been used to estimate body composition, but the available measurements in the cohort were insufficient for regression analysis. Head circumference, another relevant anthropometric measure for neurodevelopment assessment [58, 65], was not recorded during the hospital stay for critical illness in this cohort.

Lastly, the use of the pooled imputed dataset for drawing inferences may underestimate the errors around the model coefficients. However, the obtained estimates remain unbiased.

Conclusion

Weight gain during hospital stay of critically ill neonates and infants is insufficient, resulting in growth in kilograms per week less than age- and sex-appropriate norms and thus a decrease in weight-for-age Z-score throughout hospital stay.

Although neuropsychological outcome is impaired in neonates and infants after critical illness, poor growth in weight during hospital stay was not associated with poorer cognitive, emotional, or behavioral functioning 4 years after critical illness.

Further research should focus on the effects of impaired growth during infancy on development into adulthood, extending beyond weight as growth parameter and looking into other outcome measures, assessed at a later age, as well. Moreover, additional research should try to identify early factors predictive of poorer neurocognitive functioning after critical illness in neonates and infants.