When timing and dose of nutrition support were examined, the modified Nutrition Risk in Critically Ill (mNUTRIC) score did not differentiate high-risk patients who would derive the most benefit from nutrition support: a prospective cohort study

Background The timing and dose of exclusive nutrition support (ENS) have not been investigated in previous studies aimed at validating the modified Nutrition Risk in Critically Ill (mNUTRIC) score. We therefore evaluated the mNUTRIC score by determining the association between dose of nutrition support and 28-day mortality in high-risk patients who received short- and longer-term ENS (≤ 6 days vs. ≥ 7 days). Methods A prospective cohort study included data from 252 adult patients with > 48 h of mechanical ventilation in a tertiary care institution in Singapore. The dose of nutrition support (amount received ÷ goal: expressed in percentage) was calculated for a maximum of 14 days. Associations between the dose of energy (and protein) intake and 28-day mortality were evaluated with multivariable Cox regressions. Since patients have different durations of ENS, only the first 6 days of ENS in patients with short- and longer-term ENS were assessed in the Cox regressions to ensure a valid comparison of the associations between energy (and protein) intake and 28-day mortality. Results In high-risk patients with short-term ENS (n = 106), each 10% increase in goal energy intake was associated with an increased hazard of 28-day mortality [adj-HR 1.37 (95% CI 1.17, 1.61)], and this was also observed for protein intake [adj-HR 1.31 (95% CI 1.10, 1.56)]. In contrast, each 10% increase in goal protein intake in high-risk patients with longer-term ENS (n = 146) was associated with a lower hazard of 28-day mortality [adj-HR 0.78 (95% CI 0.66, 0.93)]. The mean mNUTRIC scores in these two groups of patients were similar. Conclusion When timing and dose of nutrition support were examined, the mNUTRIC did not differentiate high-risk patients who would derive the most benefit from nutrition support. Electronic supplementary material The online version of this article (10.1186/s13613-018-0443-1) contains supplementary material, which is available to authorized users.


Background
Non-volitional nutrition support is frequently required in the intensive care unit (ICU). This seemingly straightforward therapy has garnered increased attention in the literature reflecting conflicting evidence surrounding the optimal timing, dose, rate of advancement, and composition of nutrition support [1]. While several studies demonstrated delayed [2,3] or permissive underfeeding [4][5][6] to be either benign or beneficial, these modes of feedings have been reported to be detrimental in other studies [7][8][9]. A possible explanation for the disparate findings is that a one-size-fits-all approach to nutrition support is not applicable to the needs of a heterogeneous group of critically ill patients [1].
To address this issue, Heyland et al. [10] developed a score [Nutrition Risk in Critically Ill (NUTRIC)] to better determine patients in a heterogeneous ICU population who would be more likely to benefit from adequate nutrition support. While the original score comprised six components, it was subsequently revised to exclude interleukin-6 concentrations as this is rarely measured outside of research settings [9]. Consequently, the modified NUTRIC (mNUTRIC) has five components [i.e. age, Acute Physiology and Chronic Health Evaluation II (APACHE II), Sequential Organ Failure Assessment (SOFA), number of comorbidities, and days in hospital before admission to ICU] with scores "0-4" and "5-9" classified as low risk and high risk, respectively [11,12]. There are now four published validation studies [9,[11][12][13] with three showing acceptable external validity for the mNUTRIC score: high-risk patients who received higher average energy [9,11,12] and protein [11] intake were observed to have lower mortality. These results suggested that goal energy and protein intake should be achieved as soon as possible via early aggressive (i.e. high dose) nutrition support including: (1) starting enteral feeding at goal rate [14], (2) using prokinetic agents prophylactically to enhance enteral feeding tolerance [15], and/or (3) using supplemental or total parenteral nutrition support when enteral nutrition cannot meet requirements within the first few days of ICU admission [7,16].
However, recent evidence conflicts with these aggressive feeding practices. In patients with acute respiratory distress syndrome (ARDS), Braunschweig et al. [17] and Peterson et al. [18] reported that early aggressive nutrition support with a higher energy intake at the most acute phase of critical illness (i.e. around day 1 to day 7 of ICU admission) was associated with increased mortality. A close examination on the patients' characteristics of Braunschweig et al. [17] revealed that most would be classified as high risk by the mNUTRIC (score = 5 since mean age, SOFA, APACHE II, and length of hospitalisation before ICU admission were 59 years old, 8.3, 22.5, and 3 days, respectively). In addition, Doig et al. [19] showed that early aggressive feeding at the initial stage of ICU admission was associated with higher mortality in patients with refeeding syndrome (even with adequate phosphate replacement). These studies suggest that early aggressive nutrition support may not benefit all critically ill patients.
To date, the mNUTRIC recommendations would support early aggressive nutrition treatment for high-risk patients, but the concerns of harm associated with early aggressive nutrition call into question the generalisability of the mNUTRIC score to all patients. It is therefore timely to re-evaluate if early aggressive nutrition support is of benefit to all high-risk patients in a heterogeneous ICU. Since the effects of timing and dose of nutrition support have not been investigated in previous mNUTRIC studies, we therefore aimed to determine whether timing and dose of nutrition support in critically ill patients may modify the association between mNUTRIC categories (low risk and high risk) and 28-day mortality in a singlecentre cohort study.

Patient and setting
This was a prospective observational cohort study conducted in a 35-bed ICU in Ng Teng Fong General Hospital (Singapore) between August 2015 and October 2016. The ICU functions as a closed unit where board-certified intensivists and residents provide care for both medical and surgical patients. Treatment bias was minimised by blinding the intensivists and nurses to the objectives of the study.
To determine whether the association between mNU-TRIC categories and 28-day mortality was modified by not only the dose of nutrition support (as in the original study), but also its timing, all patients ≥ 21 years old and had > 48 h of mechanical ventilation and enteral or parenteral feeding planned were included in the study. In addition, these patients were not declared moribund by an intensivist and had nutritional status determined by a dietitian (using the Subjective Global Assessment [20]) within 48 h of ICU admission. Nutritional status was an inclusion criterion because it has been previously associated with mortality in ICU patients [21][22][23].
As per usual clinical practice in our unit, all patients received a nutritional assessment and were prescribed an appropriate enteral or parenteral feeding regime within 48 h of ICU admission. As the mNUTRIC was not part of the routine nutritional assessment, it was calculated at the end of the study to minimise treatment bias. For the calculation of energy and protein goals, actual body weight taken at ICU admission using weighing bed was used. In obese patients (body mass index > 30 kg/m 2 ), adjusted body weights [(actual body weight − ideal body weight) × 0.25 + ideal body weight] [24] were used. The dose of enteral and/or parenteral formulas received was recorded in the electronic medical records and verified by the attending nurse at the end of each shift.

Data collection
All data [i.e. demographics, disease severity scores (e.g. APACHE II), comorbidities, baseline nutritional status, admission diagnoses, medications, intravenous fluids, energy and protein provided by enteral and/or parenteral nutrition, and clinical outcomes] were prospectively measured and recorded in the electronic medical records. The daily energy and protein intake of patients was calculated while receiving exclusive nutrition support (ENS), either enterally and/or parenterally, from ICU admission to a maximum of 14 days, unless death occurred earlier. Energy and protein intake was calculated from enteral formulas, protein modular and readyto-use or compounded parenteral formulas. In addition, energy provided by propofol and dextrose-containing intravenous fluids were included in the calculation of total energy intake.
The dose of nutrition support was calculated by dividing the total energy and protein received by the number of days on ENS, and expressed as a percentage of the goals established at baseline [25]. Nutrition support received on the day of death was excluded in the calculation of total energy and protein intake since patients would not have received the entire prescription [10]. Ethics approval was granted by the Domain Specific Review Board (NHG DSRB Ref: 2014/00878).

Statistical analysis
The association between energy (and protein) intake and 28-day mortality was examined in two sets of multivariable Cox proportional hazard regressions. In the first set, we examined the association between each 10% increase in goal energy (and protein) intake and 28-day mortality for the entire cohort. In the second set, the effects of timing and dose of nutrition support on 28-day mortality were examined. Braunschweig et al. [17] and Peterson et al. [18] observed that the dose of energy intake at the early phase of critical illness (ICU day 1 to day 7) was positively associated with mortality, and a crossover effect was observed at the later phase in patients who required longer-term nutrition support (ICU day 8 onwards). Therefore, we determined whether this phenomenon was also present in our cohort by calculating the average goal per cent energy and protein intake from day 2 (mean of day 1 and day 2) to day 14 (mean of day 1 to day 14) in survivors and non-survivors and plotting their relationships stratified by mNUTRIC categories (low and high risk). Thereafter, we defined "short-term ENS" and "longer-term ENS" intervals by observing for crossover associations between the per cent goal energy (and protein) intake in survivors and non-survivors.
The associations between energy (and protein) intake and 28-day mortality were determined by multivariable Cox proportional hazard regression. Covariates to be adjusted were identified by comparing the patients' baseline characteristics using Student's t test, Chi-square test, or Mann-Whitney U test as appropriate. Characteristics that were significantly different (p value < 0.05) between survivors and non-survivors at the univariate level were included as covariates in the multivariable Cox proportional hazard regressions to generate the adjusted hazard ratio (adj-HR). Variance inflation factors and tolerances were used to check for multicollinearity. The above steps were repeated in multivariable logistic regressions to generate figures that depict the associations between goal energy (and protein) intake and 28-day mortality (Additional file 1: Table S1 and Figure S1). Statistical analyses were performed using STATA 14.2 (StataCorp, College Station, TX, USA). For all comparisons, associations, and interactions, p value < 0.05 was considered as significant.

Results
There were 252 patients enrolled (Fig. 1), and no patients were lost to follow-up. Mortality at day 28 following ICU admission was 33.3%, and the characteristics of survivors and non-survivors are summarised in Table 1. Nonsurviving patients had a significantly higher mNUTRIC score, were more likely to be malnourished, admitted for medical reasons, transferred from the general ward, and resuscitated before ICU admission. Patients who were excluded had similar characteristics to those enrolled apart from a lower SOFA score (median 8 vs. 9, p < 0.001) and a higher number of comorbidities (median 3 vs. 2, p < 0.001).
The cut-off intervals that defined "short-term ENS" and "longer-term ENS" were set at ≤ 6-days (n = 106) and ≥ 7-days (n = 146) of ENS, respectively, as clear separation was observed between the per cent goal energy and protein intake of survivors and non-survivors at the univariate level (Fig. 2). In patients with short-term ENS and classified as high-risk (n = 64), a large proportion of them perished in the first 6 days of ENS (Fig. 3), and enteral and/or parenteral feeding was ceased due to quick progression to oral feeding or early death. In addition, a higher proportion of them were admitted for medical reasons compared to patients with longer-term ENS. However, the median mNUTRIC score and the proportion of high-risk patients between both groups were not significantly different ( Table 1).
The mean (SD) per cent goal energy and protein intake was 65.3% (24.7) [16.6 (7.0) kcal/kg] and 61.2% (27.4) [0.71 (0.34) g/kg], respectively. To ensure a valid comparison for the associations between energy (and protein) intake and 28-day mortality in patients with short-and longer-term ENS, only the first 6 days of ENS in both groups were assessed. Patients with short-term ENS had significantly lower energy [48.0% vs. 68.7%, p value < 0.001 (12.0 kcal/kg vs. 17.5 kcal/kg] and protein intake [41.6% vs. 64.8% versus, p value < 0.001 (0.47 g/ kg vs. 0.77 g/kg], and higher incidence of feeding intolerance when compared to those with longer-term ENS (p = 0.001) ( Table 2). In addition, patients who required longer-term ENS had a significantly higher percentage of energy provided in the form of protein compared to patients with short-term ENS (17.6% vs. 14.8%, p value < 0.001).

Association between energy intake and 28-day mortality during the first 6 days of ENS
Given the crossover associations between per cent goal energy (and protein) intake and 28-day mortality at the univariate level, these associations were adjusted for covariates in multivariable Cox proportional hazard regressions (Table 3). Covariables in the Cox models include days on ENS because energy (and protein) intake increase with time [25,26] and adjustment for this immortal time bias is recommended and widely practised [9,11,25].
In the first analysis set (n = 252), where timing and dose of nutrition support were not examined, there was no significant association between each 10% increase in goal energy intake and 28-day mortality in high-risk patients [adj-HR 1.22 (95% CI 0.98, 1.53), p value: 0.081, interaction between mNUTRIC categories: 0.985].
In the second analysis set, examining the effects of timing and dose of nutrition support, both univariate and multivariable analyses were performed to determine the associations between per cent goal energy intake and 28-day mortality in low-and high-risk patients with short-and longer-term ENS. Univariate analyses-per cent goal energy intake was divided into tertiles, and the associations with 28-day mortality in low-and high-risk patient with short-and longer-term ENS are illustrated in Fig. 4. In low-and high-risk patients with short-term ENS, goal energy intake in the highest tertiles was associated with the highest 28-day mortality risk. In contrast, goal energy intake in the highest tertile was associated with the lowest mortality risk in high-risk patients with longer-term ENS. Multivariable analyses-in patients with short-term ENS, there was no significant interaction in the group (p value 0.280) (Table 3). However, highrisk patients had a 37% higher hazard (p value < 0.001) of 28-day mortality with each 10% increase in goal energy intake, while low-risk patients lost significance. Similarly, there was no significant interaction (p value 0.127) in patients with longer-term ENS. While there was an inverse association between per cent goal energy intake and 28-day mortality in high-risk patients, this was not statistically significant (p value 0.135).

Association between energy intake and 28-day mortality in patients with up to 14 days of ENS
There were significant interactions between mNUTRIC categories (p value 0.034), but the association between per cent goal energy intake and 28-day mortality in both the mNUTRIC categories was not significant. Every 10% increase in goal energy intake was associated with a nonsignificant increased hazard of 28

Association between protein intake and 28-day mortality during the first 6 days of ENS
There was no significant association between protein intake and 28-day mortality in high-risk patients when timing and dose of nutrition support were not examined [adj-HR for each 10% increase in goal protein intake for the entire cohort (n = 252): 1.14 (95% CI 0.93, 1.39), p value: 0.231, interaction between mNUTRIC categories: 0.881].
In the second analysis set, where the effects of timing and dose of nutrition support were examined, both univariate and multivariable analyses were performed to determine whether the associations between per cent goal protein intake and 28-day mortality in patients with low and high risk were stratified by short-and longer-term ENS. Univariate analysesper cent goal protein intake was divided into tertiles, and the associations with 28-day mortality in low-and high-risk patient with short-and longer-term ENS are illustrated in Fig. 4. In low-and high-risk patients with short-term ENS, goal protein intake in the highest tertiles was associated with the highest 28-day mortality    risk. In contrast, goal protein intake in the highest tertile was associated with the lowest mortality risk in high-risk patients with longer-term ENS. Multivariable analyses-in patients with short-term ENS, there was no significant interaction in the group (Table 3). However, high-risk patients had a 31% higher hazard of 28-day mortality with each 10% increase in goal protein intake (p value 0.002). In patients with longerterm ENS, the association between per cent goal protein intake and 28-day mortality varied by mNUTRIC categories with a trend of interactions (p value: 0.088): High-risk patients had a 22% lower hazard of 28-day mortality with each 10% increase in goal protein intake (p value 0.006).

Discussion
To our knowledge, this is the first study to suggest that the association between mNUTRIC score and 28-day mortality can be modified by the timing and dose of

Table 3 Association between energy (and protein) intake and 28-day mortality in low-and high-risk patients stratified by days on exclusive nutrition support
Values are hazard ratio (95% CI) adjusted for exposure to cardiopulmonary resuscitation before admission to the intensive care unit, nutritional status, and days on exclusive nutrition support a Low and high risk is defined as scores "0-5" and "6-9" of the modified Nutrition Risk in Critically Ill (mNUTRIC) score, respectively [9] Energy/protein intake Short-term exclusive nutrition support (≤ 6 days) Longer-term exclusive nutrition support (≥ 7 days) Low risk a (n = 42) High risk a (n = 64) Interaction Low risk a (n = 73) High risk a (n = 73) Interaction nutrition support. In high-risk patients with short-term ENS (≤ 6 days), energy (and protein) intake was positively associated with 28-day mortality risk. In contrast, protein intake was inversely associated with 28-day mortality in high-risk patients who required longer-term ENS (≥ 7 days).

Association between energy intake and 28-day mortality
The average energy intake achieved in our study was similar to three previous studies that determined the validity of the mNUTRIC score (58.5-64.5% of energy goal) [11,13,27] but we had different observations. When our cohort was analysed in its entirety (irrespective of timing and dose of nutrition support), energy intake was not associated with 28-day mortality. These findings are consistent with previous work by Arabi et al. [13] but differ those from Rahman et al. [9]. The reasons for this lack of concordance are unclear, but the latter was conducted as a post hoc analysis of a multicentre randomised controlled trial (RCT) originally undertaken to examine the effects of glutamine and antioxidant supplementation in critically ill patients [28]. In the post hoc analysis, the investigators examined the external validity of the mNU-TRIC score in identifying those patients who will benefit most from adequate energy intake. It demonstrated an inverse association between energy intake and 28-day mortality in high-risk patients. However, the lack of statistical adjustment for the amount of glutamine intake may limit the interpretation of this finding as it is possible that adequate energy, when combined with glutamine, may result in lower mortality per se [29,30]. This potential confounder was avoided in the post hoc analysis conducted by Arabi et al. [13] since the original study [4] was specifically designed to examine the effects of energy intake on mortality. Hence, Arabi's findings of the lack of significant association between energy intake and 28-day mortality [OR 0.93 (95% CI 0.60, 1.44), p value: 0.74] are more likely to be reliable [13], and our study concurs with this result. It is also possible that analysis of the timing and dose of energy intake in that study may also show an inverse association between energy intake and risk of 28-day mortality in high-risk patients receiving longerterm nutrition support. When timing and dose of nutrition support were considered in our analysis, a trend was observed towards an inverse association between energy intake and mortality risk in high-risk patients who had up to 14 days of ENS. We hypothesise the lack of significance could reflect the small sample size. This trend of inverse association is in agreement with the findings by Compher et al. [11]. In this large multinational prospective cohort study, energy intake was shown to be inversely associated with 60-day mortality in high-risk patients who had up to 12 days of ENS. Consequently, the investigators recommended that all high-risk patients should receive early aggressive nutrition support as they will benefit most from near-goal energy intake [11]. However, as the present study suggested a positive association between early high energy intake and 28-day mortality in high-risk patients with short-term ENS, this recommendation may need to be applied with caution at the early stage of nutrition support.
Furthermore, some studies suggested that early high energy intake is associated with increased mortality in certain groups of patients. A recent RCT (n = 78) demonstrated that high energy intake in the first 7 days of ARDS diagnosis resulted in higher mortality [6], and an energy threshold of 18 kcal/kg was significantly associated with mortality in the post hoc analysis [17]. These findings were supported by a larger cohort study (n = 298) that included ARDS patients with higher mNUTRIC characteristics [18]. In addition, Arabi et al. [31] also demonstrated that early (≤ 12 days of ENS) high energy intake in a heterogeneous ICU population was significantly associated with 90-day mortality. Collectively, these studies suggested that high energy intake in the early stage of nutrition support may not be of benefit in all high-risk patients. However, the mechanism of harm associated with early high energy intake is poorly studied. Some would attribute to mitochondrial toxicity caused by an oversupply of glucose and lipid [32], while others have linked it to the suppression of autophagy [33,34].

Association between protein intake and 28-day mortality
Although the average protein goal achieved in our study was similar to two previous studies that determined the validity of the mNUTRIC score (56.5% and 58.9% of protein goal) [11,27], we observed different results. In patients with short-term ENS, early higher protein intake was associated with increased mortality risk. It is, however, unclear whether this risk is solely attributed to protein or a reflection of the harm associated with early higher energy intake. A recent study suggested that early high protein intake (> 0.8 g/kg) was associated with a higher hazard of 6-month mortality when compared to patients who had protein restriction during the first 3 days of ENS and thereafter a higher protein intake [35]. The average proportion of energy provided by protein (protein-energy) in the enteral feeds used in our hospital is 15%. This level of protein-energy coincided with that received in patients with short-term ENS (Table 2), suggesting that protein modular was minimally used. Since it is impossible to statistically separate protein from energy in the analyses, it will be challenging to differentiate the associations between mortality and protein (and energy) intake.
In contrast, the protein-energy intake in patients with longer-term ENS was significantly higher than those with short-term ENS, suggesting that protein modular was used to increase the protein-energy ratio. This suggests that the inverse association between protein intake and mortality risk in patients with longer-term ENS is more likely, a result concordant with earlier studies [11].

Strength and limitations
There are some advantages in the current study. The consecutive recruitment, complete follow-up, and blinding of the treatment team to the objectives of the study minimised the selection, attrition, and treatment biases. In addition, the exclusion of moribund patients reduced the artificial inflation of the association between 28-day mortality and inadequate nutrition support-since these patients generally receive little nutrition due to poor tolerance or comfort feeding, and death is mainly due to disease severity.
There are, however, several limitations that must be considered before drawing conclusions with clinical implications. This was a single-centre observational study with a small sample of heterogeneous patients. The positive association between early high energy intake and 28-day mortality in high-risk patients was not expected. By indication, severely ill patients (hence short survival time) would usually receive lesser nutrition due to enteral feeding intolerance, and this may artificially inflate the inverse association between nutrition intake and mortality. However, our results were in the opposite direction despite our best efforts to adjust for known confounders. Therefore, the presence of residual confounders inherently limits our results to associations rather than causation. In addition, it was beyond the scope of our study to investigate possible causes for this observation. Therefore, our study should be considered as hypothesis generating and require further confirmation from larger or more comprehensive studies that consider extensive number of confounders.

Future research and implications for practice
In our study, energy and protein intake was positively associated with 28-day mortality in high-risk patients with short-term ENS, while it was not associated with 28-day mortality in the group with longer-term ENS. These results are disconcerting because the mNUTRIC score did not discriminate these two groups of patients (median mNUTRIC scores for both groups were 6), and it may not be possible for clinicians to accurately predict the length of ENS at ICU admission. Some characteristics of patients with short-and longer-term ENS are given in Table 1. Of note, a higher proportion of patients with short-term ENS were admitted for medical reasons, and more had cardiovascular or respiratory issues as compared to those with longer-term ENS. Conversely, patients who required longer-term ENS were those admitted with neurological issues. Therefore, there is a need to identify subgroups of patients who would likely benefit or be harmed by early aggressive nutrition support.
The present study supports the requirement for larger confirmatory studies to further investigate the modifying effect of timing and dose of nutrition support in highrisk patients. Ideally, this could be achieved by identifying biomarkers that define the different phases (i.e. acute, sub-acute, and chronic) of critical illness [36] and testing whether limiting and increasing energy and protein intake at different phases would be beneficial. Until this is achieved, we suggest a prudent approach to nutrition support. The energy and protein intake that was associated with identical mortality risks in patients with shortand longer-term ENS was 50% of goal (Additional file 1: Figure S1). Therefore, to achieve equipoise, clinicians may feed high-risk patients at 50% of the energy and protein goal in the early periods of ICU admission and intensify the provision of nutrition if ENS was required for a more extended period.