Evaluation of The ASPEN Guidelines for Refeeding Syndrome among Hospitalized Patients Receiving Enteral Nutrition: a Retrospective Cohort Study

Abstract

Background:

Until recently, refeeding syndrome (RFS) has lacked standardized diagnostic criteria. This study sought to (1) determine whether RFS, as operationalized in the 2020 ASPEN guideline definition, is associated with adverse clinical outcomes, and (2) identify key risk factors for RFS.

Methods:

This was a retrospective cohort study. Adults hospitalized from 2015 to 2019 were included in the study if they were ordered for enteral feeding during hospitalization. Data was collected for up to 30 days and RFS was operationalized as per the ASPEN 2020 guidelines as a ≥10% (corresponding to mild RFS), ≥25% (moderate), and ≥50% (severe) decline in pre-feeding serum phosphorus, magnesium, or potassium. The mortality associated with RFS was assessed, and risk factors for RFS were identified using multivariable logistic regression modeling.

Results:

Of 3,854 subjects, 3,480 (90%) developed mild RFS. Thirty-day mortality was higher in those without mild RFS (24%) than in those with mild RFS (18%) (p<0.01). When RFS was re-operationalized as a 50% decline in electrolytes, 25% of patients developed RFS with a 20% 30-day mortality. Risk factors for development of RFS included renal failure, elevated creatinine, and low platelets; additionally, pre-feeding serum phosphorus level was strongly associated with development of RFS (adjusted odds ratio 6.09, 95% confidence interval 4.95 to 7.49 for those in the highest tertile of pre-feeding phosphorus compared to the lowest).

Conclusion:

The ASPEN operationalization of RFS as a decline in baseline electrolyte values was not associated with death. Pre-feeding serum phosphorus level strongly predicted severe RFS.

INTRODUCTION

Refeeding syndrome (RFS) has been conceptualized as a set of clinical and electrolyte changes that occur in response to reintroduction of calories after a period of consistent calorie reduction or starvation⁽¹⁾. Until recently, RFS has lacked a standardized criteria for its diagnosis.⁽²⁾ Prior studies have reported wide incidence rates of RFS ranging from 0% to 80% depending on the population being studied and the criteria for RFS ⁽³⁾. The absence of a standard definition for RFS has contributed to uncertainty when it comes to determining its incidence and has hampered research related to RFS risk factors and optimal management of RFS ⁽⁴⁾. Without criteria that define RFS, it cannot be effectively studied.

To meet this need, The American Society for Parenteral and Enteral Nutrition (ASPEN) Parenteral Nutrition Safety Committee recently published consensus guidelines for the screening and management of patients who are at risk of developing or have developed RFS. ⁽¹⁾ The ASPEN guidelines, published in 2020, defined mild refeeding syndrome as a measurable reduction of 10% or more in levels of one or any combination of serum phosphorus, magnesium, and potassium from the pre-feeding baseline measurement during the five-day period following feeding, and stratified severity of RFS based on the degree of electrolyte decline from baseline and the presence of organ failure. The ASPEN guideline was intended to create an initial framework for validation and standardization of future research.

This retrospective cohort study sought to determine (1) whether RFS, as operationalized using the ASPEN guideline definition, is associated with adverse clinical outcomes, and (2) whether the risk factors for RFS proposed in the ASPEN guideline are indeed predictors of RFS.

METHODS

Overall study approach

This study was performed using a two-step approach within a single retrospective cohort. First, we sought to evaluate the performance of the ASPEN criteria for RFS within a retrospective cohort (i.e., to ask whether the ASPEN guideline operationalization of RFS associated with worse clinical outcomes). ASPEN’s operationalization of RFS hinges on electrolyte decreases from pre-feeding baseline, so we varied the cut-offs that define RFS and explored how this altered the relationship between RFS and clinical outcomes. Second, we selected the severe (>50% decline) cut-off for electrolyte changes and then evaluated risk factors for RFS using this definition.

Population

Patients were eligible for this study if they were adults 18 or more years old hospitalized from February 2015 to August 2019 at Columbia University Irving Medical Center, received an order for enteral feeding and had a baseline serum potassium, magnesium and phosphorus level recorded in the Electronic Medical Record. Patients who were fed enterally were selected because, although additional patient populations are likely at risk for RFS (e.g., those fed oral diets after prolonged fasting), it was impossible to accurately identify other patient groups. Patients were followed from the order for enteral feeding until death, hospital discharge, or for up to 30 days. This 30-day follow-up period was chosen based on the rationale that outcomes occurring after 30 days would be unlikely to be related to the initiation of feeding more than 30 days ago. This study was approved by the Institutional Review Board of Columbia University Irving Medical Center.

Performance of the ASPEN criteria for RFS

Mild RFS was defined as a 10% reduction in levels of one or any combination of serum phosphorus, magnesium, and potassium from the baseline pre-feeding measurement during the five-day period following feeding, as per the ASPEN guidelines. For all electrolytes, the baseline pre-feeding values were those obtained from the blood draw immediately preceding enteral feeding (61% of the time, these measurements were taken on the same day as feeding). If a 10% decline in any electrolyte was observed, within the five days subsequent to the initiation of feeding, refeeding syndrome was classified as present. Clinical outcomes, including death and hospital-free days, were collected using the electronic medical record (EMR).

Death was classified categorically based on documentation in the EMR of in-hospital death, or documentation of out-of-hospital death within the national social security database, which interfaces with the EMR. Additionally, hospital-free days was examined because death is not likely to fully capture the morbidity associated with RFS and because hospital-free days reflect healthcare utilization and (unlike length-of-stay) also account for death ⁽⁵⁾. Hospital-free days were calculated as the days spent alive and out of hospital during the 30-day follow-up period. For example, a patient discharged on Day 15 and alive on Day 30 would have 15 hospital-free days; a patient discharged on Day 15 who died on Day 25 would have 10 hospital-free days; a patient who died before discharge or was discharged after 30 days would have 0 hospital-free days.

After testing the ASPEN 10% electrolyte cut-off, we then tested increasing increments of electrolyte changes: 10%, 25%, and 50% roughly corresponding to mild RFS (10-20% in the guideline), moderate RFS (20-30% in the guideline), and severe RFS (>30% in the guideline). These cut-offs were selected a priori, with the lowest value (10%) and the highest value (50%) chosen to completely bracket all definitions for RFS. At each cut-off, the incidence of RFS was described and associations were tested between RFS and death within 30 days, and between RFS and hospital-free days.

Risk factors for RFS

In the second part of the study, we selected severe RFS (≥50% decline in baseline electrolytes) as the outcome. Using this definition, we then tested the ASPEN guideline risk factors for severe RFS including pre-existing conditions and being underweight at the time of feeding. The pre-existing conditions tested were: anorexia nervosa, malabsorption syndromes, mental health disorders, alcohol and substance use disorders, malignancy, history of bariatric surgery, and renal failure (all classified categorically based on International Classification of Disease (ICD) 10 Codes within the EMR, see Supplemental Table 1). The risk factor of being underweight at the time of feeding was classified based on a body mass index (BMI) recorded at the time of initiation of feeding of < 18.5 kg/m^2⁽⁶⁾. The ASPEN guideline risk factors that could not be studied because they could not be accurately captured within our dataset included unintentional weight loss of more than 15% during the previous 3-6 months, participation in athletic competitions, starvation, and little or no nutritional intake for more than 10 days preceding enteral feeding (Table S1).

Additionally, pre-feeding laboratory values were tested as risk factors for severe RFS. Laboratory values studied were serum creatinine, white blood cell count, platelets, bilirubin, and albumin. Each laboratory value was classified based on our clinical laboratory’s standard cut-offs. If no laboratory value was available within the 24 hours preceding initiation of enteral feeding, it was classified as within the normal range.

Statistical Methods

The association between RFS (operationalized using each cut-off) and death was tested in a series of chi-squared tests. The association between RFS and hospital-free days was then similarly tested using rank-sum tests (non-normally distributed data). For descriptive purposes, means were also computed, and were compared using t-tests.

For the second part of the study, the univariable relationships between risk factors and RFS were explored using chi-squared tests and then a multivariable logistic regression model for RFS was constructed. To build this model, we first included in the full model all variables that had some evidence of a relationship with RFS (p<.10) plus the following variables which were specified for inclusion a priori based on clinical experience: age, sex, BMI, anorexia nervosa and malabsorption syndromes. Variables were then dropped out stepwise, retaining in the final model only those that were specified a priori or had an independent (p<.05) relationship with RFS.

We hypothesized that the baseline pre-feeding serum values for phosphorus, magnesium, and potassium would be associated with RFS because those with low baseline values would be less likely to experience further decreases in electrolytes. To test this hypothesis, each electrolyte was divided into tertiles. The electrolytes were then tested singly within the final multivariable logistic regression model. Last, we conducted an analysis based on absolute rather than relative decline in pre-feeding serum phosphorus. All analyses were conducted using STATA version 15.0. Testing was always two-sided using the alpha .05 cut-off for determining statistical significance.

RESULTS

Population

Of the 3,854 subjects identified as receiving an order for EN, 3,480 (90%) patients met the ASPEN guideline criteria for mild RFS (10% decrease in baseline electrolytes). Among all patients, median age was 66 (IQR 54 to 76), 44% were female, and 21% had BMI < 18.5 kg/m2. Stratified by the presence or absence of mild RFS, those who developed RFS were more likely to have low serum albumin, a white blood cell count outside the laboratory reference range, normal platelets (versus platelets <150,000/mL or >400,000/mL), and low bilirubin. Otherwise, there were no differences in patient demographics, pre-existing conditions, or laboratory values comparing those who did versus those who did not develop mild RFS (Supplemental Table 2).

Incidence of RFS

Mild refeeding syndrome developed in 3,480 (90%) patients when operationalized using the 10% cut-off. A total of 2,501 (65%) patients had a decrease >= 25%, and 963 (25%) had a decrease >=50% (Figure 1).

Figure 1. Relationship between refeeding syndrome and death using different cut-offs to define refeeding syndrome.

The proportion of those who died within 30 days after the initiation of feeding is shown as bars, stratified based on RFS (left verical axis). The incidence of RFS in the cohort is shown as a line (right vertical axis).

RFS10: refeeding syndrome classified as present if there was a 10% decline from pre-feeding serum phosphorus, magnesium, or potassium.

RFS25: refeeding syndrome classified as present if there was a 25% decline from pre-feeding serum phosphorus, magnesium, or potassium.

RFS50: refeeding syndrome classified as present if there was a 50% decline from pre-feeding serum phosphorus, magnesium, or potassium.

Association between RFS and death/hospital-free days

Thirty-day mortality was 18% versus 24% in those with and without mild RFS respectively (p<0.01), 18% versus 19% in those with and without moderate RFS (p=0.60), and 20% versus 18% in those with and without severe RFS (p=0.23) (Figure 1). For hospital-free days, the median was 2 (IQR 0-17) versus 7 (IQR 0-21) in those with and without mild RFS respectively, 1 (IQR 0-16) versus 6 (IQR 0-20) in those with and without moderate RFS, and 0 (IQR 0-17) versus 4 (IQR 0-18) in those with and without severe RFS (Table 1).

Table 1.

Relationship between refeeding syndrome (RFS) and hospital free days.^*

RFS Increments **	Hospital Free Days (Median, IQR)	P-Value	Hospital Free Days (Mean, SD)	P-value
Mild RFS (10% Reduction)		<0.01		<0.01
Yes	2 (0-17)		8.7 (10)
No	7 (0-21)		11 (11)
Moderate RFS (25% Reduction)		<0.01		<0.01
Yes	1 (0-16)		8.4 (10)
No	6 (0-20)		9.9 (11)
Severe RFS (50% Reduction)		0.01		0.02
Yes	0 (0-17)		8.1 (11)
No	4 (0-18)		9.2 (10)

^*Hospital-free days were calculated as the days spent alive and out of hospital during the 30-day follow-up period.

^**Refeeding syndrome was classified categorically as present or absent: initially, it was classified as present if there was a 10% reduction in serum phosphorus, magnesium, or potassium; then as a 25% reduction in the same electrolytes, and last as a 50% reduction in these electrolytes.

Based on these results, we elected to use the 50% cut-off (severe RFS) to test risk factors associated with RFS. Our rationale for this was that (1) the prevalence of RFS when operationalized as a 10% or 25% decrease in baseline electrolytes is not consistent with clinical observation (i.e., it is implausibly high), (2) only at the 50% cut-off was there increased mortality comparing those with RFS to those without it, although this difference was not statistically significant.

Risk factors for severe RFS

Comparing patient characteristics in those with versus those without a >=50% reduction in electrolyte level, we found that patients who developed severe RFS were more likely to have alcohol/substance use, malignancy, or renal failure (Table 2). Patients who developed severe RFS were also more likely to have low albumin or to have a white blood cell count, platelets, or creatinine outside of the laboratory reference range.

Table 2.

Characteristics of patients who did and did not develop severe refeeding syndrome within 5 days after initiating feeding.

Characteristics	Severe RFS (>50% Reduction in electrolytes)	No RFS (<50% Reduction in electrolytes)	Chi-Squared P-value
Demographics
Age, years			0.39
<50	179 (25%)	552 (76%)
≥50 and <75	536 (26%)	1,539 (74%)
≥75	248 (24%)	800 (76%)
Sex			0.97
Female	422 (25%)	1,265 (75%)
Male	541 (25%)	1,626 (75%)
Race/Ethnicity			0.11
White	293 (26%)	849 (74%)
Black	106 (23%)	353 (77%)
Hispanic	126 (22%)	457 (78%)
Other/Unknown	438 (26.2%)	1,232 (74%)	0.63
BMI, kg/m2			0.63
<18.5	206 (26%)	592 (74%)
≥18.5 and <30	551 (25%)	1,641 (75%)
≥30	206 (24%)	658 (76%)
Characteristics at the time of initiation of refeeding
Anorexia Nervosa			1.0
Yes	1 (25%)	3 (75%)
No	962 (25%)	2,888 (75%)
Malabsorption			0.27
Yes	4 (40%)	6 (60%)
No	959 (25%)	2,885 (75%)
Mental Health Disorder			0.07
Yes	364 (23%)	1,188 (77%)
No	599 (26%)	1,703 (74%)
Alcohol/Substance Use			0.01
Yes	123 (20%)	477 (80%)
No	840 (25%)	2,414 (75%)
Malignancy			0.00
Yes	314 (29%)	774 (71%)
No	649 (23%)	2,117 (77%)
Renal Failure			0.00
Yes	89 (39%)	134 (61%)
No	874 (24%)	2,757 (76%)
Albumin			0.03
< 3.4 g/dL (Low)	688 (26%)	1.947 (74%)
3.4 to 5.4 g/dL	205 (23%)	667 (77%)
>5.4 dL	70 (20%)	277 (80%)
WBC			0.04
<4,500/ mL	57 (30%)	134 (71%)
4,500- 11,000/mL	345 (23%)	1,152 (77%)
>11000/mL	561 (26%)	1,605 (74%)
Platelets			0.00
<150k/mL	458 (28%)	1,153 (72%)
150k- 400k/mL	438 (23%)	1,468 (77%)
>400k/mL	67 (20%)	270 (80%)
Creatinine			0.00
<1.2 g/dL	435 (22%)	1,559 (78%)
>1.2 g/dL	528 (28%)	1,332 (72%)
Bilirubin			0.92
<1.2 g/dL	328 (25%)	990 (75%)
>1.2 g/dL	635 (25%)	1,901 (75%)

Multivariable model

A multivariable logistic regression model was constructed to identify the factors that were associated with severe RFS. The final model included alcohol/sub stance use, malignancy, renal failure, and abnormalities in pre-feeding albumin level, white blood cell count, platelets, and creatinine (Table 3). Of these, renal failure was the most important predictor of severe RFS (aOR 1.75, 95% CI 1.31 – 2.34).

Table 3.

Multivariable logistic regression model for severe refeeding syndrome.^*

Characteristics	Unadjusted Odds Ratio (95% CI)
	Full model	Final model
Age, years
<50	Reference	-
≥50 and <75	1.07 (0.88 - 1.31)	-
≥75	.96 (.77 - 1.19)	-
Sex
Male	Reference	-
Female	1.00 (.87 - 1.16)	-
Race/Ethnicity
White	.97 (.82 - 1.15)	-
Black	.85(.66 - 1.08)	-
Hispanic	.78 (.62 - .97)	-
Other/Unknown	Reference	-
BMI, kg/m2
<18.5 (Underweight)	1.04 (.86 - 1.25)	-
≥18.5 and <30 (Normal)	Reference	-
≥30 (Overweight)	.93 (.78 - 1.12)	-
Unknown		-
Anorexia Nervosa	1.00 (.10 - 9.63)	-
Malabsorption Syndromes	2.01 (.565 - .7.12)	-
Mental Health Disorder	.87 (.75 - 1.01)	-
Alcohol/Substance Use	.741 (.599 - .917)	.76 (.61 - .94)
Malignancy	1.32 (1.13 - 1.55)	1.37 (1.16 - 1.61)
Renal Failure	2.10 (1.60 - 2.78)	1.75 (1.31 - 2.34)
Bariatric Surgery	.87 (.41 - 1.84)	-
Albumin
<3.4 g/dL	.87 (.80 - 1.17)	.98 (.81 - 1.19)
3.4 to 5.4 g/dL	Reference	Reference
>5.4 dL	.72 (.54 - .94)	.79 (.56 - 1.11)
WBC
<4,500/ mL	,70 (.51 - .98)	1.21 (.86 - 1.70)
4,500- 11,000/mL	Reference	Reference
>11,000/mL	.822 (.59 - 1.14)	1.19 (1.01 - 1.39)
Platelets
<150k/mL	.75 (.65 - .87)	1.22 (1.04 - 1.44)
150k- 400k/mL	Reference	Reference
≥400k/mL	.63 (.47 - .83)	.81 (.59 - 1.12)
Creatinine
≤1.2 g/dL	Reference	Reference
>1.2 g/dL	1.42 (1.23 - 1.64)	1.33 (1.13 - 1.57)
Bilirubin
≤1.2 g/dL	Reference	-
>1.2 g/dL	.99 (.85 - 1.16)	-

^*Refeeding syndrome was operationalized as a 50% decline in serum phosphorus, magnesium, or potassium relative to prefeeding.

Sensitivity analyses

To test whether pre-feeding electrolyte levels were associated with developing severe RFS, patients were classified into tertiles based on their baseline values for these electrolytes (i.e., low, medium, high) and these variables were tested one by one in the final model. Baseline phosphorus, baseline magnesium and baseline potassium were each independently associated with severe RFS (Table 4). Of these, baseline phosphorus was the most predictive of developing severe RFS (aOR 6.09, 95% CI 4.95 – 7.49 for the highest tertile of pre-feeding phosphorus compared to the lowest).

Table 4.

Relationship between baseline electrolyte values and severe refeeding syndrome. ^*

Baseline Electrolytes	Adjusted Odds Ratio** (95% Confidence Interval)
Phosphorus
Lowest tertile (1.8 to 3.7)	Reference
Middle tertile (3.8 to 4.2)	2.92 (2.34 - 3.64)
Highest tertile (4.2 to 9.3)	6.09 (4.95 - 7.49)
Magnesium
Lowest tertile (.9 to 1.9)	Reference
Middle tertile (2 to 2.2)	1.05 (0.88 - 1.26)
Highest tertile (2.3 to 5.6)	1.47 (1.22 - 1.78)
Potassium
Lowest tertile (.7 to 2.9)	Reference
Middle tertile (3 to 3.8)	1.15 (0.96 - 1.39)
Highest tertile (3.9 to 13.6)	1.65 (1.37 - 1.99)

^*Refeeding syndrome was operationalized as a 50% decline in serum phosphorus, magnesium, or potassium relative to pre-feeding.

^**Adjusted for alcohol/substance abuse, malignancy, renal failure, albumin, white blood cells, platelets, and serum creatinine.

Last, we examined absolute (rather than relative) decline in the pre-feeding serum phosphorus level. After re-classifying RFS as an absolute decline from a normal pre-feeding serum phosphorus level to an abnormal (< 2.5 mg/dL) post-feeding serum phosphorus, the 30-day mortality rate was 8% versus 10% in those with and without RFS respectively (chi-squared p<0.01, similar in direction to the results obtained using the definition of RFS based on relative decline with a higher mortality observed among those without RFS compared to those with RFS).

DISCUSSION

In this retrospective cohort study, we found that the 2020 ASPEN guideline cut-off of a 10% decline in baseline electrolyte values was not associated with adverse clinical consequences. In fact, when RFS was operationalized as a 10% decline in baseline pre-feeding electrolytes, a higher mortality was observed among those without mild RFS compared to those with RFS, although this difference may be minor and of little clinical importance. When a 25% or 50% cut-off for RFS was used (roughly corresponding to mild and severe RFS in the ASPEN guideline), there were no mortality differences comparing those with and without RFS. When the 10% cutoff for RFS was used, a very high number (90%) of patients were classified as having mild RFS, suggesting that this criteria may be too inclusive. At all cut-offs for RFS (i.e., mild, moderate, or severe), those with RFS had fewer hospital-free days compared to those without RFS. These differences were modest—median differences of 4 to 5 hospital-free days among patients who averaged hospital stays of 20 to 30 days—yet they may nonetheless reflect substantial differences in resources given the very high cost of ICU care. This study focused on mortality as a hard endpoint to appraise RFS, and future studies may wish to look at additional outcomes.

The strongest predictor of severe RFS in this study was the pre-feeding serum phosphorus level. Those with the highest pre-feeding phosphorus levels were over 6-fold more likely to meet criteria for severe RFS. The best interpretation of this seemingly counterintuitive finding may be that it is harder to have a reduction in phosphorus levels if the levels are already very low to begin with. Hospital lab draws can have considerable test-retest variability which may have more to do with specimen transport (i.e., how long a blood sample waits before processing) than on patient characteristics and patients who initiate feeding with higher phosphorus levels have more opportunity for a decline compared to those who initiate feeding at lower phosphorus levels. The strangely protective effect of alcohol and substance abuse can be interpreted similarly. Depletion of electrolytes is frequently found in alcohol/substance abuse patients. Patients with alcohol or substance abuse driving their ICU stay may initiate feeding at lower pretreatment electrolyte levels compared to patients who do not have these problems, making it harder to drop and therefore harder to meet the ASPEN criteria for RFS since these criteria hinge on relative decline in phosphorus and other electrolytes. Overall, the results of this study suggest that relative declines in the electrolyte levels measured just prior to feeding may perform suboptimally as predictors of outcomes in patients developing RFS.

Friedli et al. have previously shown that RFS can be associated with increased mortality among malnourished hospitalized patients. Unlike our study, which measured short-term mortality (within 30 days), Friedli measured mortality up to 180 days. Additionally, the authors defined RFS based on absolute rather than relative decline in electrolyte levels (post-feeding decline to phosphate <0.31 mmol/1 regardless of pre-feeding phosphate level). Like this study, Friedli found no association between RFS and short-term mortality and no association between mild RFS and increased mortality. However, there was a significant 7.9% absolute increase in 180-day mortality rate in those with confirmed RFS compared to those without RFS. Differences in the populations studied, and in the criteria for RFS used, may explain the differences in study findings.⁽⁷⁾

Other prior studies of RFS have described it as “refeeding hypophosphatemia” and operationalized RFS based on change in serum phosphorus only. In a 1996 study, 62 patients were refed after a 48-hour starvation period.⁽⁸⁾ Twenty-one out of 62 of these patients experienced a drop in phosphorus by more than 0.16 mmol/L to below 0.65 mmol/L. In this study, the length of hospital stay was significantly longer for those who experienced hypophosphatemia than for those who did not. A 2017 systematic review confirmed heterogeneous definitions for RFS.⁽³⁾ Some studies focused on serum phosphorus,^(9-11) others on electrolyte shifts more generally (mirroring the ASPEN guideline),⁽¹²⁾ and others included clinical manifestations to establish a diagnosis.^(13-15) In this systematic review, a novel algorithm was devised to attempt to better define RFS. Like the 1996 study, hypophosphatemia was the main criterion for diagnosis, but the researchers determined that a 30% decrease from baseline or a drop below 0.6 mmol/L (within 72 hours) was a more appropriate measure. Although phosphorus was used as the main indicator of RFS, a shift in potassium and magnesium below normal range was also considered to be diagnostic of RFS. The authors suggested a two-tiered system for classifying RFS: first the electrolyte-based criteria for RFS would be met, then patients would be further classified into subgroups based on their clinical manifestations. Those who experienced clinical symptoms such as tachycardia, tachypnea and edema would be classified as “manifest RFS” and those who did not would be classified as “imminent RFS.” Another recent systematic review described a consensus algorithm for the management of RFS based on the use of clinical screening criteria to identify patients at risk of RFS and risk-stratify them to an appropriate intervention and treatment plan. Future studies may wish to explore the proposed consensus criteria, as well as novel risk factors, for RFS.⁽¹⁶⁾

There are strengths to our study. It contributes to the process of refining the definition of RFS. This is likely to be an iterative process that involves questions that cannot definitively be answered by any single study, and we believe that our results move the conversation forward in a meaningful way. The study included a large number of patients, clinical patient characteristics, and laboratory measures novel to the assessment of RFS (e.g., albumin, bilirubin). It examined mortality, a hard outcome with low risk for misclassification, and also hospital-free days, an established measure of healthcare utilization. Limitations include that it was single-center and retrospective in design. Importantly, we did not have access to the feeding protocol used including the initial rate and type of feed and we lacked clinical information related to prophylactic treatments intended to prevent RFS such as electrolyte repletion and vitamin supplementation. These potential risk factors for RFS should be considered in future studies.

In summary, in this study we found that mild RFS, as operationalized using the ASPEN guideline definition as a 10% decline in pre-feeding electrolytes, was not associated with increased mortality. As we raised the cutoffs for RFS from 10% to 25% to a 50% decline in pre-feeding electrolytes, there was still no increase in mortality associated with RFS and the incidence of RFS dropped from 90% (at the 10% cut-off) to 25% (at the 50% cut-off). Pre-feeding serum phosphorus level was the variable most predictive of RFS. We did not compare the performance of the 2020 ASPEN criteria for RFS against alternative classification schemes for predicting outcomes among patients based on RFS but our results suggest that the 2020 ASPEN criteria should be refined. How to best do so is a question for future studies.

Statement of Clinical Relevancy:

Refeeding syndrome has been conceptualized as a set of clinical and electrolyte changes that occur in response to reintroduction of calories after a period of consistent calorie reduction or starvation. Until recently, refeeding syndrome has lacked a standard criterion for diagnosis. This study contributes to the important process of refining the definition of refeeding syndrome. This is likely to be an iterative process that involves questions that cannot definitively be answered by any single study, and we believe that our results move the conversation forward towards improved definitions of RFS in a meaningful way.