The Diagnostic Criteria of Refeeding Syndrome in Critically Ill Patients from four Chinese Hospitals: A Multicenter Prospective Study

Background: Refeeding syndrome (RFS) is a group of metabolic disorders associated with refeeding after starvation. However, the diagnostic criteria of RFS are highly heterogeneous. This study aimed to identify the best diagnostic criteria of RFS in critically ill patients. Methods: A multicenter, parallel, prospective trial enrolled patients ( ≥ 18 years) with mechanical ventilation for more than 3 days. RFS, dened as new-onset hypophosphatemia (<0.87mmol/L) within 72h after feeding and a decreased concentration of serum phosphate of more than 30%, from four hospital ICU of Zhejiang provinces in China. The primary endpoint was the 28-day mortality. MH the KL, XS, XZ, collect the test data; and analyzed the data and wrote the paper. MH, KL (cid:0) WH had primary responsibility for ﬁ nal content. All approved ﬁ nal


Background
Refeeding syndrome (RFS) is a group of metabolic disorders associated with refeeding [1]. Serum hypophosphatemia is a typical clinical sign of RFS, and other metabolic disturbances, such as hypokalemia, thiamine de ciency, and uid overload. RFS is not rare. It is common in short-term fasting patients, especially critically ill patients [2].Consequently, patients who are at high nutritional risk or severely malnourished should be advanced toward improvement, while monitoring for RFS [3].
Clinical standards consensus relating to refeeding hypophosphatemia is still lacking. Besides, RFS mortality and morbidity during critical illnesses have not been well-studied because of variable de nitions, clinicians' unfamiliarity with RFS, and the complexities between acute illnesses and RFS.
Moreover, some studies included both clinical and electrolyte abnormalities in the diagnosis of RFS, while in other studies, electrolyte disturbance was the only diagnostic criterion [4,5]. This study aimed to discuss the diagnostic criteria of RFS.

Study design
We performed a multicenter, single-blind prospective clinical trial in 4 hospital ICU from Zhejiang provinces in China. Ethical approval was granted by the Institutional Review Board (NO: I2019001223). The study protocol was registered at http://www.clinicaltrials.gov (NCT04005300).
Adult patients, aged 18-year-old and over, requiring mechanical ventilation, from 4 ICU were included in the study from May 1, 2019, to April 30, 2020. All the patients were admitted for nutritional treatment for more than 3 days and their serum phosphate level was monitored regularly. Patients were excluded from the study if they had parathyroidectomy or received renal replacement therapy or phosphate binders. Readmitted patients were also excluded. The patient's nutritional characteristics were recorded, including age, gender, body mass index (BMI), nutrition risk score 2002(NRS2002), acute physiology and chronic health evaluation II (APACHEII)score, baseline blood tests, and the time before the commencement of nutritional treatment in ICU.
Based on the different serum phosphate measurements[6], three diagnostic criteria of RFS were used in this study. Criterion 1 (RFS1):serum phosphate levels between 0.65 and 0.87 mmol/L and the rate of decline was greater than 30% within 72 h after starting nutritional support).Criterion 2 (RFS2):serum phosphate levels between 0.32 and 0.65 mmol/L and the rate of change was greater than 0.16 mmol/L within 72 h after starting nutritional support).Criterion 3 (RFS3):serum phosphate levels below 0.32 mmol/L within 72 h after feeding) (Fig. 1).
Based on the result of this study, we further combined RFS2 and RFS3 and referred to as the modi ed RFS group. Moreover, RFS1 and non-RFS were combined into a modi ed non-RFS group.

Clinical Analyses
Daily caloric intake of the patients including enteral or parenteral nutrition, propofol, and sodium citrate, was recorded during the rst week after enrollment. All patients who were nally included in the study were analyzed.

Outcome
The primary endpoint was the 28-day mortality. The secondary endpoints included the 90-day mortality, nosocomial infections, AKI, LOS of ICU and/or hospital, and the duration of mechanical ventilation (MV) in days.

Data Collection And Protection
All data were obtained from the four hospital databases, veri ed by two different researchers manually, and stored in a hospital computer.

Statistical Analysis
Quantitative data were expressed as mean ± standard deviation (SD), which were dichotomized or categorized if necessary. Categorical and binary variables were reported as frequency or percentage. The equality of variances was assessed by Levene's test. The independent T-test, Mann Whitney U-test, or Kruskal Wallis test was used to analyze the quantitative data. The chi-square test or Kruskal Wallis test was to analyze the categorical variables and frequencies. Collinearity among confounding variables at different times was investigated using correlation analysis. For the primary and secondary endpoints, the 28-and 90-day, nosocomial infections, and AKI were analyzed using the chi-square test or Kruskal Wallis test, and the LOS of ICU/hospital and MV were analyzed using the independent T-test or Mann Whitney Utest. The variables considered were age, gender, body mass index (BMI), nutrition risk score 2002(NRS 2002), PNI score, acute physiology and chronic health evaluation-II (APACHE-II) score, baseline blood tests, nutrition intake (including caloric and protein intake) and time before the commencement of nutritional treatment in ICU. All analyses were performed using SPSS (version 17.0, IBM Corp., New York, USA). All tests for statistical signi cance were determined using an alpha level of 0.05.

Results
Between May 1,2019, and April 30, 2020, a total of 312 critically ill patients from 4 participating hospital ICU of the Zhejiang province in China, who accepted nutritional treatment for more than 3 days, were included. Of these, 302 patients were nally enrolled ( Fig. 1). As demonstrated in Fig. 1, the following patients were excluded:2 patient who were readmitted, 5 patient who received the CRRT therapy, 1 patient who withdrew consent, 1 patient who recently admitted to another hospital and 1 patient who discontinued treatment. All patients were divided into RFS and Non-RFS groups. According to the serum phosphate level, those in the RFS group were further divided into three subgroups: Group 1 (n = 18), Group 2 (n = 60), and Group 3 (n = 30) (Fig. 1). Table 1, the enrolled patients included medical and surgical patients, and the latter included those who underwent elective surgery or emergency surgery. Central nervous system diseases and infectious diseases ranked as the top 2 primary diseases indicated for ICU admissions. Hypertension, type 2 diabetes, and cardiovascular disease were the top three chronic complications. There were no statistical differences in the distribution of these among the four subgroups.

As shown in
Baseline characteristics are compared in Table 2. The mean age of the participants varied from 59 to 63 years of age. Women were accounted for 16.7-40%. At enrolment, the mean APACHE II score varied from 15.36 to 19.13 and this score was signi cantly higher in Group 2 (p = 0.009) and Group 3 (p = 0.01) than in the Non-RFS group.
Regarding the nutritional parameters, the mean time before starting EN treatment was between 33-43 hours and the percentage of those starting nutrition within 48 hours was 70%-83%. Except for the baseline NRS2002 and BMI index, no signi cant differences were found in the baseline data ( Fig. 3 and Table 2). The NRS2002 score in Group 2(p < 0.001) and Group 3(p = 0.001) was higher than the Non-RFS group. Moreover, the BMI index in Group 3 was lower than the Non-RFS group (p = 0.001) ( Table 2).
The electrolyte and glucose levels are shown in Fig. 2. There was a correlation at different times for electrolyte and serum phosphate levels. The levels were lower in Group 2 and Group 3 compared with the Non-RFS group. The serum potassium, magnesium, and calcium levels did not reach statistical signi cance in the different groups. The serum glucose level,3 days after starting nutrition, was signi cantly different between the groups. These levels were signi cantly higher in Group 2 and Group 3 compared with the Non-RFS group (Fig. 2).
In terms of organ injury, Pro-BNP and total blood bilirubin were signi cantly increased in the RFS group after 3 days of nutrition treatment. Of those in the RFS group, Pro-BNP was highest in Group 3. The Pro-BNP level of the RFS group was higher than that in the non-RFS group. Moreover, the total bilirubin level was signi cantly higher in Group 2 than in Group 1 ( Figure S1). There were no signi cant differences in other organ changes between the different groups( Figure S1, Tables 2 and 3). As depicted in Table 3, the rate of nosocomial infection in the RFS groups (Group 2 and 3) was higher than the Non-RFS group but the difference was not statistically signi cant.
The median time of ICU length of stay, duration of mechanical ventilation, and the 90-day mortality had no statistical signi cance ( Table 3). The length of hospital stay in Group 2 was signi cantly shorter than that in the non-RFS group. Furthermore, the 28-day mortality increased. This may be associated with the short hospital LOS, and the difference was statistically signi cant.
According to the preliminary results, the clinical prognosis of RFS2 was worse than that of the non-RFS group. The researchers had thought that RFS2 criteria may potentially be the best diagnostic criterion. To assess this, the enrolled participants were further divided into the modi ed RFS group and modi ed non-RFS group. We found that the nutrition risk indices such as NRS2002 score and BMI index were higher in the modi ed RFS group than the modi ed non-RFS group (p < 0.001 and p = 0.04, Table S1). The APACHE II score was also higher in the modi ed RFS group compared with the modi ed non-RFS group. However, the percent of high-risk RFS patients were not signi cantly different between the two modi ed groups (Table S1). Table 4, the median time of ICU/hospital LOS, duration of mechanical ventilation, and the percent of AKI were not statistically signi cant (p > 0.05). The frequency of nosocomial infection was higher in the modi ed RFS group versus the modi ed non-RFS group, p = 0.04. Furthermore, the 28-and 90-day mortality were obviously higher in the modi ed RFS group, and the difference was statistically signi cant.

Discussion
The major nding of this study is that the best criterion of RFS is de ned by a serum phosphate level of 0.65mmol/L and a rate of change greater than 0.16mmol/L within 72h after starting nutritional support. Patients with RFS were associated with higher disease severity, higher risk of nosocomial infections, and increased mortality at both 28 days and 90 days. The NICE standard has limitations in identifying highrisk RFS in critically ill patients by considering only a reduction in serum phosphate level to 0.32mmol/L. Currently, the de nition of RFS is highly heterogeneous and based on different electrolyte disturbances and/or clinical parameters. Hypophosphatemia after feeding is the hallmark and the RFS diagnosis relies on a cut-off and/or a relative decrease of serum phosphate level from the baseline. The range of cut-offs in de ning hypophosphatemia varies from serum phosphate < 1 mmol/L to < 0.32 mmol/L and are duction rate of > 30% from baseline, to > 0.16mmol/L. Of note, in the study by Rio, the diagnostic criterion of RFS was de ned as a serum phosphate level lower than 0.32mmol/L [5], which concurred with two other studies [7,8]. The diagnostic criterion of below 0.65mmol/L was rstly applied in Marek's study [9], which was also cited by another study [10], and later adapted by Marvin [11]. The above criteria have been commonly used. However, the prognosis demonstrated in those studies was different from our study, and as such, we divided our RFS group into three subgroups to further assess the different diagnostic criteria of RFS. The results of this study showed that by de ning RFS in terms of the RFS2 group, the BMI and NRS2002 scores in RFS patients were signi cantly different from that of the non-RFS patients. RFS patients had higher nutrition risk and this validated our hypothesis that patient at high nutritional risk was associated with RFS risk. Moreover, the 28-day mortality of the RFS patients signi cantly increased. These ndings suggest that clinicians must recognize the severely at-risk patients early and intervene early. A previous study by Doig [10], using the same diagnostic criteria as our modi ed RFS, found that low-calorie feeding effectively improved the 60-day survival after ICU and reduced the incidence of nosocomial infections. In contrast to the results of Doig's study, Olthof [12] found that there was no difference in clinical prognosis regardless of the presence of RFS or not and concluded that low caloric intake and APACHE II score were two independent risk factors. This further validates our conclusion and supports the credibility of using the RFS2 criterion for the diagnosis of RFS. We can also infer from these results that using stringent diagnostic criteria for RFS as suggested by Rio [5] may lead to delayed treatment and deteriorated outcomes. It is feasible to identify RFS by serum phosphorus standards in critically ill patients and this provides the theoretical basis for effective guidance of hypocaloric feeding.
The identi cation of patients at high-risk of RFS has mainly been based on the NICE standard. However, it was not a speci c evaluation index for critically ill patients. To date, there appears to be no relevant research study on the identi cation e ciency of critically ill patients.
In addition to validating the best diagnostic criteria, clinical studies at this stage also need to provide the theoretical basis for identifying patients at high-risk of RFS. In phase III of our research (NCT04005300), we veri ed the value of the NICE standard in identifying critically ill patients at high-risk of RFS. The results showed that by applying the NICE standard, the RFS3 group, but not the RFS2 group, was statistically different compared with the non-RFS group, con rming a certain predictive effect.
Furthermore, this may be associated with the NICE diagnostic criteria used for the RFS3 group and partly con rmed the NICE value standard of0.32mmol/L serum phosphate as per Rio's clinical research [5]. As mentioned, we found that the standard 0.32mmol/L serum phosphate level may lead to delays in the treatment of critically ill patients. Therefore, based on our ndings, the NICE standard was not effective in guiding the identi cation of high-risk RFS in critically ill patients. As such, the multicenter clinical trial may be discontinued at this stage.

Conclusion
The serum phosphate level of 0.65mmol/L as the diagnostic criterion for RFS has obvious clinical signi cance and may be the best diagnostic criterion at present. The NICE standard had limitations in identifying critically ill patients at high-risk of RFS. Future phase III research should focus on new criteria for identifying high-risk RFS patients.

Declarations
Availability of data and materials: The dataset used and analyzed during the current study is available from the corresponding author on reasonable request. Ethics approval and consent to participate According to the Declaration of Helsinki, this study was conducted. The clinical protocol passed ethical approval by the Ethics Committee of Second A liated Hospital, Zhejiang University School of Medicine (NO: I2019001223). All enrolled patients were informed consent.

Consent for publication Not applicable.
Competing interest: The authors declare that they have no competing interests. The authors declare that they have no competing interests.