The effects of higher versus lower protein delivery in critically ill patients: an updated systematic review and meta-analysis of randomized controlled trials with trial sequential analysis

We included RCTs of (1) adult (age ≥ 18) critically ill patients (mechanically ventilated or if uncertain, the control group mortality had to be greater than 5% to ensure including truly critically ill patients) that (2) compared protein doses with delivery via enteral (EN) formula, EN protein supplementation, parenteral nutrition (PN), or intravenous (IV) amino acids, (3) reported similar energy delivery between groups, and (4) reported clinical and/or patient-centred outcomes.

An updated systematic search in MEDLINE, EMBASE, and CENTRAL through OVID was conducted with relevant subject headings and keywords from our last search (1 April 2022) [6] to (29 May 2023) without language restrictions. Personal files and the reference list of previous SRMAs were reviewed. Additional file 1: Table S1 shows the search strategies. ClinicalTrials.gov was also searched for ongoing studies (Additional file 1: Table S2).

Search results were exported into Covidence (Veritas Health Innovation, Melbourne, Australia) to remove duplicates and screen for potential eligible studies using the title and abstract of the articles (ZYL). The potential studies were retrieved, and two authors evaluated the full text independently (ZYL, ED). Disagreements were discussed with two other authors (CCHL and CS).

Data items were collected independently by two authors (ZYL, ED) in a standardized data abstraction form and thereafter summarized into tables. Details of data handling are in Additional file 1: Supplementary Methods.

The quality of the included trials was evaluated independently by two authors (ZYL, ED) using the Canadian Critical Care Nutrition (CCN) Methodological Quality System and the Cochrane Risk of Bias version 2 (ROB2). [10]. The overall ROB2 assessment was categorized as low risk of bias, some concerns, or high risk of bias. The risk-of-bias traffic light and summary plots were generated by the Risk-of-bias VISualization (robvis) tool [11]. The use of the CCN Methodological Quality System allows us to compare critical care nutrition trials across time and topics. The scoring table is shown in Additional file 1: Table S3. Any disagreements were discussed with two other authors (CCHL and CS).

Overall mortality is the primary outcome; all other outcomes are secondary. These latter outcomes are: (i) nutritional outcomes, (ii) clinical outcomes, (iii) muscle outcomes, (iv) discharge to rehabilitation facilities, (v) quality of life (QOL) physical measurements, and (vi) biochemical outcomes (details of each outcome are in Additional file 1: Supplementary Methods). Outcomes with at least 2 studies were pooled and reported.

The following subgroup analyses were planned a priori: low vs other risk of bias, single vs multicentre trial, EN vs exclusive PN/intravenous amino acids, and AKI vs no/not known AKI. The subgroup analysis of AKI was performed in one study that enrolled exclusively AKI patients [12] and two studies that reported mortality outcomes in their subgroup of patients with AKI (Nephroprotect trial [12] and EFFORT protein trial [7]). For the Nephroprotect trial, we used the data from their secondary analysis that reported 90-day mortality outcome among patients with baseline kidney dysfunction (creatinine > 168 umol/L at the time of enrolment) and/or baseline risk of progression of AKI (creatinine increased over the previous 24 h by at least 20% to over 120 μmol/L) [13]. In both trials [7, 12], there were groups of patients with and without AKI. To ascertain the mortality count and total sample size for the no/not known AKI subgroup, the mortality count and total sample size of the AKI subgroup were subtracted from the overall mortality count and total sample size, respectively.

Post hoc, since we included studies combining higher protein and early physical rehabilitation, we added the subgroup analysis of studies with and without early physical rehabilitation. One study randomized patient to 3 groups (Group 1: usual care, Group 2: low protein + cycle ergometry, Group 3: high protein + cycle ergometry) [17], and we included groups 1 and 3 in our meta-analysis.

Dichotomous outcomes were presented as risk ratio (RR), while continuous outcomes were presented as mean difference (MD) or standardized mean difference (SMD). For AKI subgroup analysis on mortality outcome, we performed an additional analysis to present the effect measure as risk difference (RD) in order to obtain the number needed to harm (1/RD). The DerSimonian–Laird random-effect model was used to account for the different patients’ characteristics, dosing, duration, and starting time of the protein delivery. Heterogeneity was quantified by the I² measure. Publication bias was visualized by the funnel plot. Egger’s test was conducted for meta-analyses that included > 10 studies using STATA 16.1 (StataCorp LLC, Texas) [14]. All meta-analyses and tests for subgroup differences were conducted using RevMan 5.4 (Cochrane IMS, Oxford, UK). A two-sided p value of < 0.05 was considered statistically significant, and a p value of < 0.10 was considered a trend. [15]

To control for type-I and type-II errors, TSA was performed using the TSA software (0.9.5.10 Beta, The Copenhagen Trial Unit, Denmark) with pre-specified parameters detailed in Additional file 1: Supplementary Methods.

The Grading of Recommendations Assessment, Development, and Evaluation (GRADE) system was used to rate the certainty of evidence for outcomes analysed with TSA [16]. The quality of the evidence was rated as high, moderate, low, and very low by considering the risk of bias, inconsistency, indirectness, imprecision, and publication bias. The percentage of diversity-adjusted required information size (DARIS) achieved, and the TSA-adjusted 95% confidence interval for relative risk and mean difference were used to aid the assessment of imprecision in GRADE. GRADEpro was used to prepare the GRADE evidence profile table.

Our search identified an additional 853 articles (391 from MEDLINE, 350 from EMBASE, and 113 from CENTRAL). After removing duplicates and article screening and review, we included 23 RCTs (an additional 4 RCTs [7, 17‐19] from our previous SRMA). The detailed study selection flow is presented in Additional file 1: Fig. S1. The list of excluded studies and reasons for exclusion are presented in Additional file 1: Table S4. Our search on ClinicalTrials.gov and personal files identified 13 ongoing or unpublished related trials (Additional file 1: Table S2).

Twenty-three RCTs with 3,303 patients were included. The study characteristics are summarized in Table 1. Sample sizes ranged from 20 to 1,301. Patients’ baseline characteristics and the detailed nutritional data are summarized in Additional File 1: Tables S5 and S6.

The study population included mixed medical and surgical population (11 studies [7, 18, 20‐28]), patients with stroke or head injury (4 studies [29‐32]), only medical patients (1 study [33]), only surgical patients (1 study [34]), patients with non-oliguric acute renal failure (1 study [12]), patients with burn (1 study [19]), and unclear population (4 studies [17, 35‐37]). Outcomes of patients with AKI are available in 3 studies [7, 12, 20], of which 1 is reported in a separate publication [13].

Twenty studies primarily used enteral nutrition (EN), and three used exclusive parenteral nutrition (PN) [12, 20, 21] strategy to increase protein delivery. Of the 20 studies that used an EN strategy, supplemental PN was allowed in 10 studies. [7, 18, 20, 22‐25, 27, 28, 34]

Nineteen studies started the intervention within 3 days of ICU admission [12, 17‐28, 31‐33, 35‐37]. The remaining studies started the intervention within 96 h of mechanical ventilation [7], 5 days of acute stroke [30], 7–14 days after a head injury [29], and after 10 days in the ICU [34]. The duration of intervention ranged from 3 to 28 days.

Of the 23 included studies, 9 and 10 studies did not report the protein and energy delivered in g/kg BW/d or kcal/kg BW/d, respectively. The pooled mean protein delivery for the higher vs lower protein group was 1.49 ± 0.48 vs 0.92 ± 0.30 g/kg BW/d (14 studies, n = 2439), respectively, resulting in a daily MD of 0.49 g/kg BW/d (95% confidence interval [CI] 0.37–0.61, p < 0.00001; I² = 94%) more protein delivery in the higher protein group. In contrast, the pooled mean energy delivery for the higher vs lower protein group was 17.48 ± 6.85 vs 16.60 ± 6.63 kcal/kg BW/d (13 studies, n = 2258), with no difference in daily energy delivery between groups (MD 0.13 kcal/kg BW/d, 95% CI − 1.25 to 1.52, p = 0.85; I² = 91%) (Additional file 1: Fig S2).

Two studies combined high protein and early physical rehabilitation [17, 18], and one study combined high protein and neuromuscular electrical muscle stimulation (NMES) [32], which are collectively named as combined early physical rehabilitation intervention. The details of the intervention are summarized in Additional file 1: Table S7. The NMES intervention was delivered in two 30-min sessions per day for up to 14 days. [32] For cycle ergometry, one study started immediately after randomization and delivered the intervention in two 15-min sessions/day for up to 21 days. [18] Another study started cycle ergometry within 24 h of randomization and delivered the intervention for up to 28 days, either passive cycling for 20 min/day or two 10-min sessions/day if a patient was able to cycle actively. [17]

The median CCN methodological quality score of included studies was 8 (out of 14 [higher score indicates higher quality]). A total of 10 studies had a methodological quality score of > 8 [7, 17, 21‐24, 27, 29, 32, 36] (Additional file 1: Table S8). The ROB2 plots are presented in Additional file 1: Figure S3. In 21 studies that reported mortality outcomes, 4/21 (19%) studies were at low risk of bias, 14/21 (67%) had some concerns, and 3/21 (14.3%) were at high risk of bias. The biases mainly arose from the randomization process and selection of the reported results.

All outcomes are summarized in Additional file 1: Table S9 and Table S10.

No new studies were added to the meta-analysis on change in muscle mass, discharge location, and incidence of diarrhoea (Fig. 2a, 2c and 2e); therefore, findings are identical to our previous published meta-analysis [6]. Notably, higher protein delivery is associated with a muscle loss attenuation (MD − 3.44% per week, 95% CI − 4.99 to − 1.90, p < 0.0001, I² = 16%; 5 studies; Fig. 2a).

No differences in muscle strength (Fig. 2b) and self-reported quality of life physical function (Fig. 2d) measurements were detected. However, in the subgroup of studies with combined early physical rehabilitation intervention, a trend towards improvement in physical function measures (SMD 0.40, 95% CI − 0.04 to 0.84, p = 0.07, I² = 30%; 2 studies; Fig. 2d) was demonstrated, while no significant improvement was shown in studies without the combined intervention (SMD − 0.12, 95% CI − 0.28 to 0.05, p = 0.17; I² = 0%; 3 studies). The test for subgroup differences was significant (p = 0.03).

The biochemical outcomes between groups are summarized in Additional file 1: Table S10. Meta-analyses demonstrated that higher protein delivery significantly increased serum urea (MD 2.31 mmol/L, 95% CI 1.64–2.97, p < 0.00001, I² = 0%; 7 studies), urinary urea nitrogen (MD 5.55 g, 95% CI 0.87–10.23, p = 0.02, I² = 81%; 3 studies), and lymphocyte count (MD 257.43 cells per µL of blood, 95% CI 139.85–375.02, p < 0.0001, I² = 0%; 4 studies). Higher protein delivery showed a trend towards a significant increase in prealbumin level (MD 1.96 mg/dL, 95% CI 0.00–3.91, p = 0.05; I² = 23%; 4 studies) and nitrogen balance (MD 2.76 g, 95% CI − 0.38 to 5.90, p = 0.08; I² = 78%; 5 studies). No significant differences between groups were found for serum creatinine, blood glucose, insulin administration, albumin, haemoglobin, total white blood cells, C-reactive protein, interleukin-6, phosphate, and triglyceride level (Additional file 1: Fig. S6a–S6p).

No subgroup differences were detected between studies with low risk of bias and other risk of bias (Additional file 1: Fig. S7a–S7j) and studies that primarily used EN versus exclusive PN to increase protein delivery (data not shown). No subgroup differences were detected between single and multicentre studies (Additional file 1: Fig. S8a–8j).

Results of TSA are summarized in Table 2 and presented in Fig. 3 and Additional file 1: Figure S9, showing that the current systematic review did not achieve the required information sizes to detect the pre-specified effect sizes for overall mortality, infectious complications, ICU and hospital length of stay, change in muscle mass, handgrip strength, incidence of diarrhoea, and discharge to rehabilitation facilities, indicating that more trials are required for a definitive conclusion for these outcomes.

In patients with AKI, TSA confirmed the increase in mortality with high certainty. TSA revealed that further trials would be futile to detect a one-day difference in the duration of mechanical ventilation.

Higher protein delivery did not affect overall mortality in critically ill patients (low certainty of evidence). On the contrary, higher protein delivery increased mortality among patients with AKI (high certainty of evidence). The certainty of evidence of the effect of higher protein on other outcomes is low to very low (Table 3).

Our findings suggest that higher protein delivery may harm patients with AKI. Despite the heterogeneous definition of AKI in the three meta-analysed studies, the direction of the results is similar (I² = 0%), particularly from the two included multicentre RCTs [7, 13]. In this context, using isotope technique, Chapple et al. recently revealed that critically ill patients exhibited a markedly blunted muscle protein synthesis or anabolic resistance compared to healthy controls [38]. Notably, the incorporation of amino acids into the myofibrillar protein was 60% lower compared to a healthy control group. The reduced capacity to utilize protein during the acute phase of critical illness observed by Chapple et al., together with our findings of significantly higher serum and urinary urea as a result of higher protein provision, leads to a hypothesis that surplus protein may not be used for anabolism but is converted to urea for excretion. Higher urea levels may increase the metabolic burden of critically ill patients, particularly those with AKI, which may be one of the contributing factors to increased mortality in AKI patients, as demonstrated in our meta-analysis. Although a statistically significant mean increase of 2.31 mmol/L of serum urea or a mean increase of 5.55 g of urinary urea nitrogen may not be clinically significant in general, its clinical significance in critically ill patients with AKI remains unknown. Hence, the current findings have significant clinical implications, especially when considering the fact that current guidelines recommend higher protein delivery for critically ill patients with AKI, which should be carefully revised [39‐41]. In contrast, the finding of non-statistically trend towards lowered mortality of higher protein delivery in patients with no/not known AKI requires further investigations.

Recent observational studies with robust statistical adjustments have examined protein delivery to critically ill patients during their first 5–7 days in the ICU. They have found that providing higher levels of protein, as opposed to medium or standard levels, does not lead to improved clinical outcomes and may even be harmful. One study by Hartl et al. involving 16,489 patients showed that protein delivery of 0.8–1.2 g/kg BW/d after 5 days of ICU admission resulted in lower hospital mortality compared to exclusively low protein intake (< 0.8 g/kg BW/d for ≤ 11 days). However, there was no further improvement in mortality when compared to early high protein intake (> 1.2 g/kg from day 1) [42]. Similarly, Matejovic et al. studied 1,172 patients with ≥ 5 days ICU-LOS and found that moderate nutrition dose (10–20 kcal/kg for energy and 0.8–1.2 g/kg for protein) improved patient weaning and reduced 90-day mortality compared to exclusively low nutrition intake (< 10 kcal/kg BW +  < 0.8 g/kg BW/d). Yet, there was no additional benefit when comparing moderate to high nutrition dose (> 20 kcal/kg BW/day +  > 1.2 g/kg BW/d) [43]. Lastly, Lin et al. studied 2,191 patients with ≥ 7 ICU-LOS and found that both high (1.68 g/kg BW/d) and low (0.38 g/kg BW/d) protein intake, compared to medium protein intake (0.8 g/kg BW/d), were associated with increased 28-day mortality [44]. Overall, these findings align with the conclusion that higher protein intake (around 1.5 g/kg BW/d) during the first week of critical illness does not offer additional benefits in improving clinical outcomes for critically ill patients.

While no significant differences in clinical outcomes were observed, higher protein delivery may help attenuate muscle loss. Combined with early physical rehabilitation, it could potentially improve long-term self-reported quality of life physical function score. In this context, a recent systematic review among healthy and non-critically ill patients found that higher protein was associated with increased lean body mass; however, the rate of lean body mass gain plateaued beyond 1.3 g/kg BW/day without resistance training [45]. It is plausible that certain subgroups of critically ill patients, particularly those who receive early physical rehabilitation, may experience greater muscle loss attenuation, ultimately enhancing their physical function. Similar findings were evident in ICU patients with traumatic brain injury, where those with greater quadriceps muscle thickness reported better physical function. [46] Another study linked greater lean mass with improved gait speed and 6-min walk distance in survivors of acute respiratory distress syndrome [47]. However, these objective outcomes were not assessed in the studies included in our systematic review. Nevertheless, the observed muscle loss attenuation may be a type-1 error, as indicated by TSA, underscoring the need for more studies to validate this finding. Similarly, the improvement in self-reported physical function scores with combination therapy only trended towards significance and primarily originated from small studies. Ongoing trials with combinations of high protein and early physical rehabilitation (excluding patients with AKI and not on kidney replacement therapy), such as the NEXIS (NCT03021902; registered 16 Jan 2017) and EFFORT-X (NCT04261543; registered 7 Feb 2020), which assess physical function outcomes with objective measures such as the 6-min walk test and short physical performance battery test, will provide further insights into the impact of higher protein delivery on physical function outcomes in non-AKI patients.

The strength of our work lies in the comprehensive search and analysis and the predefined analysis plan for meta-analysis and TSA, all of which increase the transparency of information. In addition, excluding RCTs with different energy delivery between groups or pharmaconutrition interventions enabled us to focus solely on examining the effects of protein dosage. Furthermore, the use of TSA enabled us to detect the risk of type-1 or type-2 errors in our findings. The DARIS estimated from TSA will also inform the sample size needed for adequately powered future trials. Additionally, including extensive biochemical outcomes helped us elucidate the effects of higher protein delivery on metabolic parameters in critically ill patients.

Our work has several limitations. First, the included studies are heterogeneous in terms of the study population, dosage, timing, and routes of protein delivery. However, the included trials generally enrolled severely ill patients and primarily started intervention within 3 days of ICU admission. The subgroup analysis based on primarily EN vs exclusive PN/IV amino acids is consistent with the findings of the main analysis. The protein separation of approximately 0.49 g/kg BW/d with similar energy delivery between groups also ensures that the effect of protein was studied. Second, the three studies included in our analysis use varying definitions of AKI, which could limit the applicability of our findings in clinical practice. However, all the definitions identified AKI through an acute rise in serum creatinine levels. We recommend using the KDIGO definition of AKI [48] to guide protein delivery, as recent evidence showed that higher protein delivery is associated with increased mortality across all AKI stages, especially in patients who did not receive kidney replacement therapy [49]. Third, the number of studies with combinations of high protein and early physical rehabilitation intervention was limited, and the result is mainly attributed to one single-centre study with a high risk of bias [18]. Lastly, the certainty of evidence for most outcomes was assessed as low to very low due to the risk of bias and imprecision. Hence, more high-quality studies are warranted, especially studies with combined interventions (high protein and early physical rehabilitation).