Predicting prognosis in critically ill patients has always been a hot spot in critical areas. At present, some severity scoring systems have been established and widely used in critical practice, such as Acute Physiology and Chronic Assessment II (APACHE II), Simplified Acute Physiology Score II, and Sequential Organ Failure Assessment (1). These scoring systems use various vital signs, laboratory parameters, and imaging data. However, the scoring systems often lack accuracy and are overcomplicated due to many projects included (1, 2).

Bioimpedance analysis (BIA) is a non-invasive technology to measure the body’s electrical impedance at alternating current frequencies (3). Since the resistance and capacitive reactance characteristics are closely related to the human body composition, including muscle, fat content, and water content, it has been widely used in clinical analysis of body composition and capacity state assessment under stable conditions (4). However, BIA proved to be inaccurate in critically ill patients, leading to a significant overestimation of changes in total body water (TBW) (5). This is related to electrolyte transfer between and outside cells, and changes in fluid distribution, which commonly happen to ICU patients, can interfere with the BIA results (5, 6). Likewise, BIA results may be overestimated when extracellular water expansion occurs (i.e., heart failure, renal failure, or severe disease) (6, 7).

Interestingly, some components of BIA, such as phase angle (PA), a measure derived directly from resistance and reactance measurements, can be interpreted as an indicator of membrane integrity and water distribution between intracellular and extracellular spaces (8). A low PA reflects no fat mass loss and cellular dysfunction, while higher values (> 6 in normal subjects) reflect good cellular health or nutritional status. Theoretically, extracellular fluid composition, cell number, and membrane integrity are also closely related to disease severity, so it is also possible to use PA to predict disease severity and prognosis. PA has also been successfully used to indicate nutritional status and prognosis in patients with tumors, chronic kidney disease, and liver cirrhosis (9–12). However, several studies focus on PA in critically ill patients, and the conclusions are inconsistent (13–16). The different results might be related to the small sample size and the heterogeneity of the population among these studies. On the other hand, PA does not require parameters recall, body weights, and laboratory tests and has the advantages of simplicity, repeatability, and instantaneity (4, 16, 17). Thus, if PA can accurately reflect prognosis, it will nicely complement the deficiencies of other severity scoring systems.

Several studies have recently been published to investigate the association of PA with prognosis in ICU patients (16, 18–23), although some studies have small sample sizes. Therefore, with the help of the statistical power of meta-analysis, we aimed to conduct a systematic review and meta-analysis to explore the predictive value of PA in this patient population.

We performed the present meta-analysis according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement (24; Supplementary material 1).

The current study is the first systematic review and meta-analysis to investigate the predictive value of BIA-derived PA in the prognosis of critically ill patients. Our results showed the baseline PA varied in patients after ICU admission, ranging from approximately 3.7° to 5.9°. PA was an independent risk factor for all-cause mortality in the ICU setting with a nearly 1.5-fold increase. Further subgroup analyses and sensitivity analyses confirmed this finding. In addition, reduced PA is related to disease severity, more extended hospital LOS and longer duration of MV.

Our study has several advantages. The current meta-analysis provides strong evidence that fills a gap in previous guidelines (36). That is, clinicians can use PA to predict the prognosis in the ICU setting. Second, our findings are consistent with earlier findings in other patient populations, including advanced tumors, cirrhosis, renal failure, transplantation, and surgical patients (9–12). Therefore, our meta-analysis adds a new population of evidence. Third, most included studies focused on non-selected critically ill patients in the ICU (15–18, 21, 22, 28, 29, 34), making our findings more generalizable. Fourth, we thoroughly assessed mortality risk, including mortality between low PA and control and a linear relationship between PA and mortality. In addition, we included 20 studies of more than 3,700 patients with sufficient statistical power to perform subgroup analyses and sensitivity analyses based on different potential influencing factors. The results were consistent, further supporting the robustness of our main results.

Our results showed that ICU patients had lower PA measurements than healthy individuals, with the mean PA varied among the included studies (from approximately 3.7° to 5.9°) (13–22, 28–34). The variation was related to different patient characteristics, such as ethnicity, gender ratio, disease type, age, etc. For example, we found that the mean PA in the Asian population was 3.6° (3.0°–4.7°) (14, 16, 23, 30), lower than that of European and American patients of 5.1° (4.6°–5.9°) (13, 15, 17–22, 27–29, 31–34). Similar to the data in the healthy population, PA was significantly higher in male patients in the ICU, which was related to the higher muscle reserve in males than in females (Table 2). In addition, PA might be affected by some treatment or internal environmental changes in the body during the ICU stay (7, 8). Commonly seen was a substantial fluid transfer before ICU admission or within the first few hours of ICU admission may lead to changes in PA, which reflects inflammation-induced changes in membrane integrity and causes fluid to redistribute into the extracellular space (37). The effect of altering hydration on PA may explain why in Thibault et al.’s study (18), PA on day first but not on day five after admission could predict mortality. Thus, early PA measurement after admission may reduce the confounding of hydration changes. Finally, it should be noted that some included retrospective studies only included patients with PA measured, so that their reports may underestimate the PA incidences (13, 16, 22, 23, 28, 30, 33).

Phase angle (PA) reflects cell membrane integrity, permeability, and fat-free mass (38). Thus, lower PA can indicate severely compromised cell membrane integrity and increase cell membrane permeability due to acute disease (i.e., membrane dysfunction and fluid transfer) and the effects of underlying systemic conditions. Poorer cellular health, cellular dysfunction, and nutritional status worsen. As shown in some included studies (15–17, 19, 20, 29, 30, 33), PA is used in the assessment of nutritional status in critically ill patients and has been shown to be an accurate indicator of nutritional risk screening. In addition, PA measurement declines with age and sarcopenia, and low PA is associated with malnutrition and frailty (39, 40). In this respect, PA may reflect limited physiological reserves, which explains its association with long-term mortality (34). Therefore, reduced PA reflects acute changes and underlying poor health, muscle wasting, and fragility, which are poorly captured by other disease severity predictors commonly used in the ICU, such as the APACHE II score (14, 34).

However, our results require further discussion. First, we found a lack of a unified definition of PA cut-offs among the included studies, leading to an important source of heterogeneity in our results. Most authors adopted PA cut-off points reported in previous research to define reduced PA (15, 21, 27) or based on specific cut-off points [e.g., adopted the lowest quartile, the first quartile, or the median according to their cohorts (17, 30, 33)]. Although these studies were conducted in different cohorts, they provided consistent results, associating low PA values with lower survival rates. Therefore, these results suggest that a reduced PA cut-off point is reasonable for predicting critical illness outcomes. However, defining a unified PA cut-off point is not easy since each study contains different diseases, and data cannot be fully extrapolated to different study cohorts. On the other hand, there are differences between studies in the equipment used, the electrodes used, and the frequency of measurements (Supplementary material 3), which affect the determination of standardization threshold. Given the differences in disease, ethnicity, body size, and diet among ICU patients, the validity and accuracy of cut-offs across geographic, ethnic, and disease states still need further confirmation.

Second, most included studies assessed baseline PA at ICU admission rather than described overall PA exposure to characterize the effect of PA changes on ICU patients. Only two studies added to this gap (18, 22). In a retrospective ICU cohort study, Ellegard and colleagues found that among 26 patients who were reassessed for PA 5 days after admission, PA increased by 0.62° ± 1.24 in 17 survivors and decreased by 0.24° ± 0.82° in 9 non-survivors, which resulted in a between-group difference of 0.86° (P = 0.048) (22). The authors concluded that repeated PA measurement in ICU patients could help predict clinical outcomes. Thibault et al. also reassessed the PA in their study (18). In 540 patients with PA measurements on day 5, approximately 0.3° was higher in survivors than non-survivors (18). Therefore, assessing changes in PA over the clinical course of ICU patients may be a more effective predictor of clinical outcome assessment than a single clinical outcome. On the other hand, the previous finding of a higher mortality rate in patients without PA improvement after treatment suggests that residual PA reduction still has a predictive value (18). Thus, the predictive value of PA may be related to its treatment responsiveness, which could help assess the long-term risk of death and could be used to monitor targeted interventions aimed at improving the long-term prognosis of ICU patients.

Third, two included studies focused on the septic population and came to different conclusions (21, 28). Silva and colleagues found that PA was a good prognostic marker for patients without sepsis but not for the septic cohort (28). Meanwhile, they observed a significant negative correlation between PA and APACHE II scores only in patients without sepsis (r = −0.506; P < 0.001). In contrast, the results of Vermeulen et al. suggest that PA showed no differences in patients between patients (P = 0.179) with or without sepsis and was a useful prognostic indicator in both groups of patients (21). Of note, two studies had small sample sizes (< 50) and used sepsis-2 and sepsis-3 diagnostic criteria. Moreover, compared to Silva et al. (28), Vermeulen et al. included sicker patients (APACHE II: 22 vs. 10), less male ratio (26 vs. 63%), more from surgical settings (60 vs. 36%), and younger (55 vs. 65%) in their cohort (21). All of these may all contribute to the differences between the two studies.

Additionally, most included studies have focused on the differences in PA between men and women and among patients over 60 years of age or older, while few PA cut-off points based on gender and age have been suggested. Currently, standardized PA (SPA) normalized for age, gender, and BMI has been proposed, with the calculation = (measured PA − mean population reference PA)/standard deviation of the reference population. However, only two included studies described the associations between PA indicators obtained by BIA with mortality (15, 32). Jensen et al. reported that reduced SPA increased about three times the chance of having malnutrition (OR = 2.79, 95% CI 1.39–5.61) and two times the chance of prolonged hospital stay (OR = 2.27; 95% CI 1.18–4.34) (15). In the other study by Osuna-Padilla et al. (32), the authors found that SPA and PA were significant predictors of 60-day mortality (OR, 0.45; P = 0.001). SPA may be a better prognostic predictor and should deserve more clinical attention.

Our meta-analysis has several limitations. (1) The observational design of all included studies excluded any causal inference. Meanwhile, only patients who underwent PA measurements were recruited, prone to selection bias. (2) Most studies assessed only baseline PA levels at ICU admission, ignoring assessments of PA levels over time. (3) We included several small studies and most were single-center designs. (4) Most included studies focused on unselected critically ill patients, and the uneven distribution of different underlying diseases in these studies may also exert different prognostic values. (5) In subgroup and sensitivity analyses, we could not have considered all confounding factors that may play a role in the relationship between PA and ICU mortality, such as the timing of measurement, age, nutritional status, and the effect of artificial feeding. (6) Only a few studies have proposed the severity of PA abnormalities and their impact on prognosis, but the further investigation could not be done due to a lack of grading criteria. (7) Although we included 20 studies in our manuscript, the clinical application of PA was limited to research due to the lack of knowledge and available instrumentation. Moreover, the different instruments, frequencies, and electrodes used might potentially bias the mean values among the included studies finally, it remains unclear whether PA-guided therapy reduces mortality in ICU patients. Thus, further studies on this topic are needed.

The findings of the current meta-analysis suggest that PA may be an important prognostic factor of survival in this population and nicely complement the deficiencies of other severity scoring systems. However, it should be noted that the included studies used different cut-off values, which was the primary source of the existing heterogeneity. Therefore, further studies are needed to define the optimal cut-off value to define PA according to geography, race, and disease and to further confirm our findings.

The original contributions presented in this study are included in the article/Supplementary material, further inquiries can be directed to the corresponding author.

W-HZ contributed to the search of the scientific literature and drafted the manuscript. Y-HZ contributed to the conception, design, and data interpretation. YY helped to collect the data and performed statistical analyses. H-BH contributed to the conception, design, data interpretation, manuscript revision for critical intellectual content, and supervision of the study. All authors contributed to the article and approved the submitted version.