Malnutrition is defined as changes in body composition due to inadequate nutrient intake or malabsorption, resulting in deterioration of physical and mental function and poor clinical outcomes(Reference Cederholm, Barazzoni and Austin1,Reference Cederholm, Bosaeus and Barazzoni2) . It is a predominant cause of morbidity and mortality in clinical patients(Reference Bian, Li and Wang3) and is common in patients in the intensive care unit (ICU), with an incidence rate as high as 30 %(Reference Wang, Wang and Jiang4,Reference Javid Mishamandani, Norouzy and Hashemian5) , imposing a significant burden on individuals, society and economy(Reference Ruiz, Buitrago and Rodriguez6,Reference Inciong, Chaudhary and Hsu7) . Premorbid nutritional state and adequacy of early enteral macronutrient delivery in critically ill patients are risk factors for morbidity and mortality(Reference Elke, Hartl and Kreymann8,Reference Hill, Elke and Weimann9) . Proper nutritional support can improve nutritional state, immune status, promote early recovery and improve the quality of life for ICU patients in the later stages(Reference Buckinx10,Reference Reber, Strahm and Bally11) . In this regard, the European Society for Clinical Nutrition and Metabolism(Reference Cederholm, Barazzoni and Austin1) and American Society for Parenteral and Enteral Nutrition(Reference Ayers, Adams and Boullata12) published relevant guidelines on ICU enteral nutrition (EN) and parenteral nutrition (PN). These guidelines provide detailed nutritional management strategies aimed at optimising nutritional support for critically ill patients, reducing complications and improving survival rates and long-term outcomes.
Critically ill patients often enter the ICU due to inadequate oxygen supply and tissue hypoxia caused by hypotension(Reference Thongprayoon, Cheungpasitporn and Harrison13). Vasopressors can increase vasoconstriction and systemic vascular resistance, improve cardiac output and tissue oxygen delivery and prevent myocardial and renal injury-related occurrence of hypotension and mortality(Reference Jentzer, Coons and Link14). However, vasoconstrictor drugs may exacerbate the occurrence of gastrointestinal dysfunction in critically ill patients. Approximately 46 % of critically ill patients are intolerant to EN due to gastrointestinal dysfunction, which is associated with infection, prolonged hospital stays and elevated mortality(Reference Arunachala Murthy, Chapple and Lange15). Although guidelines recommend early provision of EN to critically ill patients(Reference Cederholm, Barazzoni and Austin1,Reference McClave, Taylor and Martindale16) , clinical practitioners may be cautious in using EN instead of PN for patients receiving vasopressor drugs due to concerns about potential risks of gastrointestinal dysfunction(Reference Loudet, Marchena and Tumino17). Interestingly, in ICU patients receiving vasopressor therapy, instead of increasing patient mortality, EN is associated with reduced mortality(Reference Khalid, Doshi and DiGiovine18). Varying methods of nutritional delivery do not cause differences in patient mortality(Reference Dorken Gallastegi, Gebran and Gaitanidis19,Reference Reignier, Boisrame-Helms and Brisard20) . Unfortunately, there is no clear conclusion in clinical practice regarding the correlation between different routes of nutritional delivery and the survival of critically ill patients receiving vasopressor therapy.
This study used patient information from the Medical Information Mart for Intensive Care (MIMIC)-IV database to select the clinicopathological characteristics of critically ill patients receiving vasopressor support and compared the differences in 28-d survival between EN and PN support. Using the Cox proportional hazards model to identify prognostic factors for patient survival, this study developed a prognostic nomogram to establish a relative systematic evaluation system and accurately predict the 28-d survival rate of critically ill patients receiving vasopressors. This work hoped to provide data support for selection of appropriate nutritional methods for ICU patients.
Methods
Data source
The MIMIC database is based on the intensive care inpatient system at Beth Israel Deaconess Medical Center of the Massachusetts Institute of Technology. It is the largest open-source and free clinical database for intensive care and emergency departments. MIMIC-IV, the latest version, records detailed information of over 70 000 de-identified patients from Beth Israel Deaconess Medical Center between 2008 and 2019, including demographics, vital signs, comorbidities and laboratory tests (https://mimic.physionet.org/about/mimic/). Additionally, the database provides information on patient mortality both within and outside the hospital from hospital or social security databases.
Patient selection
This study selected clinical data of 299 712 subjects from the MIMIC-IV database between 2008 and 2019 to investigate the survival of critically ill patients receiving vasopressor support and EN and PN. All data in the MIMIC-IV have undergone deidentification processing and cannot identify specific patients. The Beth Israel Deaconess Medical Center Institutional Review Board evaluated the collection of patient data and the development of research resources, authorised data sharing and waived the requirement for informed consent. All data collection and processing procedures strictly adhered to applicable regulations to ensure that patient privacy was fully protected(Reference Johnson, Bulgarelli and Shen21).
To ensure integrity of sample information, the following criteria were used to identify eligible patients: (a) first admission to the ICU; (b) administration of vasopressors within 24 h of ICU admission, with a duration of more than 48 h (norepinephrine, epinephrine, phenylephrine, dopamine and vasopressin) and(Reference Dorken Gallastegi, Gebran and Gaitanidis19). (c) receipt of EN and/or PN within 3 d of ICU admission.
The following patients were excluded: (a) ICU stay <3 d or death within 3 d and (b) age <18 years at admission. A total of 1046 eligible respondents were included in the final sample and divided into two groups based on different nutritional pathways: EN (n 937) and PN/PN + EN (n 109). The detailed screening process of patients is presented in Fig. 1.
Flow chart of study cohort selection.
Variable collection
This study collected information on patient demographics, severity scores, comorbidities, vital signs, biochemical indicators and treatment indicators. The basic information included gender, age, race, marital status and survival status. Severity scores include sequential organ failure assessment score and Glasgow Coma Scale. Complications include myocardial infarction, congestive heart failure, vascular diseases, dementia, chronic lung disease, rheumatic diseases, liver disease, diabetes, paralysis, kidney disease, cancer and AIDS. Vital signs and biochemical indicators included BMI, heart rate, average systolic blood pressure, respiratory rate, blood glucose, anion gap, lactate, platelets, blood potassium, partial thromboplastin time, international normalised ratio (INR), prothrombin time (PT), blood sodium, blood urea nitrogen, white blood cell count, partial pressure of oxygen (pO2), partial pressure of carbon dioxide (pCO2), pH, mean corpuscular Hg (MCH), MCH concentration (MCHC), mean corpuscular volume (MCV), red cell distribution width (RDW) and creatinine(Reference Lu, Zhang and Hong22). Treatment indicators included EN/PN and mechanical ventilation.
Statistical analysis
In statistical analysis, categorical variables were represented by sample size and percentage (n (%)), while continuous variables were represented by mean and standard deviation (mean (s d)). Group comparisons were conducted using the Mann–Whitney U test. Categorical data were presented as percentages (%) and compared using χ 2 test. Variables with P ≤ 0·05 (two-sided) were considered to have statistically significant differences. Kaplan–Meier (K-M) curves were utilised for landmark analysis to compare survival status between groups at varying time points. Samples were randomly allocated to the training cohort and test cohort in a 7:3 ratio. Univariate Cox regression model was utilised to screen potential factors leading to adverse outcomes in training cohort. By calculating the variance inflation factor (VIF), the results were further analysed in a multivariate Cox regression model using bilateral stepwise regression based on the Akaike information criterion (AIC) to identify independent prognostic factors, which were presented in the form of a forest plot. Risk score of prognostic factors was presented in the manner of nomogram. The performance of nomogram was evaluated using the receiver operating characteristic (ROC) curve. The C index was measured to quantify the discriminative performance of nomogram. Based on the nomogram model, self-sampling was performed 500 times to draw a calibration curve to evaluate the calibration of nomogram. Decision curve analysis (DCA) curve was plotted to evaluate the predictive factors of an event as the probability threshold changes.
Except for using X-tile software for optimal cut-off value analysis of sample risk scores and implementing the K–M curve method, all other statistical analyses were performed using R (4·2·3) statistical software. R packages mice(Reference van Buuren and Groothuis-Oudshoorn23), jskm (https://cran.r-project.org/web/packages/jskm/vignettes/jskm.html), tableone(Reference Panos and Mavridis24), rms (https://cran.r-project.org/web/packages/rms/index.html), survival (https://cran.r-project.org/web/packages/survival/index.html), survminer (https://rpkgs.datanovia.com/survminer/index.html), ggDCA (https://cran-e.com/package/ggDCA), timeROC (https://cran.r-project.org/web/packages/timeROC/index.html) and regplot (https://cran.r-project.org/web/packages/regplot/index.html) were utilised in this work.
Results
Baseline characteristics
This study included 1046 critically ill patients who received vasopressor support from MIMIC database during 2008–2019, with 937 patients receiving EN and 109 patients receiving PN (EN + PN). Overall, the average age of patients in the total cohort was 64·04 (16·17) years. The majority of patients were white (59·7 %) and unmarried (63·2 %). The cases of males (53·8 %) were slightly higher than that of females (46·2 %).
Clinical characteristics of participants divided according to different nutritional delivery methods showed significant statistical differences (P < 0·05) in marital status, heart rate condition, anion gap level, congestive heart failure status, liver disease condition, paraplegia and cancer condition between groups. In both cohorts, the mean age of patients receiving EN was 64·28 years (16·17); for PN patients, it was 62 years (16·06). For EN and PN (PN + EN), the proportion of unmarried patients was relatively high, accounting for 64·7 % and 50·5 %, respectively; the majority of patients had heart rates between 60 and 100 times/min, accounting for 72 % and 52·3 %, respectively. Compared with EN patients, patients receiving PN had lower levels of serum anion gap (59·4 % v. 46·8 %, P = 0·015). The incidence of congestive heart failure (35·4 % v. 24·8 %, P = 0·035) and paraplegia (8·2 % v. 0·9 %, P = 0·011) was significantly higher in the EN group compared with the PN group, while the incidence of liver disease (18·9 % v. 28·4 %, P = 0·025) and cancer (13·8 % v. 26·6 %, P = 0·001) was lower in the EN group than the PN group (Table 1).
Characteristics of patients included in stratified enteral and parenteral nutrition
EN, enteral nutrition; PN, parenteral nutrition; MBP, mean blood pressure; GCS, Glasgow Coma Scale; SOFA, Sequential Organ Failure Assessment; RDW, red cell distribution width; PTT, partial thromboplastin time; INR, international normalized ratio; BUN, blood urea nitrogen; PT, prothrombin time; pO2, partial pressure of oxygen; pCO2, partial pressure of carbon dioxide; MCH, mean corpuscular hemoglobin; MCHC, mean corpuscular hemoglobin concentration. The bolded data in the table indicate that the results are statistically significant.
Survival differences among different nutritional methods
We analysed the impact of different nutritional delivery routes on patient prognosis using K–M curves (Fig. 2). Varying ways as EN and PN had no significant impact on overall survival of patients (P > 0·05) (Fig. 2(a)). Landmark analysis results using different time cut-off points presented that although using 7 days as the cutoff point, no significant difference was seen in survival status of samples receiving EN and PN/PN + EN support (P > 0·05) (Fig. 2(b)). But when using 14 d and 21 d as cut-off points, the PN/PN + EN group showed significantly worse survival status than the EN group (P < 0·05) (Fig. 2(c) and (d)).
K–M curves assessing survival probability before and after cut-off time points based on nutritional support groups. (a) K–M curves before the landmark time; (b–d) K–M curves after the landmark time set as 7-, 14- and 21-d, respectively.
Exploration of factors for prognosis of critically ill patients receiving vasopressor support
To ensure balanced distribution of clinical characteristics between groups, patients were randomly assigned to training and testing sets in a 7:3 ratio, with no significant differences observed in any variables between the two groups (All P > 0·05) (online Supplementary Table S1). We performed univariate Cox regression analysis to assess factors influencing survival prognosis in critically ill patients receiving vasopressor support, evaluating the association between patient characteristics and survival risk factors (online Supplementary Table S2). Subsequently, we detected multicollinearity among variables by calculating VIF and conducted bidirectional stepwise regression based on AIC in a multivariable Cox regression model, ultimately identifying independent prognostic factors (Fig. 3, online Supplementary Table S2). From the results, it is evident that, age (HR = 1·032, 95 % CI:1·022, 1·042, P < 0·001), SpO2 < 95 kPa (HR = 1·690, 95 % CI:1·222, 2·339, P = 0·002), RDW > 15·5 (HR = 1·493, 95 % CI:1·135, 1·964, P = 0·004), potassium concentration >5·3 K/μl (HR = 1·384, 95 % CI:1·046, 1·830, P = 0·023), MCH > 32 pg (HR = 1·442, 95 % CI:1·030, 2·020, P = 0·033), myocardial infarction (HR = 1·432, 95 % CI:1·071, 1·916, P = 0·016) and liver disease (HR = 1·523, 95 % CI:1·103, 2·101, P = 0·011) were risk factors for dismal prognosis in critically ill patients.
Forest plot of prognostic factors in the training set patients.
Survival prediction nomogram construction and validation
We further constructed Kaplan–Meier curves predicting the 7-d, 14-d and 28-d survival prognosis of SCC patients, as shown in Fig. 4. The C-index for training cohort and test cohort was 0·678 and 0·664, respectively, indicating that nomogram had a certain discriminative ability. ROC curve illustrated that AUC values for survival status at 7, 14 and 28 d in training cohort were all 0·70 (Fig. 5(a)). AUC values of test cohort were 0·61, 0·61 and 0·69 (Fig. 5(b)), indicating favourable predictive ability of nomogram in 28-d survival status. Meanwhile, calibration curves of training cohort and test cohort were near the diagonal line, indicating a high quality of prediction results by the nomogram (Fig. 6). In addition, all DCA curves in training and test cohorts indicated that nomogram had good decision-making ability (Fig. 7).
Nomogram of participant’s survival rate at 7, 14 and 28 d.
ROC curve. (a) Training cohort; (b) Test cohort. The variables entered in nomogram are the same. ROC, receiver operating characteristic.
Calibration curve of nomogram. (a–b) Calibration curves of 7-, 14-, 28-d mortality for participants in training cohort and test cohort, respectively.
DCA curve for nomogram. (a–b) 7-, 14-, 28-d mortality benefit of nomogram in the training cohort and test cohort, respectively. DCA, decision curve analysis.
Riskscore model
According to risk score of each sample by nomogram, K–M curve results revealed significant differences in survival rates among low-risk (riskscore≤655), medium-risk (655 < risk score ≤695) and high-risk (risk score >695) groups in the training cohort (Fig. 8(a)), test cohort (Fig. 8(b)) and total cohort (Fig. 8(c)) (P < 0·001). The low-risk group had the greatest survival benefit.
Kaplan–Meier curve of survival probability for participants in different risk groups. (a–c) Survival probability for participants stratified by risk scores. a, b and c for training cohort, test cohort and total cohort, respectively.
Discussion
Vasopressor assistance is normally necessary for critically sick patients because of the severity of their illness; yet, using vasopressors is typically linked to a higher death risk(Reference Shi, Sun and He25,Reference Chou, Yeh and Chen26) . Critically ill patients who receive vasopressor medicines can benefit from nutritional assistance in reducing their death rate(Reference McClave, Taylor and Martindale16). However, we are concerned about the potential risks associated with providing nutritional support to these patients through varying means. Therefore, we conducted a retrospective analysis of patients in MIMIC-VI to compare pathological characteristics and impact of varying nutritional routes on survival of critically ill patients receiving vasopressor support. We used Cox regression to investigate factors for prognosis and developed a survival prediction tool, nomogram, for critically ill patients receiving vasopressor support and nutritional support. This tool could serve as a reference for clinicians in selecting appropriate assessment tools and types of nutritional support.
Clinicians and researchers believe that the approach, timing and quantity of nutritional support affect outcomes of critically ill patients(Reference Bear, Wandrag and Merriweather27). The main international guidelines for critical care nutrition support recommend EN over PN, even in critically ill patients receiving vasopressor support(Reference Cederholm, Barazzoni and Austin1,Reference Ayers, Adams and Boullata12,Reference Khalid, Doshi and DiGiovine18,Reference Dorken Gallastegi, Gebran and Gaitanidis19) . Our findings indicated that EN and PN had no significant impact on overall survival of critically ill patients receiving vasopressor support, which is congruous with previous results(Reference Dorken Gallastegi, Gebran and Gaitanidis19,Reference Reignier, Boisrame-Helms and Brisard20) . However, the data also presented that patients receiving EN had better survival rates than those receiving PN at the time points of day 14 and day 21. To fulfill the increased metabolic needs of intestinal cells, nutrition intake necessitates a physiological rise in intestinal blood flow(Reference Paulus, Wagner and Buehler28). Due to haemodynamic instability, most patients require vasopressors or positive inotropic agents to maintain adequate blood pressure and cardiac output(Reference Flordelis Lasierra, Perez-Vela and Montejo Gonzalez29). However, the use of vasopressors can lead to impaired intestinal blood flow, decreased motility and increased sphincter tone, increasing early (within 7 d(Reference Guo, Cheng and Li30)) EN intolerance in critically ill patients(Reference Yang, Wu and Yu31–Reference Sabino, Fuller and May33). But with time, patients’ haemodynamics start to steadily improve and stabilise, and the need for vasopressor medications will progressively lessen(Reference Hammond, McCain and Painter34,Reference Sacha, Lam and Duggal35) . This causes an increase in EN absorption and the revelation of benefits of EN in raising the survival rate of patients in critical condition. This may explain why patients who received EN outperformed those receiving PN at the 14- and 21-d time periods but did not exhibit a survival benefit in the early stages.
Age was a significant predictor of death in critically ill patients getting vasopressor therapy, as our study shown, in line with prior research findings(Reference Nour, Hegazy and Mosbah36). Age is a factor in the death rate related with critical illness(Reference Seethala, Blackney and Hou37), and older age is often linked to a worse prognosis(Reference Aldawood, Alsultan and Haddad38). In order to predict the death of critically ill patients receiving vasopressor therapy, INR was also included in the final model analysis. INR is defined as the PT of a patient divided by the mean normal PT(Reference Koenig, Pittaluga and Bogart39). In critically ill ICU patients, it is a well-accepted benchmark for predicting clinical outcomes of coagulation dysfunction(Reference Long, Tong and Miao40–Reference Tang, Chen and Liang43). In a prior retrospective analysis, INR has been identified as a promising predictive factor for critically ill patients(Reference Zhang, Yang and Tan44).
Furthermore, our investigation revealed a correlation between a worse patient survival rate and the existence of cancer, myocardial infarction and liver illness. One top cause of death and morbidity worldwide is liver disease(Reference Cheemerla and Balakrishnan45). In critically sick patients, decompensation related to acute aggravation of chronic liver failure and critical illness resulting from exacerbation of liver disease are notable causes of death(46,Reference Sarin, Choudhury and Sharma47) . One of the primary reasons of the high incidence and mortality of cardiovascular illnesses is myocardial infarction, which is defined as the death of myocardial cells brought on by prolonged ischaemia(Reference Thygesen, Alpert and Jaffe48). In ICU and coronary care unit, myocardial infarction is a prevalent condition(Reference Carroll, Mount and Atkinson49). About 66 % of myocardial infarction hospitalised patients are admitted to the ICU on the first day of admission(Reference Ohbe, Matsui and Yasunaga50). Over the past several decades, significant breakthroughs in pharmacological treatment, catheter-based therapies and surgical reperfusion have improved the prognosis of critically sick patients with myocardial infarction(Reference Yang, Huang and Zhao51,Reference Saito, Oyama and Tsujita52) . Nonetheless, patients with significant area infarction or those who do not obtain revascularisation in a timely manner are still at risk for mechanical problems(Reference Damluji, van Diepen and Katz53). Even with adequate treatment, the death rate remains high due to complications, which also greatly increase the incidence, mortality rate and hospital resource consumption(Reference Ng and Yeghiazarians54,Reference Gong, Vaitenas and Malaisrie55) . The number of cancer patients hospitalised to ICU has grown globally over the past 20 years, despite notable advancements in cancer therapy and overall prognosis(Reference Kiehl, Beutel and Boll56). However, critically ill cancer patients still have significant death rates because of the complexity of factors such the necessity for mechanical ventilation, organ failure, physical fitness and the possibility of cancer recurrence or progression(Reference Ozpinar, Gurun Kaya and Oz57,Reference Bikmaz, Gokce and Hasimoglu58) .
In this study, indicators of red blood cells were important for prognosis of critically ill patients. Elevated MCH was associated with more risk factors for prognosis in critically ill patients. MCH is useful for diagnosing anaemia in haematological examinations(Reference Velasco-Rodriguez, Blas and Alonso-Dominguez59). Critically ill individuals frequently struggle with anemia(Reference Warner, Hanson and Frank60). Anemia occurs in around 45·0 % of critically ill people during the first week of ICU admission, and it is strongly correlated with the critically ill patient death rate(Reference Lin, Liao and Wong61). RDW in red blood cell indices reflects systemic inflammation and oxidative stress(Reference Horta-Baas and Romero-Figueroa62,Reference Besedina, Skverchinskaya and Shmakov63) . Infected individuals may experience direct harm to erythrocytes by phagocytosis or red blood cell apoptosis, as well as disruptions to iron homeostasis, reduction of bone marrow-induced red blood cell formation and downregulation of erythropoietin receptor expression. These processes have the potential to raise RDW(Reference Salvagno, Sanchis-Gomar and Picanza64), and in critically sick patients, a rise in RDW is linked to a poor prognosis(Reference Sun, Zhou and Wu65). RDW value has become a predictive factor for mortality in critically ill patients(Reference Meynaar, Knook and Coolen66–Reference Chu, Yuan and Meng68). Potassium levels(Reference Tongyoo, Viarasilpa and Permpikul69) and SpO2(Reference Sun, Huang and Yin70) have been determined as risk factors for prognosis of critically ill patients. Our data also confirmed this. SpO2 is frequently used to evaluate oxygen levels in critically sick patients with cardiopulmonary failure. This can offer an early warning for hypoxemia and assist prevent unanticipated hypoxic episodes(Reference Jubran71,Reference Pretto, Roebuck and Beckert72) . The human body’s nervous system, skeletal muscles, visceral organs and cardiovascular system all depend on potassium, an essential electrolyte(Reference Gritter, Rotmans and Hoorn73,Reference Weaver74) . In severe cases, elevated potassium levels may potentially result in cardiac arrest or respiratory failure in addition to arrhythmias and neuromuscular abnormalities(Reference Jindal, Suresh and Dhakal75–Reference Grodzinsky, Goyal and Gosch77). Thus, especially in critically sick patients, prompt treatment of electrolyte imbalances and a proactive oxygenation monitoring regimen are essential.
Our study offers a number of benefits. First, to our knowledge, this study is the first to combine various parameters such as SpO2, RDW, potassium, MCH, myocardial infarction, liver disease, cancer status, INR, age and nutritional delivery method to predict the prognostic ability of 28-day mortality in ICU patients. Furthermore, this study examined the effects of EN and PN methods on survival rate of patients receiving vasopressors for the first time, offering a foundation for nutritional choices in critically ill patients. Our study does, however, have certain shortcomings. First off, because this is a single-centre retrospective study, it might be challenging to extrapolate the study’s findings to other institutions. Our data were acquired retrospectively from the MIMIC-IV database. For our findings to be broadly applicable, cohorts from other nations must undergo external validation. Additionally, it is impossible to completely rule out potential biases due to incomplete data collection and inaccurate data elements in the MIMIC-IV database. Lastly, we failed to provide the cohort energy requirements as determined by indirect calorimetry due to a lack of technology.
Acknowledgements
This research received no specific grant from any funding agency in the public, commercial or not for profit sectors.
A. Q. S. conceived of the study and participated in its design and interpretation and helped to draft the manuscript. M. L. L. and Y. S. participated in the design and interpretation of the data and drafting/revising the manuscript. Y. X. S. and J. Y. N. performed the statistical analysis and revised the manuscript critically. All the authors read and approved the final manuscript.
The authors have no conflicts of interest to declare.
The data and materials in the current study are available from the corresponding author on reasonable request.
Not applicable.
Supplementary material
For supplementary material/s referred to in this article, please visit https://doi.org/10.1017/S0007114524001399
