Clinical usefulness of C-reactive protein to albumin ratio in predicting 30-day mortality in critically ill patients: A retrospective analysis

This study aimed to examine the prognostic value of C-reactive protein (CRP)/albumin (ALB) ratio among patients who were admitted to the intensive care unit (ICU) in predicting 30-day mortality rate. This retrospective cohort study was conducted by examining the medical records of adult patients who were admitted to the ICU at Seoul National University Bundang Hospital between 1 January 2012 and 31 December 2016. Data from 6,972 individuals were included in the final analysis, and 547 of these individuals (7.1%) died within 30 days after their ICU admission. The multivariable Cox regression analysis revealed that an increase of 1 for the CRP/ALB ratio was associated with an 11% increase in the risk of 30-day mortality (hazard ratio: 1.11, 95% confidence interval: 1.09–1.14, P < 0.001). However, the area under curve of CRP/ALB ratio in receiver operating characteristic analysis was lower than that of Acute Physiologic Assessment and Chronic Health Evaluation (APACHE) II, Charlson comorbidity index, or serum albumin alone. Although an elevated CRP/ALB ratio on ICU admission was an independent risk factor for 30-day mortality rate, the predictive power of CRP/ALB ratio was lower than that of albumin alone, APACHE II, and Charlson comorbidity index.


Risks of 30-day mortality after ICU admission.
show the results of the univariable and multivariable Cox regression analyses for 30-day mortality after postoperative ICU admission. The multivariable Cox proportional hazard model revealed that an increase of 1 for the CRP/ALB ratio was associated with an  ROC analysis for 30-day mortality. Figure 2 shows the receiver operating characteristic (ROC) curve of each variable for 30-day mortality, and Table 4 Table 3. Multivariable Cox regression analysis using stepwise backward elimination method in relation to 30day mortality after ICU admission. C-reactive protein/albumin ratio is included in another multivariable cox regression model (model 2) to avoid multicollinearity with C-reactive protein and albumin (VIF > 11.0). ICU, intensive care units; IM, internal medicine; APACHE, acute physiology and chronic health evaluation; VIF, variance inflation factors.

Discussion
The present study revealed that an elevated CRP/ALB ratio at the ICU admission was independently associated with an increased risk of 30-day mortality. However, the predictive power of the CRP/ALB ratio was found to be significantly lower than that of the APACHE II or Charlson comorbidity index. Although the AUC of CRP/ALB ratio was significantly higher than that of CRP alone, AUC of albumin was higher than that of CRP/ALB ratio or CRP. As a result, this study reveals that using CRP/ALB ratio in predicting 30-day mortality after ICU admissions is not recommended instead of albumin alone, APACHE II, or Charlson comorbidity index. In other words, the clinical usefulness of the CRP/ALB ratio in predicting 30-day mortality is questionable.
To interpret results of this study, some points should be emphasized. First, we analysed the CRP/ALB ratio of a mixed ICU patient population while previous studies analysed a CRP/ALB ratio of more homogenous patient populations such those clearly diagnosed with sepsis 14 or septic shock 6 . Considering that sepsis is accompanied by severe inflammation 15 , the impact of elevated CRP on 30-day mortality might be attenuated in this study. Secondly, although we included the comorbidity of cancer at ICU admission as a covariate, the proportion of cancer patients was 21.2% at ICU admission. Considering that elevated CRP is closely associated with a risk of cancer 16 , the impact of elevated CRP might also be attenuated in this study. Third, predictive value of serum albumin was so strong in critically ill patients 17 that combining CRP with albumin (CRP/ALB ratio) was not beneficial in predicting 30-day mortality in critically ill patients.
Previous studies reported that the CRP/ALB ratio could be a useful prognostic factor in predicting mortality in patients with sepsis 14 , septic shock patients 6 , or critically ill patients requiring parenteral nutrition 8 . However, the previous studies did not evaluate the prognostic value of CRP/ALB ratio compared with that of albumin alone or other prognostic factors such as the APACHE II or Charlson comorbidity index 6,8,13,14 . By using the Delong test, we demonstrated that the ability of the CRP/ALB ratio for predicting mortality in critically ill patients is affected by a strong prognostic power of albumin at ICU admission. Furthermore, the prognostic value of a CRP/ ALB ratio is not superior to that of traditional prognostic factors such as the APACHE II or Charlson comorbidity index.
The results of this study are similar to a previous study performed by Fairclough et al. who reported that a modified early warning system for acute medical admissions has better prognostic value for a patient's outcome  than the CRP/ALB ratio 18 . These researchers also suggested that pulse, respiratory rate, temperature, urine output, and systolic blood pressure were used for the modified early warning system. Considering that the APACHE II score includes heart rate and respiratory rate, the results of our study are similar to the results of Fairclough et al. 18 In addition, we compared the difference in predictive power of 30-day mortality in critically ill patients between those assessed with the CRP/ALB ratio and Charlson comorbidity index, which are known to be useful indicators of chronic comorbidities of patients 19 .
The present study has several limitations. First, the retrospective single-centre design is associated with a risk of selection bias and/or limited generalizability of the results. Second, although patients were excluded if their CRP and ALB values were not measured during the same 24-h period, not all included patients underwent simultaneous testing of CRP and ALB. Lastly, since we analysed the mixed ICU patients in this study, applying the results of this study in specific population such as patients with cancer or sepsis is controversial. Nevertheless, the present study provides value, as it is the first study to analyse the relationship between the CRP/ALB ratio and 30-day mortality after ICU admission.
In conclusion, although an elevated CRP/ALB ratio at the ICU admission was an independent risk factor for 30-day mortality after ICU admission, the predictive power of CRP/ALB ratio is lower than albumin alone, APACHE II, and Charlson comorbidity index. Therefore, clinical usefulness of the CRP/ALB ratio in predicting 30-day mortality in critically ill patients is questionable.

Methods
This retrospective cohort study was performed with the approval of the institutional review board of SNUBH (B-1806/474-105). The requirement for written informed consent was waived by the institutional review board; this manuscript adheres to the applicable STROBE guidelines.
Patients were included in the study if they were adults (≥19 years old) and were admitted to an ICU at SNUBH between 1 January 2012 and 31 December 2016. Only the last admission was considered for patients who were admitted to an ICU more than once during the study period. Patients were excluded if they did not undergo CRP and ALB testing on the same day as their ICU admission. This study is a sequential study of previous studies conducted by our institution 13 , that assessed patients who underwent postoperative ICU admission from 2007 to 2016, while this study analysed the patients who were admitted to all ICUs from 2012 to 2016.
The SNUBH has 1,360 beds and several ICUs with a total of 102 beds (Medical, Surgical, Neurologic, Emergency I and II). Each ICU has certified intensivists (anaesthesiologists, pulmonologists, neurologists, emergency physicians, and thoracic surgeons) that provide primary care during daytime office hours, while on-duty residents and fellows provide primary care at night and on weekends.
Measurements and outcomes. Baseline data were collected regarding the patients' demographic characteristics, history of underlying diseases, laboratory test results at the ICU admission, and exact dates of death. For example, the patients' records were searched for diagnoses of hypertension, diabetes mellitus, coronary heart disease, chronic obstructive pulmonary disease, and cancer at their ICU admission. In addition, APACHE II scores and Charlson Comorbidity Indexes at the ICU admission were collected. All laboratory testing had been performed using venous or arterial sampling within 24 h after the ICU admission, and only the earliest test results were included if the same test was performed multiple times during the first 24 h. The Korean Ministry of the Interior and Safety approved the use of exact dates of death for all patients.
The primary outcome of interest was the relationship between the CRP/ALB ratio and 30-day mortality after ICU admission. In addition, the present study aimed to compare the CRP/ALB ratio with other prognostic factors (APACHE II, Charlson comorbidity index) in predicting 30-day mortality after ICU admission. Statistical analysis. Baseline characteristics of total patients were presented as numbers with percentages or means with standard deviations. First, we performed univariable Cox regression analysis to identify an individual relationship with 30-day mortality after ICU admission. Next, multivariable Cox regression analysis using a stepwise backward elimination method was performed to identify an independent relationship with 30-day mortality after ICU admission. In this multivariable Cox regression analysis, CRP, Albumin, and CRP/ALB ratio were included in another Cox regression model to avoid multicollinearity (variance inflation factors >11) within the model. All included variables fulfilled the Cox proportional hazard assumption based on a 'log minus log plot' with the CRP/ALB ratio.
Secondly, ROC analysis was performed to investigate predictability of CRP, albumin, CRP/ALB ratio, APACHE II, and Charlson comorbidity index. The AUC of the five variables were compared using Delong's test 20 . All analyses were performed using IBM SPSS Version 23.0 (IBM Corp., Armonk, NY, USA), with P-values of <0.05 being considered statistically significant.

Data Availability
The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request.