Study design
This was a retrospective cohort study based on two large publicly available critical care databases the Medical Information Mart for Intensive Care III version 1.4 (MIMIC III v 1.4) and MIMIC IV v 0.4 [16, 17]. MIMIC-III covers over 40,000 intensive care unit (ICU) admissions at the Beth Israel Deaconess Medical Center in Boston between 2001 and 2012. MIMIC-IV is an update to MIMIC-III. We passed the Protecting Human Research Participants exam and obtain the seniority to access these databases. Data were extracted by authors YZ, YYL and YHZ.
Population selection criteria
Adult patients who underwent cardiac surgery and were admitted to the ICU for the first time were included in the study. The following exclusion criteria were applied: (1) missing LDH data at admission, and (2) missing data > 5%.
Data extraction
The extracted data contained age, gender, marital status, ethnicity, body mass index (BMI), heart rate, systolic blood pressure (SBP), diastolic blood pressure (DBP), simplified acute physiology score (SAPS) II, sequential organ failure assessment (SOFA) score, comorbidities, laboratory parameters and outcomes. Comorbidities comprised hypertension, diabetes, coronary heart disease (CHD), valve disease, heart failure (HF), chronic obstructive pulmonary disease (COPD), and chronic kidney disease (CKD). Laboratory parameters included LDH, white blood cell (WBC) count, platelet (PLT) count, blood urea nitrogen (BUN), serum creatine (SCr), sodium, potassium, glucose, neutrophil count, lymphocyte count, monocyte count, and lactate. We took the preoperative laboratory indicators from the first results after admission. The NLR, LMR, and platelet–lymphocyte ratio (PLR) were calculated as follows: NLR = neutrophil/lymphocyte counts, LMR = lymphocyte/ monocyte counts, and PLR = platelet/lymphocyte counts. Cardiac surgical procedures included CABG and/or heart valve surgery. The primary endpoint was in-hospital mortality, whereas the secondary endpoints were 1-year mortality, continuous renal replacement therapy (CRRT), prolonged ventilation, and prolonged length of ICU and hospital stay. 1-year mortality referred to the time from admission to mortality from any cause within one year. Prolonged ventilation was defined as the need for mechanical ventilation for more than 24 h. Prolonged length of stay was defined as any stay beyond the 75th percentile for the total study population. Moreover, prolonged length of ICU and hospital stays were length of stay longer than 6 days and 15 days, respectively.
Statistical analysis
MIMIC III was regarded as the training cohort which was used to investigate whether LDH is a poor prognostic factor for patients undergoing cardiac surgery. We performed external validation in MIMIC IV to confirm the results obtained from MIMIC III. A total of 2325 patients were enrolled from MIMIC III, whereas 1387 patients were enrolled from the MIMIC IV (Additional file 1: Figure S1). Next, we compared the prognostic power of LDH with other indicators in patients undergoing cardiac surgery, including NLR, LMR, PLR, lactate and SAPS II. These indicators have previously been shown to be strong prognostic biomarkers in cardiac surgery patients. After excluding patients with missing NLR, LMR, PLR, lactate, and SAPS II data, there were 830 patients left in MIMIC III and 731 patients in MIMIC IV who were then subjected to further analysis (Additional file 1: Figure S1).
The relationship between categorical variables were examined by Pearson χ2 tests and reported as counts (percentage). Continuous variables were presented as means (standard deviation) or medians (range), and differences between values were examined by independent t-test or Mann–Whitney U test. Receiver operating characteristic (ROC) curve was applied to determine the best cut-off point of LDH for predicting in-hospital mortality. We performed Kaplan–Meier curves with the log-rank test to asses the one-year survival rate between groups based on the optimal cut-off value of LDH. Multivariate logistic regression analysis and Cox proportional hazards regression analysis were used to explore the predictive value of LDH for poor outcomes. Notably, LDH was tested both as a continuous and a categorical variable, and NLR, LMR, PLR, and lactate were not entered into the multivariate analysis because more than 20% of the data was missing. Significant factors associated with primary endpoint from univariate analyses were included in multivariate analysis. The following covariates in MIMIC III were adjusted as potential confounders: ethnicity, SBP, heart rate, hypertension, CHD, heart failure, BUN, SCr, sodium, potassium, glucose, SAPS II, and SOFA score (Additional file 2: Table S1). The following covariates in MIMIC IV were adjusted: hypertension, heart failure, CKD, WBC, BUN, SCr, sodium, potassium, and SAPS II (Additional file 2: Table S1). Results were expressed with odds ratios (OR) for logistic regression analysis or hazard ratio (HR) for Cox proportional hazards analysis, and their 95% confidence intervals (95% CI). To evaluate the consistency of the prognostic impact of LDH on primary endpoint, we conducted stratified analyses in different groups of gender, age, hypertension, diabetes, CHD, CKD, heart failure, and valve disease. Interaction tests between each subgroup were analyzed. We used ROC curve based on DeLong’s test to compare the predictive ability of LDH with other prognostic indicators, including NLR, LMR, PLR, lactate, and SAPS II. To assess correlations between LDH and these indicators, Pearson or Spearman analyses were performed where appropriate. All statistical analyses were examined using packages implemented in R software (version 3.6.3) and MedCalc version 19.1 (MedCalc Software, Belgium). A p < 0.05 was considered as statistically significant.