Study design, setting, population, and data collection
This single center retrospective observational cohort study was conducted at Peking University First Hospital, a giant teaching hospital in China with 1800 beds. The Ethics Committee approved this study. Due to the retrospective nature and none patient follow-up, and no patients' identification information was involved, the IRB waived the written informed consent requirement.
This study extracted data from the hospital’s perioperative database.1 Adults (age ≥ 18 years old) who underwent elective non-cardiac surgery between January 1, 2012, and December 31, 2018, were screened. Surgery identification was based on the International Classification of Diseases and Procedures, Ninth Clinical Revision volume 3 (ICD-9-v3).
The procedures included otolaryngology, general surgery, urology, gynecology, orthopedics, neurosurgery, vascular and thoracic surgery while excluding cardiac, obstetrics, and emergency surgery. Only the first procedure record was used by patients who have operated more than two times a year. If the two operations' time intervals were less than three months, the second operation would not be registered. Surgery under local infiltration anesthesia was also omitted from this analysis.
Variables Used In The Present Study
The information on each patient collected in the present study included their demographic and essential characteristics (e.g., body mass index, BMI, gender, age, smoking, and alcohol habits); preoperative co-existing disorders (e.g., hypertension, coronary artery disease, preoperative hemoglobin, albumin, creatinine levels), and intraoperative parameters (e.g., intraoperative hypotension, blood transfusion, anesthetic technique, The modified John Hopkins Hospital criteria, MJHSC, whether or not intraperitoneal surgery, mean intraoperative heart rate).10,11 The modified John Hopkins Hospital criteria (MJHSC) were used to categorize the surgical complexity.10,11 (Variables definition was shown in Table S1).
Study Endpoints
The primary endpoint was any patient with AKI within seven days after surgery in the hospital. The authors used KDIGO as the criteria for AKI, defined by the patient's postoperative serum creatinine level increase to no less than 26.5 µmol/l within 48 hours, or 1.5 times from the baseline 7 days after surgery, or initialization of blood dialysis.
2–8 GFR value or urine output was not included in the definition of the outcome as postoperative creatinine level dramatically fluctuates, which could cause an inaccurate estimate of eGFR.
Statistical analysis
Patients were separate into two groups according to the D-dimer cut-off point. Continuous variables with a normal or non-normal distribution were compared using the Student t-test or Mann-Whitney U-test. The Kolmogorov-Smirnov test was used to determine whether the data were normally distributed. Categorical variables were compared using the Chi-Square test or the continuity corrected Chi-Square test. Rank variables were compared using the Kruskal-Wallis H-test. A two-sided p-value < .05 was considered significant.
Non-linear relationship between D-dimer and on AKI
This study examined the D-dimer's unadjusted relationship and the risk of AKI using a cubic spline function by General Additive Models (GAM). The marginal effect of preoperative D-dimer on postoperative AKI was calculated and plot.
Minimum P-value approach for D-dimer threshold
This study tried to locate the inflection point dividing the D-dimer into two clinically meaningful categories.9–13 If we observed an inflation area, the optimal threshold for the D-dimer was determined using the minimum P-value approach. This approach evaluated every possible threshold of the D-dimer at intervals of 0.01 µg/mL in the multivariate logistic regression models. The D-dimer that demonstrated the smallest statistically significant P-value was selected as the optimal threshold to divide the D-dimer into two groups.
Analysis Of The Propensity Score Weighting
To increase the robustness, the present study balanced the two groups (patients’ D-dimer below or above cut-off point) using the propensity score (PS) weighting, which diminishes the effects of measured confounding factors and obtains a less biased result in observational studies. The PS weights were calculated using gradient boosted regression models, in which whether patients’ D-dimer was below or above cut-off point was the dependent variable. 12,13 Unbalanced preoperative factors (age, gender, body mass index, smoking and drinking habits, preoperative hemoglobin, albumin, creatinine level, co-existing disorders, surgery duration, cancer surgery, intraoperative blood transfusion, surgical complexity, intraperitoneal, anesthesiology experience) was included as independent variables. D-dimer's adjusted odds ratio from two new cohorts with propensity weights was calculated and considered to be robust and less biased.
Statistical Packages
All data management and statistical analysis were performed using the R programming language (v.3.5.2).