Study design and setting
This study was a secondary analysis based on a single-center retrospective study,that had been conducted a single-center retrospective study from January 1, 2012 to October 31, 2016 at the Singapore General Hospital.In the present study,it was performed to address the relationship between anemia status and perioperative prognosis.The target independent variable is anemia status obtained at baseline.
Participants and Procedures
Patients who underwent cardiac surgery, burn-related surgery, neurosurgery, and transplantation were excluded due to their categorically higher mortality rate and blood transfusion requirement,based on the original research. A total of 90785 surgical patients were recruited and selected for the study. Only surgical patients, over 18 years of age, with complete anemia data can qualified for inclusion in the study.
Covariates included in this study were specified a priori as potential confounders on the relationship of anemia and perioperative prognosis in patients, based on clinical experience and previous studies. The data collected during the preoperative anesthetic assessment visit included age, gender, race, preoperative estimated glomerular filtration rate(eGFR),presence of cerebrovascular accidents(CVA), diabetes mellitus(DM),ischemic heart disease(IHD),congestive heart failure(CHF),red cell distribution(RDW), priority of surgery, anesthesia type, surgical risk, preoperative blood transfusion with in 30days, intraoperative blood transfusion data, the RCRI score, the ASA status.Preoperative laboratory results including renal group (including eGFR) and full blood count (including hemoglobin concentration and RDW) were taken as the latest blood results within 90 days before surgery, and up to the day of surgery. RDW is the coefficient of variation (percentage) between the red blood cell volume and the normal reference range of RDW, ranging from 10.9% to 15.7%. Levels >15.7% were defined as high RDW. The severity of anemia was defined by WHO’s gender-based classification of hemoglobin concentration. Mild anemia was defined as hemoglobin concentration of 11–12.9g/dL in males and 11–11.9g/dL in females; moderate anemia was defined for both genders to be hemoglobin concentration between 8–10.9g/dL and severe anemia defined as hemoglobin concentration <8.0g/dL.Priority of surgery (emergency or elective) and surgical risk classification were based on the 2014 European Society of Cardiology (ESC) and the European Society of Anaesthesiology (ESA) guidelines[16, 17].American Society of Anesthesiologists-Physical Status (ASA-PS) follows that of the ASA-PS definitions [17].
The patients were followed up for 30 days after their index operation to identify all ICU admissions (stay time >24 hours), blood transfusion and mortality. Mortality data (the primary outcome) were synchronized with the National Electronic Health Records, ensuring a near complete follow-up [18]. The need for ICU stay (>24 hours) during surgical admission may serve as a surrogate marker for major postoperative complications.
Dataset
We downloaded the raw data for free from the DATADRYAD database (www.datadryad.org). Since Diana Xin Hui Chan et al. transferred the ownership of the original data to the DATADRYAD website, we were able to use this data for secondary data analysis based on different scientific assumptions (Dryad data package: Chan, Diana Xin Hui et al. (2018), Data from: Development of the Combined Assessment of Risk Encountered in Surgery (CARES) surgical risk calculator for prediction of post-surgical mortality and need for intensive care unit admission risk – a single-center retrospective study, Dryad, Dataset, https://doi.org/10.5061/dryad.v142481). Since our study was based on a secondary analysis of past data and the patient's personal information in the original data was anonymous, there was no need for informed consent from the participants. The ethical approval was described in the published paper [19].
Statistical analysis
Considering the differences in baseline characteristics between the two groups of eligible participants (Table 1), propensity score matching was used to identify a cohort of patients with similar baseline characteristics. Matching was performed with the use of a 1:1 matching protocol without replacement (greedy-matching algorithm), with a caliper width equal to 0.05. Covariate balances before and after PS matching was assessed using standardized differences. For a given covariate, standardized differences of less than 10.0%indicate a relatively small imbalance.
The doubly robust estimation method, the combination of multivariate regression model and a propensity score model, was also applied to infer the independent associations between anemia status and patients’ primary and secondary outcomes [20, 21]. Using the estimated propensity scores as weights, an inverse probabilities weighting (IPW) model was used to generate a weighted cohort[22] . A logistic regression was then performed on the weighted cohort, adjusting for the variables that remained unbalanced between different anemia groups in the propensity score model.
Sensitivity analysis
We conducted a series of sensitivity analyses to evaluate the robustness of the findings of the study and how our conclusions can be affected by applying various association inference models. In the sensitivity analysis, we applied three more association inference models: a propensity score-based IPW model, a propensity score-based patient-matching model, and a logistic regression-based multivariate analysis model. The calculated effect sizes and p values from all these models were reported and compared.
Continuous variables were expressed as mean ± standard deviation (normal distribution) or median (interquartile range) (skewed distribution), and categorical variables were expressed in frequency or as a percentage. In the process of multivariate regression analysis, there are some confounders with partial missing data. If it is a categorical variable, the missing data would be directly treated as a new independent group; if it is a continues variable, the missing data would be replaced with an average or median value. The T test (normal distribution), Mann-Whitney (skewed distribution) tests and chi-square tests (categorical variables) were used to determine any statistical differences between the means and proportions of the anemia groups. All of the analyses were performed with the statistical software packages R (http://www.R-project.org, The R Foundation) and EmpowerStats (http://www.empowerstats.com, X&Y Solutions, Inc., Boston, MA). P values less than 0.05 (two-sided) were considered statistically significant.