This single-center retrospective cohort study was conducted at Kyoto University Hospital, which is a 1121-bed teaching hospital. The ethics committee of the Kyoto University Graduate School of Medicine approved the study protocol (approval number: R2646, September 17, 2020) and waived the requirement for informed consent because of retrospective nature of this study. This study adhered to the Strengthening the Reporting of Observational Studies in Epidemiology statement [11]. Patients aged ≥ 18 years and who underwent major nonpulmonary and noncardiac surgery under general anesthesia at Kyoto University Hospital from June 2012 to March 2019, which was the duration of the availability of PFT data, were included. We defined major nonpulmonary and noncardiac surgery as those procedures classified as intermediate or high-risk surgery according to the 2014 European Society of Cardiology/European Society of Anaesthesiology guidelines, excluding cardiac and pulmonary surgery [12]. The specific procedures of major nonpulmonary and noncardiac surgery were as follows: abdominal aortic aneurysm repair, carotid endarterectomy, esophagectomy, gastrectomy, colorectal resection, liver resection, biliary tract surgery, pancreatic resection, splenectomy, nephrectomy, adrenalectomy, cystectomy, total pelvic organ removal, total hip arthroplasty, total knee arthroplasty, lower extremity amputation, peripheral vascular bypass surgery, and spine surgery. Patients who underwent organ transplant surgery or emergency surgery, those who were on preoperative respiratory support, and those who had not undergone a PFT within 6 months prior to surgery were excluded.

Data were collected from the Kyoto University Hospital IMProve Anesthesia Care and ouTcomes database [13], which aimed to clarify the relationship of intraoperative respiratory and cardiovascular parameters with postoperative outcomes. In addition, data from PFTs performed 6 months before surgery were collected. The definitions of the collected variables are shown in Supplemental Table 1.

The primary exposure was preoperative FEV1 within 6 months before surgery. If more than one PFT was performed within 6 months before surgery, the most recent value was used. In addition, the forced vital capacity (FVC) and the vital capacity (VC) were considered as exploratory exposure factors. FEV1, FVC, and VC were used in the analysis by determining the % predicted values, which were calculated based on age, sex, and height and have been validated in the Japanese population [14]. In the logistic regression analyses, the FEV1% predicted and FVC % predicted were reverse-coded so that the lower values were associated with higher hazard ratios.

The primary outcome was respiratory failure and/or death within 30 days after surgery. Respiratory failure was defined as mechanical ventilation by endotracheal tube or tracheostomy for more than 24 h after surgery or postoperative reintubation. The secondary outcomes were 30-day mortality, inhospital mortality, and postoperative hospital length of stay. Postoperative length of stay was recorded for patients who survived until discharge.

First, we performed an external validation of the RFRI. As reported in a previous study [10], the RFRI was incorporated into a logistic regression model after the following transformation:

ln (RFRI + 1),

where ln indicated the natural logarithm. We graphically assessed the calibration of the RFRI with a calibration plot and tested it with the Hosmer–Lemeshow test. A P-value of < 0.05 indicated a lack of good fit for the model. For model discrimination, we computed the area under the receiver operating characteristic curve (AUROC) with a 95% confidence interval (CI) using 500 bootstrap resampling.

Next, we performed a logistic regression analysis after adding FEV1 as a continuous variable to the RFRI to determine whether FEV1 was associated with respiratory failure and/or death independent of the RFRI. Because the association between preoperative FEV1 and the development of postoperative respiratory failure may not be linear, we used a restricted cubic spline of five knots and a categorized FEV1 into four based on Goldman’s classification (i.e., normal: ≥ 80%, mild decline: 70% to < 80%, moderate decline: 60% to < 70%, and severe decline: < 60%) to evaluate the existence of nonlinear relationships. In addition, we assessed the impact of adding the value of FEV1 to the RFRI in predicting respiratory failure and/or death by comparing AUROC values.

Furthermore, we examined the associations of FVC and VC with respiratory failure and/or death by adjusting for the RFRI on logistic regression analysis. In this analysis, the FVC and VC were entered into the model as continuous variables, and the odds ratios (ORs) per 10% decline were determined. According to the American College of Physicians guidelines, preoperative spirometry is recommended only in high-risk patients [4]. Therefore, according to the median RFRI, we stratified the patients into high-risk (RFRI ≥ 15) and low-risk (RFRI ≤ 14) groups and evaluated the association between FEV1 and respiratory failure and/or death in each group. In addition, we performed a subgroup analysis of abdominal and nonabdominal surgery groups, because decreased preoperative pulmonary function was reported to be associated with the occurrence of respiratory complications in abdominal surgery [7, 8].

We assessed the robustness of our findings using two sensitivity analyses. Sensitivity models were constructed as a logistic regression that was identical with the primary model described above, except for the following: (i) outcome was redefined as respiratory failure and/or death within 7 days instead of 30 days after surgery and (ii) with adjustments for factors that were reported to be risk factors for postoperative respiratory complications but were not included in the RFRI (i.e., congestive heart failure, preoperative hemoglobin level, and operative time) [4, 15].

We decided to use data from all cases included in the database to maximize the statistical power. For sample size estimation, 10 events per variable were required for reliable multivariable logistic regression analysis [16]. In our previous study on patients undergoing major abdominal surgery, 1.8% of patients received invasive respiratory support [17], and the estimated number of major nonpulmonary and noncardiac surgeries performed annually was 700. Based on these data, 4900 surgeries were calculated to meet the inclusion criteria, and 88 cases were calculated to develop the primary outcome of postoperative respiratory failure and/or death during the study period of approximately seven years. Therefore, the estimate was to perform a multivariable logistic regression with eight variables.

For missing data, the plan was to perform a complete case analysis if the percentage of missing data was < 5%; such an analysis was considered to be feasible [18]. If the percentage of missing data was > 5%, we planned to complete the missing values.