As the Global Cancer Statistics reported in 2020, lung cancer has become the second most common cancer and the highest rate of cancer-related death (1). The treatments of lung cancer mainly include radiotherapy, chemotherapy, surgery, targeted therapy, immunotherapy. However, surgical resection is still an effective and safe intervention for patients with lung cancer.

Unfortunately, the postoperative pulmonary complications (PPCs) of lung cancer surgery are still common problems and major challenges for patient recovery. And postoperative pneumonia (POP) has become the most common PPCs (2, 3). Several studies have found that the rate of POP in lung cancer patients is about 2%-25% (4–7). And POP could significantly prolong the length of hospitalization, increase hospitalization expense, and even increase perioperative mortality (4, 8–10). Thus, early identification of risk factors associated with POP among patients with lung cancer could be beneficial to forecast the risk of POP in advance.

Potential risk factors for POP existed throughout the whole perioperative period (11). Preoperative factors mainly include elderly, smoking, pulmonary function, comorbidities and nutritional status (12, 13). Intraoperative factors include surgery types, duration of surgery, anesthesia types and ventilation mode (6, 14). Postoperative factors include acute pain and other complications (15, 16). A retrospective observational cohort (n=7479) found that elderly, preoperative pulmonary infection, atrial fibrillation, obesity, and alcohol might be associated with POP among lung cancer patients (10). Deguchi et al. (5) found preoperative asthma might also be independently associated with POP in lung cancer patients. And Yendamuri et al. (17) found age >75 years, male, thoracotomy, COPD and American Society of Anesthesiologists (ASA) ≥III might be potential risk factors after analyzing a range of patients (n=12562) who underwent pulmonary lobectomy. However, most of the present studies on POP in lung cancer patients had only identified risk factors for POP, studies about prediction models for POP were very limited.

Although there had a model to predict POP for elderly patients who underwent video-assisted thoracoscopic surgery for lung cancer, the data was collected from 2012 to 2019 and other variables associating with POP such as anesthesia types, anesthetics and surgery types were scarce (4). Thus, the development of a model after fully considering the preoperative and intraoperative variables to forecast the risk of POP in advance would be beneficial to patients.

Therefore, the purpose of this study was to develop and validate a model to predict POP in patients undergoing lung cancer surgery and investigate risk factors for POP so that reasonable preventive and treatment measures could be made earlier.

This study developed and validated a nomogram to accurately forecast the risk of POP in patients who underwent lung cancer surgery. This model included six preoperative and intraoperative variables, including smoking, diabetes, preoperative chemotherapy, surgery types, ASA grade and duration of surgery, which predicted well as demonstrated by the uncorrected C index values of 0.717 and 0.726 in the training and validation cohorts. At the same time, the calibration curves showed good consistency between prediction and actual observation, respectively, and the decision curve analysis indicated this nomogram had good clinical benefit.

LASSO regression is a common method for variable selection in fitting high-dimensional generalized linear and has been widely used in clinical research (22, 23). The LASSO method selects variables via minimizing the coefficients of relatively irrelevant variables to 0 and subsequently removing these variables by constructing a penalty function, which effectively avoids the overfitting and makes the model more refined (24, 25). So we used the LASSO regression for variable selection in this study.

The nomogram we used is a superior visual tool and has been widely used in the clinical practice, which has several advantages. Firstly, the nomogram could transform predictive model into a single estimate of probability according to patient’s characteristics, making model simple to understand (26). Secondly, the scoring system has high precision and good stability characteristics in predicting results (27, 28). Therefore, we used nomogram to build the visual model to help clinicians to stratify patients and develop individual clinical treatment strategies according to patient’s conditions.

Evaluating the characteristics of the predictive model from multiple perspectives and selecting the optimal model could help promotion and application of the model (29). The calibration ability is model’s capability to demonstrate the consistency between the actual observed and the prediction by the model, which is one of the best indicators to reflect predictive performance of the model (29). Therefore, we used calibration curve and Hosmer-Lemeshow goodness-of-fit test to evaluate calibration ability of this model. And good agreements between prediction and actual observation were supported by Hosmer-Lemeshow goodness-of-fit test. However, good calibration couldn’t perfectly distinguish patients with or without POP. The ROC curve and C index had certain advantages in measuring discrimination ability of the model (30). And the results showed that the model could distinguish patients with and without POP. Furthermore, we used the decision curve analysis and net benefit to evaluate the clinical benefit of the model (31). The results suggested that using this model to assist clinical treatment strategies might help improve patient prognosis.

In the POP risk estimation nomogram, preoperative chemotherapy, thoracotomy, ASA and duration of surgery have been confirmed to increase the risk of POP (17, 32, 33). This study showed that above factors were also independent risk factors associated with POP in lung cancer patients. In addition, we illustrated that diabetes was associated with POP in patients with lung cancer surgery.

In previous reports, diabetes was associated with POP after surgery (34). The incidence of POP in type 2 diabetes mellitus patients was 21% higher than non-diabetic patients (35). In the present study, we found the risk of POP was higher in diabetes patients (OR=1.838; 95%CI:1.110-3.001) after lung cancer surgery. The potential mechanisms were diabetes could destroy innate immunity in pulmonary, impair pulmonary function and reduce cardiorespiratory fitness making patients more susceptible to infections (36).

However, this study still has some limitations. At first, the bias of patient selection could not be entirely avoided because it was a single-center retrospective study. However, we screened patients through strict inclusion and exclusion criteria, which could reduce population homogeneity to some extent. Secondly, the definition of POP in our study might be generic and sensitive to diagnose pneumonia, which made the incidence of POP a little higher than other studies. However, we used the same criteria in diagnosing pneumonia in our study which could make result reliable to some extent. Thirdly, intraoperative respiratory parameters, such as tidal volume, minute ventilation, PEEP were not included in our study because these variables could not be collected. However, some studies had found that intraoperative ventilation strategy might not be associated with postoperative pulmonary complications (37, 38). Finally, some postoperative variables, such as postoperative pain, postoperative aerosolized inhalation might influence the rate of POP among patients after lung cancer surgery, but these variables were not analyzed in the study because we aimed at predicting POP through preoperative and intraoperative variables rather than postoperative variables.

This study developed and validated a predictive model representing by the nomogram to quantify the risk of POP in patients with lung cancer surgery. This model showed good discrimination ability, calibration ability and clinical benefit which could help make better prevention and individual treatment strategies in advance.

The code of dynamic nomogram in the study is included in the article/ Supplementary Material . Further inquiries can be directed to the corresponding author.

The study involving human participants was reviewed and approved by Ethics Committee of The Fourth hospital of Hebei Medical University. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.

JF: This author helped in data acquisition, data analysis, and manuscript drafting. LW: This author helped in data acquisition, and data analysis. XQ: This author helped in data acquisition, and data analysis. JS: This author helped in data acquisition. RX: This author helped in data acquisition. H-QJ: This author helped in concept and design, data analysis, data interpretation, revision of the manuscript, and final approval of submission. All authors contributed to the article and approved the submitted version.