Neoadjuvant immunochemotherapy and postoperative acute hypoxemic respiratory failure in thoracic surgery: a multicentre cohort study

Abstract

Objectives

While immune checkpoint inhibitors (ICIs) continue to transform the neoadjuvant treatment, its association with postoperative acute hypoxemic respiratory failure (AHRF) remains unexplored. This study aims to assess the association between neoadjuvant immunochemotherapy (nICT) and postoperative AHRF risk following thoracic tumour surgeries and identify the risk subgroups.

Methods

This retrospective two-centre cohort study included 327 patients receiving nICT (n = 167) or neoadjuvant immunochemotherapy (nCT, n = 160) before thoracic tumour surgeries from December 2017 to June 2023. Data were analysed by using the propensity score matching (PSM) and multivariable logistic regressions. Subgroup and sensitivity analyses were performed to test the stability of the conclusions.

Results

The nICT group demonstrated significantly higher postoperative AHRF incidence than the nCT group (19.8% vs. 8.1%, p = 0.002). The inverse probability-weighting model (IPTW) confirmed elevated AHRF risk associated with nICT compared to nCT (OR = 2.41, 95% CI: 1.2–4.82). In patients with non-small cell lung cancer (NSCLC), the binary logistic regression analysis showed that the history of nICT was significantly associated with postoperative AHRF (OR = 4.12, 95% CI: 1.15–14.8) in patients with NSCLC. Subgroup analyses revealed elevated AHRF risks with nICT versus nCT in patients with time interval between neoadjuvant therapy and surgery within 42 days (OR = 6.68, 95% CI: 1.24–35.98), those with squamous cell carcinoma (SCC) (OR = 3.64, 95% CI: 1.41–9.44), and those who did not achieve pathologic complete response (non-pCR) (OR = 2.82, 95% CI: 1.14–6.98).

Conclusions

nICT was associated with increased postoperative AHRF risk in thoracic surgical patients, necessitating rigorous perioperative monitoring.

Introduction

Non-small cell lung cancer (NSCLC), oesophageal carcinoma and mediastinal malignancies constitute the principal indications for thoracic surgical interventions. Lung cancer is the most common malignancy in China [1] and worldwide [2], with NSCLC accounting for 80.0% to 85.0% of newly diagnosed lung cancer cases annually [3]. Neoadjuvant therapy for thoracic malignancies aims to downstage tumours, improve resectability and enhance survival outcomes [4,5]. While neoadjuvant chemotherapy (nCT) offers a modest benefit over surgery alone, emerging evidence suggests that combining immune checkpoint inhibitors (ICIs) with chemotherapy improves event-free survival (EFS) and pathological response rates [4,6].

Postoperative acute hypoxemic respiratory failure (AHRF) is a clinically critical complication associated with increased mortality, prolonged ICU stay and substantial healthcare burden [7]. The impact of neoadjuvant therapy on the development of postoperative AHRF remains controversial. Evidence is limited, and existing studies have reported inconsistent findings [8–11]. While Siegenthaler et al. observed no significant difference in major respiratory complications between patients undergoing lung resection with or without induction chemotherapy [12], Leo et al. reported a significantly higher incidence of respiratory complications following pneumonectomy in patients who received induction chemotherapy [13].

As ICIs continue to transform the neoadjuvant treatment, they may introduce new challenges requiring clinical vigilance. In the CheckMate-816 and KEYNOTE-671 trials, postoperative mortality was higher in the nICT group than in the nCT group (1.65–4.4% vs. 0.75–2.23%), with fatal pulmonary complications occurring more frequently after nICT (1.1–2% vs. ∼0.5%) [14,15]. Consistently, pulmonary complications have been one of those major safety concerns across other nICT studies [16–18]. In patients with oesophageal cancer, the incidence of postoperative pulmonary complications, such as pleural effusion, increased following nICT [19,20]. With the rapid advances of nICT in thoracic neoplasms, perioperative adverse events are receiving increasing attention. Notably, most reports of immunotherapy-related lung toxicity arise from medical oncology cohorts, whereas the potential interaction between neoadjuvant ICIs and postoperative AHRF remains underexplored.

We conducted a two-centre retrospective cohort of patients undergoing thoracic tumour resection to assess whether nICT increases postoperative AHRF risk compared with chemotherapy alone. We hypothesized that nICT is associated with a higher risk of postoperative AHRF.

Methods

Data sources and setting

This was a two-centre, retrospective cohort study. The dissemination of the findings adheres to the guidelines established by the Revised Strengthening the reporting of cohort, cross-sectional and case-control studies in surgery (STROCSS) Guideline [21]. This study was approved by the institutional review boards (IRB) of the Second Xiangya Hospital of Central South University (No. LYF2023105) and Guilin Hospital of the Second Xiangya Hospital (No. LLkt-034) in Aug. 2023. In addition, this study was registered in the ClinicalTrials.gov (NCT06566053).

Study population

This retrospective study analysed medical records from the Second Xiangya Hospital of Central South University and Guilin Hospital of the Second Xiangya Hospital, from December 2017 to June 2023. Eligible patients were adults (≥18 years) with pathologically confirmed thoracic malignancies (NSCLC, oesophageal carcinoma or mediastinal tumours) who underwent thoracic surgery following nICT or chemotherapy alone. Patients were excluded if they had incomplete neoadjuvant treatment documentation, concurrent malignancies, preoperative targeted therapy, did not undergo thoracic surgery following admission, received PD-L1 inhibitors (n = 1, excluded due to insufficient sample size and potential heterogeneity), interstitial pneumonia history, grade ≥3 treatment-related adverse events (trAEs), pulmonary embolism or perioperative heart failure. All patients received care from a multidisciplinary thoracic oncology team. Both centres implemented thoracic enhanced recovery after surgery (ERAS) in mid-2020 (Center A: June 2020; Center B: August 2020), and perioperative management criteria adhered to comparable institutional stand AHRF. The study flowchart is presented in Figure 1.

Figure 1.

Flow chart of the study population and data availability.

Data extraction

Clinical data were collected from medical records, including age, gender, smoking history, alcohol history, comorbidities, disease type, preoperative body mass index (BMI), preoperative serum albumin (ALB), preoperative creatinine, pulmonary function, neoadjuvant cycle number, time interval between neoadjuvant therapy and surgery (TTS), surgical approach, pathology type and pathological response. In lung cancer patients, the extent of resection was also collected.

Outcomes

The primary outcome was the rate of postoperative AHRF. Postoperative AHRF was defined as the acute onset of hypoxemia within 1 week postoperatively, indicated by a ratio of arterial partial pressure of oxygen to inspired oxygen fraction (PaO₂/FiO₂) ≤200 mmHg, requiring high-flow nasal cannula (HFNC), noninvasive positive pressure ventilation or invasive mechanical ventilation for a minimum of 24 h, and exclude fluid overload as the causes [8]. We conducted a retrospective chart review with medical records. Each AHRF diagnosis was independently assessed by at least two attending physicians. Secondary outcomes encompassed the length of hospital stay, length of ICU stay and length of postoperative stay.

Statistical analysis

Continuous variables were expressed as mean ± standard deviation for normal distributed data, and as median and interquartile range (IQR) for skewed distributions, while categorical variables were presented as frequency (percentage). Continuous variables were compared using Student’s t-test when normally distributed and the Wilcoxon rank-sum test otherwise. Pearson’s chi-squared test or Fisher’s exact test was used for categorical data, as appropriate. Missing data were handled using K-nearest neighbour (KNN) imputation, a widely accepted nonparametric approach [22]. A sensitivity analysis was performed after excluding patients with missing data to confirm the robustness of the results. Binary and multivariable logistic regression analyses were utilized to explore the association between nICT and the primary and secondary outcomes, using an extended logistic model to adjust for various covariates. To balance baseline characteristics between the nICT and nCT groups, we employed propensity score matching (PSM) with a 1:1 nearest neighbour matching algorithm and a calliper width of 0.2 [23]. All variables listed in Table 1 were used to estimate the propensity score. The discrimination of the PSM was assessed using the c-statistic, expressed as the area under the receiver operating characteristic curve (AUC). Covariate balance after matching was assessed using the standardized mean difference (SMD), with SMD < 0.10 indicating adequate balance. The primary outcome was further verified using the inverse probability of treatment weighting (IPTW), with estimated propensity scores as weights [24,25]. Subgroup analyses were conducted based on relevant covariates. These analyses were executed using R version 3.3.2 and Free Statistics software version 1.9. Statistical significance was determined by a two-tailed test with a p-value threshold of less than 0.05. A preprint version of this manuscript has been previously posted on Research Square (doi:10.21203/rs.3.rs-8867323/v1) [26].

Table 1.

Baseline characteristics of the included patients.

	Before PSM				After PSM
Variables	Total (n = 327)	nCT (n = 160)	nICT (n = 167)	SMD	Total (n = 200)	nCT (n = 100)	nICT (n = 100)	SMD
Age, years, mean ± SD	57.9 ± 9.3	56.0 ± 10.5	59.8 ± 7.6	0.416	58.5 ± 7.8	58.6 ± 8.1	58.5 ± 7.6	0.014
Sex, n (%)				0.132				0.031
Male	291 (89.0)	139 (86.9)	152 (91)		177 (88.5)	88 (88)	89 (89)
Female	36 (11.0)	21 (13.1)	15 (9)		23 (11.5)	12 (12)	11 (11)
Smoking history, n (%)	193 (59.0)	84 (52.5)	109 (65.3)	0.262	120 (60.0)	61 (61)	59 (59)	0.041
Alcohol history, n (%)	120 (36.7)	50 (31.2)	70 (41.9)	0.223	57 (28.5)	29 (29)	28 (28)	0.039
Comorbidity
Diabetes, n (%)	30 (9.2)	11 (6.9)	19 (11.4)	0.157	16 (8.0)	8 (8)	8 (8)	<0.001
Hypertention, n (%)	71 (21.7)	28 (17.5)	43 (25.7)	0.201	40 (20.0)	19 (19)	21 (21)	0.05
Coronary artery disease, n (%)	13 (4.0)	4 (2.5)	9 (5.4)	0.149	7 (3.5)	4 (4)	3 (3)	0.054
Disease type, n (%)				0.644				0.085
Lung	209 (63.9)	81 (50.6)	128 (76.6)		132 (66.0)	64 (64)	68 (68)
Oesophagus	99 (30.3)	61 (38.1)	38 (22.8)		66 (33.0)	35 (35)	31 (31)
Mediastinum	19 (5.8)	18 (11.2)	1 (0.6)		2 (1.0)	1 (1)	1 (1)
BMI, kg/m², mean ± SD	23.6 ± 3.0	23.3 ± 3.1	23.9 ± 2.9	0.179	23.6 ± 3.1	23.7 ± 3.5	23.5 ± 2.7	0.042
ALB, g/L, mean ± SD	39.2 ± 3.4	38.8 ± 3.4	39.7 ± 3.3	0.256	39.3 ± 3.5	39.3 ± 3.4	39.3 ± 3.6	0.004
Creatinine, mmol/L, mean ± SD	78.0 ± 18.1	74.0 ± 15.7	81.8 ± 19.4	0.441	76.4 ± 15.0	76.3 ± 16.3	76.6 ± 13.7	0.021
Pulmonary function
FEV1/FVC%, Mean ± SD	75.4 ± 10.0	76.0 ± 9.1	74.8 ± 10.7	0.12	74.9 ± 9.8	75.0 ± 9.0	74.8 ± 10.6	0.017
FEV1%pre, Mean ± SD	93.6 ± 18.1	96.0 ± 17.3	91.4 ± 18.6	0.256	94.4 ± 18.4	94.8 ± 17.1	93.9 ± 19.7	0.049
Cycle number*, mean ± SD	2.9 ± 1.2	2.7 ± 1.4	3.0 ± 1.0	0.239	2.8 ± 0.9	2.7 ± 1.0	2.8 ± 0.8	0.096
TTS#, days, median (IQR)	44.0 (37.0, 55.0)	42.5 (37.0, 52.0)	45.0 (37.0, 59.0)	0.08	43.0 (37.0, 54.0)	42.0 (37.0, 50.0)	44.0 (36.8, 56.0)	0.089
Surgical approach, n (%)				0.342				0.045
Open	95 (29.1)	59 (36.9)	36 (21.6)		56 (28.0)	29 (29)	27 (27)
MIS	232 (70.9)	101 (63.1)	131 (78.4)		144 (72.0)	71 (71)	73 (73)
Pathology type, n (%)				0.136				0.066
SCC	233 (71.3)	109 (68.1)	124 (74.3)		141 (70.5)	72 (72)	69 (69)
Non-SCC	94 (28.7)	51 (31.9)	43 (25.7)		59 (29.5)	28 (28)	31 (31)

ALB, serum albumin; BMI, body mass index; FVC, forced vital capacity; FEV1, forced expiratory volume in one second; FEV1%pre, FEV1% predicted; MIS, minimally invasive surgery; nCT, neoadjuvant chemotherapy; nICT, neoadjuvant immunochemotherapy; SCC, squamous cell carcinoma.

*Cycle number refers to neoadjuvant cycles administered prior to surgery: in the nICT group, this denotes combined concurrent cycles of chemotherapy + PD-1 inhibitor; in the nCT group, chemotherapy cycles only.

^#TTS refers to calendar days from the last neoadjuvant treatment to the date of surgery – for the nICT group, the last PD-1 inhibitor dose; for the nCT group, the last chemotherapy cycle.

^$The distribution of PD-1 agents was: pembrolizumab (n = 62), tislelizumab (n = 43), sintilimab (n = 28), camrelizumab (n = 16), nivolumab (n = 8) and toripalimab (n = 10).

Results

Patient characteristics

A total of 327 patients with thoracic cancers were finally included in our analysis, comprising 167 patients (51.1%) with a history of nICT and 160 (48.9%) with a history of nCT (Table 1). After PSM, 200 patients were successfully matched. The mean age at surgery was 57.9 ± 9.3 years, with the vast majority (89.0%) being male. The disease type distribution included lung (209 patients, 63.9%), oesophagus (99 patients, 30.3%), and mediastinum (19 patients, 5.8%) cancers. Patients who received nICT predominantly had lung cancer, higher smoking and alcohol history, higher serum albumin (ALB) and creatinine levels before surgery, lower FEV1% per, and less open surgery compared to those receiving nCT. The baseline characteristics of the two groups after PSM were balanced (Table 1).

Primary outcome

The total postoperative AHRF rate within the cohort was 14.1% (Supplemental Table 1). The PSM showed acceptable discrimination, with an AUC of 0.771 (Supplemental Figure 2). Among thoracic cancer patients, those with a history of nICT had a significantly higher postoperative AHRF rate of 19.8% compared to that of 8.1% for those with a history of nCT (p = 0.002). Initial univariable logistic regression analysis yielded an odds ratio (OR) of 2.78 (95% confidence interval [CI]: 1.41–5.51, p = 0.003, Table 2). Upon adjusting for clinically pertinent covariates, OR was still as high as 2.5 (95% CI: 1.13–5.51, p = 0.024) across multiple extended multivariable logistic regression models (Table 2). Similar to the results in the pre-matched cohort, both PSM (OR = 5.22, 95% CI: 2.04–13.39, p = 0.001) and IPTW (OR = 2.41, 95% CI: 1.2–4.82, p = 0.013) indicated that neoadjuvant immunotherapy exposure was independently associated with an increased postoperative AHRF rate.

Table 2.

Association between neoadjuvant immunotherapy and postoperative AHRF rate using PSM.

Models	OR (95%CI)	p value
Unmatched crude	2.78 (1.41 ∼ 5.51)	0.003
Multivariable adjusteda	2.5 (1.13 ∼ 5.51)	0.024
Propensity Score matchedb	5.22 (2.04 ∼ 13.39)	0.001
Propensity Score adjustedc	2.34 (1.1 ∼ 5)	0.028
Weighted IPTWd	2.41 (1.2 ∼ 4.82)	0.013

a, odds ratio from the multivariable logistic proportional model adjusted for all covariates (Table 1). b, odds ratio from a multivariate logistic proportional hazards model with the same strata and covariates matched according to the propensity score. The analysis included 200 patients (100 who received nCT and 100 who received nICT). c, odds ratio from a multivariable logistic proportional hazards model with the same strata and covariates, with additional adjustment for the propensity score. d, Primary analysis with a hazard ratio from the multivariable logistic proportional hazards model with the same strata and covariates with inverse probability weighting according to the propensity score.

Secondary outcomes

The secondary outcomes assessed included length of hospital stay, ICU stay, and postoperative stay, as presented in Supplemental Table 1. Among patients with a history of nICT, the length of hospital stay was shorter compared to those who had undergone nCT. In the propensity score-matched cohort, there were no statistically significant differences in overall hospital length of stay, ICU length of stay or postoperative length of stay between patients receiving nICT and those receiving chemotherapy alone. Multivariable logistic regression analysis for all secondary outcomes was detailed in Supplemental Table 2. Upon thorough model adjustment, we observed no significant differences in secondary outcomes among patients with a history of nICT compared to nCT group, including length of hospital stay, length of ICU stay or length of postoperative stay.

Postoperative mortality

Overall mortality was low, with 1 death observed among 327 patients (0.3%). This death occurred in a patient with AHRF but was attributed to sepsis with multiple organ failure rather than respiratory failure itself. No deaths were directly attributable to AHRF.

Subgroup analyses

Subgroup analyses, accounting for various confounders, consistently demonstrated that, compared to nCT, patients who received their final immunochemotherapy within 42 days prior to thoracic surgery had a relatively higher risk of postoperative AHRF (OR = 6.68, 95% CI: 1.24–35.98) than those with an interval of ≥42 days (Figure 2). Similarly, squamous cell carcinoma (SCC) patients and non-pCR thoracic cancer cases constituted a significantly high-risk subgroup for postoperative AHRF (OR = 3.64, 95% CI: 1.41–9.44; OR = 2.82, 95% CI: 1.14–6.98; respectively). No significant interactions were observed between subgroups (p for interaction > 0.05).

Figure 2.

Subgroup analysis of the associations between nICT and postoperative AHRF. Each stratification was adjusted for all confounders shown in Table 1, except for the stratification factor itself.

Factorial analysis

We further compared patients with (n = 46) and without AHRF (n = 281) with the aim of identifying factors related to postoperative AHRF in patients undergoing thoracic surgeries. As shown in Table 3, patients who developed postoperative AHRF had a higher history of receiving nICT, and higher rate of MIS surgery. The cycle number and TTS between the last therapy and surgery were not significantly different between the two groups. Patients experiencing postoperative AHRF typically had longer length of hospital stay, longer length of ICU stay and longer length of postoperative stay (all p < 0.001) compared to patients without AHRF.

Table 3.

Characteristics of postoperative AHRF and non-AHRF patients.

Variables	Total (n = 327)	Non-AHRF (n = 281)	AHRF (n = 46)	p value
Age, years, mean ± SD	57.9 ± 9.3	57.5 ± 9.4	60.1 ± 8.7	0.079
Sex, n (%)				0.119
Male	291 (89.0)	247 (87.9)	44 (95.7)
Female	36 (11.0)	34 (12.1)	2 (4.3)
Smoking history, n (%)	193 (59.0)	160 (56.9)	33 (71.7)	0.058
Alcohol history, n (%)	120 (36.7)	98 (34.9)	22 (47.8)	0.091
Comorbidity
Diabetes, n (%)	30 (9.2)	22 (7.8)	8 (17.4)	0.051
Hypertention, n (%)	71 (21.7)	59 (21)	12 (26.1)	0.438
Coronary artery disease, n (%)	13 (4.0)	12 (4.3)	1 (2.2)	1
Disease type, n (%)				0.352
Lung	209 (63.9)	183 (65.1)	26 (56.5)
Oesophagus	99 (30.3)	81 (28.8)	18 (39.1)
Mediastinum	19 (5.8)	17 (6)	2 (4.3)
BMI, kg/m², mean ± SD	23.6 ± 3.0	23.5 ± 3.1	23.9 ± 2.6	0.433
ALB, g/L, mean ± SD	39.2 ± 3.4	39.2 ± 3.3	39.7 ± 3.6	0.357
Creatinine, mmol/L, mean ± SD	78.0 ± 18.1	77.2 ± 18.0	82.3 ± 18.3	0.078
Pulmonary function
FEV1/FVC%, mean ± SD	75.4 ± 10.0	75.7 ± 9.9	73.9 ± 10.3	0.258
FEV1%pre, mean ± SD	93.6 ± 18.1	94.4 ± 17.7	89.0 ± 19.8	0.057
nICT, n (%)	167 (51.1)	134 (47.7)	33 (71.7)	0.002
Cycle number, mean ± SD	2.9 ± 1.2	2.9 ± 1.2	2.9 ± 0.9	0.83
TTS, days, median (IQR)	44.0 (37.0, 55.0)	44.0 (37.0, 55.0)	44.0 (37.2, 55.8)	0.535
Surgical approach, n (%)				0.026
Open	95 (29.1)	88 (31.3)	7 (15.2)
MIS	232 (70.9)	193 (68.7)	39 (84.8)
Pathology type, n (%)				0.138
SCC	233 (71.3)	196 (69.8)	37 (80.4)
Non-SCC	94 (28.7)	85 (30.2)	9 (19.6)
pCR, n (%)	72 (22.0)	58 (20.6)	14 (30.4)	0.137
Outcomes
Length of hospital stay, mean ± SD	12.6 ± 5.5	12.1 ± 5.1	15.4 ± 7.0	< 0.001
Length of ICU stay, median (IQR)	2.0 (1.0, 3.0)	2.0 (1.0, 2.0)	4.0 (3.0, 6.0)	< 0.001
Length of postoperative stay, median (IQR)	7.0 (4.0, 8.0)	6.0 (4.0, 8.0)	8.0 (7.0, 10.8)	< 0.001

AHRF, acute hypoxemic respiratory failure; ALB, serum albumin; BMI, body mass index; FVC, forced vital capacity; FEV1, forced expiratory volume in one second; FEV1%pre, FEV1% predicted; MIS, minimally invasive surgery; nICT, neoadjuvant immunochemotherapy; SCC, squamous cell carcinoma; TTS, time interval between neoadjuvant therapy and surgery; pCR, pathological complete response.

Analysis of NSCLC patients

Because NSCLC patients constituted the majority of participants, we further conducted a sensitivity analysis in NSCLC patients. The baseline characteristics and outcomes of these patients are shown in Supplemental Table 3. A total of 209 patients with NSCLC were enrolled, of whom 128 (61.2%) had a history of nICT, and 81 (38.8%) had a history of nCT. The mean age at surgery was 58.3 ± 7.9 years, with the vast majority (89%) being male. Patients with NSCLC who received nICT were older, and had more alcohol history, higher creatinine levels before surgery, longer TTS, more MIS and higher postoperative AHRF rate compared to those receiving nCT (all p < 0.05). Variables that exhibited a p-value of less than 0.1 in univariate analyses were chosen for multivariate adjustment (Supplemental Table 4). The binary logistic regression analysis showed that the history of nICT remained associated with postoperative AHRF (OR = 4.12, 95% CI: 1.15–14.8, p = 0.03) in patients with NSCLC (Table 4). However, for the secondary outcomes, including length of hospital stay, length of ICU stay and length of postoperative stay, no significant association between a history of nICT and the secondary outcome were observed in patients with NSCLC.

Table 4.

The relationship between neoadjuvant immunotherapy and outcomes in patients with NSCLC.

Outcome	Crude Coefficient (95%CI)	Crude p value	Adjusted Coefficient (95%CI)	Adjusted p value
Primary outcome
AHRF, n (%)	5.7 (1.65 ∼ 19.65)	0.006	4.12 (1.15 ∼ 14.8)	0.03
Secondary outcomes
Length of hospital stay, d	−0.48 (−1.97 ∼ 1.02)	0.533	−0.3 (−1.87 ∼ 1.27)	0.711
Length of ICU stay, d	0.54 (−0.12 ∼ 1.21)	0.111	0.43 (−0.26 ∼ 1.11)	0.227
Length of postoperative stay, d	0.69 (−0.39 ∼ 1.77)	0.213	0.6 (−0.51 ∼ 1.72)	0.289

AHRF, acute hypoxemic respiratory failure; ICU, intensive care unit.

Adjust variables from univariate analyses where the p-value was less than 0.1.

An increased risk of AHRF associated with nICT exposure in patients with NSCLC was consistently observed across key subgroups, including those younger than 65 years, those with an interval of less than 42 days between the last dose of neoadjuvant immunotherapy and surgery, and those with SCC (Supplemental Figure 1).

Sensitive analysis

Approximately 5% of relevant variables in the dataset were missing, which were pulmonary function variables. Patients with incomplete data were excluded from both the overall cohort and the NSCLC subgroup. A sensitivity analysis was subsequently performed. Consistently, nICT was significantly associated with an increased risk of postoperative AHRF in both the overall cohort and the NSCLC subgroup after exclusion of missing data (OR = 2.67, 95% CI: 1.16–6.13, p = 0.021; OR = 6.08, 95% CI: 1.33–27.8, p = 0.02, respectively; Supplemental Table 5). No statistically significant differences were observed between groups for any of the secondary outcomes.

Discussion

This retrospective cohort study demonstrated that nICT was associated with an increased risk of postoperative AHRF compared with nCT, an association that remained consistent across multiple analytic approaches, including propensity score matching. Similar patterns were observed in the NSCLC subgroup. Importantly, subgroup analyses further identified higher-risk populations, including patients with treatment-to-surgery intervals <42 days, squamous cell carcinoma, non-pCR status and younger age (<65 years) in NSCLC.

Beyond statistical significance, the observed associations in this study demonstrate clear clinical relevance. In addition to relative effect estimates, absolute risk differences were considered to enhance interpretability. Specifically, the incidence of postoperative AHRF increased from 8.1% in the nCT group to 19.8% in the nICT group, corresponding to an absolute risk increase of 11.7%. This magnitude of risk elevation is clinically meaningful in the context of thoracic surgery, where postoperative respiratory failure is associated with substantial morbidity, ICU utilization and perioperative resource burden. Consistently, the observed effect sizes (e.g. OR 4.12 in NSCLC patients and OR 6.68 in patients with shorter treatment-to-surgery intervals) indicate a marked increase in risk, supporting the robustness and clinical significance of our findings. Taken together, these results suggest that the impact of neoadjuvant immunochemotherapy on postoperative pulmonary outcomes is not only statistically significant but also clinically consequential, warranting heightened perioperative vigilance and individualized risk stratification. From a clinical perspective, this translates into approximately one additional AHRF event for every 8–9 patients treated with nICT. Although minimal clinically important difference (MCID) [27] frameworks for effect sizes such as odds ratios have been proposed, their application in perioperative surgical research remains limited. In this context, the magnitude of effect observed in our study provides a pragmatic and clinically interpretable estimate of risk.

Over the past 5 years, more than 10 randomized controlled trials (RCTs) have investigated the impact of neoadjuvant immunotherapy or immunochemotherapy on prognosis [6,14–17,28–41]. As ICIs continue to transform the neoadjuvant treatment, the investigation of perioperative complications such as AHRF is both timely and highly relevant. Perioperative trials such as CheckMate-816 and KEYNOTE-671 primarily report mortality rather than AHRF. In our cohort, overall mortality was low (1/327, 0.3%), and no deaths were directly attributable to AHRF. The single fatality occurred in a patient with AHRF but was ultimately due to sepsis with multiple organ failure rather than respiratory failure itself. In comparison, the nICT arms of CheckMate-816 and KEYNOTE-671 reported 5/179 and 4/97 deaths, respectively, including pulmonary-related fatalities [14,15]. This discrepancy may reflect differences in patient selection, perioperative management and the controlled settings of clinical trials versus real-world practice. Notably, while randomized trials primarily report mortality outcomes, AHRF as a distinct postoperative complication is less frequently captured. Our findings therefore provide complementary real-world evidence, highlighting AHRF as a clinically relevant endpoint that may not be fully reflected by mortality alone.

Postoperative hypoxemic respiratory failure is variably defined across studies, which likely contributes to the wide range of reported incidences. In a 211-patient NSCLC cohort undergoing thoracic surgery after nCT, hARF was defined by PaO₂/FiO₂ ≤200 mmHg requiring noninvasive positive pressure ventilation or invasive mechanical ventilation for ≥24 h, yielding an incidence of 5.2% [8]. In a minimally invasive oesophagectomy cohort, postoperative respiratory failure was defined using a more stringent endpoint – unplanned reintubation or tracheostomy, or delayed extubation (≥48 h) – resulting in a reported incidence of 2.4% [42]. Some studies have identified postoperative respiratory failure using discharge diagnosis codes from administrative databases, an approach that may lead to underascertainment of true positive cases [43]. In contrast, in our study, all cases were retrospectively adjudicated by two independent physicians based on detailed clinical records. This strategy allows for more accurate capture of clinically relevant events, reduces misclassification bias and ensures greater diagnostic consistency, particularly for cases that may not be fully reflected by diagnostic coding alone.

In this study, the incidence of AHRF in the nCT group was 8.1%, which reduced to 6% after PSM. Adoption of the updated AHRF definition, which included patients receiving HFNC oxygen therapy [44], was one reason for elevated total AHRF rates in our study. While prior studies identified advanced age, male sex and impaired pulmonary function as risk factors for postoperative respiratory failure [42,43], our analysis revealed a markedly higher AHRF incidence of 19.8% in thoracic tumour patients receiving nICT. Both IPTW and multivariate regression analyses confirmed a significant association between nICT and increased postoperative AHRF risk. Postoperative stress and compromised innate and adaptive immune responses may be potential factors contributing to perioperative complications [36,45]. Crucially, our findings suggest that exposure to immunotherapy may represent an important factor associated with increased postoperative AHRF risk, though its specific pathophysiological mechanisms warrant further investigation.

Existing studies have demonstrated that postoperative AHRF increases mortality and hospital stay duration [43,46,47]. Intriguingly, our findings revealed that patients who received nICT had a significantly higher incidence of AHRF, but markedly shorter hospital length of stay compared to the nCT group. This temporal disparity likely reflects the study’s retrospective design, with nICT implemented more recently compared to earlier nCT cases. Advancing perioperative care, including ERAS [48,49] protocols and surgical expertise, contributed to reduced hospitalization in later period for nICT patients. The higher MIS proportion among AHRF cases likely reflects calendar-time, as nICT – where AHRF clustered – was performed during periods of greater MIS uptake and ERAS implementation. After adjusting for surgical approach (and extent of resection in NSCLC), the nICT–AHRF association remained higher than nCT. The persistently elevated AHRF incidence in the nICT group despite these improvements suggests an intrinsic association between immunotherapy and postoperative AHRF in thoracic oncology.

Subgroup analyses revealed no association between postoperative AHRF incidence and neoadjuvant therapy cycles. While prior studies report 1 to 4 neoadjuvant immunotherapy exposure cycles (typically 3 to 4 cycles) [14,17,36,37], our subgroup analysis similarly demonstrated no cycle-dependent AHRF risk variation. However, intervals of less than 42 days between the last immunotherapy and surgery were associated with a higher incidence of AHRF, suggesting that shorter TTS may represent a potential risk factor requiring careful perioperative consideration. Current literature on nICT presents limited data concerning the TTS, with durations reported as 2 to 6 weeks [14,17,50]. Interestingly, subgroup analysis revealed that non-pCR patients exhibited increased AHRF risk, indicating that tumour response and immune activation may jointly influence postoperative outcomes. These findings highlight the importance of balancing oncologic efficacy with perioperative safety when planning surgical timing. Prospective studies are needed to further clarify the impact of treatment-to-surgery intervals on perioperative outcomes and long-term survival.

These findings have direct implications for clinical practice. Identification of higher-risk subgroups may support individualized perioperative planning, including careful timing of surgery, optimized pulmonary management and vigilant postoperative monitoring [49]. While specific biomarkers were not available in this study, emerging candidates such as PD-L1 expression and circulating cytokines may further refine risk stratification in future studies. Importantly, our results should be interpreted as hypothesis-generating and not prescriptive, pending prospective validation.

The underlying mechanism may involve PD-1 inhibitor-mediated preoperative immune activation induces bystander lung injury via immune checkpoints derepression, CD8⁺ effector T cells activation and elevated secretion of pro-inflammatory cytokines such as IL-2 and CXCL10 [51–53]. Surgical stressors, including mechanical ventilation, ischemia, amplify pulmonary infiltration of activated T-cells and inflammatory mediators, triggering systemic inflammatory response syndrome (SIRS) and compounding AHRF risk [54]. We propose a hypothesis-generating ‘dual-hit’ model, in which preoperative PD-1 blockade primes pulmonary immune responses (the ‘first hit’), and surgical trauma amplifies this activation (the ‘second hit’), potentially precipitating AHRF (Figure 3).

Figure 3.

Proposed mechanism linking nICT to postoperative AHRF.

Studies show elderly patients (≥ 65 years) receiving ICIs exhibit lower naïve cytotoxic (TcN) and helper T cells (ThN), B cells, and double-negative T cells (DNT), with attenuated cytokine responses and upregulated PD-1 expression [55], potentially mitigating systemic inflammation and thus lowering their susceptibility to AHRF. Shorter immunotherapy-to-surgery intervals (< 42 days) correlated with higher AHRF rates, likely reflecting residual PD-1 monoclonal antibodies activity from prolonged half-life [56]. Thus, these patients might still be in an actively heightened or subclinically inflamed immunologic state during surgery. Additionally, Non-pCR patients showed elevated AHRF risk despite residual tumours, possibly associated with elevated baseline T-lymphocyte levels and immune activation-related receptors [57,58], thereby predisposing them to exaggerated postoperative inflammatory cytokine release and enhanced AHRF susceptibility. Combined mechanisms, including preoperative immune activation, inflammatory mediator accumulation, lymphocyte subset imbalance and distinctive immune profiles in certain subgroups (shorter therapeutic interval, SCC, non-pCR, and patients with NSCLC who are younger than 65 years), contribute to increased AHRF risk following nICT.

Our study has several strengths. The findings are significant for perioperative practice. The retrospective two-centre design encompassing a relatively large sample size with application of multiple statistical methodologies, reflects considerable effort to reduce confounding in observational data. Furthermore, the inclusion of sensitivity analyses, subgroup analyses, and a focused assessment on NSCLC subpopulations strengthens the manuscript’s internal validity.

However, several limitations warrant consideration. Firstly, its retrospective nature introduces potential selection and ascertainment biases. Secondly, this was a two-centre study with the modest sample size, limiting the generalizability of the results. Thirdly, this study included patients with NSCLC, oesophageal cancer and malignant mediastinal tumours. While we pooled thoracic tumours in the main analysis, we justified this by their shared exposure to ICIs, common surgical setting and AHRF as a unified inflammatory endpoint. Although separate analysis within the NSCLC patient cohort yielded consistent results, the inclusion of patients with NSCLC, oesophageal carcinoma and mediastinal malignancies in the overall study population may constitute a potential source of heterogeneity. Further studies are needed within disease-specific or tumour-type–specific or tumour-staging-matched contexts to validate these findings. Fourthly, while our data suggests immune activation may contribute to AHRF, the absence of biomarker measurements (e.g. IL-6, CD8+ T cells) limits mechanistic conclusions. Future studies should correlate immunological profiles with AHRF risk. Fifthly, our male-predominant, Chinese cohort may limit generalizability across sexes and regions. Despite adjustment for sex, potential sex-specific and regional differences warrant external validation in multi-centre, multi-ethnic, more female-enriched cohorts.

Conclusion

nICT was associated with an increased risk of postoperative AHRF risk compared to nCT in thoracic oncology patients, necessitating rigorous perioperative monitoring. Notably, patients who received preoperative immunotherapy within 42 days, SCC and individuals with non-pCR thoracic malignancies, and patients with NSCLC who are younger than 65 years represented subgroups with a significantly increased risk of AHRF following nICT.

Glossary

Abbreviations

AHRF

acute hypoxemic respiratory failure

ALB

serum albumin

BMI

body mass index

FEV1

forced expiratory volume in one second

FEV1%pre

FEV1% predicted

FVC

forced vital capacity

HFNC

high-flow nasal cannula

ICIs

immune checkpoint inhibitors

ICU

intensive care unit

IQR

interquartile range

KNN

K-nearest neighbour

MIS

minimally invasive surgery

nCT

neoadjuvant chemotherapy

nICT

neoadjuvant immunochemotherapy

NSCLC

non-small cell lung cancer

pCR

pathological complete response

SCC

squamous cell carcinoma

trAE

treatment-related adverse event

TTS

time interval between neoadjuvant therapy and surgery

Funding Statement

This study was supported by the National Natural Science Foundation of China (No. 82302449), the Guangxi Natural Science Foundation (No. 2024GXNSFAA010047), the Scientific Research Launch Project for New Employees of the Second Xiangya Hospital of Central South University, the Health Research Project of Hunan Provincial Health Commission (No. W20243115), the Natural Science Foundation of Hunan Province (No. 2023JJ60081, 2024JJ9204, 2026JJ60080) and the Project of Hunan Provincial Administration of Traditional Chinese Medicine (No. B2023062).

Declaration of generative AI and AI-assisted technologies in the writing process

During the preparation of this work the authors used ChatGPT in order to enhance the clarity, language and readability of the text. After using this service, the authors reviewed and edited the content as needed and took full responsibility for the content of the publication.

Disclosure statement

The authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest, and none was reported.

Ethical approval

This study was approved by the institutional review boards (IRB) of the Second Xiangya Hospital of Central South University (No. LYF2023105) and Guilin Hospital of the Second Xiangya Hospital (No. LLkt-034) in Aug. 2023. The study was conducted in accordance with the ethical principles of the Declaration of Helsinki of the World Medical Association. Given the retrospective nature of the study, the requirement for informed consent was waived by the IRB.

Data availability statement

The data supporting this research are available from the authors on reasonable request.