Air-Test for prediction of postoperative pulmonary complications after general anaesthesia: promise, performance and limitations Authors: Kate Gallagher, Itay Bentov Published: 2026 Background and clinical relevance Postoperative pulmonary complications (PPCs) are a major clinical and financial burden, and are associated with prolonged hospital stay, increased healthcare costs, and worse postoperative outcomes (1-3). Even minor PPCs lead to increased perioperative morbidity and mortality (4). Early identification of patients who are at risk can lead to interventions that would reduce this burden or may lead to better allocation of postoperative resources (5). ARISCAT remains the best-established PPC risk score (2,6). External validation in the PERISCOPE study showed generally good discriminative performance, although calibration varied across geographic regions and specialised populations (2,6). A 2025 systematic review reported that ARISCAT remains among the most consistently validated and best-performing tools (2). The LAS VEGAS score demonstrated moderate predictive accuracy in derivation and validation cohorts, although wider external validation was recommended before routine clinical implementation (7). Although established PPC prediction models are available, they require multiple variables and may be less practical in fast-paced clinical settings (6,7). A simple bedside screening tool capable of rapidly identifying at risk patients would therefore be valuable. The Air-Test concept has been developed through a series of prior studies. Its diagnostic accuracy was examined against computed tomography (CT)-confirmed atelectasis in healthy patients after surgery, demonstrating an area under the receiver operating characteristic curve (ROC-AUC) of 0.90 with a sensitivity and specificity over 80% (8). A feasibility study in laparoscopic surgery showed that room-air oxygen saturation (SpO2) could help individualize lung recruitment and identify opening and closing pressures (9). In thoracic surgery patients undergoing one-lung ventilation, the broader iPROVE-OLV programme extended this line of research into perioperative management (10). Recently, a prospective laparoscopic surgery study of 250 patients found that a positive postoperative Air-Test was independently associated with PPCs, but discrimination was modest (AUC 0.66) (11). The key unanswered question before this study was whether the Air-Test, promising in a selected abdominal surgery cohort, would maintain performance in a broader surgical population. Marrero et al. tested whether the Air-Test could predict severe PPCs within 7 days of surgery in a mixed surgical population undergoing general anaesthesia (12). The Air-Test measures SpO2 by pulse oximetry on room-air pre- and postoperatively, using a binary cutoff (positive test SpO2 ≤96%). The study was a secondary analysis of a harmonised dataset combining three multicentre randomised controlled trials and one prospective observational study, including 3,268 patients. The cohort was randomly divided into training (n=2,288) and validation (n=980) groups, with severe PPCs occurring within 7 postoperative days as the primary outcome (12). Performance in the current study was substantially weaker than in the earlier iPROVE derivation study (Table 1), where the simplified Air-Test achieved an AUC of 0.72 and the full five-variable model reached 0.79 in intermediate-to-high-risk abdominal surgery patients (13). Table 1Comparison of the iPROVE derivation study and the current Marrero et al. analysisFeatureiPROVE derivation studyCurrent Marrero studyPopulationIntermediate-to-high risk abdominal surgery patientsHeterogeneous surgical population undergoing general anaesthesiaPPC outcomeModerate-to-severe PPCsSevere PPCs onlyAir-Test classificationGraded hypoxaemia categoriesBinary classificationPerformanceAUC 0.72–0.79AUC 0.57ValidationRandomised derivation and validation cohorts from the original study populationInternal split-sample validation within a harmonised pooled datasetAUC, area under the curve; PPC, postoperative pulmonary complication. Comparison of the iPROVE derivation study and the current Marrero et al. analysis AUC, area under the curve; PPC, postoperative pulmonary complication. Several factors likely contributed to the decline in performance, including a more varied population, a lower event rate (5%), and the replacement of graded hypoxaemia categories with a single binary cutoff, which reduced the test’s ability to capture meaningful differences between patients (12,14). This pattern of stronger performance in derivation cohorts followed by attenuation during wider testing is common in prediction-model research (14,15). Strengths of the study include the large multicentre dataset, internal validation cohort, and the Air-Test itself, which requires only room-air pulse oximetry. However, important limitations include the study design, as this is an internal split-sample analysis within a pooled dataset (14). Furthermore, there is potential heterogeneity between source studies, absence of calibration assessment or plots, and no decision-curve analysis to assess clinical utility/net benefit across plausible threshold probabilities (16). Compared with existing PPC prediction models (Table 2) such as ARISCAT and LAS VEGAS, the Air-Test demonstrated substantially weaker predictive accuracy (6,7,12). Notably, even ARISCAT, with AUC of 0.76–0.80 in general surgical cohorts, has shown reduced performance in certain specialised populations, highlighting the challenge of maintaining prediction-model performance across diverse clinical settings (2,6). Table 2Comparison of PPC prediction toolsToolStudy population and settingDiscrimination (AUC)Main strengthMain limitationARISCATMulticentre European surgical cohorts; pre-operative assessments0.76–0.80Most extensively validated PPC risk scoreReduced performance in some specialised populationsLAS VEGASInternational surgical cohorts; intraoperative assessment0.72–0.78Includes ventilation-related variablesMore complex intraoperative assessmentAir-Test (iPROVE)High-risk abdominal surgery cohorts; PACU0.72–0.79Simple, rapid bedside assessmentDerived in a narrower high-risk populationAir-Test (Marrero)Heterogeneous surgical cohorts; PACU0.57Simple, rapid bedside assessmentPoor discrimination in a heterogeneous populationAUC, area under the curve; PACU, post-anaesthesia care unit; PPC, postoperative pulmonary complication. Comparison of PPC prediction tools AUC, area under the curve; PACU, post-anaesthesia care unit; PPC, postoperative pulmonary complication. Together, these findings suggest that single-parameter tools such as the Air-Test may have limited predictive value when applied across broad surgical populations (12,14). Conclusion The Air-Test is attractive because it is simple, rapid, and requires only room-air pulse oximetry. However, in this heterogeneous surgical cohort, its ability to discriminate severe PPCs was poor [AUC 0.57; 95% confidence interval (CI): 0.50–0.64]. Although the NPV was high, this is likely influenced by the low event rate rather than by strong intrinsic test performance. Overall, these findings do not support the Air-Test as a standalone replacement for established PPC risk scores. It may still provide value as a rapid adjunct to existing risk-stratification tools, particularly in postoperative settings such as post-anaesthesia care unit (PACU) triage or bedside reassessment. Importantly, reducing PPCs depends on multimodal perioperative care from preoperative optimisation through postoperative respiratory support (12). Supplementary The article’s supplementary files as Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1731/prf