Biological and Diagnostic Reclassification of Post-Oesophagectomy Pneumonia and Pulmonary Complications

Abstract

Background: Post-operative pneumonia represents the most common major complication following oesophagectomy, occurring in 14-25% of patients and significantly impacting mortality, morbidity and healthcare costs. Despite its clinical importance, substantial heterogeneity exists in pneumonia definitions across studies, with only 14.3% of randomised trials using validated diagnostic criteria. The fundamental question of what constitutes pneumonia in this surgical population remains inadequately addressed, with diagnostic disagreement among experienced clinicians suggesting underlying biological complexity rather than methodological failure. Aims: This thesis aimed to demonstrate that diagnostic unreliability in post-operative pneumonia reflects genuine biological heterogeneity rather than inadequate clinical assessment, through four complementary investigations: (1) a systematic review documenting definitional inconsistency across intervention trials, (2) quantifying inter-observer variability among specialist clinicians in chest Xray interpretation, (3) developing a trimodal score utilising routinely collected patient parameters to categorise mild to severe risk stratification, and (4) characterising the biological entities underpinning diagnosed pneumonia cases. Methods: A systematic review of randomised controlled trials examined interventions targeting pulmonary complications after oesophagectomy, documenting diagnostic criteria consistency and microbiological confirmation rates. An inter-observer variability study assessed diagnostic agreement among four specialists (surgeon, intensivist, pulmonologist and radiologist) reviewing 200 post-operative chest radiographs, with and without standardised clinical correlation. A trimodal clinical score integrating CRP/albumin ratio, maximum FiO₂ requirement, and PaO₂/FiO₂ ratio was developed and validated in 234 patients against multiple established pneumonia classification systems. Biological characterisation employed procalcitonin-guided entity classification with comprehensive cytokine profiling of bilateral bronchoalveolar lavage fluid and serial serum samples from 43 patients, correlating biomarker patterns with microbiological findings and clinical outcomes. Results: The systematic review of 14 trials involving 1,369 patients revealed that only 14.3% employed validated diagnostic criteria, with pneumonia incidence varying from 10% to 40% among comparable populations. Microbiological confirmation was obtained in merely 30% to 40% of cases diagnosed clinically, indicating that the majority of complications are attributable to non-infectious processes. Inter-observer agreement was rated as only fair (κ = 0.270), with complete agreement observed in only 44% of cases—97.7% of which concurred on the absence of pneumonia. The application of standardized clinical correlation enhanced agreement by 56% (κ = 0.421, p<0.001), with specific discriminators such as temperature exceeding 38°C, leucocytosis greater than 11,000/μL and FiO₂ exceeding 40%, influencing diagnostic decisions in 40% of cases with uncertain diagnosis. The trimodal scoring system demonstrated excellent prognostic discrimination (AUC 0.926), allowing effective risk stratification: pneumonia incidence was 3.5% in low-risk patients (score 0-1) compared to 90.9% in high-risk patients (score 2-3), representing a twenty-six fold difference. Biological characterization indicated that 65.1% of radiographically diagnosed pneumonia cases were attributable to non-bacterial causes: SIRS pattern (44.2%) associated with elevated CRP but low procalcitonin levels (0.53-1.13 ng/mL), and tissue repair pattern (20.9%) with borderline biomarker trends. Only 34.9% exhibited bacterial infection, evidenced by markedly elevated procalcitonin levels (3.48±0.68 ng/mL, p<0.0001) and predominant enteric organisms. Cytokine analysis of bilateral bronchoalveolar lavage demonstrated an almost perfect correlation (r>0.96 for 14 out of 15 mediators), suggesting systemic rather than localised pathology and fundamentally challenging traditional paradigms of pneumonia. Conclusions: Post-operative pneumonia comprises at least two validated biological entities—bacterial infection and sterile inflammation—that require fundamentally different treatments. Moderate inter-observer agreement (κ = 0.35) among experienced specialists indicates the upper limit of syndrome-based diagnosis, reflecting appropriate recognition of biological heterogeneity rather than diagnostic failure. Current protocols focused on antibiotics inappropriately treat around 65% of patients with non-bacterial inflammatory processes. A two-stage, precision framework provides immediate clinical utility: universal risk stratification using trimodal scoring (identifying 85.9% as low-risk), followed by procalcitonin-guided therapy for high-risk patients (14.1%). This approach promotes prudent antibiotic use, risk-based monitoring and efficient resource allocation, while addressing key methodological limitations in surgical outcome research. The bilateral cytokine symmetry suggests that complications are systemic inflammatory processes affecting the lungs, calling for a shift from infection-centred to entity-specific treatment strategies. Further subdivision of non-bacterial processes into distinct entities remains hypothesis-generating, with borderline cytokine trends (p=0.075-0.097) indicating additional complexity that requires prospective validation. This study demonstrates that surgical critical care can achieve biological precision through understanding mechanisms, implemented via routine parameters, establishing practical tools for immediate use and conceptual models for investigating other surgical complications where diagnostic disagreements may also stem from unrecognised biological heterogeneity.