Neoadjuvant Therapy Context Neoadjuvant chemoimmunotherapy - combining chemotherapy with immune checkpoint inhibitors before surgery - has become a promising approach for resectable NSCLC. However, not all patients achieve meaningful pathological responses, and surgery carries risks. Identifying responders before treatment would help individualize therapy.
The Two Key Endpoints This study focused on two pathological response endpoints: pathological complete response (pCR, 0% viable tumor cells) and major pathological response (MPR, less than or equal to 10% viable cells). Both predict improved long-term survival and are accepted surrogate endpoints in clinical trials.
PET-CT as a Key Tool The study leveraged PET-CT imaging - which captures both anatomical (CT) and metabolic (PET, via FDG-SUVmax) information. Changes in metabolic activity between baseline and post-treatment PET-CT are particularly valuable as early indicators of biological treatment response before tumor shrinkage is visible.
Multicenter Design 297 NSCLC patients were enrolled from multiple centers (212 training, 85 testing). This multicenter design increases generalizability compared to single-institution studies and tests whether predictive factors are consistent across different treatment settings and patient populations.
PET-CT Metrics Extracted From PET-CT scans, multiple quantitative metrics were extracted including baseline SUVmax (B-SUVmax), post-treatment SUVmax (P-tumor-SUVmax), and the delta change (delta-Tumor-SUVmax: percent change from baseline to post-treatment). These metabolic parameters quantify how the tumor's glucose consumption changes in response to therapy.
Clinical and Hematologic Features Blood biomarkers were incorporated including baseline neutrophil-to-eosinophil count (B-NEC) and the thyroid peroxidase antibody score (TPSA). Hematologic immune parameters reflect the systemic immune environment and may predict how the immune system responds to ICI-based therapy.
LASSO Regression for Feature Selection Least Absolute Shrinkage and Selection Operator (LASSO) regression with 10-fold cross-validation was used to identify the most predictive features from a large pool of candidates. LASSO penalizes model complexity, forcing coefficients of weak predictors to zero and selecting only the most informative variables.
Logistic Regression Nomograms The selected features were incorporated into multivariate logistic regression models, presented as nomograms - visual tools that allow clinicians to calculate predicted probability of pCR or MPR by adding up point values for each predictor. Nomograms translate complex statistics into intuitive bedside tools.
pCR Prediction Performance The model predicting pathological complete response (pCR) achieved exceptional AUCs of 0.97 in the training cohort and 0.88 in the testing cohort. This level of performance - particularly the training AUC approaching 1.0 - indicates the model identifies a highly coherent set of biological signals for complete response.
MPR Prediction Performance The major pathological response (MPR) model achieved AUCs of 0.98 (training) and 0.81 (testing). Both pCR and MPR models maintained strong testing set performance, demonstrating reasonable generalizability rather than mere overfitting to training data.
Key Predictor: Delta-Tumor-SUVmax The change in SUVmax from baseline to post-treatment emerged as the most powerful predictor. Patients with large SUVmax reductions showed dramatically higher rates of pCR and MPR. This metabolic response signal precedes radiological tumor shrinkage, offering earlier actionable information.
Immune Biomarkers Contribute The B-NEC (neutrophil-to-eosinophil ratio at baseline) and TPSA added independent predictive value beyond imaging features. Eosinophil count changes may reflect ICI-associated immune activation, while TPSA relates to thyroid immune effects - suggesting systemic immune parameters capture treatment response signals not visible on imaging.
Treatment De-escalation for Low Responders Patients predicted to have low probability of pCR or MPR after initial cycles of neoadjuvant therapy might be candidates for treatment modification - switching to alternative chemotherapy regimens, escalating ICI doses, or reconsidering surgical timing to avoid the risks of major resection without meaningful tumor downstaging.
PET-CT as Response Assessment Tool The strong predictive value of delta-SUVmax supports integrating mid-treatment PET-CT scans into neoadjuvant protocols. Serial FDG-PET could serve as an early decision point, identifying responders and non-responders after 2 cycles rather than waiting for post-treatment CT or surgical pathology.
Systemic Immune Monitoring The contribution of hematologic parameters (B-NEC, TPSA) suggests routine blood count monitoring during neoadjuvant therapy carries prognostic information. These cheap, widely available tests could supplement imaging for response prediction in resource-limited settings.
Pre-surgical Risk Stratification The nomogram gives surgeons and oncologists a quantitative probability of achieving pCR or MPR before the patient goes to the operating room. This supports shared decision-making about whether to proceed with surgery or explore organ-sparing approaches in patients predicted to have deep pathological responses.
Heterogeneous Treatment Regimens Different ICI agents and chemotherapy combinations were included across centers, introducing treatment heterogeneity that may affect model performance. Future studies with standardized protocols would reduce this confounding.
SUVmax Measurement Variability PET-CT SUVmax values can vary due to differences in scanners, acquisition protocols, and time between FDG injection and scanning. Standardization of PET-CT acquisition protocols is essential before deploying these models across diverse clinical settings.
Absence of Molecular Biomarkers The model does not include PD-L1 expression, TMB, EGFR/KRAS mutation status, or gene expression profiles, which are known to influence ICI response. Integration of molecular and imaging biomarkers into a combined multimodal prediction model could further improve performance.
Future Directions Prospective validation studies using the nomogram to actively guide treatment decisions - rather than retrospective analysis - are the logical next step. Research should also explore whether radiomics features from baseline CT or PET can substitute for delta-SUVmax in settings where serial PET-CT is not available.