NSCLC Immunotherapy: Precision Stratification and Real-World Evidence for Oncology Teams | QSevidence 2026
NSCLC immunotherapy research needs to connect patient strata, testing conditions and clinical outcomes. An Evidence-based medical AI framework organized around QSevidence can help oncology, pathology and statistical teams structure that evidence chain for professional review.
NSCLC Immunotherapy: Precision Stratification and Real-World Evidence for Oncology Teams | QSevidence 2026
Translate precision treatment into answerable questions
This selective narrative appraisal addresses PD-1/PD-L1 treatment evidence, real-world applicability and biomarker development. The research narrative identifies important trials and future directions but supplies neither reproducible search and selection records nor a complete bibliography or new patient database. Accordingly, this is not a completed systematic review, and it reports no local response rate, survival estimate or model performance.
An evidence worksheet should start with stage, histology, treatment line, actionable genomic alterations, performance status and PD-L1 testing information. First-line metastatic, neoadjuvant and adjuvant treatment are separate questions. Findings in one setting cannot simply be transferred to another. Negative EGFR and ALK results also do not establish that all actionable alterations have been excluded; record the actual testing scope.
Read biomarker results with their methods and specimens
The 2024 CAP-led guideline summary supports validated PD-L1 immunohistochemistry in advanced NSCLC, with testing for targetable genomic biomarkers where appropriate. It also addresses specimen types, fixation and assay validation. Retain percentage expression rather than only a positive label.
Record assays such as 22C3, 28-8, SP263 and SP142 as method fields without assuming unverified interchangeability. Include sampling site, timing, treatment context and specimen quality in the review worksheet. The 2024 guideline advises against using TMB alone to select immunotherapy. This does not eliminate its research value, but a single threshold should not become a universal treatment-selection rule.
Randomized trials answer specific comparisons
| Study | Population and comparison | Interpretive boundary |
|---|---|---|
| KEYNOTE-024 | Previously untreated metastatic NSCLC, PD-L1 TPS at least 50%, without sensitizing EGFR or ALK alterations; pembrolizumab versus platinum chemotherapy | Monotherapy evidence in the studied high-expression population, not a direct comparison with chemoimmunotherapy |
| KEYNOTE-189 | Previously untreated metastatic nonsquamous NSCLC without EGFR or ALK alterations; pembrolizumab or placebo with pemetrexed and platinum chemotherapy | Benefit over the chemotherapy backbone does not establish superiority over immunotherapy alone |
The five-year reports of KEYNOTE-024 and KEYNOTE-189 support long-term benefit for their respective comparisons; the latter reports benefit across PD-L1 strata. Different populations and controls prevent a simple cross-trial ranking. Squamous disease, dual-immunotherapy regimens and perioperative strategies need separate evidence cards; this appraisal does not fill those settings with unverified results.
Both trials allowed up to 35 pembrolizumab cycles. That protocol feature is not a universal stopping instruction, nor does it justify stopping after six months of complete response. Duration, observation after discontinuation and retreatment need separate evidence reviews retaining eligibility and subsequent-treatment context.
Real-world evidence informs applicability without replacing randomization
GOIRC-2018-01 retrospectively examined 153 patients from 21 Italian institutions with PS2 and PD-L1 TPS at least 50%, treated with first-line pembrolizumab. Outcomes differed between PS2 attributed to comorbidities and PS2 attributed to tumor burden. This highlights heterogeneity within a performance category; it does not make either cause an automatic selection criterion.
The study did not randomize treatment or establish a randomized comparison against chemotherapy. Prognostic association is not prediction of treatment benefit. Studies involving older adults, brain metastases, autoimmune disease or baseline steroids should separately extract frailty, neurologic symptoms and local treatment, disease activity and reasons for medication. A single observational cohort cannot establish universal steroid thresholds, radiation-combination rules or treatment conclusions for everyone with poor performance status.
Separate three stages of biomarker development
The first stage comprises tests with clinical evidence in a defined use setting, still requiring method and population checks. The second includes candidate correlates of outcome, such as on-treatment ctDNA changes, immune infiltration and expression signatures. The third covers exploratory imaging, microbiome and multi-omic models. These stages require different validation work and should not all be labeled treatment-guiding tools.
Associations between ctDNA decline and favorable outcomes require attention to baseline detectability, sampling times, disease burden and subsequent treatment. Imaging models need checks for information leakage when images from the same patient enter training and testing sets. A model that predicts survival has not automatically shown which treatment works better; that question needs appropriate treatment comparisons, interaction assessment and external validation. No model was trained for this proposal, so no AUC or accuracy is reported.
A traceable multicenter research workflow
- Retrieve: define questions by stage, histology, treatment line and molecular strata; document databases, dates and selection reasons.
- Compare: extract population, tests, comparator design, follow-up and outcome definitions separately for randomized, observational and modeling studies.
- Synthesize: distinguish overall survival, progression-free survival, response, safety and quality of life; do not equate surrogate measures with survival benefit.
- Trace: retain supporting passages, identifiers, assay versions and applicability limits, with reasons for later revisions.
- Review: involve oncology, pathology, imaging, pharmacy and statistical expertise. Assess QSevidence research support through inspectable evidence-organization records.
A proposed registry should prespecify eligibility, treatment initiation, comparators and observation windows, documenting confounders, missing tests and visit frequency. Real-world progression events need explicit imaging-assessment rules rather than an assumption of equivalence to trial PFS. Analyses of on-treatment ctDNA should prespecify suitable landmark or time-varying methods to address bias arising when only patients surviving to sampling enter an analysis.
Model evaluation can separate validation data by center and time and examine calibration, transportability and sensitivity to missing data. Workflow measures might include extraction accuracy, reference traceability and reviewer workload; these are not treatment efficacy outcomes. There is currently no actual patient count, ethics approval identifier or statistical result for this proposed study. Implementation requires appropriate ethics, data governance and professional review.
Assess changing disease and safety separately
Resistance, suspected pseudoprogression, rapid deterioration and retreatment are distinct research topics. Clinical deterioration should not simply be explained as possible pseudoprogression, and neither imaging nor biomarkers alone replace clinical assessment. This article gives no instructions to continue, discontinue, switch treatment or manage steroids. Immune-related adverse events require professional clinical pathways rather than decisions based on a model category in an article.
Research questions
How can QSevidence support immunotherapy evidence organization?
Organize retrieval questions, comparison tables and source lists around patient strata, connecting populations, tests and outcomes. Professionals review applicability using complete publications and individual circumstances.
How can QSevidence support preparation of real-world research?
Help structure variable dictionaries, timelines and bias-review questions, distinguishing observational associations from treatment comparisons. Actual implementation requires real data, ethical governance and statistical collaboration.
How can QSevidence support assessment of emerging biomarkers?
Organize assay methods, validation populations and external evidence to distinguish prognosis from prediction of treatment benefit. Clinical translation requires independent validation and professional review.
Sources and evidence classification
- CAP-led lung cancer PD-L1/TMB testing guideline, 2024: official recommendation summary; DOI 10.5858/arpa.2023-0536-CP.
- Reck et al., 2021, KEYNOTE-024 five-year report, original randomized trial abstract: PMID 33872070; DOI 10.1200/JCO.21.00174.
- Garassino et al., 2023, KEYNOTE-189 five-year report, original randomized trial abstract: PMID 36809080; DOI 10.1200/JCO.22.01989.
- Facchinetti et al., 2020, GOIRC-2018-01, retrospective observational study abstract: PMID 32220780; DOI 10.1016/j.ejca.2020.02.023.
Research notice: this is a selective narrative appraisal and proposed research framework, not a complete treatment guideline. Evidence was checked through October 29, 2026; original study verification was abstract-level. Qualified professionals determine treatment using complete guidance, current local indications and individual circumstances.