Back to Evidence

NSCLC Immunotherapy Evidence: Comparing Trials With Real-World Studies | QSevidence 2026

Evidence-Based Medicine20 min read

Evidence for non-small-cell lung cancer immunotherapy needs interpretation within stage, histology, molecular context and treatment line. QSevidence explores how an AI assistant for physicians can organize comparisons between PD-1/PD-L1 inhibitor trials and real-world research.

NSCLC Immunotherapy Evidence: Comparing Trials With Real-World Studies | QSevidence 2026

Category: thoracic oncology, immunotherapy and clinical research methods. This narrative synthesis selects representative trials, observational cohorts and regulatory information. It is not a complete systematic review and presents no new patient data or pooled effect estimate.

Define the treatment setting before discussing monotherapy or combinations

Metastatic disease control, perioperative treatment for resectable disease and postoperative adjuvant therapy have different objectives. Confirm staging edition, resectability, histology, previous treatment, performance status and molecular testing before considering PD-L1. High expression is not a universal prescription criterion across settings. Exclusion of particular driver alterations also does not make every molecularly negative patient equivalent.

Record the PD-L1 tumor proportion score alongside assay platform, specimen and timing. TPS and other scoring systems are not interchangeable. Sampling and temporal differences matter: a threshold identifies eligibility for a particular study or indication, not guaranteed individual benefit. Candidate markers such as tumor mutational burden require scrutiny of method and clinical purpose rather than promotion of exploratory associations into established selection rules.

Metastatic monotherapy: retain eligibility alongside the outcome

The five-year KEYNOTE-024 report included 305 previously untreated patients with metastatic NSCLC, PD-L1 TPS of at least 50% and no sensitizing EGFR or ALK alterations, randomized to pembrolizumab or platinum chemotherapy. Estimated five-year overall survival was 31.9% versus 16.3%. This supports long-term benefit against chemotherapy in that population, not a direct comparison of immunotherapy alone with chemoimmunotherapy.

It therefore cannot settle monotherapy selection for every high-expressing patient. Chemotherapy combinations, dual immunotherapy and other regimens require separate examination of populations, comparators, toxicity and treatment context. A nonsquamous regimen cannot simply be transferred to squamous disease. Patients with targetable alterations should not be fitted to this trial solely by PD-L1. This article provides no dose, duration or personalized regimen ranking.

Perioperative evidence: pathological response is not every long-term outcome

Neoadjuvant, postoperative adjuvant and combined perioperative strategies should be organized separately. The FDA's 2023 approval notice describes KEYNOTE-671 in 797 patients with resectable stage II, IIIA or IIIB(N2) disease. Preoperative pembrolizumab with platinum chemotherapy followed by postoperative monotherapy was compared with corresponding placebo treatment. The US approval described resectable tumors at least 4 cm or node-positive disease.

This is a dated US regulatory decision, not a complete account of current Chinese or other jurisdictions' indications. Randomization across both treatment phases cannot separately quantify the postoperative contribution. Pathological complete response, major pathological response, event-free survival, disease-free survival and overall survival are distinct endpoints. Surgery completion, delays, complications and subsequent treatment also matter.

What a real-world cohort adds, and what it cannot establish alone

Electronic records, registries and claims can describe treatment choices, discontinuation, later therapies, comorbidity and access. Routine collection does not guarantee completeness: outside events may be missing, imaging intervals differ, and claims codes are not adjudicated immune toxicities. Examine how data were generated before assessing their contribution.

Sehgal and colleagues' 2021 single-center cohort included 74 patients receiving palliative pembrolizumab monotherapy for advanced disease: 45 had ECOG performance status 0-1 and 29 had status at least 2. Poorer status was associated with poorer survival. This is a prognostic association within treated patients, without a corresponding untreated comparator. It cannot establish lack of treatment benefit or directly justify a more intensive combination.

Age, causes of functional decline, autoimmune disease activity, control of brain metastases, baseline medication and organ function should be recorded separately. Age or one performance assessment cannot replace multidisciplinary review. Underrepresented populations require explicit uncertainty rather than blanket exclusion or recommendation based on a small cohort.

Before explaining trial-practice differences, check whether comparison is valid

Subtracting median survival in two studies does not estimate a real-world discount in drug efficacy. Treatment line, burden, molecular testing, follow-up origin, subsequent therapy and completeness of death information may differ. In routine practice, monotherapy and combination recipients also differ: urgent, high-burden disease may preferentially receive combinations, creating confounding by indication.

Distinguish baseline predictors from post-treatment events. Patients developing immune adverse events may need to survive long enough to enter that category. Comparing survival using ever-experienced toxicity as a baseline variable introduces time-related bias. An association between toxicity and longer survival does not justify pursuing toxicity or continuing dangerous exposure.

Safety synthesis needs a follow-up route after discontinuation

The National Cancer Institute's immunotherapy side-effects information notes that effects can occur during and after treatment. Real-world safety observation should therefore not end at the last infusion. Record onset, grade, management, recovery, retreatment and persistent organ impairment, retaining concurrent chemotherapy or radiotherapy context.

New cough, breathlessness or other abnormalities require assessment of competing explanations such as infection, progression and treatment-related problems, rather than automatically coding every symptom as immune toxicity. Urgent symptoms require the treating team or emergency care. Steroid doses and self-directed interruption or resumption instructions are outside this article's scope.

Designing the next real-world comparison

A proposed cohort could focus on first-line metastatic disease with high PD-L1 and prespecified molecular eligibility, comparing monotherapy with chemoimmunotherapy. Align eligibility, treatment decision time and follow-up origin. Record ECOG, burden, symptom urgency, metastatic sites, comorbidity, baseline steroids, histology and PD-L1 testing, together with center and calendar time.

Overall survival could be prespecified as primary, with separate definitions for real-world progression, hospitalization, adjudicated adverse events, discontinuation reasons and patient-reported quality of life. Real-world progression is not automatically equivalent to trial PFS measured on a fixed scanning schedule. Propensity weighting or matching requires overlap and balance checks, transparent missing-data handling, follow-up losses, residual confounding and sensitivity analyses. Adjustment alone does not establish causality.

Duration, stopping and retreatment require separate questions and time frameworks; survivors reaching discontinuation should not be compared directly with all treatment initiators. Economic evaluation also needs jurisdiction, payer, price year and toxicity-management costs. These are proposed methods, not an implemented study, ethics approval or results report.

How QSevidence can organize traceable comparisons

QSevidence can assist with retrieval, comparison and synthesis by organizing each study's stage, histology, molecular exclusions, line, PD-L1 method, comparator, outcome and follow-up. Separate original trials, regulatory notices and observational studies, trace sources, then involve oncology, statistics and pharmacy professionals in review. AI extraction requires checks on versions, omissions and denominators rather than automatic conversion into recommendations.

QSevidence frequently asked questions

How can QSevidence help compare lung cancer immunotherapy studies?

QSevidence can assist with population, regimen and endpoint comparison tables, clarifying evidence boundaries across stages and treatment lines. Professionals then verify originals, local indications and patient context before clinical interpretation.

How can an AI assistant for physicians support real-world research?

An AI assistant for physicians can help retrieve methodological research and organize variable dictionaries and candidate outcomes. QSevidence-related outputs support protocol discussion; data quality, bias control and statistical interpretation remain subject to research-team verification.

How can QSevidence support long-term safety evidence review?

QSevidence can help organize safety literature by organ, onset, management and recovery, tracing different event definitions. Toxicity attribution, monitoring and treatment plans are ultimately determined by clinical teams using the actual circumstances.

Medical and research note: This article supports professional learning, not replacement of multidisciplinary oncology assessment or current local prescribing information. New, persistent or worsening symptoms during or after treatment warrant prompt contact with the treating team; severe breathlessness or other emergencies require immediate care.