Immune checkpoint inhibitors (ICIs), including antibodies targeting programmed death 1(PD-1), programmed death-Ligand 1(PD-L1), and cytotoxic T-lymphocyte-associated protein 4 (CTLA-4), have transformed the treatment landscape of multiple malignancies by unleashing antitumor immune responses. However, these therapies can precipitate immune-related adverse events (irAEs) due to unintended activation against self-tissues. Among these, immune checkpoint inhibitor–related pneumonitis (CIP) is one of the most serious pulmonary toxicities, often associated with considerable morbidity and mortality (Pozzessere et al., 2021).

The incidence of CIP varies across studies and populations. In meta-analyses of clinical trials, the incidence of any-grade pneumonitis with ICIs has been reported at ~3–5%, with higher rates of severe (grade ≥3) events observed for PD-1 inhibitors in comparison to PD-L1 inhibitors (e.g. 3.6% vs 1.3%) (Sternschein et al., 2018). Real-world cohorts frequently report higher incidence rates, possibly reflecting broader patient heterogeneity and comorbidities (Lin et al., 2024). The onset of CIP is variable, often within weeks to months after ICI initiation (range from ~9 days up to >19 months). Risk factors identified in multiple studies include preexisting lung disease (e.g. interstitial lung disease), prior thoracic radiotherapy, combination immunotherapy regimens, and baseline pulmonary inflammation markers (Zhou et al., 2025).

Crucially, there is currently no well-validated diagnostic approach that can reliably exclude infectious etiologies in patients with suspected CIP, despite the central importance of doing so for accurate diagnosis and appropriate management. Without such tools, clinicians must often rely on conventional microbiological testing (CMT), which has limited sensitivity and lengthy turnaround times, particularly for fastidious, rare, or coinfecting pathogens. This lack of validated exclusionary diagnostics contributes directly to diagnostic uncertainty, delays in correct therapy, and potential harm from either unnecessary immunosuppression or inappropriate antimicrobial use.

Clinically, CIP may present with nonspecific symptoms (cough, dyspnea, or low-grade fever) and radiographic features such as ground-glass opacities, consolidation, or interstitial abnormalities—patterns that substantially overlap with infectious pneumonia (Li et al., 2025a). Because of this overlap, CIP is often a diagnosis of exclusion, requiring systematic evaluation to rule out infection, tumor progression, radiation pneumonitis, or other interstitial lung diseases (Lin et al., 2024). Conventional microbiological testing—including cultures, multiplex PCR, antigen/serology tests, and standard panels—is routinely used to assess infectious etiologies in this context. However, such methods have several limitations in this setting: many fastidious, rare or coinfecting pathogens escape detection by targeted methods, limiting sensitivity and coverage; culture-based techniques often require days to weeks, delaying clinical decisions; patients already on empiric antimicrobials or with low pathogen burden may yield negative results; invasive sampling such as Bronchoscopy/BAL or lung biopsy can increase sensitivity, but are traumatic to patients (Lavalle et al., 2024; Xu et al., 2025b).

Given these drawbacks and the absence of validated diagnostic tools to reliably exclude infection in suspected CIP, clinicians frequently face uncertainty in distinguishing infectious pneumonia from CIP, leading to treatment dilemmas (e.g., starting steroids in unrecognized infection, or withholding immunosuppression for fear of infection).

Metagenomic next-generation sequencing (mNGS) is an emerging, unbiased approach that sequences all nucleic acids in clinical specimens (e.g. BALF, sputum, lung tissue), enabling simultaneous detection of bacteria, viruses, fungi, and parasites—without prior targeting. Several recent studies in immunocompromised patients and severe pneumonia cohorts have demonstrated that mNGS can substantially improve pathogen detection rates compared with conventional methods (Peng et al., 2021; Xu et al., 2023; You et al., 2024). For example, in a cohort of lower respiratory tract infections (LRTIs), mNGS identified pathogens in ~93.3% of cases versus ~55.6% by culture methods (Lai et al., 2025; Li et al., 2025c). Another recent study confirmed high positive rates of mNGS relative to culture/traditional PCR in both immunocompromised and immunocompetent patients with pulmonary infections (Li et al., 2025c). Furthermore, in immunocompromised ICU patients, mNGS has been shown to help refine antimicrobial regimens and inform targeted therapy (Zhao et al., 2024).

To address the critical gap in diagnostic tools for discriminating infectious pneumonia from CIP, we conducted a retrospective cohort study in cancer patients receiving ICIs who developed pulmonary infiltrates suggestive of either CIP or infectious pneumonia. Our objective was to compare the diagnostic performance (sensitivity, specificity, positive/negative predictive values), turnaround time, and incremental diagnostic yield of mNGS versus conventional microbiological testing in the discrimination of CIP from infectious pneumonia. We hypothesized that mNGS would outperform conventional methods in both accuracy and speed, thereby improving clinical decision-making in this challenging scenario.

In this retrospective cohort of cancer patients receiving ICIs with suspected pneumonitis, we found that mNGS markedly outperformed conventional microbiological testing in discriminating CIP from infectious pneumonia: its sensitivity reached 88% and specificity 94%, far exceeding the 69% and 33% achieved by traditional methods. To our knowledge, this is among the first studies to directly compare mNGS and standard diagnostics in the context of ICI-related pneumonitis. These findings highlight two crucial advantages of mNGS in this setting: superior diagnostic accuracy and significantly shorter time to etiologic determination. Importantly, mNGS’s ability to more confidently exclude infection addresses a core clinical challenge in CIP—namely, that CIP diagnosis often requires robust exclusion of infectious etiologies as a prerequisite for safe initiation of immunosuppression, an issue repeatedly emphasized in CIP diagnostic consensus and expert reviews.

Although numerous prior investigations have documented the added value of mNGS in improving pathogen detection in immunocompromised or critically ill patients, most focused on general pneumonia cohorts rather than the specific challenge of differentiating CIP. For example, in immunocompromised hosts, BALF mNGS has been shown to boost microbial detection rates by 20~40 percentage points over conventional tests, particularly in diagnosing opportunistic infections such as Pneumocystis jirovecii (with sensitivity improvements from ~28% to ~100%) (Xin et al., 2025). However, these studies did not address the competing diagnostic scenario where absence of pathogen detection itself carries diagnostic weight in favor of noninfectious immune-mediated lung injury—a central requirement in CIP diagnostic pathways highlighted by recent research.

By contrast, the literature on CIP strongly emphasizes the clinical and radiographic overlap between immune-mediated pneumonitis and infectious pneumonia and warns of frequent misclassification, which can lead to either overt immunosuppression in occult infection or delayed steroid therapy in true CIP (Li et al., 2025a). Prior reviews and expert consensus acknowledge that CIP diagnosis often hinges on exclusion of infectious etiologies, yet effective tools to aid that exclusion remain limited (Wang et al., 2022; Xu et al., 2025b). In other words, while mNGS has been advocated for infection diagnosis, its direct role in ruling out infection to support CIP diagnosis has been relatively unexplored in existing studies.

Our study helps fill this gap by demonstrating that in a cancer population treated with ICIs and presenting with suspected lung injury, mNGS not only detects pathogens when present but also yields negative results with high confidence, thereby offering strong support for CIP in appropriate clinical contexts. In our cohort of cancer patients treated with ICIs who developed suspected lung injury, mNGS markedly outperformed conventional microbiological methods in diagnostic accuracy. With a sensitivity of 88% and specificity of 94% (versus 69% and 33% for CMT), mNGS is capable not only of reliably identifying true CIP cases (by “excluding” pathogens in immune-related pneumonitis) but also of detecting infection when present. Practically, a negative mNGS result (no pathogen detected) carried a positive predictive value for CIP of ~93%, which is of critical importance: misdiagnosing CIP in a patient with occult infection and proceeding with immunosuppression could have serious consequences. At the same time, mNGS’s high specificity (94%) means that it seldom misses infection—in our series, only one patient with a real infection had a false-negative mNGS, whereas conventional testing missed 12 infections. Conversely, a positive mNGS result (pathogen identified) effectively ruled out CIP in nearly all cases (negative predictive value ~89%), alerting clinicians to an underlying infectious etiology. Conventional methods, with their NPV of ~55%, left much greater ambiguity: negative cultures or antigen tests could not reliably suggest noninfectious pneumonitis, because over half of true infections in our series would have been missed by such tests. This supports a clinical decision-making pathway where early mNGS results can meaningfully shorten the period of diagnostic uncertainty and guide whether to initiate immunosuppressive therapy or targeted antimicrobial treatment.

These findings underscore that, in an ICI-treated cancer population with suspected lung injury, mNGS not only enhances pathogen detection when infection is present, but more crucially offers a high-confidence “no pathogen” result to support a diagnosis of CIP in the appropriate clinical setting. This superior performance is especially relevant in light of the historically low yields of conventional microbiological methods (e.g. culture-based detection rates often reported around 30–35% in immunocompromised patients), and in our series, CMT detected pathogens in only ~32% of cases—leaving many infections undiagnosed by traditional approaches (Lu et al., 2022; Miao et al., 2018).

Another major advantage of mNGS observed in our study was its substantially shorter turnaround time. The median interval from sample collection to result was 24 hours (IQR 22–30 h), compared with approximately 121 hours (IQR 80–156 h) for conventional microbiological testing—nearly a fivefold difference. This acceleration is consistent with prior reports in immunocompromised pneumonia, where mNGS reduced time-to-diagnosis from 4–6 days (for culture or PCR workflows) to roughly 1 day (Li et al., 2025b).

This difference carries profound clinical importance. Patients with CIP may deteriorate rapidly and often require timely initiation of corticosteroid therapy to control immune-mediated inflammation. Yet clinicians typically hesitate to start high-dose steroids until infection has been confidently excluded. With traditional diagnostics taking nearly a week, this creates a prolonged period of uncertainty in which broad-spectrum antimicrobials are empirically prescribed, or steroids are delayed—both of which can worsen outcomes. In current practice, this often leads to a combined empiric treatment strategy that includes both high-dose steroids and broad-spectrum antibiotics, reflecting the difficulty in distinguishing immune-mediated lung injury from infection based on clinical and radiographic features alone (Tiu et al., 2022). Faster and more accurate mNGS diagnostics have the potential to shift this paradigm. A rapid negative mNGS result—available within ~24–48 hours in our cohort—can provide clinicians with greater confidence in excluding infection, supporting earlier and more targeted use of corticosteroids while reducing unnecessary antibiotic exposure. Conversely, early identification of a pathogen by mNGS can justify prioritizing pathogen-directed antimicrobial therapy and withholding or de-escalating immunosuppression to avoid exacerbating an underlying infection. For instance, in two of our cases—one involving a COPD patient with CIP complicated by Pseudomonas aeruginosa pneumonia and another with CIP plus Aspergillus infection—mNGS rapidly detected the respective pathogens, prompting prompt initiation of targeted antimicrobials; however, standalone anti-infective therapy failed to fully alleviate respiratory deterioration, as immune-mediated lung injury from CIP persisted. This necessitated the tailored addition of glucocorticoids to control inflammation, leading to significant clinical improvement.

In our cohort, for example, initiation of corticosteroids was delayed in several patients until pathogen exclusion by conventional methods—a process that often took multiple days—leading to prolonged empiric antibiotic use. By contrast, in patients where mNGS results were available early, clinicians were able to tailor therapy more rapidly: early negative results supported timely steroid initiation, and positive mNGS findings prompted focused antimicrobial therapy and avoidance of unnecessary immunosuppression. This represents a clinically actionable decision pathway, where mNGS results directly informed the balance between anti-inflammatory and anti-infective strategies, potentially reducing the risks associated with both overt immunosuppression and unnecessary antibiotic usage.

Quantitatively, this acceleration parallels recent studies in critically ill and immunosuppressed populations, where mNGS shortened diagnostic turnaround by 72–80 hours on average and improved time-to-appropriate therapy by 2–3 days (Li et al., 2025b) (Sears et al., 2019). Such gains may be especially impactful for CIP, where earlier diagnostic clarity directly informs whether to escalate immunosuppression or intensify antimicrobial coverage.

In our cohort, mNGS identified a range of bacterial, fungal, and viral pathogens that are clinically relevant and similar to those encountered in other infectious respiratory diseases. Detected organisms included Streptococcus pneumoniae, Nocardia spp., Aspergillus spp. (including Aspergillus), Pneumocystis jirovecii, Staphylococcus aureus, SARS-CoV-2, Pseudomonas aeruginosa, and Haemophilus influenzae. These taxa span common bacterial, fungal, and viral agents that contribute to lower respiratory tract infections in both immunocompromised and immunocompetent populations, reflecting the broad detection capability of mNGS in diverse infectious settings (Diao et al., 2022).

Despite its advantages, mNGS faces key limitations that merit careful consideration. First, cost and accessibility remain barriers in many centers, though prices are trending downward as sequencing becomes more common. Second, interpretation of positive findings is complex: mNGS may detect colonizers, contaminants, or latent organisms rather than true pathogens, particularly in low-microbial burden samples. Distinguishing clinically relevant signals often requires correlating read counts, abundance metrics, and clinical context. In a recent review, Simner et al. highlight that detection of microbial DNA does not always equate to infection, due to background laboratory or reagent contamination (Simner et al., 2018). Xu et al. similarly emphasize that positive mNGS reports may include incidental organisms requiring expert adjudication. In our study, two CIP patients had mNGS-detected species ultimately judged incidental (Xu et al., 2025a). Overall, the false-positive rate was low, but expert interpretation—often with infectious disease consultation—is essential to avoid misattribution.

There are also limitations in our study. First, our cohort included only 34 patients, reflecting the rarity of CIP and the single-center design; such a modest sample size may reduce the precision and generalizability of our sensitivity and specificity estimates, despite the observed effect sizes being large. Second, the retrospective design introduces inherent selection bias. For instance, mNGS may have been more often ordered in atypical or more severe cases, and relies on the completeness and accuracy of medical records in determining the reference diagnosis. We attempted to mitigate this by applying rigorous diagnostic criteria and employing blinded, multidisciplinary adjudication, but residual misclassification bias cannot be excluded. Third, although we compared the diagnostic performance of mNGS with CMT, the heterogeneity of CMT methods and the limited number of positive culture-based results precluded meaningful threshold variation and robust ROC-based analyses, restricting more granular comparisons across diagnostic modalities. Finally, although we measured diagnostic turnaround times, we did not formally evaluate downstream effects on clinical outcomes (e.g. time to appropriate therapy, duration of hospital stay, or mortality), limiting inferences about the ultimate patient benefit of mNGS. Future prospective, multicenter studies with larger sample sizes should investigate whether mNGS-guided diagnosis yields tangible improvements in clinical endpoints (shorter hospitalization, improved survival, cost-effectiveness), and validate standardized reporting thresholds and interpretation frameworks to optimize its use in CIP management.

In conclusion, our study highlights the effectiveness of mNGS as a diagnostic tool for suspected ICI−related pneumonitis. Relative to conventional microbiological testing, mNGS demonstrated substantially greater accuracy in differentiating CIP from infectious pneumonia and provided results in a fraction of the time, potentially enabling earlier, evidence−based therapeutic decisions. In patients receiving immunotherapy who present with pulmonary symptoms, early application of mNGS on BALF or other respiratory specimens may help rapidly exclude or confirm infection, thereby guiding optimal use of immunosuppression and antimicrobial therapy. However, this study’s retrospective design and single−center cohort with a modest sample size limit the generalizability of our findings, and further prospective, multicenter research with economic evaluation is needed to validate these results and clarify clinical impact. Taken tosgether, our experience suggests that mNGS can meaningfully improve both the precision and timeliness of diagnosis in this challenging clinical scenario, supporting its integration into diagnostic protocols for CIP with appropriate contextual interpretation.