Introduction
Chronic rhinosinusitis and chronic rhinosinusitis with nasal polyps are high-prevalence inflammatory disorders of the upper airways that collectively affect an estimated 5–12 per cent of adults worldwide.Reference Dietz de Loos, Lourijsen, Wildeman, Freling, Wolvers and Reitsma1 Beyond persistent nasal obstruction, rhinorrhoea, facial pressure and hyposmia, these conditions impose a disproportionate burden on daily functioning, health-related quality of life and productivity. Recurrent healthcare use, lost workdays and long courses of medical therapy translate into substantial societal costs.Reference Albu2 Given this broad impact, timely and accurate diagnosis is essential to guide appropriate therapy, minimise avoidable radiation exposure and optimise surgical planning when needed.
Clinically, chronic rhinosinusitis is defined by two or more symptoms – one of which must be nasal blockage and/or obstruction or nasal discharge – lasting at least 12 weeks, together with objective evidence of mucosal inflammation obtained on nasal endoscopy or cross-sectional imaging.Reference Lohiya, Patel, Pawde, Bokare and Sakhare3 Importantly, consensus guidelines state that computed tomography (CT) is not essential for diagnosis and should be reserved for severe disease, immunocompromised patients, surgical planning or evaluation of complications.Reference Raz, Win, Hagiwara, Lui, Cohen and Fatterpekar4 Histopathology is similarly a complementary test rather than a diagnostic ‘gold standard’. Magnetic resonance imaging (MRI) is specifically not recommended for routine chronic rhinosinusitis diagnosis because of its limited specificity, although it provides superior soft-tissue contrast and is useful when neoplasm or fungal disease is suspected. These recommendations underscore that the definitive diagnosis of chronic rhinosinusitis and/or chronic rhinosinusitis with nasal polyps remains clinical and that objective tests should augment, not replace, clinical assessment.
Nevertheless, distinguishing chronic rhinosinusitis from mimics such as allergic or non-allergic rhinitis can be challenging when symptoms wax and wane or findings are subtle. Consequently, clinicians frequently rely on a combination of endoscopy and imaging to refine diagnostic probability and tailor treatment decisions. This practice has fostered ongoing debate about the relative diagnostic accuracy of endoscopy versus imaging and the appropriate thresholds for ordering CT or MRI.
Diagnostic nasal endoscopy is a minimally invasive, office-based procedure that allows direct assessment of mucosal oedema, purulence, crusting and polypoid change. Its strengths are availability, repeatability and the opportunity to perform therapeutic manoeuvres – such as debridement or targeted culture – during the same visit.Reference Bachert, Marple, Schlosser, Hopkins, Schleimer and Lambrecht5, Reference Rosenfeld, Piccirillo, Chandrasekhar, Brook, Ashok Kumar and Kramper6 However, consensus remains variable regarding its diagnostic accuracy relative to imaging. Some patients with normal endoscopy demonstrate clinically meaningful sinus disease on imaging, while others with prominent endoscopic oedema have limited radiographic involvement.Reference Tecimer, Kasapoglu, Demir, Ozmen, Coskun and Basut7 Clarifying the performance characteristics of diagnostic nasal endoscopy relative to commonly used references is therefore important for determining when endoscopy alone is sufficient and when additional imaging is warranted.
Computed tomography is frequently regarded as the reference imaging modality for chronic rhinosinusitis because it depicts the extent and distribution of mucosal disease and delineates critical bony anatomy that governs ventilation and drainage pathways.Reference Zhang, Liu, Yang, Wang, Tan and Huang8 Computed tomography is indispensable for pre-operative mapping,Reference Rathor and Bhattacharjee9 detection of anatomic variants and evaluation of orbital or skull-base proximity. Nevertheless, CT entails ionising radiation, and it may over-call reversible mucosal thickening or retained secretions, limiting its specificity for chronic inflammatory remodelling.Reference Schulz, Potente, Zangos, Friedrichs, Bauer and Kerl10, Reference Pirimoglu, Sade, Sakat, Ogul, Levent and Kantarci11 In routine care, clinicians must weigh these trade-offs when ordering imaging for patients with persistent symptoms following initial medical management.
Magnetic resonance imaging offers superior soft-tissue contrast without radiation and can better discriminate polypoid tissue, fungal elements, inspissated secretions and neoplastic processes.Reference Raz, Win, Hagiwara, Lui, Cohen and Fatterpekar4 Magnetic resonance imaging is particularly attractive when CT findings are atypical or discordant with the clinical picture, when radiation minimisation is a priority or when tissue characterisation will directly influence management. Despite these theoretical advantages, MRI is less accessible in many settings, and its diagnostic accuracy in typical chronic rhinosinusitis populations has not been comprehensively quantified across studies. These uncertainties hinder clear guidance on when MRI should complement or substitute for CT.
Prior diagnostic studies have reported heterogeneous estimates for diagnostic nasal endoscopy, CT and MRI, reflecting small sample sizes,Reference Papadopoulou and Marinou12, Reference Chong, Piromchai, Sharp, Snidvongs, Webster and Philpott13 variable patient selection (community versus tertiary-care populations) and differences in index-test thresholds and reference standards. Many studies benchmark diagnostic nasal endoscopy or MRI against CT rather than histopathology, raising concerns about imperfect references and partial-verification bias.Reference Fritz and Runge14 Others report accuracy metrics without providing sufficient 2 × 2 data for meta-analysis, hampering quantitative synthesis. Consequently, evidence has remained fragmented and clinicians have lacked pooled estimates that can be applied in Bayesian probability-updating frameworks at the point of care.Reference van der Geest, Sandovici, Brouwer and Mackie15
Given this fragmented evidence base and the recognised limitations of current reference standards, we undertook a systematic review and meta-analysis to synthesise available data on the diagnostic performance of nasal endoscopy, CT and MRI in adults with suspected chronic rhinosinusitis and/or chronic rhinosinusitis with nasal polyps. We focused on estimating pooled sensitivity, specificity and likelihood ratios while recognising that CT and histopathology are not gold standards but comparators, and that MRI is rarely used for routine diagnosis. By presenting prediction intervals and exploring heterogeneity, we aim to provide transparent, evidence-based guidance for tiered diagnostic decision-making and to highlight the paucity of high-quality data. The following sections describe the detailed methodology and present our pooled results, which should be interpreted as measures of concordance between tests rather than definitive diagnostic accuracies.
Methods
Search strategy and selection criteria
We conducted a comprehensive search of PubMed/Medline, Embase (via Ovid/Elsevier) and Web of Science Core Collection from January 1990 to 27 August 2025. For PubMed, the following query was used: (‘rhinosinusitis’[MeSH Terms] OR ‘chronic rhinosinusitis’[Title/Abstract] OR CRS[Title/Abstract]) AND (‘endoscopy’[Title/Abstract] OR ‘nasal endoscopy’[Title/Abstract] OR ‘rigid endoscopy’[Title/Abstract]) AND (‘computed tomography’[Title/Abstract] OR CT[Title/Abstract] OR ‘paranasal sinus CT’[Title/Abstract]) AND (sensitivity[Title/Abstract] OR specificity[Title/Abstract] OR ‘predictive value’[Title/Abstract] OR ‘2x2’[Title/Abstract] OR ‘contingency table’[Title/Abstract]), with filters for humans, adults (≥19 years), English or Chinese language and publication years 1990–2025.
Equivalent search terms were adapted for Embase using controlled vocabulary and text words: (‘chronic rhinosinusitis’/exp OR ‘chronic rhinosinusitis’:ab,ti) AND (‘nasal endoscopy’/exp OR ‘diagnostic nasal endoscopy’:ab,ti) AND (‘computed tomography’/exp OR ‘paranasal sinus CT’:ab,ti OR ‘magnetic resonance imaging’/exp OR MRI:ab,ti) AND (sensitivity:ab,ti OR specificity:ab,ti OR ‘predictive value’:ab,ti), limited to humans, adults, 1990–2025 and English or Chinese language.
The Web of Science query was: TOPIC: (chronic rhinosinusitis OR CRS) AND TOPIC: (‘nasal endoscopy‘OR endoscopy) AND TOPIC: (CT OR ‘computed tomography’OR MRI OR ‘magnetic resonance imaging’) AND TOPIC: (sensitivity OR specificity OR ‘predictive value’ OR ‘2x2‘), with timespan 1990–2025. We additionally searched reference lists of included studies and relevant reviews, scanned conference abstracts and theses for grey literature, and contacted field experts; no additional eligible data were identified. After deduplication, two reviewers independently screened titles and abstracts, retrieved full texts of potentially eligible studies and resolved disagreements by consensus.
To address concerns about the comprehensiveness of our search, we subsequently repeated the strategy in Scopus and the Cochrane Library using equivalent keywords and subject headings, and performed an additional hand-search of conference proceedings and grey literature. These supplementary searches did not yield any studies meeting the inclusion criteria.
Inclusion criteria were: (1) adult patients with suspected chronic rhinosinusitis and/or chronic rhinosinusitis with nasal polyps; (2) evaluation of diagnostic nasal endoscopy, CT or MRI as the index test; (3) CT or surgical histopathology as the comparator; (4) availability of data to construct a 2 × 2 contingency table; and (5) cross-sectional, prospective or comparative cohort design. We acknowledge that neither CT nor histopathology constitutes a perfect reference standard for chronic rhinosinusitis and/or chronic rhinosinusitis with nasal polyps; these tests were treated as comparators because they were consistently reported across studies. Studies without diagnostic outcomes, paediatric cohorts or insufficient data were excluded. A 2022 prospective study comparing an adjusted MRI protocol with CT for pre-operative planning met criteria for narrative synthesis but lacked disease-level 2 × 2 data and therefore was not included in the quantitative pooling.Reference Landsberg, Schneider, Masalha, Margulis, Guindy and Luckman16
Data extraction and quality assessment
Two reviewers independently extracted study characteristics, including author, year, country, study design, sample size, patient demographics (mean age and sex distribution), clinical diagnostic criteria for chronic rhinosinusitis and/or chronic rhinosinusitis with nasal polyps, index and reference tests, test thresholds and prespecified cut-off values, and the numbers of true positives, false positives, false negatives and true negatives. When exact counts were not reported, we reconstructed them from reported sensitivity, specificity and sample size. To assess the applicability of each study we also recorded whether symptom definitions adhered to consensus guidelines and whether participants were recruited from community clinics or tertiary referral centres. Risk of bias was evaluated using the Quality Assessment of Diagnostic Accuracy Studies 2 tool,Reference Whiting, Rutjes, Westwood, Mallett, Deeks and Reitsma17 which examines four domains: patient selection, index test conduct and interpretation, reference standard, and flow and timing. Disagreements were resolved through discussion. Inter-rater agreement was strong (Cohen’s kappa = 0.86 for study inclusion and 0.92 for extraction of 2 × 2 data).
Statistical analysis
For each study, sensitivity (true positives/(true positives + false negatives)) and specificity (true negatives/(true negatives + false positives)) were calculated, and exact 95 per cent confidence intervals (CIs) for individual proportions were obtained using the Clopper–Pearson method. To obtain summary estimates, sensitivities and specificities were logit-transformed and the DerSimonian–Laird random-effects model was applied. Between-study heterogeneity was quantified with the I 2 statistic. To convey the expected range of results in future clinical settings we computed 95 per cent prediction intervals on the logit scale and back-transformed them to probabilities using the method of Higgins and Thompson.Reference Higgins, Thompson and Spiegelhalter18 Pooled positive and negative likelihood ratios were derived from the pooled sensitivity and specificity. Subgroup analyses were prespecified according to study region (India vs other) and sample size (<75 vs ≥75 participants), and leave-one-out analyses assessed the influence of individual studies.
All analyses were conducted in Python (statsmodels and metafor packages). We initially attempted to fit hierarchical summary receiver-operating characteristic and bivariate random-effects models – which jointly model sensitivity and specificity, and account for their correlation – but with fewer than five studies contributing to each comparison these models did not converge and produced unstable covariance estimates. Therefore, separate random-effects models for logit-transformed sensitivity and specificity were used as a pragmatic alternative in this small-sample setting.
Results
Study selection
The search yielded 62 records (PubMed, n = 30; Embase, n = 25; Web of Science, n = 7). After removal of duplicates, 52 records were screened; 29 were excluded after title and abstract screening. Twenty-three full-text articles were assessed for eligibility and 15 were excluded for insufficient diagnostic data or failure to meet inclusion criteria. Ultimately, nine diagnostic datasets met the inclusion criteria: five compared diagnostic nasal endoscopy with CT,Reference Lohiya, Patel, Pawde, Bokare and Sakhare3, Reference Uwaneme, Asoegwu, Adekoya and Nwawolo19–Reference Amine, Lininger, Fargo and Welch21Reference Inchingolo, Maino, Cannella, Vernuccio, Cortese and Dezio,35 two datasets (at different Lund–Mackay cut-offs) compared CT with surgical histopathologyReference Deosthale, Khadakkar, Harkare, Dhoke, Dhote and Soni22 and two compared MRI with CT.Reference Bhattacharyya and Fried23, Reference Cohen-Kerem, Marshak, Uri, Gruber, Huberfeld and Paz24 All data were obtained from publicly available databases and no additional ethics approval was required. The selection process is shown in Figure 1.
Figure 1. Diagram of study selection. Records identified: PubMed (n = 30), Embase (n = 25), Web of Science (n = 7). Records after de-duplication, n = 52; title and/or abstract excluded, n = 29; full-text assessed, n = 23; full-text excluded, n = 15 (no extractable 2 × 2 data, non-adult populations or non-diagnostic outcomes); studies included in qualitative and quantitative synthesis, n = 9.
Characteristics of included studies
Nine diagnostic datasets with 50–301 participants were included. Five compared diagnostic nasal endoscopy with CT, two compared CT with histopathology at different Lund–Mackay thresholds (>2 and >4)Reference Deosthale, Khadakkar, Harkare, Dhoke, Dhote and Soni22 and two compared MRI with CT. Datasets originated from India (n = 3), Nigeria (n = 1), Oman (n = 1), the United States (n = 3; two thresholds from the histopathology dataset) and Canada (n = 1). Endoscopy was evaluated using rigid or flexible scopes against CT scored by the Lund–Mackay system. Histopathology datasets evaluated CT at two cut-offs (>2 and >4) against surgical specimens (Table 1).Reference Deosthale, Khadakkar, Harkare, Dhoke, Dhote and Soni22
Table 1. Characteristics of included diagnostic datasets
One study contributed CT–histopathology data at two Lund–Mackay cut-offs. CT = computed tomography; MRI = magnetic resonance imaging.
Diagnostic nasal endoscopy versus computed tomography
Five studies (n = 399 participants) compared endoscopy with CT.Reference Uwaneme, Asoegwu, Adekoya and Nwawolo19, Reference Kolethekkat, Paul, Kurien, Kumar, Al Abri and Thomas20, Reference Deosthale, Khadakkar, Harkare, Dhoke, Dhote and Soni22 Study-level sensitivity ranged from 0.73 to 0.94 and specificity from 0.29 to 0.86. Random-effects meta-analysis yielded a pooled sensitivity of 0.87 (95 per cent CI = 0.78–0.92) and specificity of 0.63 (95 per cent CI = 0.38–0.83). Between-study heterogeneity was substantial (I 2 ≈ 70 per cent for sensitivity; I 2 ≈ 64 per cent for specificity). The pooled positive likelihood ratio was 2.24 (95 per cent CI = 1.25–4.00) and the negative likelihood ratio was 0.22 (95 per cent CI = 0.13–0.38). Thus, a positive endoscopic finding approximately doubles the odds of CT-confirmed disease, while a negative endoscopy finding lowers post-test probability but does not exclude sinus pathology. Forest plots and the hierarchical summary receiver-operating characteristic appear in Figure 2.
Figure 2. Diagnostic performance of nasal endoscopy versus computed tomography (CT). (A) Study-level sensitivities with pooled estimate. (B) Study-level specificities with pooled estimate. (C) Hierarchical summary receiver-operating characteristic (ROC) curve with 95 per cent confidence interval and prediction regions. Pooled effects from random-effects models; symbol size proportional to study weight.
CT versus histopathology
One dual-cohort study evaluated CT against histopathology using Lund–Mackay thresholds of more than 2 and more than 4.Reference Bhattacharyya and Fried23 Sensitivity and specificity values at more than 2 were 0.94 and 0.41 and at more than 4 were 0.85 and 0.59. Pooled across thresholds, sensitivity was 0.90 (95 per cent CI = 0.77–0.96) and specificity was 0.50 (95 per cent CI = 0.33–0.67). Positive likelihood ratio values were approximately 1.6–2.1 and negative likelihood ratio values were 0.14–0.26, indicating high sensitivity but modest rule-in performance as a result of false positives. Forest plots and hierarchical summary receiver-operating characteristics are shown in Figure 3.
Figure 3. Diagnostic performance of computed tomography (CT) versus histopathology (dual-cohort at thresholds >2 and >4). (A) Sensitivity forest plot. (B) Specificity forest plot. (C) Hierarchical summary receiver-operating characteristic (ROC) with threshold-specific estimates. Random-effects models; symbol size proportional to precision.
Magnetic resonance imaging versus computed tomography
Two datasets provided paired accuracy data for MRI versus CT. The earlier dataset by Lin et al.reported moderate sensitivity (0.67) and high specificity (0.90) compared with CT.Reference Bhattacharyya and Fried23 In 2024, Parker et al. evaluated adjusted MRI as a low-radiation alternative to CT and reported sensitivity of 0.71 and specificity of 0.88.Reference Cohen-Kerem, Marshak, Uri, Gruber, Huberfeld and Paz24 Pooled estimates should be interpreted cautiously given the small number of datasets and heterogeneity in MRI protocols. Forest plots and hierarchical summary receiver-operating characteristics are shown in Figure 4.
Figure 4. Diagnostic performance of magnetic resonance imaging (MRI) versus computed tomography (CT). (A) Sensitivity forest plot. (B) Specificity forest plot. (C) Hierarchical summary receiver-operating characteristic (ROC) with threshold-specific estimates. Random-effects models; symbol size proportional to precision.
Subgroup analyses
Subgroup analyses explored whether study characteristics influenced diagnostic performance (Table 2). Pooled sensitivity was higher and specificity lower among the Indian studies (sensitivity = 0.895, specificity = 0.577) compared with non-Indian studies (sensitivity = 0.833, specificity = 0.695). Studies with fewer than 75 participants exhibited slightly higher sensitivity (0.910) and specificity (0.728) than larger studies (sensitivity = 0.849, specificity = 0.574). While these differences hint at possible spectrum effects (referral populations vs community samples) and operator expertise, confidence intervals overlapped substantially because of small numbers.
Table 2. Subgroup analysis of diagnostic nasal endoscopy versus CT
CT = computed tomography; CI = confidence interval.
Sensitivity analyses
Leave-one-out analyses were conducted by sequentially omitting each diagnostic nasal endoscopy study and recalculating pooled estimates. The pooled sensitivity varied from 0.857 to 0.898 and specificity from 0.524 to 0.717, indicating that no single study dominated the results. Exclusion of the Nigerian study produced the largest increase in sensitivity (from 0.873 to 0.898) and decrease in specificity (from 0.640 to 0.524), suggesting that population differences may partly explain heterogeneity.
Risk of bias and applicability
Methodological quality was moderate to low. Patient selection often involved symptomatic individuals at tertiary clinics, risking spectrum bias. Consecutive sampling was rarely stated. Index test interpretation was seldom blinded to clinical findings or reference results, raising potential diagnostic review bias. Computed tomography served as the reference for endoscopy and MRI, but may misclassify reversible mucosal changes as disease; histopathology was only available in surgical cohorts. Flow and timing were generally acceptable, but the MRI study allowed up to three months between scans. These limitations likely contributed to heterogeneity and limit generalisability.
The Quality Assessment of Diagnostic Accuracy Studies 2 assessment is summarised graphically in Figure 5. Most studies had high or unclear risk of bias for patient selection and index test interpretation, whereas reference standards and flow and timing were generally low risk. Applicability concerns were low across domains, except for the use of CT rather than histopathology as the reference in most studies.
Figure 5. Risk of bias and applicability summary (Quality Assessment of Diagnostic Accuracy Studies 2). Stacked bar chart of domain-level judgements (patient selection, index test, reference standard, flow and/or timing) across included studies.
Detailed search results, reasons for exclusion and individual risk-of-bias assessments are presented in Supplementary Tables S1–S3. The supplementary file is provided as a separate document for readers to access.
Discussion
Main findings
This systematic review synthesises the diagnostic accuracy of diagnostic nasal endoscopy, CT and MRI for chronic rhinosinusitis and/or chronic rhinosinusitis with nasal polyps. Diagnostic nasal endoscopy remains highly sensitive relative to CT and a negative endoscopy significantly reduces the probability of CT-confirmed disease. However, diagnostic nasal endoscopy’s moderate specificity means that positive findings only modestly increase post-test probability; clinicians should continue to interpret endoscopic oedema or purulence in the context of symptom burden and risk factors.Reference Amine, Lininger, Fargo and Welch21, Reference Cohen-Kerem, Marshak, Uri, Gruber, Huberfeld and Paz24
Computed tomography, benchmarked against surgical histopathology, demonstrates excellent sensitivity but limited specificity.Reference Goo and Goo28 The choice of Lund–Mackay threshold has a predictable trade-off: lower cut-offs maximise sensitivity and are appropriate when ruling out disease before surgery, whereas higher cut-offs increase specificity and may be preferable when aiming to rule in disease or minimise overdiagnosis.Reference Bhattacharyya and Fried23 These findings underscore that CT should not be considered a perfect gold standard; rather, its results must be integrated with clinical features and, when available, histopathology.
Pooling the two available MRI studies yielded a sensitivity of 0.710 and a specificity of 0.876 compared with CT, indicating high specificity and moderate sensitivity.Reference Parker, Beyea, Rioux, King, Abdolell and Reeve27 The early study reported strong correlation between MRI and CT scores (r ≈ 0.84).Reference Lin and Bhattacharyya26 Magnetic resonance imaging’s superior soft-tissue contrast enables discrimination of polypoid tissue, inspissated secretions and neoplasms, and the absence of ionising radiation is particularly valuable in young patients, those requiring repeated imaging and problem-solving scenarios.Reference Maas, Menon, Manley, Abrams, Akerlund and Andelic29 The 2022 study using adjusted MRI protocols demonstrated that MRI can visualise critical bony landmarks as well as CT in most cases,Reference Landsberg, Schneider, Masalha, Margulis, Guindy and Luckman16 suggesting that MRI could eventually replace CT for selected surgical planning. However, the evidence base remains thin; both included MRI studies used CT rather than histopathology as the reference and required reconstruction of 2 × 2 counts from summary data. To date no study has paired MRI and CT with histopathology in the same cohort, and cost and availability remain barriers.
Sources of heterogeneity and implications
Substantial heterogeneity across diagnostic nasal endoscopy studies likely reflects differences in patient populations, operator expertise, index-test thresholds and reference standards. Tertiary-care studies from India reported higher sensitivity but lower specificity than studies from Nigeria and Oman, possibly because referral centres enrol patients with more severe or refractory disease, shifting the implicit diagnostic threshold. Variation in endoscopic equipment (rigid vs flexible), sedation and reporting standards also contributes, and radiologic thresholds influence CT versus histopathology estimates.Reference Bhattacharyya and Fried23
Another important source of heterogeneity is geographic diversity: the included studies originated from India, Nigeria, Oman, the United States and Canada, and differences in patient demographics, healthcare infrastructure and disease prevalence across these regions may affect test performance and limit generalisability. Our subgroup and sensitivity analyses suggest these factors influence point estimates but do not fundamentally alter the qualitative conclusion that diagnostic nasal endoscopy is a good rule-out test and CT is a sensitive but non-specific imaging modality. Future studies should clearly report recruitment methods, index-test protocols and thresholds, and should avoid using CT as the sole reference; composite references incorporating histopathology and clinical follow up would improve specificity estimates.Reference Yang, Gu, Hong, Zou, Zhang and Yuan30
Strengths and limitations
The strengths of this review include a comprehensive, multi-database search with de-duplication, explicit inclusion criteria requiring extractable 2 × 2 data and use of random-effects modelling on the logit scale to obtain pooled sensitivities and specificities. To our knowledge, this is the first meta-analysis of chronic rhinosinusitis and/or chronic rhinosinusitis with nasal polyps to provide pooled positive and negative likelihood ratios for diagnostic nasal endoscopy, CT and MRI. We also generated summary receiver-operating characteristic curves and conducted a structured risk-of-bias assessment using Quality Assessment of Diagnostic Accuracy Studies 2. By translating sensitivity and specificity into clinically interpretable likelihood ratios and post-test probabilities, the review provides clinicians with novel, directly actionable information at the point of care.
Nonetheless, several limitations should be noted. First, we did not formally evaluate small-study effects or publication bias (e.g. Deeks’ funnel-plot asymmetry test) because fewer than 10 studies contributed to the endoscopy–CT meta-analysis, the CT–histopathology comparison derived from a single centre using two thresholds and only a single study assessed MRI versus CT; under these conditions such tests are underpowered and potentially misleading. Second, we did not fit hierarchical bivariate and/or hierarchical summary receiver-operating characteristic models because too few studies (fewer than five per comparison) reported paired sensitivity and/or specificity estimates to permit stable estimation of the between-study covariance and the summary operating point; instead, we meta-analysed logit-transformed sensitivity and specificity in separate random-effects models and displayed non-hierarchical summary receiver-operating characteristic curves for visualisation. Third, despite conducting a comprehensive and meticulous literature search, we located only two MRI studies with extractable data; both lacked histopathology and required reconstruction of 2 × 2 tables from reported sensitivities and specificities. Many recent MRI reports focus on technique or anatomy rather than diagnostic accuracy, limiting quantitative synthesis. We could not adjust for differences in CT slice thickness or radiation dose, which may influence sensitivity and specificity. Finally, we relied on aggregate data; individual participant data meta-analysis would allow modelling of continuous CT scores and patient-level covariates.
Clinical implications and tiered diagnostic pathway
Our findings support a tiered diagnostic approach to chronic rhinosinusitis and/or chronic rhinosinusitis with nasal polyps. In routine practice, diagnostic nasal endoscopy should serve as the first-line investigation because it is inexpensive, office-based and radiation-free; a negative endoscopy markedly reduces the likelihood of CT-confirmed disease, whereas a positive endoscopy modestly increases it. Computed tomography should be reserved for cases with persistent symptoms despite optimal medical therapy, for pre-operative planning or when the endoscopic examination is inconclusive, given its high sensitivity but limited specificity and the associated radiation exposure. Magnetic resonance imaging should be considered in selected scenarios where soft-tissue characterisation is critical (e.g. distinguishing tumour from inflammatory tissue), when invasive fungal disease or neoplasm is suspected, or when radiation avoidance is prioritised (e.g. in pregnant patients or those requiring repeated imaging).Reference Poto, Pelaia, di Salvatore, Saleh, Scadding and Varricchi31 These recommendations align with professional guidelines that advocate objective confirmation with endoscopy or CTReference Kulich, Long, Reyes Orozco, Yi, Hao and Han32, Reference Garcia-Garrigos, Arenas-Jimenez, Monjas-Canovas, Abarca-Olivas, Cortes-Vela and J33 while reserving MRI for complications or indeterminate cases.Reference Tsukamoto, Mavrogenis, Tanaka and Errani34–Reference Flannery, Abouharb and McKinstry36 However, our review underscores that the evidence base for MRI remains sparse and that existing studies used CT rather than histopathology as the reference standard, therefore clinicians should interpret MRI findings cautiously until further paired-imaging studies become available.
Recommendations for future research
To strengthen the evidence base, prospective paired-imaging studies are needed that apply diagnostic nasal endoscopy, CT and MRI to the same patients, use standardised test protocols and verify findings against surgical histopathology or long-term clinical outcomes. Studies should report 2 × 2 data at predefined thresholds and explore how combining modalities modifies post-test probabilities. Economic evaluations are warranted to assess the cost-effectiveness of MRI in scenarios such as recurrent disease or suspected complications. Artificial intelligence may enhance diagnostic performance by integrating imaging with symptom scores and biomarkers, but algorithms must be externally validated.
• Chronic rhinosinusitis and chronic rhinosinusitis with nasal polyps are commonly diagnosed using nasal endoscopy, computed tomography (CT) and magnetic resonance imaging (MRI)
• No consensus exists on the relative roles of diagnostic nasal endoscopy, CT and MRI in the diagnostic pathway for chronic rhinosinusitis and chronic rhinosinusitis with nasal polyps
• This systematic review and meta-analysis demonstrates that diagnostic nasal endoscopy has high sensitivity but only moderate specificity for diagnosing chronic rhinosinusitis and/or chronic rhinosinusitis with nasal polyps
• Computed tomography shows good sensitivity but limited specificity when compared with histopathology, contributing to potential false positives
• Magnetic resonance imaging appears to offer higher specificity than diagnostic nasal endoscopy or CT, although supporting evidence is limited
Conclusion
Our results suggest that diagnostic nasal endoscopy is highly sensitive but only moderately specific relative to CT; a negative endoscopy therefore substantially lowers the post-test probability of radiologic chronic rhinosinusitis, whereas a positive endoscopy modestly increases it. Computed tomography, when benchmarked against histopathology, appears highly sensitive but limited in specificity; clinicians should interpret mild mucosal thickening within the broader clinical context rather than equating it with definitive disease. Preliminary evidence indicates that MRI may provide high specificity and moderate sensitivity compared with CT, but these estimates are derived from very few studies and rely on CT rather than histopathology as the reference. As such, MRI should currently be reserved for selected scenarios, such as assessing soft-tissue complications or avoiding radiation, and its diagnostic role remains to be clarified by larger, prospective, histology-verified studies. Collectively, these observations suggest a tiered diagnostic pathway that begins with endoscopy, progresses to CT when necessary and considers MRI for problem-solving or radiation-sparing indications, but they require confirmation. High-quality prospective studies are therefore needed to validate or refine these suggestions and to determine how best to incorporate MRI into clinical guidelines for the diagnosis and management of chronic rhinosinusitis and chronic rhinosinusitis with nasal polyps.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S002221512610437X.
Data availability statement
The datasets presented in this study can be found in online repositories. The names of the repository and/or repositories and accession number(s) can be found in the article and/or Supplementary Material.
Competing interests
None declared.
