PubMed/MEDLINE was searched as the primary database. The search combined terms related to specimen type, HPV testing, and diagnostic accuracy, including “urine”, “urine‑based”, “vaginal self‑sampling”, “HPV testing”, “human papillomavirus”, “cervical cancer screening”, “diagnostic accuracy”, “sensitivity”, “specificity”, and “concordance”. MeSH terms and free‑text keywords with Boolean operators were used to maximize retrieval. In addition, the reference lists of relevant articles and identified systematic reviews were manually screened to identify additional studies. The search was limited to human studies published in English, with no geographical restrictions.

Studies were included if they reported diagnostic accuracy of HPV testing in urine samples, self‑collected vaginal samples, or both, compared with clinician‑collected cervical samples. The reference standard could be HPV DNA detection in cervical specimens or histologically confirmed cervical intraepithelial neoplasia grade 2 or worse (CIN2+). Eligible studies involved human participants from screening, referral, or high‑risk populations and were published as original research articles in peer‑reviewed English‑language journals up to March 2026. Systematic reviews and meta‑analyses were retained for background context. Studies were excluded if they were editorials or lacked sufficient information to extract diagnostic performance metrics. Narrative reviews, scoping reviews, and any review that did not apply a systematic search strategy or provide quantitative meta‑analytic estimates were excluded.

Data were extracted from each included study. Extracted variables included first author, publication year, population characteristics, sample size, risk category, sample types tested (urine, self‑collected vaginal, clinician‑collected cervical), HPV assay platform, reference standard details, and diagnostic accuracy metrics such as sensitivity, specificity, percent agreement, Cohen’s kappa, and relative sensitivity or specificity when reported. For meta‑analyses used for background, pooled estimates and 95% confidence intervals were retrieved.

Population groups were assigned based on the recruitment setting and clinical context described in each primary study. Screening populations comprised asymptomatic individuals participating in organized or opportunistic cervical cancer screening programs, with no prior abnormal cytology or known HPV infection. Referral populations included individuals referred for colposcopy due to an abnormal screening test (e.g., positive HPV test, High‑grade squamous intraepithelial lesion [HSIL]) or directly from colposcopy clinics, thereby representing a higher pretest probability of CIN2+/CIN3+. High‑risk populations were defined as individuals with known risk factors for HPV persistence or cervical disease, such as HIV infection or previous treatment for cervical lesions, but not necessarily referred for colposcopy at enrolment. For diagnostic accuracy estimates, the reference standard was specified as either 1) HPV DNA detection from clinician‑collected cervical samples (using the same PCR‑based assay as applied to urine or self‑collected vaginal samples), or 2) histologically confirmed endpoints (CIN2+, CIN3+, or cancer) based on colposcopy‑directed biopsy or excisional procedure. HPV detection endpoints were used only when histology was not available or when the outcome of interest was agreement for HPV positivity rather than lesion identification. This classification was applied consistently during data extraction to allow narrative comparison across subgroups.

No meta-analysis was performed. Data from primary studies were synthesized narratively and organized thematically into overall diagnostic accuracy, direct comparisons between urine and vaginal self-sampling, population differences, assay technology and cut-off effects, pre-analytical factors, and agreement statistics. Comparable outcomes across studies were summarized using ranges and patterns rather than pooled estimates. Sources of heterogeneity, including population characteristics, collection methods, and assay differences were explored qualitatively.

3.1.1. Meta-Analytic Evidence: Pooled Summary Estimates for CIN2+/CIN3+ Detection (Table 1)

For urine-based HR-HPV testing targeting histologically confirmed CIN2+, Jeevika et al. [17] reported a pooled sensitivity of 74% (95% CI: 67 - 80) and specificity of 52% (43 - 61) from five studies. Ye et al. [18] found similar estimates across 31 studies: sensitivity 79% (72 - 84) and specificity 58% (50 - 65) for CIN2+. Using a broader endpoint that combined HSIL+ and CIN2+, Hsiao et al. [19] reported a pooled sensitivity of 83.0% (77.5 - 87.3) and specificity of 51.3% (39.2 - 63.3) for urine nucleic acid amplification tests. For vaginal self-sampling, Arbyn et al. [7] showed a pooled sensitivity of 76% (69 - 82) for CIN2+ and 84% (72 - 92) for CIN3+, with specificities of 86% (83 - 89) and 87% (84 - 90), respectively. In an updated analysis, Arbyn et al. [8] demonstrated that PCR-based HR-HPV assays on self-collected vaginal samples had a relative sensitivity for CIN2+/CIN3+ of 0.99 (0.97 - 1.02) and a relative specificity of 0.98 (0.96 - 0.99) compared with clinician collected samples. For mRNA testing on clinician collected samples, Arbyn et al. [20] reported relative sensitivities for CIN2+ and CIN3+ of 0.98 (0.95 - 1.01) each; however, on self-collected vaginal samples the same mRNA assay (APTIMA) showed a lower relative sensitivity of 0.84 (0.74 - 0.96) for CIN2+. Cho et al. [14] calculated a pooled relative sensitivity for urine HPV testing of 0.84 (0.78 - 0.91) for CIN2+. In a head-to-head meta-analysis, Li et al. [21] found that for >CIN2 lesions, vaginal self-sampling achieved a sensitivity of 98% (96 - 99) and specificity of 63% (48 - 77), while urine testing gave 95% (91 - 97) sensitivity and 62% (31 - 86) specificity.

3.1.2. Primary Studies: Urine versus Cervical Samples for CIN2+/CIN3+ Detection (Table 2)

In a screening population of 1517 women, Davies et al. [22] reported that first void urine detected 24 of 25 CIN2+ cases (sensitivity 96.0%, 95% CI 79 - 100) with a specificity of 85% (83 - 87). Xu et al. [23] compared two assays in 2,038 screening participants: careHPV gave a CIN2+ sensitivity of 85.7% (48.7 - 97.4) and specificity 86.8% (84.5 - 88.8), whereas cobas4800 yielded 69.2% (42.4 - 87.3) sensitivity and 82.3% (79.8 - 84.5) specificity. In referral populations, Van Keer and colleagues reported relative sensitivity ratios (urine/cervical) for CIN2+ of 0.96 - 0.98 with Cobas assays [24], 0.91 - 0.96 with Abbott Alinity m [25], 1.00 (0.93 - 1.07) with BD Onclarity [26], and 0.95 (0.88 - 1.01) with Abbott RealTime [27]. Zhao et al. [28] found that urine had a CIN2+ sensitivity of 87.7% (vs. cervical 85.5%) but low specificity (33.5% vs. 26.9%) in a referral cohort of 732 women. Cuzick et al. [15] reported a high CIN2+ sensitivity of 88.3% (81.9 - 93.0) for urine but very low specificity (24.7%, 20.3 - 29.5) in a colposcopy referral population. Lorenzi et al. [29] observed a CIN2+ sensitivity of 83.4% (78.4 - 87.6) and specificity of 50.8% (33.7 - 59.0) in an enriched referral cohort. For CIN2/3 detection, Sahasrabuddhe et al. [30] reported urine sensitivity of 80.8% (60.6 - 93.4) and specificity of 42.2% (27.7 - 57.8). Daponte et al. [31] found that for cervical cancer, urine sensitivity was 88.8% (51.8 - 99.7), and for high grade lesions it was 76.5% (50.1 - 93.2). Punyashthira et al. [32] reported a urine sensitivity for CIN2+ of 86.2% (cervical 94.8%) in women referred with abnormal Pap smears. Davies et al. [33] found a CIN2+ sensitivity of 90.3% (83.7 - 94.9) in a colposcopy referral cohort.

3.1.3. Primary Studies: Urine and Vaginal Self-Sampling versus Cervical Samples for CIN2+/CIN3+ Detection (Table 3)

In the VALHUDES framework, Cuschieri et al. [16] reported relative sensitivity ratios for CIN2+ of 0.95 (0.88 - 1.03) for urine and 1.05 (1.01 - 1.09) for vaginal self-samples, with corresponding relative specificities of 0.95 and 0.81. van den Borst et al. [34] found identical relative sensitivity of 0.96 for both sample types, with specificities of 0.94 (urine) and 0.98 (vaginal). Latsuzbaia et al. [35] reported relative sensitivity of 0.95 for both urine and vaginal self-samples and relative specificities of 0.95 and 0.93, respectively. In a screening population of 2228 women, Xue et al. [36] found that for CIN2+ detection, urine sensitivity was 70.0% (35.4 - 91.9) and vaginal sensitivity 90.0% (54.1 - 99.5), with very high specificities (>98% for both). Yan et al. [37] reported CIN2+ sensitivities of 86.7% (urine) and 90.6% (vaginal) with low specificities (36.5% and 30.5%, respectively) in a referral cohort of 1,588 women. Stanczuk et al. [38] demonstrated that vaginal self-sampling (Cobas 4800) had a CIN2+ sensitivity of 94.6% (90.7 - 98.5) compared with 63.1% (54.6 - 71.7) for urine, while specificity was 85.4% (84.4 - 86.3) for vaginal and 89.8% (89.0 - 90.7) for urine. Martinelli et al. [39] found CIN2+ sensitivities of 90.9% (urine) and 95.5% (vaginal), with low specificities (~40% and 36%). Senkomago et al. [40] reported CIN2+ sensitivities of 89.9% (urine) and 79.1% (vaginal) but very low specificities (34.8% and 46.2%) in a small referral cohort (n = 37). Ornskov et al. [41] showed relative sensitivities for CIN2+/CIN3+ of 1.00 (0.96 - 1.04) for vaginal and 0.96 (0.91 - 1.03) for urine. Multiple other VALHUDES related studies [42][43][44] consistently reported relative sensitivity ratios for CIN2+ between 0.92 and 1.00 for both sample types, with modestly lower specificity for vaginal self-samples in some analyses. Xu et al. [45] evaluated vaginal self-sampling alone for CIN2+ detection, reporting sensitivities of 77.4% (careHPV) and 100% (GenPlex), with specificities of 71.0% and 58.7%, respectively. Mathews et al. [46] found that for CIN2+, vaginal self-sampling using cobas HPV DNA or APTIMA mRNA achieved relative sensitivities of 0.77 - 0.94 (depending on workflow) and relative specificities of 0.87 - 0.99.

Few studies used histologically confirmed high grade squamous intraepithelial lesion or worse (HSIL+) as the exclusive reference. In Table 1, the meta-analysis by Hsiao et al. [19] combined HSIL+ and CIN2+ outcomes, reporting a pooled sensitivity of 83.0% and specificity of 51.3% for urine NAATs; however, the endpoint was not reported separately for HSIL+ alone. In primary studies (Table 2), Maged et al. [47] evaluated urine HPV testing in 1375 women referred for colposcopy, using HSIL on histology as the reference. Urine testing achieved a sensitivity of 97.8% (92.1 - 99.7) and a specificity of 100% (99.7 - 100) for HSIL detection. No other primary study in this review reported HSIL+ as a distinct endpoint without mixing with CIN2+.

In this category, the reference standard is the detection of HR-HPV DNA in clinician collected cervical samples, without histological confirmation. These estimates reflect assay agreement rather than clinical disease detection.

3.3.1. Meta-Analytic Evidence: Pooled Summary Estimates for High-Risk Human Papillomavirus (HR-HPV) Genotypes (Table 1)

Park et al. [48] reported a pooled sensitivity of 82% (78 - 86) and specificity of 91% (87 - 94) for urine real time PCR assays targeting HR-HPV detection. Ye et al. [18] also provided a pooled estimate for cervical HPV detection of 76% sensitivity (72 - 80) and 90% specificity (87 - 92). Pathak et al. [13] found that urine testing had a sensitivity of 77% (68 - 84) for HR-HPV and 73% (56 - 86) for HPV16/18, with specificities of 88% (58 - 97) and 98% (91 - 100), respectively. For any HPV detection, Pathak et al. reported 87% sensitivity and 94% specificity. For vaginal self-sampling, Li et al. [21] reported HR-HPV detection sensitivities of 92% (90 - 94) for vaginal and 83% (77 - 88) for urine, with specificities of 80% (58 - 92) and 81% (65 - 91), respectively.

3.3.2. Primary Studies: Urine versus Cervical Samples for HR-HPV Detection (Table 2)

In high-risk HIV positive women, Kaoma et al. [49] showed an HR-HPV detection sensitivity of 84.8% (68.1 - 94.9) and specificity of 92.3% (83.0 - 97.5) using first void urine. Intan et al. [50] compared urine HR-HPV ReadyMix qPCR with cervical Cobas 6800 in 876 colposcopy referrals, showing urine sensitivity of 80.9% and specificity of 100% for HR-HPV detection. Lee et al. [51] evaluated the Cobas HPV test on urine in a screening based diagnostic validation (n = 67), reporting HPV16 sensitivity of 81.3%, HPV18 sensitivity of 100%, and other HR-HPV sensitivity of 96.4%, with specificities of 98.0%, 95.3%, and 100%, respectively. Vergara et al. [52] found that urine had a sensitivity of 82.7% (78.0 - 86.8) and specificity of 94.6% (90.9 - 97.1) for any HR-HPV detection in a screening population of 543 women. Tshomo et al. [53] reported urine sensitivity of 80% (67 - 90) and specificity of 61% (43 - 76) for 21 HPV types in a follow up cohort. Lim et al. [54] compared two assays on urine in 100 screening women: Roche cobas 4800 gave HPV16/18 sensitivity of 79.2% (57.9 - 92.9) and Abbott RealTime 81.8% (59.7 - 94.8), with 100% specificity for both. Combita et al. [55] showed urine sensitivity of 90.6% (87.0 - 94.0) for any HPV and 62% - 100% for type specific detection, with specificity of 74.1% (68.0 - 80.0) for any HPV and 96% - 99% for types 6/11/16/18. Tanzi et al. [56] found near perfect agreement in HIV positive women: urine sensitivity 98.6% (93.1 - 99.9) and specificity 97.4% (87.7 - 99.9) for HPV detection.

3.3.3. Primary Studies: Urine and Vaginal Self-Sampling versus Cervical Samples for HR-HPV Detection (Table 3)

Giubbi et al. [57] reported that for HR-HPV detection, urine had a sensitivity of 93.4% and specificity of 83.5%, while vaginal self-sampling achieved 99.1% sensitivity and 85.0% specificity (κ = 0.773 and 0.870, respectively). Mizuno et al. [58] evaluated BD Onclarity in an abnormal cytology cohort (n = 121): urine sensitivity was 46.4% (36.0 - 57.1) and specificity 77.1% (60.2 - 88.5), whereas vaginal self-sampling gave 75.7% (65.7 - 85.6) sensitivity and 69.0% (52.2 - 85.7) specificity. Kuriakose et al. [59] compared urine directly with vaginal self-sampling without a cervical reference, reporting a urine sensitivity relative to vaginal of 57.9% (47.5 - 67.7) and specificity of 84.6% (66.4 - 93.8). Tranberg et al. [60] studied 150 women referred with ASC-US positive cytology: urine vs. vaginal self-sample gave a sensitivity of 59.1% (43.2 - 73.7) and specificity of 99.1% (94.9 - 99.7) for HR-HPV detection; urine vs. cervical samples gave 63.9% (46.2 - 79.2) sensitivity and 96.5% (91.3 - 99.0) specificity.

To help interpret the diagnostic accuracy estimates we assessed the included primary studies using four QUADAS 2 domains: patient selection, index test, reference standard, and flow/timing. We did not apply formal scoring; instead, we summarize the main risk-of-bias and applicability concerns qualitatively.

Many studies enrolled women from colposcopy referral cohorts with a high CIN2+ prevalence, including Van Keer et al. [24], Zhao et al. [28], Punyashthira et al. [32], Lorenzi et al. [29], and Cuzick et al. [15]. These referral-enriched samples may inflate sensitivity and limit specificity relative to true screening populations. By contrast, fewer studies were conducted in screening settings, such as Davies et al. [22], Xu et al. [23], Vergara et al. [52], and Stanczuk et al. [38], and these are more applicable to population-based screening. High-risk cohorts, including HIV-positive women in Kaoma et al. [49] and Tanzi et al. [56], further restrict generalizability. For example, in a screening population, Davies et al. reported urine sensitivity of 96.0% for CIN2+ and specificity of 85.2% [22], whereas in a referral cohort, Cuzick et al. found sensitivity of 88.3% but specificity of only 24.7% [15], illustrating the effect of disease enrichment on predictive performance. Among HIV-positive women, Kaoma et al. reported sensitivity of 84.8% and specificity of 92.3% for CIN2+ [28], while Tanzi et al. found nearly perfect agreement for overall HPV detection (sensitivity 98.6%, specificity 97.4%) [56]. In younger women (aged 19 - 25 years), Combita et al. found that first‑void urine produced HPV prevalence estimates similar to cervical samples [55]. Overall, the large variation in CIN2+ prevalence across studies indicates substantial spectrum bias, so direct numerical comparisons should be interpreted cautiously.

Two main reference standards were used: clinician-collected cervical HPV DNA testing, as in [50], [51], and [54], and histologically confirmed CIN2+, as in [22], [26], [38], and [47]. Cervical HPV comparison mainly reflects assay agreement rather than clinical disease detection, which may overstate performance for CIN2+ detection. Where histology was used, many studies verified only HPV-positive or abnormal cytology cases, creating partial verification bias, particularly in referral cohorts such as Van Keer et al. [27] and Lorenzi et al. [29]. Only a few large studies, including Stanczuk et al. [38] and Xu et al. [23], applied colposcopy and biopsy to all participants and therefore avoided this bias.

Most studies used commercial PCR-based assays, including Cobas, Abbott, BD Onclarity, Anyplex, and GeneXpert, but these platforms were not always formally validated for urine or vaginal self-samples. Pre-analytical procedures also varied, including first-void versus random urine, preservation methods, and extraction kits. Assay selection and threshold setting strongly influence diagnostic performance. Commercial PCR‑based assays generally show better and more consistent results than less standardized platforms. Van Keer et al. [25] found that the Abbott Alinity m HR-HPV assay on first‑void urine achieved a relative sensitivity of 0.91 and relative specificity of 1.04 at the manufacturer’s cut‑off; after cut‑off adjustment, performance improved to 0.96 and 1.00, respectively. Similar patterns were observed with the OncoPredict HPV QT assay in Latsuzbaia et al. [43], where sensitivity was already high but specificity improved after optimization. Cuschieri et al. [16] and Giubbi et al. [42] also showed that cut‑off optimization improved the clinical performance of self‑samples, particularly by restoring specificity toward the cervical benchmark. The meta‑analysis by Cho et al. [14] further emphasized assay dependence, noting that some PCR‑based assays showed relative sensitivity close to 1.00, whereas the overall pooled relative sensitivity for urine was 0.84. Regarding specimen type and target, Mathews et al. [46] suggested that mRNA assays may perform at least as well as DNA assays on self‑collected vaginal samples, although evidence remains limited.

Pre‑analytical conditions also have a substantial effect on urine HPV detection. First‑void urine consistently outperforms random or midstream urine because it contains more exfoliated cervicovaginal material. Senkomago et al. [40] showed that initial‑stream urine performed similarly to first‑void urine, achieving 89.9% sensitivity for CIN2+, whereas midstream specimens were less informative. Davies et al. [33] demonstrated that a dedicated first‑void device (Novosanis Colli-pee® 10 mL with urine conservation medium) improved sensitivity for CIN2+ to 90.3%, compared with 73.4% using a standard pot. Preservation and processing also matter: Kaoma et al. [49] used EDTA‑preserved first‑void urine and obtained strong performance in HIV‑positive women. The tested urine fraction (pellet vs. supernatant) is another relevant variable; both may contain HPV DNA, but yield depends on collection and processing methods. Overall, the evidence supports standardized first‑void collection, prompt preservation, and validated extraction workflows as prerequisites for reliable urine‑based HPV testing.

Agreement between self‑collected and clinician‑collected samples ranges from fair to excellent, depending on assay, population, and study design. The lowest agreement in this evidence set was κ = 0.29 [59] (Kuriakose et al., 2020), while Tanzi et al. [56] reported near‑perfect agreement for HPV detection in HIV‑infected women (κ = 0.96). Most recent VALHUDES studies showed moderate to substantial agreement, e.g., κ = 0.68 - 0.87 for Cobas assays [24] and κ = 0.68 - 0.87 for Alinity m HR-HPV [44]. Importantly, agreement statistics alone do not always reflect clinical usefulness; some studies showed only moderate κ while still achieving non‑inferior CIN2+ detection [22][23].

Studies using the same assay across urine, vaginal, and cervical samples, such as Latsuzbaia et al. [43], and Giubbi et al. [42], have stronger internal validity. In contrast, studies using different assays for urine and cervical samples, such as [50], may reflect assay disagreement as well as true sample differences.

Most studies were cross-sectional and collected paired urine, vaginal, and cervical samples at the same visit, which is appropriate for diagnostic accuracy assessment. However, blinding of the reference standard to index test results was rarely reported, and the interval between sampling and biopsy was not always stated. Overall, the evidence is limited by referral-enriched cohorts, mixed reference standards, and partial verification. Accordingly, studies with screening populations, histology applied to all participants, and the same assay across specimen types provide the most robust estimates, whereas the reported ranges in sensitivity and specificity, including urine sensitivity for CIN2+ from 63% to 96%, should be interpreted in light of these limitations.