Accuracy of human papillomavirus tests on self-collected urine versus clinician-collected samples for the detection of cervical precancer: a systematic review and meta-analysis.

OBJECTIVE: The human papillomavirus (HPV) test is an effective screening tool to prevent cervical cancer. Urinary sampling for HPV detection improves the accessibility and participation of screening services and reduces the cost and burden on physicians. The clinical accuracy of urinary HPV test has yet to be determined via meta-analysis. This study assessed the clinical accuracy of these tests to detect cervical intraepithelial neoplasia (CIN) 2 or worse.
METHODS: Relevant studies were identified using the PubMed, Embase, and Cochrane databases. Research eligibility was based on the clinical accuracy of HPV test on clinician-collected samples as a comparator test, and urine as an index test. The reference standard was the presence of CIN2 or worse. The pooled absolute, relative sensitivity, and specificity of the urinary HPV test versus clinician-collected samples were assessed using a bivariate model.
RESULTS: The pooled sensitivity of urinary HPV test was significantly lower than that of clinician-collected samples (ratio=0.84, 95% confidence interval [CI]=0.78-0.91). However, some polymerase chain reaction (PCR)-based HPV test such as GP5+/6+ (relative sensitivity=0.98, 95% CI=0.91-1.05), SPF10 (relative sensitivity=0.98, 85% CI=0.88-1.08) and non GP5+/6+ PCR (relative sensitivity=1.00, 95% CI=0.88-1.14) showed similar sensitivity in both the urine and clinician-collected samples.
CONCLUSION: Our findings indicate that HPV test with some PCR-based assay on urine versus clinician-collected samples demonstrate similar clinical accuracy to detect CIN2 or worse. It suggests that urinary HPV test may present itself as a decent alternative screening tool for the detection of cervical pre-cancer. TRIAL REGISTRATION: PROSPERO identifier: CRD42021227901.

INTRODUCTION

The human papillomavirus (HPV) test is an effective screening tool for cervical cancer used as a primary screening tool in several countries. Previous randomized controlled trials demonstrate that screening with high-risk HPV (hrHPV) is more sensitive in detecting precancerous lesions as opposed to cytological screening methods although its specificity is relatively low [1]. In 2018, the United States Preventive Services Task Force recommended screening for hrHPV alone, once every 5 years, as a preferred strategy [2]. European guideline also recommends primary testing for HPV to improve cervical cancer prevention and control.

Despite these powerful screening tests, more than 570,000 women are diagnosed with cervical cancer, and 311,000 women have succumbed to this disease in 2018 worldwide [3]. Importantly, inequalities in access to screening tools result in ethnic, racial, and social disparities in the incidence and mortality of cervical cancer [45]. Low participation is also a barrier to to cervical cancer screening, even in countries with well-established screening programs [6].

There is a major interest in showing that self-collected samples including urine for cervical screening is just as effective as clinical sampling both in developing countries due to its poor accessibility to screening units and developed countries due to low screening uptake rates, especially in light of the coronavirus disease 2019 pandemic. Self-sampling has been suggested to increase accessibility and uptake of screening services, reduce the burden on health workers, and save the costs of screening. In particular, urine sampling for HPV detection offers a more accessible and acceptable method [789]. Vaginal fluid containing exfoliated HPV infected cells are washed away with urine which allows identification of HPV DNA in urine [10]. In 2014, meta-analysis of 14 studies has shown that the urinary HPV test had a pooled sensitivity of 77% and specificity of 88% compared with clinician-collected cervical HPV test (cervical HPV test) [11]. However, this meta-analysis has not been updated for 6 years, and clinical accuracy was not assessed to detect cervical intraepithelial neoplasia (CIN) 2 or worse. Although several recent studies have demonstrated that the urinary HPV test shares similar accuracy in detecting cervical precancer compared with cervical HPV test, clinical outcomes for urinary HPV vary widely between studies [121314]. As such, there is a need to update the meta-analysis in order to comprehensively evaluate the clinical performance of urinary HPV test for secondary prevention of cervical cancer.

This study evaluated relative accuracy of the urinary HPV test versus the cervical HPV test that uses clinician-collected cervical samples to detect CIN2 or worse. We also assessed the absolute clinical accuracy of the HPV test on urine. We included a subgroup meta-analysis to evaluate the relative accuracy of urinary HPV test according to the income status of country, hrHPV assay, sampling method, sampling device, and storage medium.

MATERIALS AND METHODS

This study was registered with PROSPERO (CRD42021227901) and performed according to the standard Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement.

1. Search strategy and eligibility criteria

We conducted an electronic search for English-language literature in PubMed, Embase and the Cochrane databases from August 1, 1968 to July 15, 2020. The detailed study scheme and search strategies are described in the Fig. 1 and Table S1. Studies were deemed eligible for inclusion if they satisfied 3 key criteria: 1) the study assessed the clinical accuracy of HPV test on clinician-collected cervical and urine samples as an index test in women; 2) the reference was the presence of CIN2 or worse via colposcopic biopsy or conization; and 3) the study provided numbers of true positive, false positive, false negative, and true negative results, or this data were derived from the published results of studies.

Fig. 1

Flow chart of study identification and selection.

CIN, cervical intraepithelial neoplasia; HPV, human papillomavirus.

2. Study selection, data extraction, and quality assessment

All titles and abstracts for relevant studies were reviewed. Two authors (H.W.C and S.R.S) had independently reviewed the full text for study selection, and data extraction. Information regarding the study participants, the setting, hrHPV assays, sampling devices, sampling, storage methods, and clinical outcomes (absolute numbers of true positive, false positive, false negative, and true negative results) were collected in a comprehensive table (Table 1). We accepted the cutoff proposed by the manufacturer to define HPV positivity by tests. Two authors independently evaluated the quality of the method using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool. Disagreements between the authors were settled by consensus involving third author if necessary.

Table 1

Characteristics of studies included in the meta-analysis

Study	Study design	Population/setting	Study size (total/number for analysis)	Age	HPV assays	Urine collection	Sampling device	Preservative for urine transport	Gold standard	Clinical outcome (number)
Alameda et al. [15]	Cross-sectional	Primary screening	50	Mean, 36; range, 28–55	Urine: (PCR) using consensus primers (MY09/MY11)	Unknown	Unknown	presevCyt	Colposcopic cervical biopsy	CIN2+ (11)
					Clinician: (PCR) using consensus primers (MY09/MY11)				: All participants
Asciutto et al. [16]	Cross-sectional	Follow-up (colposcopic clinic)	218	Mean, 35.2; range, 19–71	Urine: Cobas	First stream	Plastic cup	Cobas PCR media	Colposcopic cervical biopsy	CIN2+ (112)
					Clinician: Cobas				: All participants
Asciutto et al. [17]	Cross-sectional	Follow-up (colposcopic clinic)	209	Mean, 33.7; range, 20–68	Urine: Aptima	Initial stream urine	Plastic container	No (aptima transport media)	Colposcopic cervical biopsy	CIN2+ (67)
					Clinician: Aptima				: All participants
Bernal et al. [18]	Cross-sectional	Follow-up (colposcopic clinic)	120	Median, 35.5; range, 21–65	Urine: Cobas	First stream	Sterile container	Unknown	Colposcopic cervical biopsy or conization	CIN2+ (20)
					Clinician: Cobas				: All participants
Buchegger et al. [19]	Cross-sectional	Follow-up (colposcopic clinic)	190	Median, 34; IQR, 15–75	Urine: GP5+/6+	First stream	Sterile container	No (10% crystal violet)	Colposcopic cervical biopsy or conization	CIN2+ (61)
					Clinician: GP5+/6+				: All participants	CIN3+ (33)
Cuzick et al. [12]	Cross-sectional	Follow-up (colposcopic clinic)	501	Median, 30; IQR, 27–34	Urine: Trovagene	Unknown	Unknown	Preservative solution	Colposcopic cervical biopsy	CIN2+ (145)
					Clinician: Trovagene				: All participants	CIN3+ (81)
Leeman et al. [13]	Cross-sectional	Follow-up (colposcopic clinic)	91	Range, 18–60	Urine: SPF10-DEIA-LiPA25 assay, GP5+/6+-EIA-LMNX	Morning first	Colli-Pee	4 mL of a buffered lithium dodecyl sulfate solution containing RNA preservative	Colposcopic cervical biopsy	CIN2+ (19)
					Clinician: SPF10-DEIA-LiPA25 assay, GP5+/6+-EIA-LMNX	Random first void			: All participants	CIN3+ (13)
Arias et al. [20]	Cross-sectional	Follow-up (colposcopic clinic)	433	Mean, 36; range, 21–74	Urine: Aptima	First void	Unknown	No or aptima transport media (ATS)	Colposcopic cervical biopsy or conization	CIN2+ (66)
					Clinician: Aptima				: All participants	CIN3+ (19)
Padhy et al. [21]	Cross-sectional	Follow-up (colposcopic clinic)	189	Median, 41	Urine: Aptima	Initial stream urine	Sterile container	No (aptima transport media)	Colposcopic cervical biopsy	CIN2+ (33)
					Clinician: Aptima				: group 1&2 (abnormal pap or hrHPV positive)
Piyathilake et al. [22]	Cross-sectional	Follow-up (colposcopic clinic)	502/468	<25: 292	Urine: Linear array	Unknown	Unknown	Unknown	Colposcopic cervical biopsy	CIN2+ (72)
				≥25: 210	Clinician: Linear array				: All participants	CIN3+ (34)
Rohner et al. [23]	Cross-sectional	Follow-up (colposcopic clinic)	434	Median, 36; IQR, 31–45	Urine: Onclarity	Initial stream	Unknown	0.2 mL of a proprietary preservative	Colposcopic cervical biopsy	CIN2+ (83)
					Clinician: Onclarity				: All participants
Sahasrabuddhe et al. [24]	Cross-sectional	Follow-up (colposcopic clinic)	72/71	Median, 28; range, 24–34	Urine: Linear array	Random first void	Collection cup	10 mL EDTA	Colposcopic cervical biopsy	CIN2+ (26)
					Clinician: Linear array				: All participants	CIN3+ (10)
Sahasrabuddhe et al [25]	Cross-sectional	Follow-up (colposcopic clinic)	72/71	Median, 28; range, 24–34	Urine: Trovagene	Random first void	Collection cup	10 mL EDTA	Colposcopic cervical biopsy	CIN2+ (26)
					Clinician: Linear array				: All participants	CIN3+ (10)
Sargent et al. [26]	Cross-sectional	Follow-up (colposcopic clinic)	79	40% of patients between 25–29	Urine: Cobas	First stream	Sterile dry pot	Unknown	Colposcopic cervical biopsy or conization	CIN2+ (18)
					Clinician: Cobas				: All participants	CIN3+ (8)
Sellors et al. [7]	Cross-sectional	Follow-up (colposcopic clinic)	200	Mean, 31.5; SD, 9.4	Urine: Hybrid Capture II	First void	Plastic bottle	Unknown	Colposcopic cervical biopsy	CIN2+ (58)
					Clinician: Hybrid Capture II				: All participants
Senkomago et al. [14]	Cross-sectional	Follow-up (colposcopic clinic)	37/37	Median, 42; range, 30–63	Urine: Trovagene	Morning first void	Collection cup	8 mL EDTA	Colposcopic cervical biopsy	CIN2+ (11)
					Clinician: Aptima	Initial stream			: All participants
						Mid-stream
Sorbi et al. [27]	Cross-sectional	Follow-up (conization)	134	Unknown	Urine: Linear array	First void	Unknown	Unknown	Colposcopic cervical biopsy or conization	CIN2+ (103)
					Clinician: Linear array				: All participants
Stanczuk et al. [28]	Cross-sectional	Follow-up (colposcopic clinic)	100/100	Median, 27.5; range, 21–60	Urine: Cobas	Unknown	Universal container	Unknown	Colposcopic cervical biopsy	CIN2+ (65)
					Clinician: Cobas				: All participants
Stanczuk et al. [29]	Cross-sectional	Primary screening	5,318/5,003	Median, 41.3; range, 18–76	Urine: Cobas	Random void	Universal container	6 mL was mixed with 3 mL of Roche PCR media (Roche Molecular Systems)	Colposcopic cervical biopsy	CIN2+ (130)
					Clinician: Cobas				: high-grade abnormalites	CIN3+ (68)
									: 2 low-grade or 3 borderline smears
									: 3 consecutive unsatisfactory or a subsequent abnormal smear
									: cyto−/hrHPV+ women if HPV 16 and/or 18 positive
Tshomo et al. [30]	Cross-sectional	Follow-up (colposcopic clinic)	89	Median, 39; 5–95%, 30–54	Urine: E7 MPG, GP5+/6+	First void	Colli-pee	7 mL urine preservative medium	Colposcopic cervical biopsy or conization	CIN2+ (5)
					Clinician: E7 MPG, GP5+/6+				: All participants	CIN3+ (5)
Xu et al. [31]	Cross-sectional	Primary screening	1,952/989,983	48.2±7.3 yr	Urine: Cobas, CareHPV	Random void	Sample bottle	10 mL EDTA	Colposcopic cervical biopsy	CIN2+ (20)
					Clinician: Cobas				Triage: HPV (+) and or abnormal cytology

EDTA, ethylenediaminetetraacetic acid; HPV, human papillomavirus; IQR, interquartile range; PCR, polymerase chain reaction; SD, standard deviation.

3. Data analysis

Absolute sensitivity and specificity of HPV test on the urine and clinician-collected samples were calculated using a bivariate model using midas, a STATA module for the meta-analysis of diagnostic accuracy studies. The ratio of sensitivity and specificity of urinary HPV test compared with cervical HPV test using clinician-collected samples was estimated, and a bivariate random effect model with metan functions in STATA, was used to pool the meta-analysis. Sub-analysis of relative accuracy and meta-analysis was conducted to evaluate variations in the tests accuracy based on the study setting (primary screening or follow-up) and covariates including hrHPV assays, urine collection time, urinary stream, urine collecting device, and storage medium. We conducted meta-regression and sub-analysis using the covariate to identify significant factors that had an impact on the accuracy of urinary HPV test. The accuracy of diagnostic tests, including sensitivity, specificity, and diagnostic odds ratios (DORs), were described as point estimates with 95% confidence intervals (CIs). A DOR was defined as the ratio of the odds of the test being positive if the subject has a disease relative to the odds of the test being positive if the subject does not have the disease. Statistical heterogeneity between studies was evaluated using I2 and the Cochrane Q test based on random-effects analysis. Publication bias was assessed using the Deeks' funnel plot asymmetry test for pooled absolute accuracy measurements. Statistical tests were -sided, and statistical significance was defined as p<0.05. We used Meta-DiSc (version 1·4) for meta-regression analysis, whilst all other statistical analysis was undertaken using STATA (version 16·1).

RESULTS

The literature search process is described in Fig. 1; a total of 297 studies from PubMed, 506 studies from Embase, and 48 studies from Cochrane were identified. Following review, 21 studies were deemed eligible for inclusion. All included studies were observational studies. As some studies contained up to 2–4 combinations of HPV assays, urine collection method, device and preservative use [131420263031], each of these combinations were analyzed as independent studies. Thus, data from 30 combinations for urine and 25 combinations for clinician-collected samples were included in the meta-analysis, totaling to 11,159 and 10,774 women for each collection method, respectively. For the urinary HPV test studies, generally healthy women were enrolled in 4 combinations from 3 studies [152931], whilst women who were referred to colposcopic clinics were enrolled in 26 combinations from 17 studies [712131416171819202122232425262830]. Women undergoing conization for CIN were enrolled in only one study [27].

The summary of the study design, population, demographics of participants, and details of the hrHPV assay, sample collection, sampling device, and storage methods are available in the Table 1, Tables S2 and S3.

1. Quality assessment

Fig. S1 shows the QUADAS-2 assessment in the meta-analysis. The overall quality of studies appeared to be adequate; all studies used consecutive enrollments of participants. The risk of bias for patient selection was considered low in 3/30 (10%) patients. The risk of bias for the index/comparator test and reference test was low across all studies (100%). Risk of flow and timing was considered low in 24 combinations as index, comparator and reference test were done on the same day (80%). Time interval between test was unknown in 6 (20%). Most studies were evaluated as having low applicability for patient selection (29/30, 96·7%), index/comparator test (28/30, 93·3%), and reference tests (25/30, 83·3%).

2. Relative clinical accuracy of HPV test on urine versus clinician-collected samples

The pooled relative sensitivity and specificity of HPV test on urine versus clinician-collected samples is demonstrated in Fig. 2. The overall relative sensitivity and specificity of HPV test on urine versus clinician-collected samples was 0.84 (95% CI=0.78–0.91) and 1.06 (1.03–1.10), respectively. Although the variation based on the clinical setting was not substantial, significant heterogeneity was observed in the sensitivity between studies (p<0.001). Table 2 and Fig. S2 shows that relative sensitivity of urinary versus cervical HPV are widely varied by type of hrHPV assays used. Some PCR-based HPV test such as GP5+/6+ and SPF10 showed similar levels of sensitivity and specificity in both the urine and clinician-collected samples. In contrast, Aptima and HC2 tests were less sensitive, although more specific in terms of urine compared to clinician-collected samples. Although there was no significant difference, the urinary HPV test was less sensitive than the cervical HPV test when Trovagene, Abbott, Linear array, Cobas, and CareHPV were used.

Fig. 2

(A) Relative sensitivity; and (B) specificity of urinary versus cervical HPV test, by clinical setting, for the detection of cervical intraepithelial neoplasia 2 or worse.

CI, confidence interval; HPV, human papillomavirus.

Table 2

Relative sensitivity and specificity of urinary versus cervical HPV test, using HPV assays, for the detection of cervical intraepithelial neoplasia 2 or worse

Type of Test		No. of combinations (study)	Relative sensitivity (95% CI)	Relative specificity (95% CI)
PCR-based assays
	PCR GP5+/6+	4 (3)	0.98 (0.91–1.05)	1.03 (0.90–1.18)
	PCR-SPF10*	2 (1)	0.98 (0.88–1.08)	0.92 (0.65–1.29)
	Other non GP5+/6+ PCR	3 (3)	1.00 (0.88–1.14)	1.00 (0.88–1.14)
	Trovagene†	5 (3)	0.94 (0.89–1.00)	0.94 (0.77–1.15)
	Linear Array‡	3 (3)	0.84 (0.69–1.01)	1.27 (1.11–1.45)
	Cobas§	6 (6)	0.80 (0.64–1.10)	1.03 (1.01–1.04)
Others
	CareHPV∥	1 (1)	0.87 (0.59–1.27)	1.02 (0.98–1.06)
	Aptima¶	4 (3)	0.52 (0.41–0.67)	1.39 (1.07–1.81)
	Hybrid Capture II**	1 (1)	0.46 (0.34–0.61)	1.34 (1.11–1.62)
	Abbott††	1 (1)	0.94 (0.72–1.22)	1.33 (0.86–2.07)

CI, confidence interval; HPV, human papillomavirus; PCR, polymerase chain reaction.

*SPF10 PCR-DEIA-LiPA25 version 1 (Labo Bio-medical Products, Rijswijk, the Netherlands); †Trovagene (Trovagene Inc., San Diego, CA, USA); ‡Linear Array (Roche Molecular Systems, Pleasanton, CA, USA); §Cobas 4800 HPV (Roche Molecular Systems, Pleasanton, CA, USA); ∥careHPV (QIAGEN Corporation, Germantown, MD, USA); ¶Aprima (Gen-Probe Inc., San Diego, CA, USA); **Hybrid capture II HPV (Digene Corporation, Gaithersburg, MD, USA); ††Abbott RT PCR hrHPV (Abbott Molecular Inc., Des Plaines, IL, USA).

Table 3 and Table S4 shows the relative accuracy of hrHPV assays on urine compared to that of the clinician-collected samples; accuracy was dependent on urine collection time, urinary stream, sampling device and the type of preservative used for storage. When morning first urine (ratio of sensitivity=0.97, 95% CI=0.88–1.07/ratio of specificity 1.02, 95% CI=0.77–1.34) or Colli-pee (ratio of sensitivity=0.98, 95% CI=0.91–1.05/ratio of specificity 1.03, 95% CI=0.90–1.19) was used, there was no significant difference of sensitivity and specificity between urine and clinician-collected samples. In studies of middle- and low-income country, the sensitivity (ratio=0.93 [95% CI=0.84–1.02]) and specificity (ratio=1.01 [95% CI=0.99–1.04]) of urinary HPV test were comparable with cervical HPV.

Table 3

Relative accuracy of urinary versus cervical human papillomavirus test, by other covariates, for detection of cervical intraepithelial neoplasia 2 or worse

Variables		No. of combinations (study)	Relative sensitivity (95% CI)	Relative specificity (95% CI)
Urine collection time
	Morning first	3 (3)	0.97 (0.88–1.07)	1.02 (0.77–1.34)
	Others	27 (18)	0.83 (0.77–0.90)	1.07 (1.03–1.11)
Urinary stream
	Initial stream	23 (16)	0.83 (0.76–0.91)	1.10 (1.03–1.18)
	Others	7 (6)	0.87 (0.75–1.02)	1.03 (1.00–1.05)
Collecting device
	Colli-pee	6 (2)	0.98 (0.91–1.05)	1.03 (0.90–1.19)
	Others	24 (19)	0.82 (0.75–0.89)	1.07 (1.03–1.11)
Preservative for storage
	Yes	18 (11)	0.91 (0.85–0.97)	1.03 (1.01–1.04)
	No or unknown	12 (10)	0.74 (0.63–0.88)	1.16 (1.06–1.27)
National income status
	Middle and low	5 (3)	0.93 (0.84–1.02)	1.01 (0.99–1.04)
	High	25 (18)	0.84 (0.78–0.92)	1.11 (1.04–1.18)
Population/setting
	Primary screening	4 (3)	0.79 (0.61–1.03)	1.03 (1.01–1.04)
	Follow-up/high-risk	26 (18)	0.85 (0.79–0.92)	1.11 (1.04–1.18)

CI, confidence interval.

Table 4 shows the results of multivariate meta-regression to evaluate the moderator effect of HPV assays and covariates. It suggests that the sensitivity of urinary HPV versus cervical HPV differ significantly based on the hrHPV assay type. Aside from these factors, there were no other covariates that significantly impacted the clinical accuracy of the HPV test. In addition, preservative use of urine storage was significant factor to negatively affect the specificity of urinary HPV (p=0.001).

Table 4

Effects of covariates on the relative accuracy of urinary versus cervical human papillomavirus test (multivariate meta-regression)

Variables		Relative sensitivity			Relative specificity
		Coefficient	SE	p-value	Coefficient	SE	p-value*
Urine collection time
	Morning first vs. others	0.168	0.658	0.802	0.011	0.212	0.960
Collecting device
	Colli-pee vs. others	−0.111	0.705	0.876	−0.088	0.196	0.659
Preservative for urine storage
	Yes vs. no/unknown	0.173	0.283	0.548	−0.473	0.134	0.002
Income status of country
	Middle and low vs. high	0.094	0.574	0.872	−0.066	0.159	0.684
hrHPV assays
	PCR-base vs. others	0.737	0.330	0.037	−0.028	0.122	0.819
Population/setting
	Follow-up/high-risk vs. primary screening	0.148	0.659	0.828	−0.133	0.216	0.544

hrHPV, high-risk human papillomavirus; PCR, polymerase chain reaction; SE, standard error.

*p-value from random effects meta-regression using restricted maximum likelihood.

3. Absolute clinical accuracy of HPV test on urine and clinician-collected samples

In primary screening studies, the pooled sensitivity of HPV test was 76% (95% CI=56–95) in urine, while it was 97% (95% CI=0.93–1.00) in clinician-collected samples for detection of CIN2 or worse. The pooled specificity of HPV test to exclude CIN2 or worse was similar between urine (87%, 95% CI=0.83–0.89) and clinician-collected samples (85%, 95% CI=0.79–0.91). In the follow-up studies, pooled sensitivity for detection of CIN2 or worse was 79% (95% CI=0.72–0.86) and 93% (95% CI=0.89–0.96) in urine and clinician-collected samples, respectively. The specificity in urine (48%, 95% CI=0.42–0.54) and clinician-collected samples (42%, 95% CI=0.36–0.48) were comparable. Fig. S3 presents a forest plot of absolute clinical accuracy for HPV test across all studies included in the meta-analysis.

Fig. S4 shows the summary receiver operating characteristics curves for the detection of CIN2 or worse. It suggests that the area under the curve were 0.86 (95% CI=0.83–0.89) for cervical HPV and 0.74 (95% CI=0.69–0.77) for urinary HPV. The pooled DOR was 14 (95% CI=7–30) for clinician-taken samples and 5 (95% CI=3–7) for urine samples.

4. Publication bias

The Deeks' funnel plot asymmetry test indicated that there is no significant publication bias for urinary HPV test (p=0.24) and clinician-collected samples (p=0.40) (Fig. S5).

DISCUSSION

This study demonstrates that urinary HPV test is capable of detecting ≥79% of CIN2 or worse and urinary HPV test using a some PCR-based assay such as GP5+/6+, SPF10 and non GP5+/6+ PCR had comparable sensitivity to that of clinician-collected samples. In the primary screening and follow-up setting, the sensitivity of urinary HPV was 21% and 14% lower than that of cervical HPV test using clinician-collected samples, respectively. The specificity to exclude CIN2 or worse was 87% for primary screening studies, and 48% for follow-up studies. A 2% and 6% higher specificity of urinary HPV test was observed compared to the HPV test using clinician-collected samples, respectively. In terms of relative accuracy, the sensitivity of the urinary HPV test was significantly lower than that of its clinician-collected counterpart, whilst the specificity of the urinary HPV test was slightly higher than that of the HPV test on clinician-collected samples.

This study demonstrated that the type of HPV assay used is the most important factor affecting tests accuracy. In addition, meta-regression suggests that only the type of HPV test is a significant factor impacting on sensitivity of urinary HPV compared with cervical HPV. Overall, the HPV assay based on target amplification by PCR showed a higher accuracy in urine than the HPV assay based on signal amplification or messenger RNA (mRNA) tests with Aptima. Some PCR amplifications such as GP5+/6+, other non-GP5+/6+, and SPF10 shared a similar sensitivity and specificity in both urine samples and clinician-collected samples. For Abbott or CareHPV, sensitivity was not significantly lower in urine; however, both hrHPV assays were only used in only one study [2631]. The HC2 and APTIMA urinary HPV showed considerably lower sensitivity than that of the HPV test on clinician-collected samples. These findings are consistent with studies on HPV test on self-collected vaginal samples. In meta-analyses, the PCR based HPV assays showed similar levels of sensitivity for self-collected samples compared to that of clinician-collected samples, whereas sensitivity was significantly lower when HPV test was based on signal amplification or mRNA tests [432]. In a meta-analysis of urinary HPV test, the influence of the HPV assay type had not been assessed, as PCR-based hrHPV assays were used in most studies [11]. The lower sensitivity of urinary HPV test based on signal amplification or mRNA tests may be attributable to the relatively low viral load of hrHPV from urine. As urine may contain insufficient exfoliated cervical cells, variations in sensitivity between hrHPV assays may have a significant impact on test accuracy. HPV assays based on target amplification, such as a PCR-based HPV test, are more sensitive for the detection of hrHPV compared with a HPV assay based on signal amplification [33]. In addition, several technical factors which are varied by HPV assays, including primer and DNA extraction, may contribute to the variability of urine test accuracy even among PCR-based hrHPV assays [1034]. Moreover, whether whole urine was processed instead of urine supernatant to prevent the loss of cell-free DNA may affect the outcome of urinary HPV in DNA extraction [10]. It was not assessed due to lack of information in this study.

Our findings did not demonstrate any clear moderator effects of urine collection time, urinary stream, sampling device, and preservative use for storage. When the initial urinary stream in the morning was collected or a Colli-pee or preservative solution was used, a higher sensitivity in the urinary HPV test was generally observed. However, meta-regression suggested that these covariates are not significant factors to affect the test accuracy after controlling effect of hrHPV assays. Previous meta-analysis on self-collected vaginal sample did not show an effect from the sampling device on relative sensitivity and specificity, which is in line with our findings [432]. However, finding that the relative sensitivity did not vary with urine collection time had not been anticipated. Previous comparative studies showed the favorable results for morning first or initial stream of urine. The meta-analysis demonstrated that the accuracy of HPV test on first-void urine was significantly higher than that of the midstream or random urine sample [11]. Senkomago et al. [14] compared the HPV test accuracy between the morning-first void at home, initial stream, and midstream at a clinic, finding that the hrHPV positivity rate was similar in the morning-first void and initial stream, and lower for midstream samples. Leeman et al. [13] demonstrated that HPV test in very first-void urine in the morning and another first void urine at clinic showed high sensitivity for detecting CIN2 or worse without any significant differences. Although there is no significant impact of urine sampling time in our findings, we should be cautious about drawing conclusions from these results. There is a considerable variance in the study design and settings in few comparisons of urine collection time from the included studies. Similarly, there were no comparative studies on the use of sampling devices or preservative for hrHPV DNA test and one studies on storage medium [20]. Therefore, it is difficult to conclude that the urine collection time, urinary stream, sampling device, or use of preservatives does not affect the test results.

This study suggested that sensitivity of urinary HPV versus cervical HPV was slightly higher in low- and mid-income countries than high-income countries, but national income status was not a meaningful factor that impact clinical accuracy of HPV test. Higher sensitivity of urine HPV in low- and mid-income countries may be due to high prevalence of HPV infection or multiple HPV infection [353637]. However, as only 3 clinical trials of urinary HPV test in middle and low-income countries were obtained, there is no certainty in determining as to whether the incomes of these countries materially affect the accuracy of urinary HPV test.

Urine HPV may be useful in specific population who have cultural preference or acceptance. A Thailand study demonstrated that the 3 most common reasons of women for not undergoing screening are no symptoms, fear of pain, and embarrassment [38]. In a Korean study evaluating satisfaction with cervical cancer screening modalities, satisfaction of urine HPV was significantly higher than that of cervical HPV [8]. Urine HPV may help to improve the uptake of cervical cancer screening by eliminating emotional or cultural barriers due to clinician sampling.

There are some limitations to this study; as it included primary screening and follow-up studies, the absolute sensitivity and specificity varied based on the study setting. For this reason, the relative accuracy of comparing urinary tests versus tests clinician-collected samples was assessed, and it was found that there was no substantial variance based on the study setting. There were limitations in identifying clinical factors that affect test accuracy due to the lack of detailed data from the studies that were included. Meta-regression failed to show a significant effect from urine collection, sampling device or storage medium. A well-designed clinical trial to evaluate the effect of sample handling or storage is urgently needed. Another limitation is relatively small number (21) and small sample sizes (<500) of the included studies. Especially, since there were only 3 studies in primary screening setting, further research will be needed to determine the clinical accuracy of HPV test on urine for screening purpose. In addition, the effect of sample postal conditions and clinical accuracy of HPV test on urine for detecting CIN3 were not assessed due to lack of information from included studies. Finally, there is a minor confusion regarding the definition of first void. While it generally refers to the initial stream of urine, some studies refer to it as morning first urine. To prevent misunderstandings arising from this difference, we regarded first void as initial stream of urine and if there is no mention that urine was collected in the morning, we did not regard first void as morning first urine.

Nevertheless, this study has several strengths. To the best of our knowledge, this study is the first meta-analysis that assesses the clinical performance of urinary HPV test to detect CIN2 or worse. Most previous systematic reviews have been on self-collected vaginal sample and one meta-analysis assessed only virological accuracy of urinary HPV test. In addition, we tried to examine the impact of various factors such as study population, national income status, urine collection time, urinary stream, sampling device, and preservative use for storage on clinical relative accuracy of urinary HPV test.

Our findings suggest that urinary HPV test with appropriate HPV assays, collection methods, sampling devices, and storage may serve as an alternative strategy for cervical cancer screening. This is particularly the case in certain populations that cannot be used to screen for cervical cancer and have a heavy burden of cervical cancer. This study confirmed the need for a randomized non-inferiority trial to evaluate the clinical accuracy of urinary HPV test within an organized screening setting using optimal HPV assays, collection methods, and storage. In terms of HPV test on self-collected samples, a meta-analysis, the randomized trial (IMPROVE study), confirmed that the clinical performance of HPV self-sampling is not inferior to that of clinician-collected samples to detect CIN2 or worse [39]. Furthermore, well-designed comparative studies to assess the effect of urine collection methods, collecting devices, and storage media are necessary to improve the clinical performance of HPV test in urine.

In conclusion, although clinician-collected sampling should be recommended in screening program using HPV test, urinary HPV test may be offered as an additional test method for women who do not participate in a cervical cancer screening program if validated PCR-based assays are used. Some PCR-based HPV test share similar accuracy on urine and clinician-collected samples to detect CIN2 or worse. Urinary HPV test based on signal amplification or mRNA tests should not be recommended.