Diagnostic accuracy of HPV16 early antigen serology for HPV-driven oropharyngeal cancer is independent of age and sex

Abstract

A growing proportion of head and neck cancer (HNC), especially oropharyngeal cancer (OPC), is caused by human papillomavirus (HPV). There are several markers for HPV-driven HNC, one being HPV early antigen serology. We aimed to investigate the diagnostic accuracy of HPV serology and its performance across patient characteristics. Data from the VOYAGER consortium was used, which comprises five studies on HNC from North America and Europe. Diagnostic accuracy, that is, sensitivity, specificity, Cohen's kappa and correctly classified proportions of HPV16 E6 serology, was assessed for OPC and other HNC using p16^INK4a immunohistochemistry (p16), HPV in situ hybridization (ISH) and HPV PCR as reference methods. Stratified analyses were performed for variables including age, sex, smoking and alcohol use, to test the robustness of diagnostic accuracy. A risk-factor analysis based on serology was conducted, comparing HPV-driven to non-HPV-driven OPC. Overall, HPV serology had a sensitivity of 86.8% (95% CI 85.1-88.3) and specificity of 91.2% (95% CI 88.6-93.4) for HPV-driven OPC using p16 as a reference method. In stratified analyses, diagnostic accuracy remained consistent across sex and different age groups. Sensitivity was lower for heavy smokers (77.7%), OPC without lymph node involvement (74.4%) and the ARCAGE study (66.7%), while specificity decreased for cases with <10 pack-years (72.1%). The risk-factor model included study, year of diagnosis, age, sex, BMI, alcohol use, pack-years, TNM-T and TNM-N stage. HPV serology is a robust biomarker for HPV-driven OPC, and its diagnostic accuracy is independent of age and sex. Future research is suggested on the influence of smoking on HPV antibody levels.

1 |. INTRODUCTION

With over 900 000 new annual cases and 450 000 deaths worldwide,¹ head and neck cancer (HNC) is a serious health concern. HNC comprises a diverse group of cancers, including those of the oral cavity, pharynx, larynx and sinonasal cavity.^2,3 Traditionally, the main causes of HNC have been long-term tobacco use and alcohol consumption. More recently, it has been recognised that oncogenic human papillomavirus (HPV) types, especially HPV16, cause a subset of HNC as well.^2,4 Much remains unclear about the progression of HPV infection to HNC. It is suggested that patients with HPV-driven HNC are slightly younger than those with non-HPV-driven HNC.⁴ HPV-driven HNC incidence varies between countries and populations, as it is more common in North America and Northern Europe, more common among men, and has an increasing incidence in recent decades.^4,5

HPV-driven HNC has a unique biology, pathology and clinical features.^6,7 Therefore, patients with HPV-driven HNC might have different treatment needs⁸: HPV-positive HNC patients usually respond better to radiotherapy and chemotherapy.^6,9 For oropharyngeal cancers (OPC) the difference in prognosis is even stronger, for which an increase in HPV-driven cancer is seen in recent years.¹⁰ The five-year survival of HPV-positive OPC is 70% to 80% compared to 25% to 40% for HPV-negative OPC.^6,7 Whether the difference in survival is due to molecular pathogenesis, or related to age and overall health of patients is unclear.⁹ Additionally, HPV-positive cancer patients are less likely to experience recurrence of OPC.⁶ Therefore, being able to distinguish HPV-driven HNC from non-HPV-driven HNC is of clinical and epidemiological importance.

Several methods are available to define whether an HNC is HPV-driven, each with their own strengths and limitations. HPV DNA detection by polymerase chain reaction (PCR) only suggests the presence of viral DNA, but does not identify biological activity of the virus driving carcinogenic processes.^11–13 The detection of both HPV DNA and RNA by PCR is considered more reliable, although these methods are laborious and most clinical routine laboratories might not have access to both DNA and RNA PCR.^11,12 HPV DNA detection by in situ hybridisation (ISH) not only detects the presence of HPV, but also the location of the viral genome in the host cell.¹⁴ Nevertheless, ISH tends to have a lower sensitivity, than for example PCR, and is prone to cross-contamination.^12,15 A widely used marker is p16^INK4a immunohistochemistry (p16). The HPV E7 protein degrades the retinoblastoma tumour suppressor protein, leading to the overexpression of p16.¹³ Therefore, p16 is used as a surrogate marker for active HPV. This inexpensive method is easy to interpret, well studied, became the primary method in routine clinical diagnostics for OPC,^13,14 and has further been incorporated into the most recent version of the HNC staging.¹⁶ However, this surrogate marker is not HPV-specific, resulting in suboptimal specificity, especially outside the oropharynx.^12–14,17

Serum antibodies against HPV16 early antigens have emerged as a highly sensitive and specific blood-based biomarker to distinguish HPV-driven from non-HPV-driven HNC, especially for OPC.¹¹ The main advantages of this method include that it only requires a blood draw, not tumour tissue, that it is less susceptible to cross-contamination, and it can be used in epidemiological studies where tumour tissues are not available.¹¹ Serology could additionally be used as a rapid screening method for subsites of the oropharynx that are not routinely tested, particularly when attributable fractions (AFs) are low. A recent systematic review showed that HPV16 E6 serology had the best overall performance, with a sensitivity of 83.1% and specificity of 92.5%,¹¹ making this the best single antibody marker for HPV-driven OPC. In some rare cases, HPV-driven OPC cases are HPV16 E6 negative, but positive for other HPV16 early (E) antibodies, such as E1, E2 or E7.¹⁸ Based on this, OPC can be also considered HPV-driven when having E6 seropositivity (>1000 MFI), or seropositivity to at least three E proteins.¹⁸ It is unclear whether the sensitivity and specificity of this blood-based biomarker varies by patient characteristics, such as age and sex. Immune responses differ in ageing men and women, which might result in a different diagnostic accuracy.^19,20 Therefore, our study, based on data from an international consortium, aimed to determine the diagnostic accuracy of HPV16 early antigen serology as marker for HPV-driven HNC, and to estimate robustness of the diagnostic accuracy across population characteristics such as age and sex.

2 |. MATERIALS AND METHODS

2.1 |. Study population

For our study, data from the NIH-funded VOYAGER (Human Papillomavirus, Oral and Oropharyngeal Cancer Genomic Research) consortium were used. This consortium consists of five studies on HNC patients across 10 countries in North America and Europe: (a) the Alcohol-Related Cancers and Genetic Susceptibility in Europe (ARCAGE) study, (b) the Carolina Head and Neck Cancer Study (CHANCE), (c) Head and Neck 5000 (HN5000), (d) the University of Pittsburgh head and neck cancer case-control study (Pittsburgh), and (e) the Mount Sinai Hospital-Princess Margaret (MSH-PMH) study in Toronto (Toronto). The details of these studies have been described previously.^21–25 Four are case-control studies^21–23,25 and one is a case-series.²⁴ For our analyses, we only used data from HNC cases. In all studies, demographic and lifestyle information were obtained via administered questionnaires. In this analysis, only HNC cases with serology data available were included.

2.2 |. Tumour classifications

All cancers were classified based on the International Classification of Diseases Volume 10 (ICD-10). Cancers were classified as HNC if they were at the oropharynx (including base of tongue/lingual tonsil), larynx, oral cavity and lip (including salivary glands), nasopharynx, hypopharynx and sinonasal cavities. Other sites (eg, oesophagus, thyroid or jaw) were excluded from our analysis. If the tumour had overlapping sites within the head and neck region without a clear designation of the primary site, it was labelled as having overlapping sites. Finally, tumours were labelled as unspecified if the location was within the head and neck region, but the exact location was unclear (eg, tongue, not otherwise specified).

Tumour stage was classified using the seventh edition of the cancer staging system manual of the American Joint Committee on Cancer (AJCC).^26,27 This staging is based on the tumour size (TNM-T), regional lymph node involvement (TNM-N) and metastasis (TNM-M). We discussed using the new AJCC eighth edition and decided that this would not be appropriate for these analyses given that the new edition already incorporated HPV status into its staging classification.²⁸

2.3 |. HPV serology procedures

All serological analyses were conducted at the German Cancer Research Center (DKFZ, Heidelberg, Germany). HPV antibodies were measured using multiplex serology, a bead-based assay that allows analysis of large numbers of serum samples for antibodies against multiple viral antigens simultaneously.²⁹ Antibodies against the major capsid protein (L1), early oncoproteins (E6, E7), and regulatory early proteins (E1, E2) were measured for six oncogenic HPV-types: HPV16, HPV18, HPV31, HPV33, HPV45 and HPV52. Antibody levels were further dichotomised into positive and negative, based on predefined cut-offs for median fluorescence intensity (MFI) units (Table S3).^18,29,30

2.4 |. Markers of HPV-driven tumours

In our analysis, we used three reference methods to determine if HNC was HPV-driven: (a) p16 immunohistochemistry on tumour tissue was used in patient subsets of all five studies; local determinations for p16 positivity followed standard guidelines; (b) ISH was used in subsets of both the HN5000 study and the Pittsburgh study populations; and (c) PCR was used in a subset of the HN5000 study population. The selection of methods per patient was done as part of the patient’s care, so not specifically for our study. More details can be found in Figure 1 and Tables S1 and S2.

FIGURE 1.

Venn diagram of the study population having had p16^INK4a immunohistochemistry, HPV DNA by ISH and/or HPV DNA and RNA by PCR. DNA, deoxyribonucleic acid; HPV, human papillomavirus; ISH, in situ hybridisation; p16, p16^INK4a immunohistochemistry; PCR, polymerase chain reaction; RNA, ribonucleic acid.

Finally, HPV serology was determined for all participants. Tumours were considered to be HPV-driven based on serology (ie, HPV seropositive) if HPV16 E6 was positive (with cut-off >1000 MFI), or at least three HPV16 E-proteins were positive (using a lower cut-off >484 MFI for E6); serological results not meeting the aforementioned data but meeting quality control were considered HPV seronegative.¹⁸

2.5 |. Statistical analysis

Study population characteristics were described in relation to HPV serology using descriptive statistics, for the combined study population and for each individual study. HPV AFs were calculated per anatomical location, according to HPV serology.

The overall diagnostic accuracy, as measured by sensitivity, specificity, Cohen’s kappa and correctly classified proportion with corresponding 95% confidence intervals (CIs), were calculated for OPC with p16, ISH and PCR as reference methods, separately. Additionally, a combination of markers was used as a reference, as this is sometimes used in clinical settings; if both p16 and either ISH or PCR were positive, an OPC was considered to be HPV-driven. Cohen’s kappa was considered low if the kappa value was below 0.20, fair between 0.20 and 0.40, moderate between 0.40 and 0.60, high between 0.60 and 0.80, and very high if kappa values were higher than 0.80.³¹ To evaluate appropriateness of HPV serology to assess HPV-attributability of tumours at other anatomic locations than the oropharynx, diagnostic accuracy was also calculated for cancer at the larynx and oral cavity, using only p16 status as reference test, as this was the only test with sufficient participants. Finally, this was also done for sub-entities of the oropharynx, being the tonsils and base of the tongue.

To assess variation in diagnostic accuracy across patient characteristics, stratified analyses were conducted. Estimates were stratified for variables potentially related to HPV-driven OPC (compared to non-HPV-driven OPC based on HPV serology), being: year of diagnosis (categories based on tertiles), age in categories of 5 years, sex, body mass index (BMI) status at diagnosis, current alcohol use, pack-years, TNM-T, TNM-N and TNM-M stage. Additionally, estimates were stratified by study. For the stratified analyses, p16 was used as a reference method.

Finally, among OPC patients, univariable and multivariable logistic regression models were used to identify risk factors for HPV-driven (based on HPV serology) OPC, compared to non-HPV-driven OPC. For the multivariable model, backwards selection was done based on the Akaike Information Criterion (AIC). Independent variables considered for the model are the same variables as for the stratified diagnostic accuracy, and were tested on collinearity before entering the model with help of a multicollinearity matrix. No strong collinearity was observed. All statistical analyses were conducted using RStudio (Version 1.3.959). For all analyses, statistical significance was defined as P < .05.

3 |. RESULTS

3.1 |. Study population

The study population comprised 6809 participants with HNC diagnosed between 2002 and 2017, of which 2234 (32.8%) were HPV seropositive (Table 1). The majority of the participants were male (76.6%), ever smoker (65.3%) and consumed alcohol (57.6%). The median age of the study participants was 60.5 years old (interquartile range: 53.1-68.0). The HN5000 study contributed the most participants (60.5%). Characteristics per study are presented in Table S1.

TABLE 1.

Characteristics of the VOYAGER study population, consisting of patients with HNC diagnosed between 2002 and 2017, according to HPV serology.

Variables	Total, n (%)	Sero−, n (%)	Sero+, n (%)	P-value
Total	6809 (100.0)	4575 (67.2)	2234 (32.8)
Study				<.001
ARCAGE	945 (13.9)	874 (19.1)	71 (3.2)
CHANCE	509 (7.5)	333 (7.3)	176 (7.9)
HN5000	4119 (60.5)	2934 (64.1)	1185 (53.0)
Pittsburgh	371 (5.4)	129 (2.8)	242 (10.8)
Toronto	865 (12.7)	305 (6.7)	560 (25.1)
Geographical location				<.001
Continental Europea	848 (12.5)	792 (17.3)	56 (2.5)
North Americab	1745 (25.6)	767 (16.8)	978 (43.8)
United Kingdom	4216 (61.9)	3016 (65.9)	1200 (53.7)
Year of diagnosisc				<.001
2002-2011	2182 (32.0)	1557 (34.0)	625 (28.0)
2012-2013	2664 (39.1)	1802 (39.4)	862 (38.6)
2014-2017	1963 (28.8)	1216 (26.6)	747 (33.4)
Age in yearsd	60.5 (53.1, 68.0)	62.0 (54.7, 70.0)	57.9 (52.0, 64.0)	<.001
Sex				<.001
Male	5215 (76.6)	3372 (73.7)	1843 (82.5)
Female	1594 (23.4)	1203 (26.3)	391 (17.5)
BMI at diagnosis in kg/me	25.9 (23.1, 29.4)	25.1 (22.3, 28.6)	27.4 (24.5, 31.0)	<.001
Unknown/NA	1326	1008	318
Current alcohol use				<.001
Yes	3923 (57.6)	2657 (58.1)	1266 (56.7)
No	1791 (26.3)	1100 (24.0)	691 (30.9)
Unknown/NA	1095 (16.1)	818 (17.9)	277 (12.4)
Drink intensity, units per dayd	2.6 (1.2, 5.7)	3.4 (1.3, 5.9)	2.0 (0.9, 3.5)	<.001
Unknown/NA	2515	1679	836
Smoking status				<.001
Current smoker	1793 (26.3)	1498 (32.7)	295 (13.2)
Former smoker	2651 (38.9)	1666 (36.4)	985 (44.1)
Never smoker	1227 (18.0)	575 (12.6)	652 (29.2)
Unknown/NA	1138 (16.7)	836 (18.3)	302 (13.5)
Smoking packyearsd	22.5 (0.0, 43.0)	30.0 (8.0, 48.0)	7.0 (0.0, 29.5)	<.001
Unknown/NA	1837	1363	474
Anatomical location tumour				<.001
Hypopharynx	262 (3.8)	245 (5.4)	17 (0.8)
Larynx	1254 (18.4)	1230 (26.9)	24 (1.1)
Nasopharynx	102 (1.5)	89 (1.9)	13 (0.6)
Oral cavity	1635 (24.0)	1581 (34.6)	54 (2.4)
Oropharynx	3266 (48.0)	1152 (25.2)	2114 (94.6)
Sinonasal	97 (1.4)	94 (2.1)	3 (0.1)
Overlapping sites	27 (0.4)	26 (0.6)	1 (0.0)
Unspecified	166 (2.4)	158 (3.5)	8 (0.4)
TNM-T				<.001
Tx/Tisf	198 (2.9)	169 (3.7)	29 (1.3)
T0	23 (0.3)	12 (0.3)	11 (0.5)
T1	1828 (26.8)	1322 (28.9)	506 (22.6)
T2	2304 (33.8)	1380 (30.2)	924 (41.4)
T3	1110 (16.3)	703 (15.4)	407 (18.2)
T4	1346 (19.8)	989 (21.6)	357 (16.0)
TNM-N				<.001
Nxf	209 (3.1)	188 (4.1)	21 (0.9)
N0	2991 (43.9)	2722 (59.5)	269 (12.0)
N1	763 (11.2)	481 (10.5)	282 (12.6)
N2	2702 (39.7)	1125 (24.6)	1577 (70.6)
N3	144 (2.1)	59 (1.3)	85 (3.8)
TNM-M				<.001
Mxf	321 (4.7)	293 (6.4)	28 (1.3)
M0	6409 (94.1)	4224 (92.3)	2185 (97.8)
M1	79 (1.2)	58 (1.3)	21 (0.9)
P16 status				<.001
Negative	1201 (37.4)	1149 (75.6)	52 (3.1)
Positive	2007 (62.6)	370 (24.4)	1637 (96.9)
Unknown/NA	3601	3056	545
HPV DNA by ISH status				<.001
Negative	129 (30.3)	84 (74.3)	45 (14.4)
Positive	297 (69.7)	29 (25.7)	268 (85.6)
Unknown/NA	6383	4462	1921
HPV DNA/RNA by PCR status				<.001
Negative	299 (52.8)	287 (80.4)	12 (5.7)
Positive	267 (47.2)	70 (19.6)	197 (94.3)
Unknown/NA	6243	4218	2025

Note: P-value based on χ²-test or Wilcoxon rank sum test; Sero−: HPV serology negative; Sero+: HPV serology positive; Unspecified or overlapping sites: HNC cancer without a clear primary; BMI not overweight: <25; BMI overweight: ≥25 and <30; BMI obese: ≥30; TNM classification is a classification to describe tumour size, lymph node involvement and metastasis.

Abbreviations: BMI, body mass index; DNA, deoxyribonucleic acid; HNC, head and neck cancer; HPV, human papillomavirus; ISH, in situ hybridisation; NA, not available; PCR, polymerase chain reaction; RNA, ribonucleic acid.

^aCountries for Continental Europe: Croatia, Czech Republic, Germany, Greece, Ireland, Italy and Spain.

^bCountries for North America: Canada, United States of America.

^cCategories based on tertiles.

^dMedian (IQR).

^fTx/Nx/Mx: diagnosed with HNC, but T, N or M status not assessed.

OPC was the most common HNC in our dataset, with 3266 cases (48.0%), followed by oral cavity (24.0%) and laryngeal cancer (18.4%). Of the HPV seropositive participants, 94.6% had OPC (Table 1). Table 2 presents an overview of the anatomical locations, with HPV AFs based on HPV serology. The AF of HPV for OPC was 64.7%. For laryngeal cancer and oral cavity cancer, the HPV AFs were only 1.9% and 3.3%, respectively (Table 2).

TABLE 2.

HPV attributable fractions in HNC cases, according to HPV serology.

Anatomical location tumour	Totals n (%)	Sero− n (%)	Sero+ n (%)
Oropharynx	3266 (100)	1152 (35.3)	2114 (64.7)
Oral cavity	1635 (100)	1581 (96.7)	54 (3.3)
Larynx	1254 (100)	1230 (98.1)	24 (1.9)
Hypopharynx	262 (100)	245 (93.5)	17 (6.5)
Nasopharynx	102 (100)	89 (87.3)	13 (12.7)
Sinonasal	97 (100)	94 (96.9)	3 (3.1)
Unspecified or overlapping sites	193 (100)	184 (95.3)	9 (4.7)

Note: Unspecified or overlapping sites: HNC cancer without a clear primary. Sero−: HPV serology negative; Sero+: HPV serology positive.

3.2 |. Overall diagnostic accuracy of HPV serology for OPC

Table 3 presents the overall diagnostic accuracy of HPV serology to detect whether an OPC is HPV-driven or not, using three reference methods. Across all three reference methods, HPV serology had a high sensitivity. Against p16 as a reference, sensitivity was 86.8%, against ISH 90.6% and against PCR 83.8%. Specificity of HPV serology was also high against p16 (91.2%) and PCR (93.8%), but lower against ISH (58.0%). With the combined markers as a reference, HPV serology had a sensitivity of 90.2% and specificity of 67.0%.

TABLE 3.

Diagnostic accuracy of HPV serology for HPV-driven OPC, compared to the reference methods p16 status, HPV DNA by ISH and HPV DNA/RNA by PCR.

Oropharyngeal cancer
	p16−	p16+	Sensitivity (95% CI)	Specificity (95% CI)	Cohen’s kappa (95% CI)	Correctly classified proportion (95% CI)
Sero−	510	241	86.8 (85.1-88.3)	91.2 (88.6-93.4)	0.70 (0.67-0.73)	87.8 (86.4-89.1)
Sero+	49	1580
	ISH−	ISH+	Sensitivity (95% CI)	Specificity (95% CI)	Cohen’s kappa (95% CI)	Correctly classified proportion (95% CI)
Sero−	58	27	90.6 (86.6–93.7)	58.0 (47.7-67.8)	0.51 (0.41-0.61)	82.2 (78.0-85.9)
Sero+	42	260
	PCR−	PCR+	Sensitivity (95% CI)	Specificity (95% CI)	Cohen’s kappa (95% CI)	Correctly classified proportion (95% CI)
Sero−	45	35	83.8 (78.2-88.4)	93.8 (82.8-98.7)	0.62 (0.51-0.72)	85.6 (80.8-89.6)
Sero+	3	181
	Combined−	Combined+	Sensitivity (95% CI)	Specificity (95% CI)	Cohen’s kappa (95% CI)	Correctly classified proportion (95% CI)
Sero−	75	39	90.2 (86.8-92.9)	67.0 (57.4-75.6)	0.48 (0.57-0.65)	85.1 (81.7-88.1)
Sero+	37	358

Note; Sero−: HPV serology negative; Sero+: HPV serology positive. Combined: If for OPC patients both P16 and either ISH or PCR was positive, then positive, otherwise negative.

Abbreviations: DNA, deoxyribonucleic acid; HPV, human papillomavirus; ISH, in situ hybridisation; OPC, oropharyngeal cancer; PCR, polymerase chain reaction; RNA, ribonucleic acid.

Only a subset of patients received either an ISH test or p16 test, and the availability of each test was not random. Thus, OPC cases with ISH results might have differed from cases with p16 results. Therefore, a sensitivity analysis was conducted for OPC patients who had both p16 and ISH test results (n = 375), which resulted in similar diagnostic accuracy as the overall estimate (Table S5). Unfortunately, due to a small sample size, the corresponding comparison was not possible to interpret for individuals with both p16 and PCR test results.

Diagnostic accuracy for serology in tonsillar cancer and cancer of base of the tongue was similar to the overall estimates (Table S5). Finally, Table S5 also shows the diagnostic accuracy of HPV serology for larynx cancer and oral cavity cancer with p16 as a reference method, showing a low sensitivity for both anatomical locations.

3.3 |. Stratified diagnostic accuracy analyses

The stratified diagnostic accuracy was determined for participants with OPC, for which both HPV serology and p16 were available (details of this subpopulation in Table S4). Of the 3266 OPC cases, there were 2380 (72.9%) for whom p16 stats was available. Information on ISH and PCR was available for a much smaller number of study participants, 387 and 264, respectively and could therefore not be used in the stratified analyses.

Stratified diagnostic accuracy analyses were performed for strata of several variables (see Figure 2 and Table S6). Sensitivity estimates did not vary substantially across the age and sex strata. Additionally, they were consistent across all strata of year of diagnosis, BMI status at diagnosis, current alcohol use and TNM-T stage, with sensitivities ranging from 82.6% to 89.6% for those strata, which is similar to the overall estimate against p16 (86.8%). Sensitivities across the studies were inconsistent, with the ARCAGE study, involving cases from Europe, showing a lower sensitivity of 66.7%. Additionally, a lack of confirmed lymph node involvement (TNM–Nx-N0) resulted in a drop of sensitivity to 74.4%, compared to the 90.0% sensitivity in participants with documented lymph node involvement (TNM–N1-N3). Finally, among smokers with 40 or more packyears, sensitivity was lowered to 77.7%.

FIGURE 2.

Forest plot of the overall, and stratified analyses, for sensitivity and specificity of HPV serology to detect HPV-driven OPC. For these analyses, p16^INK4a immunohistochemistry was used as reference method. Categories for year of diagnosis are based on tertiles. BMI not overweight: <25; BMI overweight: ≥25 and <30; BMI obese: ≥30. T_x/N_x, diagnosed with OPC, but T or N status not assessed. BMI, body mass index; CI, confidence interval; OPC, oropharyngeal cancer.

For specificity, similar estimates were seen across the strata for study, year of diagnosis, age, sex, BMI status at diagnosis, current alcohol use, TNM-T and TNM-N. Specificities ranged from 85.1% to 98.6% for those strata, which was similar to the overall estimate of 91.2% (Figure 2, Table S6). The specificity differed between the smoking strata, with a lower specificity of 72.1% in the stratum of <10 packyears.

Cohen’s kappa varied from 0.42 (for individuals with <10 packyears of smoking) to 0.80 (for participants who had TNM–N1-N3 disease). Of the 30 strata examined, 27 had a kappa of at least 0.60. Correctly classified proportions were all considered high, ranging from 80.8% to 92.1% (Table S5).

3.4 |. Risk-factor model

Univariable and multivariable logistic regression models were generated for all participants with OPC with HPV serology (n = 3266; Table 4). The univariable logistic regression model showed increasing odds ratios (OR) for having an HPV-driven OPC when having a more recent year of diagnosis, being of lower age, being a male, having a higher BMI at diagnosis, and having a higher TNM-N score. A decreased OR was seen for OPC patients who had greater packyears and higher TNM-T and TNM-M stages.

TABLE 4.

Risk factors for HPV-driven OPC vs non-HPV-driven OPC according to HPV serology, results from univariable and multivariable analyses.

		Unadjusted OR		Adjusted OR
Variable	n/N	OR	(95% CI)	OR	(95% CI)
Study
ARCAGE	66/237	0.18	(0.13-0.24)	0.27	(0.17-0.41)
CHANCE	171/304	0.59	(0.46-0.76)	0.58	(0.39-0.86)
HN5000	1097/1599	1	1	1	1
Pittsburgh	238/320	1.33	(1.02-1.75)	1.00	(0.7-1.45)
Toronto	542/806	0.94	(0.78-1.13)	0.77	(0.6-0.98)
Year of diagnosisa
2002-2011	599/1098	1	1	1	1
2012-2013	813/1172	1.89	(1.59-2.24)	1.39	(1.05-1.85)
2014-2017	702/996	1.99	(1.66-2.38)	1.48	(1.11-1.96)
Age in years
<50	353/517	1.91	(1.46-2.50)	1.80	(1.32-2.45)
50-54	421/577	2.39	(1.84-3.13)	2.22	(1.64-3.02)
55-59	474/707	1.81	(1.41-2.31)	1.84	(1.38-2.44)
60-64	389/602	1.62	(1.26-2.09)	1.61	(1.21-2.15)
65-69	255/444	1.20	(0.92-1.57)	1.17	(0.86-1.59)
≥70	222/419	1	1	1	1
Sex
Male	1744/2610	1.56	(1.31-1.85)	1.52	(1.23-1.86)
Female	370/656	1	1	1.52	(1.23-1.86)
BMI at diagnosisb
Not overweight	522/1047	1	1	1	1
Overweight	733/1013	2.63	(2.19-3.17)	2.12	(1.72-2.61)
Obese	569/728	3.60	(2.91-4.47)	2.72	(2.14-3.47)
Unknown/NA	290/478	1.55	(1.25-1.93)	1.56	(0.99-2.46)
Current alcohol use
Yes	1199/1849	1	1	1	1
No	663/1003	1.06	(0.90-1.24)	0.98	(0.8-1.19)
Unknown/NA	252/414	0.84	(0.68-1.05)	0.53	(0.29-0.98)
Packyears
<10 pack-years	701/819	1	1	1	1
10-39 pack-years	642/1044	0.27	(0.21-0.34)	0.35	(0.27-0.45)
≥40 pack-years	503/978	0.18	(0.14-0.22)	0.21	(0.16-0.27)
Unknown/NA	268/425	0.29	(0.22-0.38)	0.47	(0.25-0.89)
TNM-T
Tx/Tisc	27/57	0.42	(0.24-0.73)	2.10	(0.56-9.99)
T0	10/15	0.94	(0.33-3.05)	0.60	(0.18-2.18)
T1	480/706	1	1	1	1
T2	894/1272	1.11	(0.91-1.36)	1.10	(0.87-1.39)
T3	378/621	0.73	(0.58-0.92)	0.80	(0.61-1.04)
T4	325/595	0.57	(0.45-0.71)	0.53	(0.4-0.69)
TNM-N
Nxc	19/53	0.94	(0.52-1.67)	1.51	(0.31-5.89)
N0	230/618	1	1	1	1
N1	272/428	2.94	(2.28-3.8)	3.18	(2.4-4.23)
N2	1511/2045	4.77	(3.95-5.78)	4.34	(3.5-5.38)
N3	82/122	3.46	(2.31-5.26)	2.99	(1.9-4.78)
TNM-M
Mxc	26/77	0.27	(0.16-0.42)
M0	2068/3147	1	1
M1	20/42	0.47	(0.26-0.87)

Note: Sero−: HPV serology negative; Sero+: HPV serology positive; BMI not overweight: <25; BMI overweight: ≥25 and <30; BMI obese: ≥30; TNM classification is a classification to describe tumour size, lymph node involvement and metastasis. Totals might vary due to missings.

Abbreviations: 95% CI, 95% confidence interval; BMI, body mass index; DNA, deoxyribonucleic acid; HNC, head and neck cancer; HPV, human papillomavirus; ISH, in situ hybridisation; NA, not available; OPC, oropharyngeal cancer; OR, odds ratio; PCR, polymerase chain reaction; RNA, ribonucleic acid.

^aCategories based on terciles.

^bBMI in kg/m².

^cTx/Nx/Mx: diagnosed with OPC, but T, N or M status not assessed.

The final multivariable model based on backward selection included the following factors associated with being HPV positive (compared to being HPV negative): study, year of diagnosis, age, sex, BMI at diagnosis, current alcohol use, packyears, TNM-T and TNM-N stage (Table 4). The HN5000 study had the highest odds of having an HPV-driven OPC, followed by Pittsburgh, Toronto, CHANCE and finally ARCAGE. Additionally, increased odds were found for male sex, more recent year of diagnosis, younger age, having a higher BMI and not being a current smoker. Having lymph node involvement (>TNM-N0) also increased the odds of having HPV-driven OPC. TNM-M stage was not included as this did not improve the performance of the multivariable model.

4 |. DISCUSSION

In our study, the diagnostic accuracy of HPV16 early antigen serology was determined as a marker for HPV-driven HNC. Additionally, the robustness of the diagnostic accuracy across strata of patient characteristics was assessed. HPV serology was found to have good diagnostic accuracy with an overall sensitivity of 86.2% and specificity of 91.2% against p16 as a reference. Although immune responses differ in ageing men and women,^19,20 diagnostic accuracy was robust across age and sex, but also across year of diagnosis, BMI at diagnosis, current alcohol use and primary tumour size. However, diagnostic accuracy was found to be potentially influenced by smoking status and lymph node involvement.

To the best of our knowledge, this is the first study assessing the diagnostic accuracy of HPV serology in a real-world setting in which variation across patient characteristics is assessed. Knowing the variability across these characteristics, such as age or sex, is of great importance for interpretation of these results in clinical or research settings. OPC itself is a relatively uncommon condition and therefore most studies on diagnostic accuracy of markers for molecular HPV tumour stage had rather small sample sizes.^18,32–35 By combining studies, from multiple centres and 10 countries, we were able to assess diagnostic accuracy of HPV serology for HPV-driven OPC in one of the largest study populations so far.

HPV serology showed to have a good diagnostic accuracy against p16, even within the sub-entities of the oropharynx. The overall sensitivity and specificity in our study were slightly lower than the first study defining the HPV16 early antigen serology algorithm as used in our study (sensitivity: 97%; specificity: 98%).¹⁸ Main differences include that their study used HPV RNA PCR as reference method, had a small study sample (HNC n = 214, OPC n = 120), and a larger proportion never smoked (35%). Another study using an HPV serological algorithm also found slightly higher diagnostic accuracy (sensitivity: 93.1%; specificity 96.0%).³² Our study also had a small sample size (n = 112), used as reference p16 combined with ISH, and excluded indeterminate cases (n = 49). Finally, a systematic review and meta-analysis evaluated only HPV16 E6 as a marker for HPV-driven OPC, and found an overall sensitivity of 83.1% and specificity of 94.6%, results which are similar to our results.¹¹

Through stratified analyses, our results suggest a robust diagnostic accuracy for HPV serology across the majority of patient subsets defined by clinicodemographic variables. Nevertheless, a variation in diagnostic accuracy was observed between smoking strata: the sensitivity was lower for higher packyears, and specificity decreased for lower packyears. How this discordance between p16 expression and HPV serology is related to smoking is yet unclear. A possible explanation could be that the antibody response could be influenced by smoking, leading to discordance between HPV serology status and p16 status. Laboratory research found that current smokers have statistically significantly higher HPV viral loads than nonsmokers.³⁶ Most likely this would have led to higher antibody responses. Using the same cut-offs for smokers and nonsmokers, this could lead to a classification bias for HPV serology status for borderline cases, as smokers are more likely to surpass the cut-off than nonsmokers. Nevertheless, other studies showed no strong and consistent associations between smoking behaviour and HPV antibody expression.^37,38 Additionally, p16 is suggested to be more strongly overexpressed for nonsmokers than among smokers.³⁹ This may result in a classification bias for nonsmokers, as they might be falsely classified as being HPV-positive, or for smokers if they are falsely classified as being HPV-negative based on their p16. Therefore, the interaction of smoking and the expression of p16 and/or HPV16 antibodies is recommended to be further studied. If differences are found, cut-offs might be set differently for smokers and nonsmokers.

The slightly lower sensitivity for participants without lymph node involvement, might be due to cancer cells which have metastasized to the lymph nodes are more likely to be exposed to the immune system, in addition to immune cell infiltrates.⁴⁰ Consequently, lack of lymph node involvement might be a surrogate for a less pronounced immune response, leading to a slightly lower sensitivity of HPV serology. The specificity, however, was not influenced by the lack of lymph node involvement.

HPV serology in the ARCAGE study demonstrated a lower sensitivity. ARCAGE included a small proportion of HPV seropositive patients, which explains the broad CIs. The proportion of smokers and participants without lymph node involvement among those with HPV-driven tumour is higher in our study compared to the other four studies, which might explain the lowered sensitivity. The association between HPV serology and smoking should be further explored.

We performed a multivariable logistic regression model regression analysis to identify factors within patients with OPC that were associated with being HPV-driven. As HPV-driven HNC prevalence is higher in North America and Northern Europe,^4,5 could explain our findings of the larger odds ratios for studies from the United Kingdom and North America. The proportion of HPV-driven OPC has been increasing over the years,² which explains why a more recent year of diagnosis is related to higher chance for OPC to be HPV-driven. This might also explain partially why younger patients might have a greater chance of being HPV-positive through a cohort effect.⁴ Another explanation is that HPV-driven OPC is often found in younger patients than non-HPV-diven OPC.^2,41 Additionally, male sex and being a nonsmoker are known risk factors for HPV-driven OPC compared to non-HPV-driven OPC.^2,4 Finally, lymph node involvement increased the odds, and a larger tumour size was associated with lower odds for HPV-driven OPC. The fact that patients with HPV-driven OPC often have smaller primary tumours, and patients with those tumours frequently present with lymph node metastasis,^9,42 is a possible explanation for this.

Some limitations should be mentioned. First, all participants in our study originated from Europe or North America. As HPV prevalence varies around the world, with the highest numbers in North America and Europe,⁵ the generalisability of our results outside these regions may be limited, especially since important risk factors such as packyears vary across the world.⁴³ Additionally, data on sexual behaviour were not available for all studies. In previous studies, earlier age of sexual debut, higher number of lifetime partners, higher number of oral sexual partners, and same-sex contact have been suggested to be risk factors for HPV-driven HNC.^4,44–46 Another important limitation is that p16 was used as a reference for the diagnostic accuracy in most analyses. Although this is the most common method for determining HPV status in OPC clinically and in research, and is widely accepted, it has a lower specificity than desired.^14,17 This might have resulted in an underestimation of the true diagnostic accuracy. In an atempt to adjust for this, a combined marker was used in one of the analyses. Nevertheless, this might have resulted in a biased subset, as only specific patients will have had more than one diagnosis method. In the sensitivity analyses, p16 was also used as a reference method for HNC outside of the oropharynx, as opposed to previous studies suggesting p16 should not be used outside the oropharynx as a prognostic marker for HPV-driven tumours.⁴⁷ PCR might have served as a better reference for all anatomical locations, but unfortunately, very few patients across all studies of the VOYAGER consortium had PCR-based results for HPV. Finally, not all participants in our dataset had p16 (or any other marker method) been performed. Therefore, our results might not be generalisable to all OPC or HNC patients. Nevertheless, HPV serology could serve as a feasible substitute for current HPV detection methods Additionally, the diagnostic accuracy was shown to be robust across most risk factors, which is in favour of the generalisability.

In conclusion, in this large multinational and multicentre study, we demonstrated that HPV serology can be used for assessing whether OPC is HPV-driven, due to the high sensitivity and specificity. This marker is suggested to be robust across age and sex, but also across year of diagnosis, BMI, current alcohol use and tumour size. Further research is recommended on the interaction of smoking and p16 expression, and on the interaction of smoking and HPV antibody expression, to explain the variation in sensitivity and specificity. In addition, definitions for seropositivity might be further optimised in further studies for different groups, to increase the diagnostic accuracy even further. Nevertheless, due to the robustness across all other variables, this marker is recommended to be used in clinical and epidemiological settings, especially when other methods are not available. Serology is minimally invasive, and can be used when tumour tissue is not available. Currently, different treatment options are being discussed for HPV-driven OPC patients, for example, by de-intensifying therapy, HPV serology could be a useful tool to distinguish HPV-driven OPC from non-HPV-driven OPC. Finally, more research on HPV serology for other HNC than OPC, should be explored, so that this tool may be used for those sites as well.

What’s new?

Human papillomavirus (HPV) is driving an increasing proportion of head and neck cancers (HNC), and patients with HPV-driven cancers may have different treatment needs. Serum antibodies against HPV16 early antigens appear to be a highly sensitive biomarker for HPV-driven HNC. Here, the authors investigate the accuracy of HPV early antigen serology across populations of varying characteristics, including age and sex. Overall, they found that HPV serology showed high sensitivity and specificity which was retained independent of age, sex, and other characteristics.

Abbreviations:

AF

attributable fraction

AIC

Akaike Information Criterion

AJCC

American Joint Committee on Cancer

BMI

body mass index

CI

confidence interval

DNA

deoxyribonucleic acid

HNC

head and neck cancer

HPV

human papillomavirus

ICD-10

International Classification of Diseases Volume

ISH

in situ hybridization

MFI

median fluorescence intensity

OPC

oropharyngeal cancer

OR

odds ratio

p16

p16^INK4a immunohistochemistry

PCR

polymerase chain reaction

RNA

ribonucleic acid

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.