Sex- and type-specific genital human papillomavirus transmission rates between heterosexual partners: a Bayesian reanalysis of the HITCH cohort

Abstract

Background:

It is unclear whether sexual transmission rates of human papillomaviruses (HPV) differ between sexes and HPV types. We estimate updated transmission rates from the final HITCH cohort study, and propose an estimation method that accounts for interval-censored data and infection clearance.

Methods:

We enrolled young women 18–24 years old and their male sex partners ≥18 years old in Montréal, Canada between 2005–2011. We followed women over 24 months and men over 4 months. We tested genital samples with Linear Array for HPV DNA detection and genotyping. We calculated infection transmission rates between partners using a multistate Markov model via a Bayesian approach. We report the posterior median and 2.5%−97.5% percentile intervals (95%PI).

Results:

We observed 166 type-specific incident HPV transmission events in 447 women and 402 men. The estimated median transmission rate from an HPV-positive to a negative partner was 4.2 (95%PI 3.1–5.3) per 100 person–months. The transmission rate from men to women was 3.5 (95%PI 2.5–4.7) and from women to men was 5.6 (95%PI 3.8–7.0) per 100 person–months, corresponding to a rate ratio of 1.6 (95%PI 1.0–2.5). Partners reporting always using condoms had a 0.22 (95%PI 0.05–0.61) times lower HPV transmission rate than those reporting never using condoms. HPV16/18 did not have particularly high transmission rates relative to other HPV types.

Conclusion:

Our updated analysis supports previous research suggesting higher women-to-men than men-to-women HPV transmission rates and a protective effect of condoms in heterosexual partnerships. Our results also suggest that crude incidence rates underestimate HPV transmission rates due to interval-censoring.

Introduction

Human papillomaviruses (HPV) of the Alphapapillomavirus genus are among the most prevalent sexually transmitted infections worldwide.¹ Several HPV types are oncogenic and cause cervical, anogenital, and oropharyngeal cancers in men and women.² Accurate estimates of sex- and type-specific HPV transmission rates between partners are needed to inform public health decisions and cost-effectiveness models. We have previously estimated HPV transmission rates between heterosexual partners in the HPV Infection and Transmission Among Couples through Heterosexual Activity (HITCH) study based on 179 couples up to 2010.³ The study initially found very little difference in the directionality of HPV transmission rates between men and women.³ Since then, follow-up has been completed in the full cohort of 548 couples. We recently performed a systematic review and meta-analysis of all couples’ studies of HPV transmission, and found some evidence that HPV transmission might be higher from women-to-men than from men-to-women across studies,⁴ though the confidence interval could not exclude higher men-to-women transmission rates.

Most HPV types cause commensal infections without symptoms,⁵ which complicates studies of HPV incidence and transmission. Observations of HPV infection are interval-censored, meaning that the timing of an incident infection or clearance is known only to lie within an interval between two tests. Unfortunately, most traditional epidemiologic methods to calculate incidence rates assume exact knowledge of the timing of an event through right-point or mid-point imputation.⁶ Right-point imputation makes the implicit assumption that individuals are at risk of infection during the entire interval between observations. However, this likely overestimates the time at risk of infection, because individuals who become infected during the interval are not at risk during the entire interval. Because HPV infections can be cleared fairly quickly, the number of incident infections may also be underestimated due to their clearance during the interval. Furthermore, it is likely that many incident HPV detections do not represent new transmission events but rather the reactivation of latent pre-existing infections.⁷ This makes it difficult to assess whether incident HPV detection can be attributed to transmission.

The objective of this analysis was to update HPV transmission rates estimates based on the final HITCH data, in order to evaluate potential sex differences and assess predictors of transmission such as HPV type and sexual behaviors. We also propose an alternative estimation method for infection transmission rates that accounts for issues with interval-censored data and latent infections, and compare this method to traditional methods used in previous studies.

Methods

Study design & setting

The HITCH cohort study enrolled young female university and college students aged 18–24 years old and their male sex partners ≥18 years old in Montréal, Canada. Participants were recruited between 2005–2011, and follow-up ended in September 2013. The full study protocol, procedures, and data collection instruments have been described previously.^3,8–10 Participants were recruited through promotional materials distributed on campus and student venues. Eligible couples needed to have initiated sexual activity with each other within the past 6 months. If a couple terminated their relationship during the study, the participants were encouraged to enrol any new eligible partner. The ethical review committees of McGill University, Concordia University, and the Centre Hospitalier de l’Université de Montreal approved the study. All participants provided written informed consent.

Women were followed-up over 24 months for 6 clinic visits scheduled 4–6 months apart. Men were followed up over 2 clinic visits scheduled 4 months apart. At each clinic visit, participants filled out a self-administered questionnaire and provided genital samples for HPV testing. Participants were instructed to abstain from oral, vaginal, or anal sex for 24 hours prior to the clinic visit in order to reduce the probability of false positive HPV test results due to deposition from recent sex. However, many participants deviated from protocol and reported having sex within the last 24h.

HPV DNA and genital samples

Women self-collected vaginal samples using a Dacron swab. The research nurse provided women written instructions for the collection of self-samples, reviewed the instructions with them and addressed any questions or concerns. Self-collected HPV vaginal samples have been shown to have sensitivity and specificity equivalent to clinician-collected samples.¹¹ For men, the nurse collected epithelial cells from the penis and scrotum in separate sample containers using gentle exfoliation with ultra-fine emery paper followed by swabbing with a Dacron swab. The Dacron swabs were placed into vials with Preservcyt™ (Hologic, Marlborough, MA, USA), agitated to release cells, and then discarded; the emery papers were also placed in the respective vials.

Samples were processed and DNA extracted as previously described.¹⁰ Genital samples were tested by a real-time polymerase chain reaction using the Linear Array HPV genotyping assay (LA-HPV) (Roche Molecular Systems, Alameda, CA, USA),¹² which detects DNA from 36 HPV genotypes of the Alphapapillomavirus genus. β-globin DNA was co-amplified to assess DNA integrity of samples and to control for the presence of cells. A sample was considered valid if β-globin DNA was detected.

Statistical analysis

Participant restriction

We restricted our analysis to the participants with at least two visits with valid genital HPV samples, and whose HITCH partner also had a valid concurrent genital HPV sample at the first of the two visits. We refer to the time between these two visits as an eligible interval. Due to their shorter follow-up, most men only had a single eligible interval for the analysis, corresponding to their first and second visit. One man deviated from protocol by coming for a third visit, and 13 men recruited new female partners into the study: these men had two intervals potentially eligible for the analysis. Women could contribute multiple intervals to the analysis due to their longer follow-up and recruitment of new male partners, provided that their partner’s HPV status was known at the start of each interval. For ascertainment of an individual’s partner HPV status at a given visit date, we only included their partner’s visits that could be matched within ±1 day of the individual’s visit.

Crude analyses

The unit of analysis was the HPV type-specific infection. We defined transmission as occurring when a participant who was negative and whose partner was positive at the start of an interval became positive by the end of the interval for a given HPV type. We defined background incidence as occurring when a participant who was negative and whose partner was negative at the start of an interval became positive by the end of the interval for a given HPV type; background incidence is expected to occur due to sex with non-study partners and reactivation of latent infections. We defined clearance as occurring when a participant who was positive at the start of an interval became negative at the end of the interval for a given HPV type. In crude analyses, incidence rates were calculated by dividing the number of events by the person–time accrued in the interval between visits, with exact confidence intervals (CI) around a Poisson mean.¹³

Multistate Markov model analyses

We used a multistate Markov model to estimate HPV transmission and background incidence rates via a Bayesian approach. A multistate Markov model defines a system in terms of the transition rates between a set of defined health states. We modeled two type-specific health states: HPV-positive and HPV-negative (Figure 1). The model allows for the possibility that unobserved incidence and clearance events might have occurred in the interval between two visits by allowing a backwards transition rate from HPV-positive to HPV-negative states during each interval.

Figure 1.

Multistate Markov Model structure. Boxes represent HPV type-specific health states; λ_0i, λ_1i, and λ_2i represent transition rates between health states; β₀, β₁, and β₂ represent multiplicative effects of predictive variable X on transition rates. HPV=human papillomavirus.

We defined transitions between type-specific HPV-negative and -positive states in terms of individual-specific and type-specific transition rates λ_0i, λ_1i, and λ_2i. The background incidence rate of infection (λ_0i) represents the background expected rate of incident type-specific genital positivity due to sex with non-study partners and reactivation of latent infections. The transmission rate of infection (λ_1i) represents the increase in incident positivity rate in individuals exposed to an HPV-positive HITCH partner which is additional to the background rate; this rate was only applied to individuals who had an HPV-positive HITCH partner at the start of an interval. The clearance rate (λ_2i) applied to all HPV-positive individuals.

Each individual could contribute multiple observations to the analysis due to infections with multiple HPV types across multiple time intervals. We accounted for these correlated observations by allowing each individual to have their own specific rates i. This allowed an individual propensity for infection and clearance across all HPV types and intervals. These individual-specific rates were modeled as random effects, log-normally distributed around the population geometric means $\overline{{\lambda }_0},\overline{{\lambda }_1}$, and $\overline{{\lambda }_2}$.

The model uses the observed HPV positivity data of HITCH participants at each visit and the time elapsed between visits to estimate incidence and clearance rates. The probability of an individual being HPV-positive at the end of an interval therefore depends on the relative strength of incidence and clearance rates and on their state at the start of the interval, and was specified in terms of Kolmogorov’s forward equations:¹⁴

$$P_{neg\to pos,i}(t)=1-\frac{({\lambda }_{0i}+{\lambda }_{1i})e^{-({\lambda }_{0i}+{\lambda }_{1i}+{\lambda }_{2i})t}+{\lambda }_{2i}}{{\lambda }_{0i}+{\lambda }_{1i}+{\lambda }_{2i}}$$

$$P_{pos\to neg,i}(t)=1-\frac{({\lambda }_{0i}+{\lambda }_{1i})+{\lambda }_{2i}{e}^{-({\lambda }_{0i}+{\lambda }_{1i}+{\lambda }_{2i})t}}{{\lambda }_{0i}+{\lambda }_{1i}+{\lambda }_{2i}}$$

Where $P_{neg\to pos,i}(t)$ is the probability an HPV-negative individual will be positive after an interval of time t, and $P_{pos\to neg,i}(t)$ is the probability an HPV-positive individual will be negative after an interval of time t.

The association between rates and predictive variables were modeled through coefficients β₀, β₁, and β₂, which we exponentiated to calculate incidence rate ratios (IRR). We assessed as univariate predictors sex, HPV type, positivity for other HPV types, and self-reported sexual behaviors. We also assessed whether the persistence of a participant’s HITCH partner’s HPV infection influenced transition rates. This latter analysis was restricted to individuals whose partner’s HPV status is known both at the start and end of each interval; this was only a subset of participants because many partners were lost to follow-up and did not attend the second visit. We also allowed clearance rates to vary depending on whether an individual’s HITCH partner was positive or not at the start of the interval for a given HPV type.

The models were implemented using WinBUGS version 1.4.3 and the R2WinBUGS package in R version 3.6.3.^15,16 WinBUGS uses a Bayesian Markov Chain Monte Carlo (MCMC) simulation approach with a Gibbs sampling algorithm to estimate parameters. Prior IRR were centered on 1 with very large variances, in order to integrate a weak conservative prior assumption of absence of effect of sex and other predictors on incidence rates. For each model we ran three chains over 50,000 iterations. We discarded the first 20,000 iterations, and kept only every fifth iteration for analysis to decrease autocorrelation between iterations and computational costs. This led to a sample of 18,000 iterations for the posterior. We visually assessed chain convergence using iteration histories (eFigure 1 of Supplemental Digital Content). We present as results the median of the posterior distribution, and its 2.5^th and 97.5^th percentiles as the 95% posterior probability interval (95%PI); we use the term posterior probability interval instead of credibility interval in order to avoid confusion with confidence intervals when using abbreviations. We provide sample R code in the Supplemental Digital Content.

Sensitivity analyses

For comparability with previous HITCH results and other couples’ studies, we repeated the analyses using a frequentist random effects Poisson regression model, which does not account for interval-censoring or the possibility of clearance during the interval. The outcome definitions were the same as those used for the crude analyses. We used SAS 9.4 for this analysis.

We also performed an analysis with an informative prior for the IRR of transmission rates by sex. We reanalyzed the data from our previous meta-analysis of HPV transmission rates excluding the HITCH study,⁴ which yielded a pooled random effects IRR for transmission from women to men compared with men to women of 1.7 (95%CI 1.1–2.8). Based on this, we assumed a prior normal distribution with a mean of 0.55 and a variance of 0.055 for β₁.

Results

Characteristics of included participants

There were 502 women and their 548 male partners recruited into HITCH. Of these, we retained 447 women and 402 men for our analysis. We excluded more men than women for having only a single clinic visit (22% vs. 10%) or an invalid genital sample (4.0% vs. 0.4%) (eFigure 2 of the Supplementary Digital Content). The average age at baseline was 21.1 (standard deviation 2.1) for included women and 23.0 (standard deviation 3.6) for included men. Other baseline characteristics of included participants are further described in eTable 1 of the Supplementary Digital Content. Women contributed a median of two eligible intervals (range one to four) to the analysis, while men contributed a median of one eligible interval (range one to two) to the analysis. The median follow-up time was 8.6 months (interquartile range 6.2–11.6) for women and 5.2 months (interquartile range 4.1–6.5) for men.

Transmission rates by sex

There were 166 type-specific incident HPV transmission events (Table 1); 106 of these were in women and 60 were in men. After accounting for background incidence and clearance, the model-estimated median rate of genital transmission of a given HPV type from a positive to a negative sex partner was 4.2 (95%PI 3.1–5.3) per 100 person–months. This median rate was 3.5 (95%PI 2.5–4.7) per 100 person–months for men-to-women transmission, and 5.6 (95%PI 3.8–7.9) for women-to-men transmission, a difference of 2.0 (95%PI 0.1–4.4) per 100 person–months. The median IRR of women-to-men compared with men-to-women transmission rates was 1.6 (95%PI 1.0–2.5). Based on the posterior distribution, there was a 98% probability this IRR>1, a 60% probability this IRR>1.5, and a 15% probability this IRR>2. The median background incidence rate of HPV detection was 0.12 (95%PI 0.09–0.14) per 100 person–months.

Table 1.

HPV transmission rates, background infection rates, and clearance rates, for all HPV types combined.

	Crude estimates				Multistate model
Event	Na	Person–months	Rate (/100 person–months)	95% CI	Rate (/100 person– months)	95% PI	IRR	95% PI
Transmission incidenceb
Overall	166	4736	3.5	(3.0–4.1)	4.2	(3.1–5.3)	–	–
Men-to-women	106	3282	3.2	(2.6–3.9)	3.5	(2.5–4.7)	1.0	(ref)
Women-to-men	60	1454	4.1	(3.1–5.3)	5.6	(3.8–7.9)	1.6	(1.0–2.5)
Background incidencec
Overall	378	218256	0.17	(0.16–0.19)	0.12	(0.09–0.14)	–	–
Women	208	139644	0.15	(0.13–0.17)	0.10	(0.08–0.13)	1.0	(ref)
Men	170	78612	0.22	(0.18–0.25)	0.14	(0.10–0.18)	1.3	(1.0–1.8)
Clearanced
Overall	589	10301	5.7	(5.3–6.2)	8.7	(7.5–10)	–	–
Women	401	6403	6.3	(5.7–6.9)	9.8	(8.3–12)	1.0	(ref)
Men	188	3898	4.8	(4.2–5.6)	7.3	(6.0–8.8)	0.74	(0.60–0.92)

CI=confidence interval; HPV=human papillomavirus; IRR=incidence rate ratio; PI=posterior interval

^aObservation level is the HPV type.

^bMultistate model transmission rates correspond to the model parameter $\overline{{\lambda }_1}$, the geometric mean of individual-level transmission rates.

^cMultistate model transmission rates correspond to the model parameter $\overline{{\lambda }_0}$, the geometric mean of individual-level background incidence rates.

^dMultistate model transmission rates correspond to the model parameter $\overline{{\lambda }_2}$, the geometric mean of individual-level clearance rates.

Transmission rates by HPV type

There was moderate evidence that transmission rates varied by HPV type (Table 2). HPV types of the α3 and α14 species were estimated to have the highest transmission rates while HPV types of the α10 species had the lowest transmission rates. The estimated median transmission rates of HPV16 (3.6 per 100 person–months) and HPV18 (2.2 per 100 person–months) were not particularly high relative to other HPV types, and their 95%PI included the overall median of 4.2 per 100 person–months.

Table 2.

HPV species- and type-specific transmission rates.

	Crude estimates			Multistate model
HPV type	N	Person–months	Rate (/100 person–months)	Rate (/100 person– months)	95% PI	IRR	95% PI
Species groupb
α7	25	686	3.6	5.1	(3.0–8.2)	1.6	(0.86–2.8)
α9	39	1037	3.8	3.8	(2.4–5.8)	1.2	(0.69–1.9)
α10	6	290	2.1	1.9	(0.4–5.0)	0.59	(0.14–1.6)
α3/14	42	995	4.2	5.4	(3.5–7.9)	1.6	(0.99–2.7)
Other	54	1728	3.1	3.3	(2.1–4.8)	1.0	(ref)
Type-specific
HPV6	5	192	2.6	2.5	(0.3–8.0)	0.40	(0.05–1.5)
HPV11	0	22	0.0	2.9	(0.0–539)	0.47	(0.0–90)
HPV16	14	358	3.9	3.6	(1.6–7.0)	0.57	(0.23–1.4)
HPV18	2	110	1.8	2.2	(0.2–9.6)	0.35	(0.04–1.8)
HPV26c	0	0	–	6.3	(0.0–2875)	1.0	(0.0–459)
HPV31	5	104	4.8	4.4	(1.1–13)	0.69	(0.17–2.4)
HPV33	2	67	3.0	3.1	(0.2–25)	0.50	(0.03–4.3)
HPV34	0	14	0.0	0.8	(0.0–30)	0.12	(0.0–5.0)
HPV35	0	17	0.0	0.6	(0.0–14)	0.10	(0.00–2.5)
HPV39	10	220	4.5	5.6	(2.2–12)	0.88	(0.31–2.5)
HPV40	2	136	1.5	1.2	(0.1–5.2)	0.19	(0.02–0.88)
HPV42	8	256	3.1	1.9	(0.4–5.3)	0.30	(0.06–0.97)
HPV44	1	77	1.3	0.4	(0.0–3.8)	0.06	(0.00–0.62)
HPV45	3	72	4.2	5.3	(0.8–21)	0.85	(0.12–3.8)
HPV51	10	329	3.0	2.3	(0.8–5.3)	0.37	(0.12–0.98)
HPV52	9	162	5.6	5.1	(1.7–13)	0.81	(0.25–2.4)
HPV53	7	222	3.2	2.5	(0.5–6.8)	0.39	(0.08–1.3)
HPV54	6	155	3.9	3.7	(1.0–11)	0.59	(0.15–2.0)
HPV56	4	233	1.7	0.9	(0.0–3.7)	0.15	(0.01–0.66)
HPV58	2	116	1.7	0.7	(0.0–3.9)	0.11	(0.00–0.66)
HPV59	6	158	3.8	5.6	(1.7–16)	0.89	(0.25–2.9)
HPV61	3	130	2.3	2.0	(0.3–6.8)	0.32	(0.05–1.3)
HPV62	7	194	3.6	3.0	(0.9–7.7)	0.48	(0.13–1.5)
HPV66	9	167	5.4	6.8	(2.3–17)	1.1	(0.34–3.3)
HPV67	7	212	3.3	2.9	(0.9–7.6)	0.46	(0.13–1.4)
HPV68	1	83	1.2	1.1	(0.1–6.4)	0.18	(0.01–1.1)
HPV69c	0	0	–	6.2	(0.0–2791)	1.0	(0.00–442)
HPV70	3	44	6.8	13	(2.2–78)	2.0	(0.33–13)
HPV71	0	13	0.0	1.5	(0.0–110)	0.24	(0.00–17)
HPV72	1	20	5.0	5.0	(0.2–63)	0.78	(0.03–11)
HPV73	5	114	4.4	3.5	(0.7–10)	0.55	(0.11–1.8)
HPV81	2	22	9.1	26	(2.3–276)	4.2	(0.35–47)
HPV82	3	106	2.8	4.7	(0.8–19)	0.76	(0.13–3.3)
HPV83	0	16	0.0	0.8	(0.0–21)	0.13	(0.00–3.6)
HPV84	14	374	3.7	3.8	(1.6–7.8)	0.60	(0.23–1.6)
HPV89	15	226	6.6	6.3	(3.0–12)	1.0	(ref)

HPV=human papillomavirus; IRR=incidence rate ratio; PI=posterior interval

^aMultistate model transmission rates correspond to the model parameter $\overline{{\lambda }_1}$, the geometric mean of individual-level transmission rates.

^bHPV types included in each species. α7: HPV18, 39, 45, 59, 68, 85. α9: HPV16, 31, 33, 35, 52, 58, 67. α10: HPV6, 11, 13, 44, 74. α3/14: HPV61, 62, 71,72, 81, 83, 84, 86, 87, 89.

^cNo observed person-time data for these HPV types. Model estimates reflect assumed prior distributions.

Predictors of transmission and background incidence

Univariate predictors of genital transmission rates between partners are presented in Table 3. The most important predictor of transmission was the persistence of HPV infection in the transmitting partner; the transmission rate of infection was 6.1 times higher (95%PI 2.9–13) if the initially positive partner was persistently positive than if he or she became negative by the end of the interval. Partners who reported always using condoms had 0.22 (95%PI 0.05–0.61) times lower transmission rates than those who reported never using condoms; this association was stronger for men than for women. The frequency of vaginal sex per week was positively associated with transmission. Time since last vaginal sex was also associated with transmission, with those reporting vaginal sex in the last day having the highest transmission rates. Transmission rates were generally higher in men than in women across most strata of predictors.

Table 3.

Univariate predictors of incident genital HPV transmission rates, stratified by sex acquiring the infection.

	Men					Women					All
Characteristics	N	Person–months	Rate (/100 person–months)	IRR	95% PI	N	Person–months	Rate (/100 person–months)	IRR	95%PI	Rate (/100 person–months)	IRR	95% PI
Lifetime # sex partnersa
1	2	50	53	10	(0.79–186)	3	78	3.9	1.1	(0.25–4.6)	7.8	2.0	(0.49–9.1)
2–4	13	275	6.4	1.2	(0.49–2.9)	14	502	3.2	1.0	(0.40–2.1)	4.6	1.2	(0.61–2.1)
5–9	15	476	4.0	0.74	(0.30–1.7)	34	1008	3.4	1.0	(0.55–1.9)	3.7	0.91	(0.55–1.5)
≥10	29	628	5.4	1.0	(ref)	55	1694	3.4	1.0	(ref)	4.0	1.0	(ref)
HITCH male circumciseda
Yes	28	758	4.8	1.0	(ref)	46	1630	3.0	1.0	(ref)	3.6	1.0	(ref)
No	32	680	6.5	1.3	(0.69–2.7)	60	1628	4.3	1.4	(0.85–2.4)	4.9	1.4	(0.92–2.1)
Positive for other HPV typesa
No	12	412	3.1	1.0	(ref)	20	806	3.3	1.0	(ref)	3.2	1.0	(ref)
Yes (≥1)	48	1043	6.5	2.1	(0.94–5.0)	86	2477	3.7	1.1	(0.60–2.0)	4.4	1.4	(0.89–2.1)
Reported other sex partners during intervalb
No	48	1037	6.4	1.0	(ref)	69	2202	3.3	1.0	(ref)	4.4	1.0	(ref)
Yes	12	418	3.1	0.48	(0.18–1.1)	37	1081	3.8	1.2	(0.64–2.0)	3.9	0.88	(0.55–1.4)
HITCH partner persistently positive at second visitb
No	7	371	2.2	1.0	(ref)	4	485	1.0	1.0	(ref)	1.4	1.0	(ref)
Yes	40	524	11	5.1	(2.1–14)	41	748	7.1	7.4	(2.5–22)	8.7	6.1	(2.9–13)
Condom use with HITCH partnerb,c
Never (0%)	21	533	5.7	1.0	(ref)	62	1552	4.8	1.0	(ref)	5.0	1.0	(ref)
Rarely (1–25%)	11	288	4.7	0.82	(0.32–2.0)	12	610	2.4	0.50	(0.23–0.99)	3.1	0.62	(0.35–1.1)
Sometimes (26–75%)	17	190	14	2.4	(1.0–5.8)	11	358	3.9	0.81	(0.34–1.8)	7.2	1.4	(0.80–2.6)
Most times (76%-99%)	8	190	5.5	0.96	(0.31–2.7)	9	282	3.6	0.76	(0.31–1.6)	4.4	0.88	(0.44–1.7)
Always (100%)	0	174	0.1	0.02	(0.00–0.24)	5	233	1.9	0.41	(0.11–1.1)	1.1	0.22	(0.05–0.61)
Frequency of vaginal sex with HITCH partnerb,c
≤2/week	13	516	2.8	1.0	(ref)	25	882	3.0	1.0	(ref)	2.9	1.0	(ref)
2–4/week	26	614	5.2	1.9	(0.82–4.6)	44	1288	3.7	1.2	(0.68–2.3)	4.1	1.4	(0.86–2.4)
>4/week	19	250	19	6.7	(2.6–18)	30	863	4.3	1.4	(0.73–2.9)	6.4	2.2	(1.3–4.2)
Cessation of relationshipb,d
No	49	958	7.0	1.0	(ref)	85	2605	3.5	1.0	(ref)	4.3	1.0	(ref)
Yes	11	457	2.9	0.41	(0.16–0.93)	21	665	3.8	1.1	(0.59–2.0)	3.5	0.82	(0.49–1.3)
Days since last vaginal sexb,e
0–1	16	211	14	1.0	(ref)	21	490	5.1	1.0	(ref)	7.4	1.0	(ref)
2	16	293	8.7	0.63	(0.23–1.7)	31	865	4.0	0.77	(0.38–1.6)	5.0	0.68	(0.38–1.2)
3	4	66	6.7	0.49	(0.10–1.9)	11	288	4.1	0.79	(0.30–2.0)	4.7	0.65	(0.28–1.4)
4	2	73	2.2	0.16	(0.01–0.87)	2	138	1.1	0.21	(0.03–0.88)	1.5	0.20	(0.05–0.62)
5–7	4	136	3.7	0.27	(0.05–1.1)	4	259	1.0	0.19	(0.03–0.67)	1.7	0.23	(0.08–0.60)
8–14	2	79	2.3	0.16	(0.01–0.83)	4	166	2.5	0.48	(0.07–1.8)	2.8	0.38	(0.10–1.1)
≥15	14	569	2.4	0.18	(0.06–0.49)	29	1008	2.9	0.55	(0.26–1.2)	2.8	0.37	(0.20–0.70)

HPV=human papillomavirus; IRR=incidence rate ratio; PI=posterior interval

^aReported at the start of the interval.

^bReported at the end of the interval.

^cAverage of both partners’ responses if both attended the visit, or of only one individual if they attended the visit alone.

^dWe considered the relationship ended if both partners reported having ended the relationship, or if only one individual reported having ended the relationship if they attended the visit alone.

^eLatest date reported by either partner.

Predictors of background incidence rates (Table 4) were different than predictors of transmission rates. The lifetime number of sexual partners was strongly correlated with background incidence rate. The incidence rate of new background infections was 2.9 (95%PI 2.1–4.0) times higher in those who were already positive for other HPV types, and was 2.4 (95%PI 1.8–3.2) times higher in those who reported having sex with partners other than their HITCH partner in the interval. Frequency of vaginal sex and time since last vaginal sex with their HITCH partner were not associated with background incidence. However, individuals whose HITCH partner became newly positive for an HPV type during an interval had a median background incidence rate of 7.0 (95%PI 4.8–10) per 100 person–months, which was comparable to estimated transmission rates. Background incidence rates were generally higher in men than in women across most strata of predictors.

Table 4.

Univariate predictors of background HPV incidence rates from a multistate Markov model, stratified by sex acquiring the infection.

	Men					Women					All
Characteristics	N	Person–months	Rate (/100 person–months)	IRR	95% PI	N	Person–months	Rate (/100 person–months)	IRR	95%PI	Rate (/100 person–months)	IRR	95% PI
Lifetime # sex partnersa
1	6	6828	0.13	0.48	(0.10–5.5)	4	14886	0.02	0.10	(0.03–0.28)	0.03	0.16	(0.07–0.40)
2–4	18	20611	0.05	0.18	(0.09–0.34)	27	42259	0.05	0.27	(0.15–0.46)	0.05	0.23	(0.15–0.35)
5–9	50	24661	0.14	0.51	(0.32–0.83)	76	38590	0.15	0.81	(0.52–1.3)	0.15	0.67	(0.47–0.94)
≥10	95	26166	0.28	1.0	(ref)	101	43913	0.19	1.0	(ref)	0.22	1.0	(ref)
HITCH male circumciseda
Yes	77	34762	0.14	1.0	(ref)	89	62728	0.10	1.0	(ref)	0.12	1.0	(ref)
No	90	43580	0.13	0.90	(0.57–1.4)	117	76224	0.10	1.0	(0.70–1.5)	0.11	0.96	(0.71–1.3)
Positive for other HPV typesa
No	40	35080	0.07	1.0	(ref)	51	64811	0.06	1.0	(ref)	0.06	1.0	(ref)
Yes (≥1)	130	43530	0.22	3.4	(2.1–5.5)	157	74838	0.16	2.6	(1.7–3.9)	0.18	2.9	(2.1–4.0)
Reported other sex partners during intervalb
No	109	62740	0.12	1.0	(ref)	118	107682	0.09	1.0	(ref)	0.10	1.0	(ref)
Yes	59	15714	0.28	2.3	(1.4–3.7)	90	31966	0.21	2.5	(1.7–3.6)	0.23	2.4	(1.8–3.2)
HITCH partner became positive at second visitb
No	81	52316	0.13	1.0	(ref)	49	51899	0.09	1.0	(ref)	0.11	1.0	(ref)
Yes	36	517	8.2	64	(38–107)	36	715	5.9	68	(40–116)	7.0	64	(44–94)
Condom use with HITCH partnerb,c
Never (0%)	79	25351	0.23	1.0	(ref)	83	49518	0.13	1.0	(ref)	0.16	1.0	(ref)
Rarely (1–25%)	19	18876	0.07	0.28	(0.14–0.52)	40	29796	0.10	0.76	(0.47–1.2)	0.09	0.52	(0.35–0.76)
Sometimes (26–75%)	23	11729	0.12	0.51	(0.26–0.96)	29	19574	0.11	0.83	(0.49–1.4)	0.11	0.70	(0.46–1.1)
Most times (76%-99%)	12	10295	0.08	0.33	(0.14–0.70)	20	17537	0.08	0.60	(0.32–1.1)	0.08	0.48	(0.29–0.77)
Always (100%)	17	9268	0.12	0.53	(0.24–1.1)	9	14167	0.05	0.35	(0.15–0.74)	0.08	0.46	(0.26–0.78)
Frequency vaginal sex with HITCH partnerb,c
≤2/week	49	21852	0.14	1.0	(ref)	43	38410	0.08	1.0	(ref)	0.10	1.0	(ref)
2–4/week	59	32131	0.12	0.82	(0.47–1.4)	71	55362	0.09	1.1	(0.70–1.8)	0.10	0.97	(0.68–1.4)
>4/week	41	21424	0.15	1.0	(0.56–1.9)	67	36822	0.13	1.7	(1.0–2.8)	0.14	1.3	(0.91–1.9)
Cessation of relationshipb,d
No	126	64086	0.13	1.0	(ref)	159	113719	0.10	1.0	(ref)	0.11	1.0	(ref)
Yes	36	14022	0.17	1.3	(0.73–2.4)	47	24595	0.13	1.3	(0.81–2.0)	0.14	1.3	(0.89–1.8)
Days since last vaginal sexb,e
0–1	21	10242	0.14	1.0	(ref)	26	14599	0.12	1.0	(ref)	0.13	1.0	(ref)
2	49	20548	0.15	1.1	(0.52–2.2)	60	37810	0.10	0.86	(0.50–1.5)	0.12	0.90	(0.59–1.4)
3	16	7999	0.12	0.82	(0.32–2.0)	21	13080	0.12	1.0	(0.52–2.0)	0.12	0.93	(0.53–1.7)
4	5	4484	0.07	0.46	(0.12–1.5)	8	8740	0.07	0.62	(0.24–1.5)	0.07	0.56	(0.26–1.2)
5–7	12	7946	0.10	0.72	(0.26–1.9)	17	12007	0.09	0.74	(0.35–1.5)	0.10	0.72	(0.39–1.3)
8–14	12	5305	0.12	0.83	(0.29–2.2)	14	9935	0.11	0.95	(0.42–2.0)	0.12	0.88	(0.46–1.7)
≥15	48	20258	0.14	0.96	(0.46–2.0)	60	40796	0.10	0.84	(0.47–1.5)	0.11	0.85	(0.53–1.4)

HPV=human papillomavirus; IRR=incidence rate ratio; PI=posterior interval

^aReported at the start of the interval.

^bReported at the end of the interval.

^cAverage of both partners’ responses if both attended the visit, or of only one individual if they attended the visit alone.

^dWe considered the relationship ended if both partners reported having ended the relationship, or if only one individual reported having ended the relationship if they attended the visit alone.

^eLatest date reported by either partner.

Sensitivity analyses

The Poisson model HPV transmission, background infection, clearance rates, and univariate regression model estimates by sex are presented in eTables 2–4 of the Supplementary Digital Content. Results were similar to those from the multistate Markov model, but the Poisson model incidence rate estimates were lower.

When we used an informative prior based on previous studies in our model, the median IRR of women-to-men compared with men-to-women transmission rates was 1.6 (95%PI 1.2–2.3). The estimated IRR was slightly higher with the informative prior due to the higher pooled average IRR from previous studies.

Discussion

In this study we calculated genital HPV transmission rates between new young heterosexual partners using a method accounting for interval-censoring, background incidence, and clearance. We observed higher HPV transmission rates from women to men than from men to women. Men also had higher background incidence rates and lower clearance rates of infection than women. We observed some variation in transmission rates across HPV types, but the most oncogenic HPV types 16 and 18 did not have particularly high transmission rates relative to the median. Expectedly, the variables associated with increased transmission rates were markers of recent and ongoing sexual activity, and lack of condom use between HITCH partners. The variables associated with background incidence rates were markers of sex with other partners and previous exposure to HPV (lifetime number of partners, positivity to other HPV types). The strong association with partner incident infections suggests some detections we considered to be background infections may also be due to partner transmission.

We had previously estimated in HITCH genital HPV transmission rates from men-to-women of 3.5 per 100 person–months and from women-to-men of 4.0 per 100 person–months.³ Our current estimates differ from our previous estimates because of the inclusion of newly accrued observations and more follow-up time, but also because we implemented a different method to calculate transmission rates. The traditional method for calculating HPV transmission rates in previous couples’ studies was to divide the number of incident events by person–time of follow-up between visits.^3,17–22 For comparability with previous studies on this topic we provided incidence rates using crude data and Poisson regressions in Table 1 and the appendix, respectively. The traditional method imputes the infection event as occurring at the end of the interval, and does not account for uncertainty in the timing of the event during the interval. This method likely underestimates HPV incidence rates, because it overestimates the person–time at risk for infection, and it underestimates the number of events that may have occurred within an interval because of HPV clearance. This method also does not account for the background incidence that would have occurred even without a partner. An important strength of our study is that the analysis method we used accounts for these biases which are inherent to interval-censored data. However, a limitation is that models which allow backwards transitions between health states, such as ours, suffer from problems of parameter identifiability when there are few observations. While the convergence of MCMC chains suggests the model was identifiable, we cannot exclude the possibility that there may be other combinations of transmission, background, and clearance rates that are compatible with the data.

Whether men or women are at higher risk of HPV infection has been a matter of debate. Most previous couples’ studies have observed higher HPV transmission rates from women to men than from men to women,^17–19,22 though two studies from China reported a higher transmission rate from men to women.^20,21 Most previous studies have had low statistical power to test these differences due to their small sample size. Our recent meta-analysis found a higher pooled average woman-to-man transmission rate, but the confidence interval could not exclude a higher man-to-woman transmission rate.⁴ Our analysis suggests that in our population there was a high probability that the average woman-to-man transmission rate was higher than the average man-to-woman transmission rate.

There are several hypotheses for why HPV transmission rates might differ between sexes. Firstly, the difference may be due to a sampling artefact, where HPV is more likely to be detected in penile than in cervico-vaginal samples. We do not believe this accounts for the difference we observed, because we had more invalid penile than invalid vaginal samples. If our ability to detect HPV had been compromised in men, it would have biased results towards a higher male-to-female transmission. Secondly, the difference may be caused by confounding due to differential exposure to HPV in men and women. Men were slightly older than women in our study and reported slightly more lifetime sexual partners, and may therefore have had more exposure to HPV prior to the study. However, we found that sex differences persisted across stratifications of multiple sexual behaviors associated with HPV transmission, which does not suggest confounding. Finally, there may be a true biologic transmission rate difference between men and women. While this difference could be due to anatomy, we hypothesize that it is more likely to be due to immunity. There is substantial evidence that sex hormones and genetics influence the innate, humoral, and cell-mediated immune responses to viral vaccines.²³ Lower HPV geometric mean antibody titers have been reported in vaccinated adult men than women,²⁴ though this was not observed in children.²⁵ Men are less likely to seroconvert following a genital HPV infection than women.^26–28 Women display a marked decline in HPV prevalence with age that is likely partly attributable to the development of immunity against reinfection;²⁹ there is no similar marked decline of HPV prevalence with age in men.^30,31 We therefore hypothesize the lower transmission rates of infection from men to women may be due to women developing more effective natural immunity against HPV reinfection.

The difference in HPV incidence rates between sexes was surprising, because we previously observed nearly identical cross-sectional genital HPV prevalence and type-specific concordance between male and female partners in HITCH^8,9,32 and in other couples’ studies.^19,33 It is possible that sex differences in HPV transmission rates and clearance rates may not translate into sizeable prevalence differences within partnerships due to the homogenizing effects of transmission within partnerships, which leads to a dependence in prevalences between men and women. In other words, the transmission rates in both sexes may be sufficiently high to maintain comparably high infection prevalences in both partners, and a loss of HPV positivity in one partner may be offset in many cases by reinfection from continued exposure to their HPV-positive partner.

HPV16 is the most prevalent oncogenic HPV type worldwide in women with both normal cytology and cervical cancer.^1,34 However, HPV16 did not have particularly high transmission rates relative to other Alphapapillomavirus types in our analysis. This suggests that its high population prevalence and force of infection may be instead driven by its longer infection duration and persistence.³⁵ There were approximately 16% of women in HITCH who reported being vaccinated against HPV. However, because HPV-16, 18, 6, and 11 represent only a small fraction of all HPV types, this is unlikely to have a strong impact in our study.

Even when studies use the same method to calculate transmission rates, there is substantial heterogeneity across studies,⁴ suggesting that HPV sexual transmission rates vary between populations. Transmission rates from HITCH are on the higher end of estimates. This is likely because we recruited a population of young adults who were near the start of their sexual relationship and who reported frequent sexual activity. Lower transmission rates have been reported in heterosexual couples with older age distributions and in more established partnerships,^20–22 possibly due to a lower frequency of sexual activity and higher levels of acquired natural immunity. It is also possible that some HPV detections in HITCH are false positives from depositions due to many of the participants reporting recent sexual activity.³² The presence of depositions may have led to an overestimation of HPV infection transmission rates in HITCH. Finally, we also only recruited heterosexual partnerships; it is very likely that transmission rates differ in non-heterosexual partnerships due to differences in sexual behaviors.

In conclusion, in this updated analysis we found a larger difference in HPV transmission rates between men and women than we had previously observed, and detected some variability in transmission rates across HPV types. Consistent condom use was associated with lower HPV transmission rates between partners. Importantly, we propose a method to calculate HPV transmission rates between partners which accounts for various biases in interval-censored data, and which could also be applied to estimate the transmission rates of other asymptomatic infections.

Data availability

Sample code used to generate the results can be found in the Supplemental Digital Content. Population-level data are provided in the tables. To access individual-level HITCH data, please contact eduardo.franco@mcgill.ca.