Effect of a multicomponent HPV intervention on self-reported HPV vaccine uptake and intention among French adolescents and parents: results from the national, cluster-randomised PrevHPV trial

Abstract

Background

Human papillomavirus (HPV) infections cause several cancers, including nearly all cervical cancer cases. While there are safe and effective vaccines, the adolescent HPV vaccine coverage in France is low (<50%). Thus, we studied the effect of municipality-wide interventions to increase HPV vaccine uptake and intention among middle school students.

Methods

This cluster-randomised trial used an incomplete factorial design to arrange three components (in-school education, motivation, mobilisation (EMM); in-school vaccination; and local general practitioner (GP) trainings) into six intervention conditions, which were randomly assigned to 91 participating French municipalities. We assessed HPV vaccine status using online self-reported questionnaires that students (typically aged 13–14) and parents completed at baseline and 5-month follow-up. Using adjusted linear regression, we estimated differences in (1) vaccine uptake and (2) uptake and intention to vaccinate, both by randomisation arm and by intervention component. We explored subgroup effects by at-home multilingualism, gender, age and parental education.

Results

2047 of 2664 (74%) students were unvaccinated against HPV at baseline. The 5-month probability of first-dose vaccine uptake in the control group was 0·09 (95% CI 0.06 to 0.11). Vaccine campaigns alone contributed a 24-percentage-point (0.18, 0.30) increase in uptake compared with the control group. EMM was only effective in increasing vaccine uptake among monolingual francophone students, and we detected no effect from GP training. Vaccine campaigns and EMM both increased the combined outcome of vaccine intention and first-dose uptake, but EMM had the same subgroup effect. Questionnaire data from parents (N=236) were sparser but demonstrated similar overall trends.

Conclusions

Our results demonstrate that in-school HPV vaccine campaigns effectively increase HPV vaccine uptake among adolescents without exacerbating extant disparities in vaccine uptake. EMM can also be a useful tool to increase HPV vaccine intention but must be improved to bridge disparities in its effectiveness. GP training results were inconclusive.

Trial registration number

NCT04945655.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Nearly 100% of cervical cancer cases are due to human papillomavirus (HPV) infection. While a safe and effective HPV vaccine is recommended for adolescents, uptake is still low in many countries. In many countries, school-based vaccination has been a successful tool to overcome barriers to HPV vaccination.

WHAT THIS STUDY ADDS

• The PrevHPV trial, concluding in 2022, is the first large-scale study in France to examine the effectiveness of school-based vaccination, among other vaccine uptake tactics.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• To accelerate HPV vaccine uptake among adolescents in France, the French government announced in March 2023 that vaccine campaigns would be conducted in French middle schools for students in fifth grade (5ème; approximately age 12, equivalent to US grade 7 and UK grade 8) starting the fall of 2023. Our study results support this policy decision, demonstrating that in-school free-of-charge HPV vaccination is an effective strategy for increasing HPV vaccine uptake in a short period.

Introduction

An estimated 80% of people are infected with human papillomavirus (HPV) at some point in their lives, typically in early adulthood at the time of sexual debut.¹ While the majority of cases (80%–90%) of HPV infection naturally clear without individual health consequences, the remaining people with a persistent infection can suffer long-term health consequences, including anogenital warts and/or precancerous lesions and cancers of the cervix, vulva, vagina, anus, oropharynx, larynx or penis. Nearly 100% of cervical cancers can be attributed to HPV infection.2,4 Safe and effective vaccines can prevent persistent HPV infections and sequelae associated with persistent infections, namely cervical cancer.^{5 6} These vaccines are offered during early adolescence to ensure full protection before any exposure to the virus.

In 2006, the first HPV vaccine became available for distribution and was initially marketed for adolescent girls to protect against cervical cancer.⁷ During the two decades since its introduction, new HPV vaccines have been introduced to protect against more genotypes. Over this time, these vaccines have been thoroughly evaluated and exhibit excellent safety profiles and high effectiveness.^{6 8 9} The WHO now recommends HPV vaccination for older children and adolescents, with girls as the primary target and boys as a secondary target.¹⁰ Many European countries are approaching sufficiently high levels of HPV vaccine coverage to eliminate cervical cancer.¹¹ However, France still lags in HPV vaccine coverage, ranking 27 on the continent in 2021.^{12 13} In France, where HPV vaccination is recommended for all adolescents aged 11–14 years, only 48% of girls in 2022 had received a first dose of the HPV vaccine by the age of 15, and only 13% of boys.^{14 15} Social disparities also persist in HPV coverage, with notable differences in vaccine coverage by gender and socioeconomic status.16,19 Additionally, a study of HPV vaccine coverage in France identified a correlation between low HPV vaccine coverage in geographic units and higher percentage of people born outside of France, tentatively suggesting that there may be sociocultural drivers to HPV vaccination that may be similar to those found in other countries.²⁰

To address the low HPV vaccine coverage in France, the PrevHPV Consortium was established in 2019 with the objective of conducting an interventional research programme targeting three populations: (1) adolescents aged 11–14 years; (2) parents of adolescents aged 11–14 years (as parental consent is mandatory for vaccination) and (3) general practitioners (GPs), who prescribe most HPV vaccines in France and play an essential role in families’ health decision-making process. After exploring knowledge, attitudes, practices and preferences that functioned as barriers or facilitators to vaccine uptake, the consortium coconstructed an intervention with three components: school-based education and motivation targeting students and their parents, school-based vaccine campaigns and training and support to GPs; these components were subsequently evaluated in a cluster-randomised controlled trial.²¹ A priori, the study hypothesised that each of the three components would improve vaccination, and an intervention including all three would be most effective. In the present article, we present the short-term (5 month) effect of these intervention components on self-reported HPV vaccine uptake and intention among French adolescents and their parents. Based on prior research suggesting social disparities in HPV vaccine coverage,16,19 we further examined whether the intervention had a differential impact on HPV vaccine uptake when stratified by preselected sociodemographic characteristics.

Materials and methods

Study design

Ninety-one municipalities, meeting the following criteria, were randomly selected across 9 French regions and 14 school districts: (1) only contained one middle school (students typically aged 11–14 years) and (2) at least 2/3 of adolescents aged 11–14 years residing in the municipality attended this middle school. Using an incomplete factorial design, three intervention components were arranged into six intervention conditions, including one control group, which were randomly assigned to the participating municipalities and subsequently implemented between December 2021 and April 2022 (online supplemental table S1).

Ethics statement

The trial was registered at ClinicalTrials.gov with the following identifier: NCT04945655 (https://classic.clinicaltrials.gov/ct2/show/NCT04945655). The study protocol received ethics approval from the French Ethics Committee ‘CPP Sud-Est VI’ on 22 December 2020 (ID-RCB: 2020-A02031-38). Per French law, no individual informed consent was required for this type of research. Nevertheless, all participants were informed of their rights, particularly their right to not participate or to oppose the collection of their data. No identifying information was collected, and by not requiring written consent, we avoided collecting any participants’ names, thereby protecting all participants, including the minors who were responding to the questionnaires. Additional PrevHPV trial protocol details have been published elsewhere.²¹ Here, we report the study according to the Consolidated Standards of Reporting Trials Checklist (online supplemental material).

Participants

Participants were enrolled through the participating middle schools of the randomised municipalities. While the primary trial outcome and the overall intervention targeted adolescents in sixth to third grades (typically aged 11–14 years), the in-class education component and the questionnaire-based evaluation targeted students attending the fourth or third grades only (typically corresponding to ages 13–14 years, equivalent to US grades 8 and 9 or UK grades 9 and 10). The focus on students in fourth and third grades was justified by the fact that the national education programme addresses topics such as immunity and vaccination in these academic years, and that age-appropriate reproductive health education has occurred.²² This offered a better opportunity to integrate content on HPV vaccination. Students who were outside of the typical age range (13–14 years old) for the fourth or third grade, but who were indeed in these grades at participating middle schools, were still welcome to participate. Eligible students were invited to complete a baseline and a follow-up questionnaire, irrespective of gender or participation in the intervention components planned for their school. Separately, all parents of students attending all grades (French sixth through French third grade) of participating middle schools were eligible to complete the two questionnaires, irrespective of the gender of their child or their participation in the intervention components planned for their school. No exclusion criteria were defined. While the present manuscript reports only secondary outcomes of the PrevHPV trial, the study’s target sample size was determined by power calculations to detect at least a 10 percentage-point difference in vaccine coverage (using French healthcare databases, the primary outcome of the PrevHPV trial). This calculation assumed a municipality-level intraclass correlation of 0.05, 90% power, and an alpha value of 5%.²¹

Randomisation and masking

Participants were cluster randomised to one of six intervention conditions based on their municipality (randomisation unit). Of 1205 eligible municipalities, 351 were randomly sampled and contacted; the first 91 municipalities whose respective middle schools agreed to participate were ultimately selected and randomised, 1:1. Using computer-based random allocation, a senior study member (NT) block randomised these 91 municipalities (block size=6), stratified by school district and the municipality’s social deprivation index.²³ Given the geographic spread of municipalities, there was little risk of spillover effects. Due to the nature of the intervention components, this was an open label trial, and neither the schools nor the study team members offering the interventions were blinded. However, some analyses were conducted while the lead author (JD) was blinded to intervention assignments; masked versus unmasked analyses are identified in the statistical analysis section.

Procedures: intervention components

Three intervention components were provided in different combinations within the six intervention scenarios, with the sixth scenario functioning as the control (online supplemental table S1). These intervention components were implemented following the PrevHPV study’s codevelopment stage in collaboration with stakeholders from the key groups being targeted. These components ultimately included: 1) In-school education, motivation and mobilisation (EMM; December 2021- January 2022), which provided two age-appropriate lessons on HPV to adolescents (only fourth and third grades; typically aged 13–14 years) during regular class time; via the EMM component, students could also access eHealth tools (eg, videos, a serious video game); parents of middle schoolers (between sixth and third grade) were also offered HPV education via web-conference and educational group sessions in school, led by speakers who were trained online; (2) GP trainings (January 2022–April 2022), delivered individually via e-learning sessions, to increase knowledge of HPV infection, vaccination and motivational interviewing techniques to propose vaccination to patients and their parents; (3) facilitation of access (FA; March–April 2022) to HPV vaccination by offering vaccination days where health professionals offered free HPV vaccination on school premises to eligible adolescents who had parental consent. In the present paper, we frequently refer to this FA component as vaccine campaigns. Briefly, EMM was designed to increase vaccine demand, while GP trainings were expected to improve the vaccine offer, and FA was intended to make the vaccine more geographically and financially accessible. All three intervention components were thus hypothesised to increase vaccination via different mechanisms. In the absence of any component, the status quo would be that the participating schools implemented the National Education Program and any previously existing activities, did not hold any vaccine campaigns, and GPs only received any regularly scheduled trainings. For a more complete picture of the trial’s implementation, the PrevHPV group has also published the study protocol,²¹ codevelopment of intervention components²⁴ and implementation evaluation results.²⁵ As people were voluntary participants, their compliance was never forced; instead, the study team measured participants’ degree of exposure, fidelity and comprehension as secondary study outcomes to understand the as-treated intervention effect. The study’s implementation evaluation results are included in a forthcoming publication that describes programme fidelity, dose, reach, acceptability and sustainability using a mixed-methods approach harnessing data from activity reports, self-administered questionnaires and group discussions; it further assesses the effectiveness of the interventions given their dose.

Public involvement

GPs, middle school parents and other stakeholders were involved in the coconstruction process of designing the intervention components during an earlier stage of the PrevHPV trial development. They were not involved in the study design, analysis or dissemination of results described in the present paper.

Data collection

Adolescent participants (in fourth and third grades) completed questionnaires during scheduled class time using a unique and anonymised numeric code. Parents were invited by the school to participate and were provided information to complete the questionnaire online at home using the family’s anonymous numeric code. Parents were asked to complete the questionnaire in reference to their oldest child enrolled in the middle school as we hypothesised that conversations with the oldest child may be more advanced on the subject of HPV. The surveys for adolescents and parents broadly asked for sociodemographic information (eg, parental education, at-home multilingualism, gender, etc), along with knowledge, attitudes, beliefs and practices associated with HPV and the HPV vaccine. The same procedures were used for baseline and follow-up questionnaires. In the interest of completing follow-up within the same academic year, the follow-up survey occurred approximately 2 months after the conclusion of intervention activities within each municipality and approximately 5 months after the baseline survey. Because students were allotted class time to complete the questionnaires, there was much higher questionnaire completion among adolescents compared with parents.

Outcomes: HPV vaccine uptake and intention to vaccinate

The primary PrevHPV trial endpoint, first-dose HPV vaccine coverage (registry-based) 2 months after the conclusion of intervention activities, was reported in a separate manuscript.²⁶ The present manuscript addresses three secondary outcomes. The first (a priori) outcome of the present analysis was first-dose HPV vaccine uptake among individuals unvaccinated at baseline; it is a self-reported, binary variable. The second outcome then combined reported intention to receive a first HPV dose or having received a first dose into one binary variable. Our study protocol planned to measure intention to vaccinate in recognition of the short follow-up period, which meant that many adolescents likely did not have adequate time to visit their primary care providers to receive the vaccine, even if they had the intention to do so. Our decision to combine intention with uptake was made a posteriori, recognising that many students who had never heard of the vaccine or had previously stated that they were not interested in getting vaccinated (during the baseline questionnaire) had subsequently reported that they were vaccinated (during the follow-up questionnaire), meaning that they had passed through the intention stage during the follow-up period and taken action to get vaccinated. Thus, by excluding those who got vaccinated from the outcome definition, we would have underestimated the number of students whose intention to get vaccinated increased during the follow-up period. We derived both outcomes from questions on the follow-up questionnaire that separately asked if the student: (1) was vaccinated against HPV or (2) intended to get vaccinated against HPV (online supplemental table S2). To capture exclusively new vaccinations, adolescents who reported that they had already been vaccinated against HPV on the baseline questionnaire were excluded from the analysis sample. Variables in the parent database had parallel questions to the adolescent database, and outcomes were defined equivalently to the adolescent outcomes described above. Questions referred to the parents’ oldest middle-school child.

A third outcome was defined as self-reported receipt of the second dose during the follow-up period. This outcome was only measured among students (in the adolescent database) who reported having previously received only one dose of the HPV vaccine at the time of the baseline questionnaire (see online supplemental table S2). This outcome aimed to understand the effect of the intervention components on HPV vaccine series completion.

Statistical analysis

While blinded to intervention assignments, author JD conducted a series of analyses on the first outcome of interest, first-dose HPV vaccine uptake. Only individuals with linked baseline and follow-up questionnaires and with complete data on all analysis variables were included. First, the vaccine uptake outcome was regressed onto the six intervention arms, represented as disjoint indicator variables with the control group as the reference. Second, the same regression model was used but regressing the vaccine uptake outcome onto the three intervention components to see the individual effects of each component. These components were modelled as disjoint indicator variables; the control group was still the reference. Finally, an a posteriori exposure definition grouped intervention arms into three categories: (1) no EMM and no FA; (2) EMM but no FA; (3) both EMM and FA. Consequently, this a posteriori model paired the GP-only intervention group together with the true control group under the umbrella of ‘no EMM and no FA’. This post hoc decision was made under the assumption that the outside-school GP intervention had low probability of substantial spillover into the adolescent and parent sample within the 2 months between the intervention and the follow-up questionnaire. This exposure coding was similarly modelled as disjoint indicator variables, with ‘no EMM and no FA’ as the reference.

After unblinding, author JMD applied the same three intervention coding schemes to model the second and third outcomes of interest: (2) first-dose HPV vaccine uptake or intention to vaccinate and (3) second-dose uptake.

All analyses were conducted using linear regression to estimate risk differences (RD), which have been demonstrated to have more public health significance than ORs when the outcome is not rare.^{21 27 28} Models only included fixed effects and accounted for clustering at the school level in the robust estimation of the variance. The fixed-effects model was selected after testing a series of mixed-effects models and finding that the intracluster correlation for each outcome was still quite small (<0.02). An overview of the model selection process, comparing eight different models (including generalised estimating equations), is found in online supplemental table S3. Ultimately, the fixed-effects model achieved greater parsimony and easier interpretability while offering similar accuracy and precision as the other models.

Data analysis was conducted using Stata V.17²⁹; the R tidyverse package was used for final data visualisation.³⁰

Bias adjustment

We posited that people who were more interested in the HPV vaccine were likely more engaged in responding to the questionnaire accurately and completely (including providing their unique ID number correctly) compared with those who were less interested in HPV vaccination; similarly, people who were more amenable to vaccination were also more likely to complete the questionnaires, producing a specific form of selection bias due to collider bias, where the observability of the data is the collider (online supplemental figure S1).³¹ The result of such bias is a confounding path that otherwise would not have existed in randomised data. Therefore, in lieu of assuming exchangeability of participants per intervention arm based solely on randomisation, we elected a priori to control for bias associated with respondents’ baseline propensity towards vaccination. Rooted in the transtheoretical model (TTM) of health behaviour change,^{32 33} we identified five exclusive categories to represent distinct stages that culminate in vaccine uptake. Briefly, the TTM posits that there are six temporal stages to behavioural change: precontemplation, contemplation, preparation, action, maintenance and termination. This model was originally proposed in the context of repetitive preventive behaviours (eg, smoking cessation or other forms of harm reduction), and in the context of HPV vaccination, these stages can be condensed due to the limited maintenance required after taking action.³⁴ Thus, we identified the following five categories to reflect the stages leading to vaccine uptake: vaccine ignorance (precontemplation), vaccine indecision (contemplation), vaccine refusal (precontemplation), vaccine intention (preparation) and vaccine uptake (action/maintenance/termination). Using questions from the baseline questionnaire (online supplemental table S2), respondents were captured into one of these five categories, which we henceforth call vaccine propensity. Respondents in the ‘vaccine uptake’ category at baseline were already excluded from analysis of the first two outcomes. Therefore, this category was excluded from regression models and the other four categories were modelled as disjoint indicator variables. Disjoint indicator variables allowed the analytical model to relax the assumption of linearity between these levels of vaccine propensity³⁵; vaccine ignorance was the reference category. The third outcome of interest, second-dose uptake, was analysed only among adolescents who had taken a first vaccine dose (thus considered to be in the ‘uptake’ category); therefore, no additional adjustment was applied to the regression models of the second-dose uptake analysis, as this was more a question of maintenance of the action (vaccine uptake to complete the HPV vaccine schedule) among people who had already taken the initial action (first dose). Analysis of parents’ data was crude due to small sample size and non-convergence.

Subgroup analysis by sociodemographic determinants

After analysis of the average intervention effects, we explored how these effects may vary within certain sociodemographic subgroups. Based on prior knowledge of disparities in HPV vaccine coverage in France,¹⁸ we selected four sociodemographic indicators to examine for modification of intervention effect: gender, school grade (as a proxy for age), multilingualism at home (prompt: ‘Do you speak another language other than French fluently with your parents?’) and parental educational attainment. To this end, we produced stratified estimates of the effect of intervention components by applying interaction terms to the fixed-effects models described above; each interaction term occurred in a separate model. Most demographic variables were dichotomised (ie, girl/boy; monolingual/multilingual; third/fourth grade). Parental education offered three options: less than or equal to baccalauréat, greater than the baccalauréat or the student did not know (the most frequent response). All stratified analyses of the first outcome (first dose vaccination alone) did not adjust for baseline vaccine propensity to allow convergence on the linear scale and were conducted while author JD was blind to intervention assignments. Stratified analyses of the second outcome (intention or first dose vaccination) adjusted for baseline vaccine propensity and were conducted after unblinding. We did not explore subgroup effects for the third outcome due to small sample size. Likewise, no subgroup effects were explored in the parent questionnaire data.

Role of the funding source

The PrevHPV study was conducted with the support from the Institut pour la Recherche en Santé Publique and Alliance Nationale pour les Sciences de la Vie et de la Santé/National Alliance for Life Sciences & Health (ITMO Cancer AVIESAN). JMD received postdoctoral funding from the Gillings Foundation. The funders of the study had no role in study design, data collection, data analysis, data interpretation or writing of the report.

Results

Characteristics of the adolescent sample

In total, 91 middle schools, with 19 885 students, were randomised. After 31 schools formally dropped out (many citing COVID-19 pandemic-related reasons), 13 233 students (67%) attended the 60 remaining schools. In total, we received 7632 (58%) baseline questionnaire responses from eligible students at the 60 remaining schools. After deduplicating and linking the baseline and follow-up questionnaire data, 2758 (36%) students from 39 schools remained (figure 1); 94 of these students were missing baseline vaccine status data and were subsequently excluded, resulting in a final sample of 2664 students. Among these students, baseline vaccine coverage was similar between arms. The arm with the fewest schools was arm 1, with four schools remaining; the arm with the most schools was arm 3 (nine schools; online supplemental table S4). The median time between adolescents’ baseline (November 2021–March 2022) and follow-up survey responses (May 2022–June 2022) was 4.5 months (IQR: 4.2–4.9). At baseline, 829 of 2664 students (31%) had not heard of the HPV vaccine, 303 (11%) indicated that they did not intend to get vaccinated, 514 (19%) were not sure whether they would get vaccinated, 398 (15%) reported that they did intend to get vaccinated, and 620 (23%) had already received at least one HPV vaccine dose at baseline.

Figure 1. Participant flowchart. ¹Estimates of 30% class non-participation were extrapolated based on the EMM sample of classrooms. ²Identification numbers (IDs) in questionnaires allowed unique identification of schools but not classrooms, therefore, class participation in questionnaires cannot be directly calculated. ³Not included in calculation of loss to follow-up. EMM, education, motivation,

In this sample of students who had no known HPV vaccine (n=2044), there were similar proportions of students by school year and gender across all arms. However, school area deprivation level, language spoken at home and parental education had wide variation across intervention arms (table 1). Baseline vaccine propensity also varied by arm, with HPV vaccine ignorance ranging from 48% in arm 6 (control) to 30% in arm 1, which received all intervention components. Similarly, baseline vaccine intention was highest in arms 1 and 3, with 28% and 26% of students having intention before any intervention activities began (table 1).

Table 1. Participant characteristics (overall and by randomisation arm) among middle-school students with no reported history of HPV vaccination at baseline (N=2044)^*.

	Arm 1 (EMM, FA, GP)	Arm 2 (EMM, GP)	Arm 3 (EMM, FA)	Arm 4 (EMM)	Arm 5 (GP)	Arm 6 (Control)	Overall
	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)
Gender
Girls	65 (45)	125 (44)	183 (39)	190 (49)	116 (52)	237 (44)	916 (45)
Boys	81 (55)	158 (56)	285 (61)	200 (51)	105 (48)	299 (56)	1128 (55)
School area deprivation level
Lower deprivation	16 (11)	171 (60)	210 (45)	57 (15)	206 (93)	430 (80)	1090 (53)
Higher deprivation	130 (89)	112 (40)	258 (55)	333 (85)	15 (7)	106 (20)	954 (47)
School year
4th (age~13 years)	80 (55)	165 (58)	264 (56)	245 (63)	125 (57)	298 (56)	1177 (58)
3rd (age~14 years)	66 (45)	118 (42)	204 (44)	145 (37)	96 (43)	238 (44)	867 (42)
Language spoken at home
Multilingual	15 (10)	76 (27)	89 (19)	92 (24)	84 (38)	111 (21)	467 (23)
Monolingual French	131 (90)	207 (73)	379 (81)	298 (76)	137 (62)	425 (79)	1577 (77)
Highest level of education achieved by either parent
≤ Baccalauréat	32 (22)	33 (12)	77 (16)	74 (19)	46 (21)	79 (15)	341 (17)
> Baccalauréat	40 (27)	127 (45)	160 (34)	121 (31)	89 (40)	220 (41)	757 (37)
Student does not know	74 (51)	123 (43)	231 (49)	195 (50)	86 (39)	237 (44)	946 (46)
Baseline vaccine propensity
HPV vaccine ignorance	44 (30)	95 (34)	160 (34)	178 (46)	97 (44)	255 (48)	829 (41)
HPV vaccine refusal	24 (16)	48 (17)	75 (16)	59 (15)	37 (17)	60 (11)	303 (15)
HPV vaccine indecision	37 (25)	93 (33)	112 (24)	94 (24)	46 (21)	132 (25)	514 (25)
HPV vaccine intention	41 (28)	47 (17)	121 (26)	59 (15)	41 (19)	89 (17)	398 (19)
Newly vaccinated at follow-up
Received first dose	56 (40)	39 (14)	192 (42)	62 (17)	24 (11)	62 (12)	435 (22)
Unvaccinated	85 (60)	236 (86)	264 (58)	308 (83)	187 (89)	461 (88)	1541 (78)
Total	146	283	468	390	221	536	2044

^*Where the stratified cells do not add to the column totals, this indicates missing data for the selected variable among participants within the selected treatment arm.

EMM, education, motivation and mobilisation; FA, facilitation of access (ie, vaccine campaigns); GP, GP training component; HPV, human papillomavirus.

Among previously unvaccinated students who provided complete vaccine status information at follow-up (n=1976), 435 (22%) reported receiving their first HPV vaccine dose during the follow-up period (table 1); 249 (47%) of these students attended schools that were assigned to vaccine campaigns. Of these 249 students, 149 (60%) reported that they received their HPV vaccine at school.

Regression results: first-dose vaccine uptake, reported by adolescents

Linear regression models, adjusting for baseline vaccine propensity, estimated the following 5-month probabilities of first-dose HPV vaccine uptake: 0.09 (95% CI 0.06–0.11) in the control arm, 0.34 (0.18–0.49) in arm 1 (with EMM, FA and GP) and 0.37 (0.33–0.41) in arm 3 (with EMM and FA). Thus, arms 1 and 3 yielded a 25 percentage-point (0.09–0.41) and a 28 percentage-point (0.24–0.32) net increase in first-dose vaccine uptake, respectively. We did not observe any comparable differences in first-dose vaccine uptake in intervention arms without FA compared with the control group. When decomposed, the school-based vaccine campaign (FA) component was associated with a 24 percentage-point (0.18–0.30) net increase in first-dose uptake compared with the control group (table 2), while EMM and GP trainings did not show any meaningful net increase.

Table 2. HPV vaccine uptake during follow-up among adolescents who were not vaccinated against HPV at baseline (N=1976)^*.

	Risk of first-dose HPV vaccination during follow-up		5-month risk difference of first-dose HPV vaccination†
	Estimate	95% CI	Estimate	95% CI
Randomisation group
Arm 1 (EMM, FA, GP)	0.34	(0.18 to 0.49)	0.25	(0.09 to 0.41)
Arm 2 (EMM, GP)	0.11	(0.08 to 0.15)	0.03	(−0.01 to 0.06)
Arm 3 (EMM, FA)	0.37	(0.33 to 0.41)	0.28	(0.24 to 0.32)
Arm 4 (EMM only)	0.13	(0.05 to 0.20)	0.04	(−0.03 to 0.11)
Arm 5 (GP only)	0.10	(0.07 to 0.13)	0.01	(−0.02 to 0.04)
Arm 6 (Control)	0.09	(0.06 to 0.11)	Ref
Decomposed Intervention Components
GP Training (GP)	0.09	(0.05 to 0.13)	−0.00	(−0.04 to 0.03)
Vaccine campaigns (FA)	0.33	(0.27 to 0.39)	0.24	(0.18 to 0.30)
In-school education (EMM)	0.13	(0.08 to 0.18)	0.03	(−0.01 to 0.08)
None	0.09	(0.06 to 0.12)	Ref
Collapsed component clusters
EMM & FA	0.36	(0.31 to 0.41)	0.27	(0.22 to 0.32)
EMM but no FA	0.12	(0.07 to 0.17)	0.03	(−0.01 to 0.08)
No EMM & No FA	0.09	(0.07 to 0.12)	Ref

Model adjusted for baseline HPV vaccine propensity.

^*Due to complete case analysis, the effective number of adolescents in this regression analysis is smaller than the analysis sample represented in table 1.

^†The risk difference represents the absolute increase in percentage points of first-dose vaccine coverage among previously unvaccinated, per intervention/exposure.

EMM, education, motivation and mobilisation; FA, facilitation of access (ie, vaccine campaigns); GP, GP training component; HPV, human papillomavirus.

No subgroup effects were observed on the relationship between vaccine campaigns and first-dose uptake. Furthermore, there were no subgroup effects by gender, school grade and parental education on any relationship between any intervention component and first-dose vaccine uptake (figure 2). However, at-home multilingualism did modify EMM intervention effects: monolingual francophone students experienced a gain in first-dose uptake from the in-school education component (RD: 0.07 (95% CI 0.03–0.11)), while multilingual students did not experience any gain (RD: −0.03 (−0.08–0.01); figure 2 and online supplemental figure S2).

Figure 2. Risk differences (relative to control arm) in first-dose uptake per intervention component, stratified by adolescents’ gender (A), grade (B), multilingualism (C), and parental education (D). ¹Reference level for risk difference is the control arm (no intervention) within each stratum; no baseline vaccine propensity adjustment. In (C) the CI for multilingual students in municipalities receiving GP training was 0.04 (95% CI: −0.003 to 0.09). The point estimate for monolingual students in municipalities receiving GP training was −0.04 (−0.07 to −0.001). EMM, education, motivation, mobilisation; FA, facilitation of access; GP, general practitioner; HPV, human papillomavirus.

Regression results: vaccine intention and uptake, reported by adolescents

In similar analyses on the second outcome, all intervention arms except for the GP training-only arm (arm 5) increased the probability of first-dose uptake or intention to vaccinate (table 3). Comparing the decomposed components to the control group, the school vaccine campaigns (FA) were associated with an increased probability of first-dose uptake or intention (RD: 0.14 (0.06 to 0.22)) compared with the control, as was in-school education (EMM, RD: 0.12 (0.05 to 0.19); table 3). In subgroup analysis, the gain in intention/vaccination from the in-school education was only observed among monolingual francophone students (RD: 0.17 (0.10 to 0.23)), while multilingual students experienced no benefit from the EMM component (RD: −0.01 (−0.08 to 0.07)), even when accounting for baseline vaccine propensity (online supplemental figure S3). This trend persisted even when at-home multilingualism and parental education were stratified together: the differences in subgroup effects were still driven by at-home multilingualism and not parental education (online supplemental figure S4).

Table 3. First-dose uptake or intention to vaccinate at 5-month follow-up questionnaire among adolescents who were not vaccinated against HPV at baseline (N=1976)^*.

	Risk of HPV vaccination or intention (linear model)		Risk difference†
	Estimate	95% CI	Estimate	95% CI
Randomisation group
Arm 1 (EMM, FA, GP)	0.48	(0.36 to 0.60)	0.30	(0.18 to 0.42)
Arm 2 (EMM, GP)	0.30	(0.25 to 0.35)	0.12	(0.06 to 0.17)
Arm 3 (EMM, FA)	0.43	(0.37 to 0.48)	0.24	(0.19 to 0.30)
Arm 4 (EMM only)	0.30	(0.19 to 0.41)	0.12	(0.01 to 0.22)
Arm 5 (GP only)	0.17	(0.13 to 0.21)	−0.01	(−0.05 to 0.03)
Arm 6 (Control)	0.18	(0.16 to 0.21)	Ref
Decomposed intervention components
GP training (GP)	0.19	(0.14 to 0.23)	0.01	(−0.04 to 0.06)
Vaccine campaigns (FA)	0.32	(0.24 to 0.40)	0.14	(0.06 to 0.22)
In-school education (EMM)	0.30	(0.22 to 0.38)	0.12	(0.05 to 0.19)
None	0.18	(0.15 to 0.20)	Ref
Collapsed component clusters
EMM and FA	0.44	(0.38 to 0.50)	0.26	(0.20 to 0.32)
EMM but no FA	0.30	(0.23 to 0.37)	0.12	(0.06 to 0.19)
No EMM and No FA	0.18	(0.15 to 0.20)	Ref

Model adjusted for baseline HPV vaccine propensity.

^*Due to complete case analysis, the effective number of adolescents in this regression analysis is smaller than the analysis sample represented in table 1.

^†The risk difference represents the absolute increase in percentage points of first-dose vaccine uptake or intention to vaccinate among previously unvaccinated.

EMM, education, motivation and mobilisation; FA, facilitation of access (ie, vaccine campaigns); GP, GP training component; HPV, human papillomavirus.

Regression results: second-dose uptake, reported by adolescents

The analysis sample for the third outcome, second-dose uptake, included the 175 adolescents who reported receipt of only one HPV vaccine dose at baseline. Most students declaring receipt of one dose were girls and lived in monolingual households (online supplemental table S5). There were no detectable effects of the intervention components on second-dose uptake (online supplemental table S6).

Analysis of survey responses from parents of middle-school students

Analyses of the parent questionnaire data had a lower sample size compared with the adolescent data set, with 1948 parents responding to the baseline questionnaire, but only 517 parents responding to the follow-up questionnaire. Ultimately, 236 parents used the same ID code and could therefore be linked between their baseline and follow-up questionnaire responses. Of these, 143 (61%) reported that their oldest middle-school child was unvaccinated against HPV at baseline. A large proportion of parents in this sample expressed in the baseline questionnaire that they intended to vaccinate their children (online supplemental table S7). In crude linear regression analyses, results mirrored those of the adolescent sample (RD (arm1): 0.30 (–0.04 to 0.64); RD (arm3): 0.32 (0.09 to 0.56)), although with wide CIs (online supplemental table S8). Given the small sample size, analyses of the other two outcomes could not be replicated in the parent data.

Discussion

In this cluster-randomised trial, we found that free at-school HPV vaccination rapidly increased first-dose vaccine uptake among adolescents in the third and fourth grades at French middle schools (typically aged 13–14 years), with a net increase of 24 percentage points compared with the control (ie, no intervention). Expressed as a number needed to treat, for every four unvaccinated adolescents offered the HPV vaccine through these campaigns at their middle school, one student got vaccinated who would not have otherwise been vaccinated during our study period. In-school vaccination also had a homogenous effect across subgroups, that is, subgroups that were known to have lower HPV vaccine coverage at baseline benefited equally from this intervention. This result suggests that free school-based vaccination is an effective approach to increase HPV vaccine coverage among adolescents. While it does not bridge pre-existing gaps in coverage due to social inequities, it also does not appear to widen such gaps. This aligns with findings in many other countries where school-based vaccine campaigns have been adopted. For example, in Europe, a comprehensive review of HPV vaccination policies and vaccine coverage in 31 countries demonstrated that countries with high vaccine coverage commonly had facilitated (ie, free, on-site) access to vaccination.³⁶ Furthermore, in New Zealand, a publicly funded school-based vaccination programme began in September 2008 and successfully avoided gaps in HPV vaccine coverage among Māori people, who often experience greater disparities in vaccine coverage for other infectious diseases.³⁷ Thus, these results aligned with other evidence and our initial hypotheses.

In this analysis, counter to expectation, we did not observe an effect from the GP training component. One explanation of this null effect may be the low levels of GP participation: while there were 199 eligible GPs in the 45 municipalities randomised to receive GP training, we ultimately had just 43 GPs participate in the training across only 26 municipalities. While a full analysis of intervention fidelity is forthcoming,²⁵ one reason for low participation may be the timing of the study in late 2021: after 2 years of working during the COVID-19 pandemic, many physicians were experiencing high workload and limited capacity.^{38 39} Another explanation for our study results may be that the modality of the training, online rather than in-person, reduced the effectiveness of the training. However, several studies have demonstrated that in-person and online medical training can be equally effective, particularly for continuing medical education.40,42 One additional explanation for the null effect of the GP training component is that the follow-up period, with a median 4.5 months between baseline and follow-up questionnaires (collected only 2 months postintervention), was too granular to capture the downstream effects of this component. Studies have shown that 75%–90% of French adolescents visit their doctor at least once a year.^{18 43 44} However, given that the follow-up period only covered approximately a third of a year, many students likely did not have consultations during that time. Understanding whether GP trainings might improve HPV vaccine uptake while targeting disparities in uptake is still a question of importance that requires further research to answer. To address this question and assess the long-term effects of the study intervention components, the PrevHPV group intends to compare the municipalities’ vaccine coverage (using French healthcare databases) at 6 months and 12 months postintervention. Previous studies in the context of other vaccines (eg, COVID-19 vaccination) have demonstrated that doctor’s communication of their trust in a vaccine can influence individual uptake.⁴⁵

While we initially anticipated that in-school education would also increase intention and vaccination, our study results demonstrated that in-school education had a heterogeneous effect, where it was only effective in increasing HPV vaccination or intention among adolescents who spoke exclusively French at home. The effect on first-dose uptake in this subgroup, representing 77% of the adolescent sample, was small (RD: 0.07 (0.03 to 0.11)) but impressive given the short follow-up period. However, by contrast, in-school education did not result in any net gains in first-dose vaccination or intention to vaccinate among adolescents also speaking another language (besides French) fluently at home (23% of the sample) during the follow-up period. We further observed that this subgroup effect persisted when stratifying language with parent education, suggesting that socioeconomic status cannot fully explain the observed trend. Given that this multilingual subgroup is highly heterogeneous, there are likely several explanations for these detected subgroup effects, and deeper qualitative research is needed. While our study’s measure captured only whether the students spoke another language fluently (besides French) with their parents, this could be considered as a proxy for recent immigration, lower linguistic competencies and/or less access to health resources, which can limit engagement in preventive healthcare. Broadly, however, our findings align with other research on HPV vaccine initiation in France that have highlighted how the proportion of recent immigrants in a geographic unit is highly predictive of the level of HPV vaccine uptake.^{16 20 46} Our findings, placed within the context of the larger body of evidence on this subject,^{16 18 20 46} suggest that this is a group requiring more tailored approaches to increase uptake. We again identify qualitative, coconstructive work with intentional recruitment of stakeholders from diverse language and cultural backgrounds as a way forward to better realise the full potential of in-school EMM for all adolescents. In doing so, we hope to avoid deepening disparities in HPV vaccination, and instead closing extant gaps to achieve equity.

This study has several limitations. First, a large proportion of baseline participants were lost to follow-up, which led to imbalances in certain variables, such as multilingualism, between intervention groups. This loss-to-follow-up may have induced selection bias. We have addressed this potential bias as much as possible in our regression models by adjusting for baseline vaccine propensity, and we do not anticipate that any residual bias impacted the direction of our results. However, this residual selection bias may have influenced the validity of the reported estimates in the event that participants were more likely to respond to the follow-up questionnaire if they had gotten vaccinated or experienced new intention to vaccinate between baseline and follow-up. Nonetheless, we do not anticipate such a trend to have been sufficiently extreme to produce a large bias in the analysis of the adolescent data, as the study was designed for them to complete the questionnaire during designated class time. A second implication of the loss to follow-up, along with the specific selection criteria of the study municipalities, is reduced generalisability to all middle schools in France. Rather, the present study results more likely reflect the results that would be seen in middle schools with leadership who are particularly interested in offering HPV vaccination and education and remained in the study despite the unique constraints of conducting the interventions during the 2021–2022 academic year. A similar issue with generalisability is recognisable in the parents’ data, which is a small and highly motivated group. Despite these limitations, we believe it is important to present the parent data analysis as parents are important in the vaccine decision-making process for adolescents in France, particularly because two parents/legal guardians must provide consent for a minor to have any non-obligatory medical procedure (eg, HPV vaccination) in France.^{47 48} Another possible limitation was social desirability bias, which we attempted to reduce by using anonymous online questionnaires, and which should not affect our estimates under the assumption that the prevalence of social desirability bias did not significantly differ across our six (randomised) groups. Also, due to convergence issues in modelling the subgroup effects for the first outcome of this analysis, we ran crude effects models that did not adjust for baseline vaccine propensity, suggesting the possibility of confounding (due to collider bias on data observability; see online supplemental figure S1) in these subgroup estimates. Because this subgroup analysis was interested in observing heterogeneity of effect estimates (rather than exactness of estimates), any confounding would have to be differential by subgroup strata in order to bias our interpretation of the results. Moreover, because this specific subgroup analysis found primarily homogenous effects, we would expect that any differential confounding induced by our modelling approach biased our results towards a null (ie, homogenous) result. However, we do not anticipate that there was substantial differential confounding in this subgroup analysis, as the results are quite similar to those of the subgroup analysis for the second outcome, which did adjust for baseline vaccine propensity. Finally, the present analysis assumed homogeneity in questionnaire participants’ dose of exposure to interventions per cluster and does not account for the variation in intervention fidelity and delivery among participating middle schools and GPs. Future analyses will model vaccine uptake and intention among adolescents, accounting for the varying levels of adherence to the intervention protocols and whether GPs in their municipality truly engaged in training.

The present study also has several strengths. First, we have collected a broad amount of data via a randomised control study design that included a geographically diverse swath of adolescents and parents of adolescents in France. Moreover, using individual-level survey data with a large sample size allowed us to investigate subgroup effects. Furthermore, the evaluation of individuals’ baseline vaccine propensity (ranging from never having heard of the HPV vaccine to already having the intention to get vaccinated) provided useful information to adjust for individual-level predictors of the outcome and for selection into the final analysis sample within a trial that was randomised by school clusters. Finally, our approach of modelling the RD (vs ORs), for both the randomisation arms and the decomposed intervention components, produces policy-relevant insight into the contributions of each intervention component.

To accelerate HPV vaccine uptake among adolescents in France, the French government announced in March 2023 that vaccine campaigns would be conducted in French middle schools for students in fifth grade (5ème; approximately age 12, equivalent to US grade 7 and UK grade 8) starting the fall of 2023.⁴⁹ Our study results support this policy and suggest that HPV vaccine campaigns function well to increase in-school HPV vaccine uptake among adolescents in France and should not exacerbate social disparities in vaccine uptake. Moreover, in the time since our initial analysis and presentation of the PrevHPV study results in 2023,⁵⁰ new results from the inaugural HPV vaccine campaigns (2023–2024 academic year) in middle schools across all of France have recently been released by the National Public Health Agency. Their results demonstrate that vaccine campaigns had a strong influence on HPV vaccine coverage, with a 24-percentage-point increase in first-dose coverage among girls after two vaccine campaign phases over the course of the school year. The estimate for boys was 22 percentage points. They also note disparities among geographic regions, identifying low gains in vaccine coverage in France’s overseas territories compared with mainland departments.⁵¹ These results align with our study results, as the estimates in gains in vaccine coverage are similar, with a relatively homogenous effect among girls and boys. Nevertheless, the disparities in vaccine coverage for residents in overseas territories require further investigation as there are both socioeconomic and cultural differences between the overseas territories and mainland France that could influence the effectiveness of in-school vaccine campaigns, particularly when implemented in schools that typically have less resources than many regions of mainland France.^{52 53}

Finally, from an ethical perspective, it is our view that vaccination campaigns should be conducted with substantial information and education for parents and adolescents. However, the current vaccine promotion tools do not equally address vaccine intention and uptake among adolescents and may expand existing disparities in HPV vaccine coverage. Using the approach of proportionate universalism, which proposes ‘actions of sufficient scale and intensity to be universal but also proportionately targeted’,⁵⁴ may be an ideal way to move forward in reducing inequalities in HPV vaccination in particular target groups that may need additional, tailored vaccine promotion programming. For such an approach to be successful, more education and communication materials must be developed with the support of a demographically diverse cohort of parents, students and teachers who can offer greater insight into the ways in which the current materials can be improved to increase effectiveness.

Data availability statement

Data are available upon reasonable request.