An Evaluation of Dose-Related HPV Vaccine Effectiveness Using Central Registries in Michigan

More than 200 human papillomavirus (HPV) types have been identified; approximately 40 are sexually transmitted, and 12 (HPV16/18/31/33/35/39/45/51/52/56/58/59) have been classified as human carcinogens (1, 2). Together, HPV16 and 18 cause 70% of HPV-attributable cancers worldwide (3). Compared with other types, HPV16 is more likely to persist and progress to cervical precancer and invasive cervical cancer and does so more quickly (4).

The advent of HPV vaccination has created the opportunity to reduce the incidence of HPV types that cause cervical neoplasia. HPV vaccines were licensed based on clinical trials, which demonstrated high efficacy for prevention of vaccine-type (i.e., associated with HPV16 or 18) cervical precancer among women who did not have evidence of infection at the time of vaccination (5). These vaccines, originally evaluated, licensed, and recommended as a 3-dose schedule (0, 1–2, and 6 months), have been introduced into national vaccination programs in nearly 100 countries (6). In the United States, routine vaccination of girls ages 11 to 12 years, with catch-up vaccination through age 26, was first recommended in 2006 (7). Three vaccines have been licensed; all target HPV16/18. Recommendations have changed over time as new vaccines were licensed and indications approved (7). Noninferiority immunogenicity studies provided data accepted for licensure of 2-dose schedules in persons who start the series before age 15 years (8–11). More recently, there has been interest in single-dose vaccination (12).

Studies conducted to evaluate vaccine effectiveness (VE) in real-world programs have generally shown lower effectiveness with <3 doses than with 3 doses, although methodologic challenges have been noted (13). Previously, a pilot study investigated the feasibility of linking Michigan's statewide cancer and immunization registries, with a goal of creating a framework for monitoring vaccine impact and effectiveness (14). The inclusion of adults in the immunization registry and of in situ lesions in the cancer registry, continuously since the HPV vaccination program began in the United States, presents a unique opportunity for a record linkage VE study. Here, we use the registry linkage to evaluate HPV VE against in situ and invasive cervical cancers, including by number of vaccine doses and by age at vaccination. Additionally, we compare results of a case–cohort study design, which included verification of continuous residency to ensure complete vaccination history and cervical cancer status, to a less labor-intensive record linkage cohort study without continuous residency determination.

We established a cohort of women born in Michigan, 1980–1995, by linking records from multiple statewide data systems using methods piloted in 2009 (14); this study was approved by the Michigan Department of Health and Human Services (MDHHS) Institutional Review Board.

Females born in Michigan were identified from the Michigan Live Birth Registry (hereafter, “birth records”). The Michigan Care Improvement Registry is Michigan's immunization information system; in 2006, this transitioned from a childhood immunization registry to a lifespan registry including children and adults. The Michigan Cancer Surveillance Program is the state's cancer registry. In addition to invasive cervical cancers, diagnoses of certain cervical precancers have been reported continuously since prior to the 2006 licensure of HPV vaccines, in contrast to most U.S. state cancer registries. For this study, invasive cervical cancers and in situ cervical lesions including histologically confirmed cervical intraepithelial neoplasia grade 3 (CIN3), and adenocarcinoma in situ (AIS) comprised cases of CIN3 or worse (CIN3+). Terminology used by pathologists has evolved during the study period, including adoption of the two-tier Lower Anogenital Squamous Terminology in 2012 (15); efforts were made to report similar cases to cancer registries over time. CIN3+ includes high-grade squamous intraepithelial lesions and lesions diagnosed as CIN2 or CIN2–3 with positive p16.

Females in the immunization registry and women with CIN3+ in the cancer registry were each linked to birth records, establishing a linkage between CIN3+ and HPV vaccination history. The immunization registry and birth records were linked using probabilistic and deterministic methods that considered patient names, birth dates, parent names, and cities of residence, as previously described (14). Methods used for linkage varied over the study period, and percentage linked between the birth and immunization registries increased from 47% for the 1980 birth cohort to 83% for the 2004 birth cohort (Supplementary Fig. S1). Next, to link cancer registry CIN3+ cases to birth records, probabilistic methods using name and date of birth were used. Subsequently, CIN3+ cases were limited to those diagnosed 2009–2016 to include only women who could have been vaccinated ≥24 months before CIN3+ diagnosis. The cohort data set includes birth year and race (as described below), HPV vaccination doses and dates from the immunization registry, and CIN3+ diagnoses and dates from the cancer registry.

From the full cohort of Michigan-born women in the immunization registry, we selected a subcohort of women who had resided in Michigan continuously to limit bias that would result if vaccinations or CIN3+ diagnoses occurred while residing elsewhere. Determining residential history for the entire cohort was cost-prohibitive, so a stratified random sampling strategy was initially planned to estimate vaccination history by birth year group (1980–1984, 1985–1989, 1990–1995), with a 5% margin of error for each proportion. It was important to consider different time periods because vaccine eligibility varied by birth year (i.e., by age in 2006 when HPV vaccine was initially recommended) and to reflect the erosion of continuous residency over time. The sampling fractions per thousand women were 4.1, 2.6, and 2.2 for 1980–1984, 1985–1989, and 1990–1991, respectively. Lexis Nexis Accurint was used to obtain address history information, as described in the feasibility study (14), of the sampled subcohort and all linked CIN3+ cases. Information on name at birth, name as recorded in the immunization registry (if available), and name at CIN3+ diagnosis was obtained to determine address history for the woman, her mother, and the father as listed on the linked birth certificate. If an address within any such history included a non-Michigan address, the woman's residence was classified as noncontinuous.

All HPV vaccine doses [quadrivalent (CVX code 62); bivalent (CVX code 118); nonavalent (CVX code 165); and HPV, unspecified (CVX code 137)] administered and reported to the immunization registry by May 26, 2017, were considered for inclusion. These doses were assessed for validity following Advisory Committee on Immunization Practices guidance (10, 16, 17). Invalid doses and vaccines doses administered <24 months before (i) a cervical lesion diagnosis or (ii) December 31, 2016, if not diagnosed, were excluded. A status of unvaccinated had zero valid doses reported, up-to-date (UTD) had either 3 valid doses or 2 valid doses where dose 1 was administered before the 15th birthday and dose 2 was administered ≥5 months—4 days after dose 1, not UTD had either 1 valid dose or 2 valid doses that did not meet UTD criteria.

Data on race were obtained from birth records and were classified as white, black, and other. Race was derived from the bridged single race of the reported race(s) for the named parents using bridging methodology from National Center for Health Statistics (18). Hispanic ethnicity in the birth record is classified separately, and the Hispanic population in Michigan was deemed too low for a statistically informative analysis.

The distributions of birth year group, vaccination status in 2016, and race were analyzed for the full cohort, all noncases and CIN3+ cases; and for continuous residents, the subcohort noncases, and CIN3+ cases. The distributions of birth year group and race were analyzed by vaccination status for the samples used in the cohort and case–cohort analyses. Chi-square tests were used to evaluate statistical differences in proportions.

We performed a VE analysis using the full cohort with all CIN3+, regardless of continuous residency determination. The relative risk of CIN3+ by vaccination status was calculated using log-binomial regression adjusted for race and continuous birth year. VE, calculated as (1 − aRR) × 100, was estimated for ≥1 dose of HPV vaccine and by number of doses (1-dose, 2-dose, not UTD; and UTD). VE was also estimated for ≥1 dose stratified by birth year group and by age at vaccination (<20, ≥20 years).

We compared the cohort analysis with a case–cohort design, using the continuously resident subcohort and continuously resident CIN3+ cases (19). Subcohort members were weighted using the inverse of the stratum-specific sampling fractions, and subcohort cases were analyzed as cases, which had a weight of 1. Because subcohort sampling fractions were low and only 9 (0.65%) subcohort members developed CIN3+, survival analysis methods using time-varying weights were considered unnecessary (20). Using cases versus subcohort noncases as the dependent variable, we used logistic regression to estimate the odds ratio (OR) by vaccination status, and to estimate adjusted OR (aOR) adjusted for race and continuous birth year. Variances were calculated using methods appropriate for weighted analyses. The resulting aOR approximates the adjusted risk ratio (aRR), defined as the ratio of the cumulative incidence in the exposed by the cumulative incidence in the unexposed. VE was estimated as (1 − aOR) × 100. The exposures and covariates evaluated in the cohort analysis were repeated.

All analyses were performed using SAS 9.4.

Among 1,926,433 women with immunization registry records born during 1980–2004, 1,312,116 (68%) were linked to birth records (Fig. 1). After excluding 538,923 women whose birth year was not in 1980–1995, a cohort of 773,193 women remained. For the subcohort, 2,110 records were selected from the cohort. Of these, 1,383 (66%) were continuous residents and formed the study subcohort.

Of the 14,428 CIN3+ cases diagnosed during 1995–2016 and reported to the cancer registry among women born in 1980–1995, 11,036 (76%) were linked to a birth record; of these, 7,010 (64%) were also linked with the immunization registry (Fig. 1). After excluding 3,172 CIN3+ cases diagnosed before 2009, 3,838 CIN3+ cases diagnosed during 2009–2016 among cohort women remained. By histologic type, these included 3,549 (92.5%) squamous cell neoplasias, 133 (3.5%) adenocarcinomas (invasive or in situ), 132 (3.4%) carcinomas (invasive or in situ) not otherwise specified, and 24 (0.6%) other specified carcinomas. Overall, 3,750 (97.7%) were precancers and 88 (2.3%) were invasive cancers.

Of the 3,838 CIN3+ cases arising from the cohort, 2,900 (76%) occurred among continuous Michigan residents; these included cases among 9 women who were sampled into the subcohort. The subcohort included 1,383 continuous Michigan residents (1,374 noncases and 9 cases). Therefore, the case–cohort analysis included 2,900 cases and 1,374 noncases.

Characteristics of the cohort and case–cohort analytic populations are shown in Table 1. The cohort was not evenly distributed among birth years; about 1 in 5 cohort women were born in 1980–1984, whereas nearly half were born in 1990–1995. By race, most women (78.7%) were white, 19.6% were black, and 1.7% were other race. The majority (72.4%) of the cohort was unvaccinated, and 17.6% were UTD on vaccination; of these, most (97%) had 3 doses and 3% had 2 doses initiated prior to 15th birthday with appropriate spacing. A total of 10% of the cohort had initiated vaccination but were not UTD, including 4.4% with 2 doses and 5.6% with 1 dose (Table 1). Among 213,404 women who initiated vaccination, most (80.2%) did so before age 20 years. However, no women born 1980–1984 initiated vaccination before age 20, whereas 38.8% of vaccinated women born 1985–1989 and 94.2% of those born 1990–1995 did. The median ages at vaccination also differed by birth year group: 25 years [range, 22–27, interquartile range (IQR) 24–26] for women born 1980–1984, 20 (range, 17–28; IQR 19–23) for women born 1985–1989, and 16 (range, 11–25; IQR 14–17) for women born 1990–1995. By number of doses, the median age at vaccination for UTD was 16 (range, 11–27; IQR 14–18), for 2 doses-not UTD was 18 (range, 11–28; IQR 16–20), and for 1 dose was 18 (range, 11–27; IQR 17–21). By cohort, the median age at vaccination initiation was lower for UTD than 1-dose or 2-dose not UTD groups for the 1990–1995 cohort [15 (range, 11–24; IQR 14–17) vs. 17 (range, 11–25; IQR 16–18) years] and the 1985–1989 cohort [20 (range, 19–23; IQR 17–27) vs. 21 (range, 17–28; IQR 19–23) years].

For the cohort, most characteristics evaluated differed between cases and noncases (Table 1). Cases were more likely than noncases to be born in earlier years. The median age at CIN3+ diagnosis differed by birth year group: 29 (range, 24–35, IQR 27–31) years for women born 1980–1984, 24 (range, 19–31; IQR 23–26) years for 1985–1989, and 22 (range, 15–26; IQR 20–23) years for 1990–1995. Cases and noncases also varied by race. A higher proportion of cases than noncases was unvaccinated, and among vaccinated, a higher proportion of cases than noncases initiated vaccination at age ≥20 years. For the case–cohort population, similar associations between characteristics and case status were evident, although the association with race was not statistically significant. The distributions of characteristics were similar between all cases and continuously resident cases, as well as between cohort noncases and subcohort noncases (P > 0.1 for all comparisons).

Vaccination status varied by birth year group and race (Table 2). Among cohort women, only 2.1% of those born in 1980–1984 were UTD, 2.2% initiated vaccination but were not UTD, and 95.8% were unvaccinated; 30.4% of women born 1990–1995 were UTD, 14.6% initiated but were not UTD, and 55% were unvaccinated. By race, women who were white or other races had higher proportions with UTD vaccination, but black women had higher proportions who initiated vaccination but were not UTD. Associations were similar for the case–cohort sample.

Table 3 shows VE results from the cohort analyses. The adjusted relative risk of CIN3+ for vaccination with ≥1 dose was 0.46 (95% CI, 0.41–0.52), corresponding to a VE estimate of 54%. VE was higher for UTD (66%) than not UTD (37%), but significant for both. Among women who were not UTD, VE was significant for 2 doses (33%) and 1 dose (40%). Stratified by birth year group, significant VE was observed for women born 1980–1984 only if UTD (39%). Significant VE was observed for women born 1985–1989 and 1990–1995 for both UTD and not UTD, but VE point estimates were consistently higher for UTD than for not UTD.

Compared with the cohort analysis, the case–cohort analysis yielded higher VE estimates with wider confidence intervals (Table 4). The aRR of CIN3+ for vaccination with ≥1 dose was 0.39 (95% CI, 0.32–0.48), corresponding to a VE estimate of 61%. VE was higher for UTD (72%) than not UTD (44%), but significant for both. Among women not UTD, VE was significant for 2 doses (39%) and 1 dose (48%). Stratified by birth year group, no significant VE was observed for women born 1980–1984; however, few women in this group were vaccinated. Significant VE was observed for women born 1985–1989 and 1990–1995 for both UTD and not UTD, but VE point estimates were higher for UTD. Results for the cohort and case–cohort analyses were compared graphically, highlighting the lower VE estimates with narrower CIs for the cohort analysis (Fig. 2).

In this analysis of linked statewide registries in Michigan, we found significant effectiveness of HPV vaccination against CIN3+, the outcome most proximal to invasive cervical cancer. To our knowledge, this is the first U.S. study to report significant VE against CIN3+ after 1 or 2 doses and only the second worldwide to do so (21–26). The two analytic designs used in this report concurred on the direction and statistical significance of findings, although the magnitude and precision differed.

We found that both incomplete vaccination with 1 and 2 doses and UTD vaccination with 2 or 3 doses of HPV vaccine were effective against CIN3+, although effectiveness was lower for those who did not complete the recommended series. A handful of prior studies have included evaluation of VE against the highest grades of precancer (21–26); only one found significant protection against CIN3/AIS with <3 doses (26). That study was conducted in Denmark using a data linkage cohort design, included 5,258 CIN3+ (303 vaccinated), and focused on women who initiated vaccination at age ≤16 years. The study found similar VE for 1, 2, and 3 doses (62%–63%). No other prior studies have reported significant VE against CIN3+ with <3 doses; the other studies that have evaluated this outcome were smaller, and predominantly included women who were vaccinated at older ages as part of catch-up vaccination programs, as our study did (21–25).

Most prior VE studies have included the lower-grade CIN2 lesions, which are more numerous than CIN3/AIS and diagnosed at younger ages, on average (21–23, 26–32). Some recent studies reported similar effectiveness by number of doses against CIN2+ (26, 27, 30, 32). The accumulating body of literature underlines the importance of age at vaccination; studies that have focused on or stratified analyses to include women vaccinated at younger ages have found higher VE and/or more similar VE by number of doses (26, 30–32, 33), whereas other studies that included predominately catch-up vaccinees have found greater differences and little to no effectiveness with <3 doses (21–23, 25, 29, 34). Although our study did find higher VE among women vaccinated at younger versus older ages (<20 vs. ≥20 years), we were unable to perform an age-restricted analysis. In addition, ideally, we would have evaluated VE with a younger age at vaccination cutpoint such as ≤16 years, but too few women were vaccinated in this age range to perform a robust analysis. An unexpected finding was similar VE estimates for women in the 1985–1989 and 1990–1994 cohorts, which had different median ages at vaccination (20 vs. 16 years). We note that ≥1 dose HPV vaccination coverage increased between the 1985–1989 and 1990–1995 cohorts, from 17% to 45%. Herd effects could occur with higher coverage; whether this may have affected VE estimates for the 1990–1995 cohort is unknown.

The overall VE estimates from this study (61% using case–cohort methodology and 54% using cohort analysis) were similar to those observed in Verdoodt and colleagues (26), and plausible but high, given the prevalence of HPV16/18 in CIN3/AIS lesions observed in U.S. women in the prevaccine or early vaccine eras (range, 57%–63%; refs. 35–38). The observed VE was even higher in some of our analyses, especially in women who were UTD for all recommended doses, whose age at first dose was <20 years, or who were UTD and born in 1990–1995. Characteristics of the study population associated with higher prevalence of HPV16/18 among CIN3+ cases might explain these higher VE estimates; these include younger age, high proportion of white non-Hispanic women, and inclusion of AIS and invasive cancers (35, 38–40).

The design of this study, using linked statewide registries, is unique for the United States thus far, and more typical of studies conducted in countries that have robust national data with linkable personal identifiers. We found a similar pattern of results in both the cohort analysis and the analysis of continuous Michigan residents. Evaluating continuous residency for the entire cohort would have been cost-prohibitive; the case–cohort design is one strategy for improving exposure ascertainment using fewer resources while retaining the clear time order and reduced information biases of a cohort study. The cohort design yielded more precise VE estimates (i.e., narrower CIs), because the sample size was larger; the lower VE estimates with this method could result from misclassification of exposures and outcomes, since women with noncontinuous Michigan residence could have been vaccinated or diagnosed with CIN3+ in another state. The results of the cohort analysis were closer to null, but the inferences were similar to the more laborious case–cohort approach, in which continuous residency determination strengthened our confidence in vaccination history and CIN3+ status. Similar registry linkage studies may be feasible in other U.S. states. Determination of continuous residency on a subcohort would be preferred for limiting misclassification; however, in states with levels of outmigration similar to or less than Michigan, naïve cohort linkage approaches without using case–cohort methodology might provide useful information, particularly if misclassification bias can be quantified.

Studies including cervical precancers diagnosed during the HPV vaccination era should be interpreted in the context of cervical precancer screening and management recommendations. From 2009 to 2012, several medical organizations changed recommendations to start screening at age 21 years (41–43) rather than at 3 years after onset of sexual activity or age 21 years, whichever was earlier (44). In addition, in 2012, there were recommendations from all major medical organizations to extend the screening intervals from every year to every 3 years, if using cytology alone, or 3 to 5 years, if using cytology + HPV co-tests for women aged ≥30 years. Revised management guidelines were less aggressive for following women aged 21 to 24 years with low-grade cytology results (45). Because cervical precancers are only detected by screening, studies of population impact of the HPV vaccination program on precancers typically use screened women as the denominator when calculating incidence rates to avoid attributing declines due to less frequent screening to the vaccination program. The influence of changes in screening on VE estimates, on the other hand, is less clear. Some studies have shown that HPV vaccination and participation in cervical cancer screening are positively associated (46, 47); therefore, unvaccinated women may have fewer opportunities to be diagnosed with precancers, which could make vaccines appear less effective. Some VE studies have restricted analyses to screened women, or evaluated screening rates among vaccinated and unvaccinated women, to reduce or evaluate potential ascertainment biases; our study did not have access to screening data. We assume all precancers included in this study were screen-detected, but it is possible that differential adherence to screening guidelines by vaccination status introduced some biases.

Additional limitations should also be considered. Our study, like most previous post-licensure effectiveness studies, was conducted in the setting of a national 3-dose recommendation. As such, girls who received <3 doses were often different from those completing the schedule, which could bias results toward lower VE with <3 doses if girls who received less than a 3-dose schedule had a higher risk of infection at the time of vaccination due to age at vaccination (as shown in our data) or risk during follow-up (13). The statewide data systems included no data on sexual behavior such as number of partners or age at sexual debut; therefore, we could not evaluate this potential bias. We found that the unvaccinated group disproportionately represented the earlier birth cohorts, who also had higher CIN3+ incidence due to higher attained age during the study period. Although we controlled for birth year in analyses that included the birth year groups combined, residual confounding could remain. Because earlier and later cohorts differed strongly in their CIN3+ risk and likelihood of vaccination at the recommended ages, we conducted cohort-stratified analyses; in the two more recent cohorts, there was also significant VE among women not UTD. Another limitation is that there were too few UTD 2-dose vaccinees to evaluate the current 2-dose recommendation directly (9 in subcohort), so these were included in the UTD group, which mainly comprised 3-dose vaccinees. Although the immunization registry has incorporated adult vaccine doses since 2006, immunization data may be incomplete. Inputting doses for people ages ≥20 years was optional; some women who received doses in their 20s could have been misclassified as unvaccinated, and doses received at younger ages could also be missing. Additionally, the linkage was not 100% complete and varied by birth year. Systematic differences between linked and unlinked women could have biased results in either direction and could not be determined. Finally, although our study sample included some women with invasive cervical cancer, there were too few to analyze separately; we cannot conclude significant VE against invasive cancer from this study, as was recently shown in Sweden (48).

In conclusion, we report HPV VE against CIN3+ using state cancer and immunization registries in Michigan. Vaccination at younger ages or being part of a more recent cohort that was vaccinated at younger ages increased this protective effect. Maximum protection was associated with UTD vaccination. Future evaluations by number of doses for women vaccinated at younger ages, including at ages 11 to 12 years as recommended, may further clarify dose-related effectiveness.

M. You reports grants from CDC Immunization Grant and National Program of Cancer Registries during the conduct of the study. R. Potter reports grants from Centers for Disease Control and Prevention during the conduct of the study. No disclosures were reported by the other authors.

The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the Centers for Disease Control and Prevention.

J.W. Gargano: Investigation, visualization, methodology, writing–original draft, writing–review and editing. M. You: Software, formal analysis, validation, investigation, visualization, methodology, writing–review and editing. R. Potter: Conceptualization, resources, software, validation, investigation, writing–original draft, project administration, writing–review and editing. G. Alverson: Resources, data curation, supervision, funding acquisition, investigation, writing–review and editing. R. Swanson: Resources, data curation, supervision, funding acquisition, writing–review and editing. M. Saraiya: Resources, supervision, writing–original draft, project administration, writing–review and editing. L.E. Markowitz: Conceptualization, supervision, methodology, writing–review and editing. G. Copeland: Conceptualization, resources, supervision, funding acquisition, investigation, writing–original draft, project administration, writing–review and editing.

The authors thank Jean Shapiro and Paran Pordell, Division of Cancer Prevention and Control, National Center for Chronic Disease Prevention and Health Promotion, Centers for Disease Control and Prevention, for leadership of the CIN 3 project, which coordinates data collection on cervical precancers in Michigan and selected other states and territories. This study was supported by immunization grant funds under section 317 of the Public Health Service Act (42 USC Sect. 247b), Reference Number H23IP000854-01, to M. You, and National Program of Cancer Registries grant awards 5U58DP000812 and 1U58DP003921, to M. You.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.