Expanding HPV Vaccination to Include School-aged Boys
Expanding HPV Vaccination to Include School-aged Boys
The study methods followed the Burden of Disease Epidemiology, Equity and Cost-Effectiveness Programme (BODE) Protocol. We adapted a previous Markov model on the cost-utility of girls-only HPV vaccination, to estimate quality-adjusted life years (QALYs) gained and net health system costs, for girls-only and girls and boys vaccination. The QALY metric captures both years of life lost from premature death, and loss of quality of life through morbidity. To value this health loss, we used disability weights (DWs), as outlined below. We formally use the term QALY in this paper's Methods and Results, but shorten it to QALY in other sections.
The two main adaptations were adding vaccine and administrative costs per vaccinated boy, and additional reductions in HPV infection due to additionally vaccinating boys.
Health system costs included both direct intervention costs and downstream health system costs incurred/averted. A 3% discount rate applied to costs and quality-adjusted life years (QALYs) gained. Currency conversions were conducted using the OECD purchasing power parity (PPP) estimates, to eliminate the differences in price levels between countries. In 2011, the PPP values were US$1 = NZ$1.48.
HPV vaccination was modeled as preventing CIN I-III, cervical, anal, oropharyngeal and vulval cancers and anogenital warts (Figure 1) using rates of all-cause mortality, excess mortality rates from cancer, and incidence rates for cancer and morbidity states. Vaginal and penile cancers were excluded due to their small contribution to the HPV16/18-related cancer burden (<3%). Recurrent respiratory papillomatosis (RRP) was excluded due to sparse data. Cervical cancer screening programs were assumed to continue in New Zealand, but these costs were excluded from our analyses.
(Enlarge Image)
Figure 1.
Stylized Markov model for HPV-related disease states. r1 = rates of all-cause mortality from population lifetables, by sex, age, ethnicity (Māori, non-Māori) and area deprivation (approximate tertiles), and projected to future. Source: [22]. r2i = excess mortality rates of death from cancer i, by sex, age, ethnicity and deprivation, and by time since diagnosis. Source: [23] r3i = incidence rates for cancer i, by sex, age, ethnicity and deprivation. Source: [24] r4j = incidence rates for morbidity states j, by sex and age (and ethnicity for CIN I, CIN II/III and anogenital warts). Source: [20].
Input parameters are shown in Table 1 (HPV prevalence reduction and vaccination costs) and Additional file 1: Tables S1 and S2 http://www.biomedcentral.com/1471-2334/14/351/additional, and briefly described below.
Disability weights (DWs) for each disease state were used, with uncertainty, on a scale from 0 (full health) to 1.0 (death) (Additional file 1: Table S1 http://www.biomedcentral.com/1471-2334/14/351/additional). Expected population morbidity (i.e., due to disease other than HPV-related disease) was allowed for by using the average ethnic and age-specific prevalent years of life lived in disability from the New Zealand Burden of Disease Study, limiting the maximum QALYs gained with increasing age. Consider a non-Māori male aged 50–54 with oropharyngeal cancer in the remission phase (expected DW = 0.248, although actually modeled with uncertainty), with an expected background morbidity of 0.112; his expected QALYs awarded in that year would be (1–0.248) × (1–0.112) = 0.668.
Cancer incidence rates were those predicted for 2011 with trends projected into the future based on regressions on New Zealand Cancer Registry data. Therefore, when the model was run without intervention effects, the model produced the same disease data as the input dataset. Incidence rates for other diseases and the proportion of disease due to HPV 16/18 or 6/11 were compiled from various sources, including national cancer registration data and screening program information. Australian burden of disease models were used to allocate durations for each cancer in diagnosis and treatment, remission, pre-terminal and terminal states, with attendant DWs sourced from the Global Burden of Disease 2010 with modification to the New Zealand distribution of cancers. Cancer survivors were modeled in the cancer state for five cycles accumulating QALYs with morbidity adjustment via the DWs, then returned to the healthy state.
Just as QALYs were awarded to each individual as they traveled through states, so were health system costs. We used routine, linked administrative health data for the entire New Zealand population with costs attached, as described elsewhere. We assigned health system costs by sex and age to the healthy state (see Additional file 1: Table S2 http://www.biomedcentral.com/1471-2334/14/351/additional). The added cost for cancer patients at different stages of care (i.e., diagnosis, remission, terminal) were estimated using gamma regression, as per the previous research in the BODE Program for girls-only vaccination. The added cost for other disease states (CIN I-III and warts) were simple averages. All health system costs were measured in 2011 New Zealand dollars.
Effectiveness: Vaccination Coverage and Future Reduction in HPV Prevalence. The vaccination coverage levels were assumed to be the same for both boys and girls for two interventions: (1G and 1G + B) the current girls' 3-dose coverage level in 12–13 year olds in New Zealand (45–56%); and (2G and 2G + B) Australia's coverage level of 12–13 year old girls (73%), with no catch-up vaccination modeled. We suspect that similar coverage is a reasonable assumption for school-based vaccination programs, provided that sufficient information is communicated to parents and providers about the direct benefits to males of HPV vaccination. We note that a recent review found a preference to vaccinate females over males among both health care providers and parents, which could lead to lower coverage for boys than girls. However, the review authors noted that in many studies the direct benefits to males were not communicated and the main reason given for refusal was the lack of perceived direct benefit for males.
The reduction in HPV infections for varying programs of girls-only and girls and boys vaccination was estimated by meta-regressions on outputs from Brisson et al's (2011) dynamic Canadian model, which allowed for changing likelihood of acquiring HPV over time and herd immunity effects. This model used gender and age-specific sexual behavior characteristics (e.g., partner acquisition rates, mixing between age groups) as risk factors for HPV infection. Thus, our Markov macro-simulation models assumed similar sexual behaviors to that in the Canadian model. We considered this generally reasonable on the basis of available comparable data on age of sexual debut and the epidemiology of genital warts. While data on adolescent sexual behavior are sparse, age of sexual debut are similar between Canada and New Zealand. Likewise, incidence of genital warts peak in the same age groups (<25 years old) in both countries.
Briefly, we fitted regression models to their output for the median, 10 and 90 percentile reductions in HPV prevalence with vaccine coverage as an independent variable for two types of HPV (6/11 and 16/18) for girls-only vaccination, doubled the uncertainty range on the logistic scale to account for 'structural' uncertainty when using the results in New Zealand, and then used these median and widened uncertainty intervals to generate Beta distribution parameters to sample in the model (method details described in an Appendix to the parallel study on girls-only vaccination). For the marginal impact of boys, we assumed a correlation of 0.5 between the HPV prevalence impacts of girls-only vaccination and the marginal impact of adding boys, and calibrated Beta distributions to achieve the uncertainty reported in for the marginal impact of adding boys' vaccination. Examples of future long-term HPV reduction for specified vaccine coverage levels are shown in Table 1. We assumed that the vaccine efficacy was 99% and had a duration of 20 years, that the effect of HPV vaccination was in a 'steady state' and that vaccine immunogenicity was the same in both boys and girls.
The vaccination costs were calculated per fully vaccinated girl/boy, based on the annual vaccine cost paid by the Ministry of Health (in 2011), resulting in the vaccine cost per dose of NZ$113 (Table 1). The delivery/administration costs of three main interventions were NZ$141 or NZ$126 per dose depending on the method of delivery (i.e., in schools and primary care settings or in schools only). These costing data were based on official Ministry of Health data which include funding to cover program management (a component which is often omitted from 'administration costs' of other studies). The vaccination costs were multiplied by three doses, and then assigned for each fully vaccinated member of the cohort (allowing for vaccine coverage); incomplete vaccine courses were not considered. We used a discount rate of 3% and as for health system costs, all intervention costs were measured in 2011 New Zealand dollars.
In scenario analyses, we explored reductions in vaccine cost-per-dose at several levels of reduction down to $1 and lower administrative costs (NZ$19 per dose). We also conducted a scenario analysis which removed the herd immunity cancer reduction benefits for males in the girls-only vaccination program (1G). We compared these results with the benefits of adding boys to the vaccination program (1G + B) and included MSM-attributable warts and cancers in the disease incidence data that populates the model. We also performed a threshold analysis to estimate the maximum cost per delivered vaccine dose (including vaccine and administration costs) which would allow vaccination of boys in addition to girls to be cost-effective compared to girls only vaccination.
Monte Carlo simulation was used, with 2,000 draws from input parameter distributions. Incremental QALYs, costs and incremental cost-effectiveness ratios (ICERs) were calculated for each draw of Monte Carlo simulation. All modeling and uncertainty analyses were undertaken in TreeAge Pro 2012. ICERs were calculated for each intervention including boys (1G + B and 2G + B), compared to the equivalent intervention for girls-only (1G and 2G) and compared to no vaccination program. In line with contextualizing ICER results with GDP per capita comparisons, we made such comparisons for the New Zealand setting. Because there is no universally accepted threshold in New Zealand for describing interventions as being "cost-effective" or not, we relied on the WHO definition and used a nominal GDP per capita of NZ$45,000 in 2011 (US$29,600) as being such a threshold. Using the GDP per capita level is based on the WHO description of interventions below this value as being 'very cost-effective' (for the WPROA Region to which New Zealand belongs).
Methods
Model Overview
The study methods followed the Burden of Disease Epidemiology, Equity and Cost-Effectiveness Programme (BODE) Protocol. We adapted a previous Markov model on the cost-utility of girls-only HPV vaccination, to estimate quality-adjusted life years (QALYs) gained and net health system costs, for girls-only and girls and boys vaccination. The QALY metric captures both years of life lost from premature death, and loss of quality of life through morbidity. To value this health loss, we used disability weights (DWs), as outlined below. We formally use the term QALY in this paper's Methods and Results, but shorten it to QALY in other sections.
The two main adaptations were adding vaccine and administrative costs per vaccinated boy, and additional reductions in HPV infection due to additionally vaccinating boys.
Health system costs included both direct intervention costs and downstream health system costs incurred/averted. A 3% discount rate applied to costs and quality-adjusted life years (QALYs) gained. Currency conversions were conducted using the OECD purchasing power parity (PPP) estimates, to eliminate the differences in price levels between countries. In 2011, the PPP values were US$1 = NZ$1.48.
HPV vaccination was modeled as preventing CIN I-III, cervical, anal, oropharyngeal and vulval cancers and anogenital warts (Figure 1) using rates of all-cause mortality, excess mortality rates from cancer, and incidence rates for cancer and morbidity states. Vaginal and penile cancers were excluded due to their small contribution to the HPV16/18-related cancer burden (<3%). Recurrent respiratory papillomatosis (RRP) was excluded due to sparse data. Cervical cancer screening programs were assumed to continue in New Zealand, but these costs were excluded from our analyses.
(Enlarge Image)
Figure 1.
Stylized Markov model for HPV-related disease states. r1 = rates of all-cause mortality from population lifetables, by sex, age, ethnicity (Māori, non-Māori) and area deprivation (approximate tertiles), and projected to future. Source: [22]. r2i = excess mortality rates of death from cancer i, by sex, age, ethnicity and deprivation, and by time since diagnosis. Source: [23] r3i = incidence rates for cancer i, by sex, age, ethnicity and deprivation. Source: [24] r4j = incidence rates for morbidity states j, by sex and age (and ethnicity for CIN I, CIN II/III and anogenital warts). Source: [20].
Model Input Parameters
Input parameters are shown in Table 1 (HPV prevalence reduction and vaccination costs) and Additional file 1: Tables S1 and S2 http://www.biomedcentral.com/1471-2334/14/351/additional, and briefly described below.
Morbidity
Disability weights (DWs) for each disease state were used, with uncertainty, on a scale from 0 (full health) to 1.0 (death) (Additional file 1: Table S1 http://www.biomedcentral.com/1471-2334/14/351/additional). Expected population morbidity (i.e., due to disease other than HPV-related disease) was allowed for by using the average ethnic and age-specific prevalent years of life lived in disability from the New Zealand Burden of Disease Study, limiting the maximum QALYs gained with increasing age. Consider a non-Māori male aged 50–54 with oropharyngeal cancer in the remission phase (expected DW = 0.248, although actually modeled with uncertainty), with an expected background morbidity of 0.112; his expected QALYs awarded in that year would be (1–0.248) × (1–0.112) = 0.668.
Incidence and Survival
Cancer incidence rates were those predicted for 2011 with trends projected into the future based on regressions on New Zealand Cancer Registry data. Therefore, when the model was run without intervention effects, the model produced the same disease data as the input dataset. Incidence rates for other diseases and the proportion of disease due to HPV 16/18 or 6/11 were compiled from various sources, including national cancer registration data and screening program information. Australian burden of disease models were used to allocate durations for each cancer in diagnosis and treatment, remission, pre-terminal and terminal states, with attendant DWs sourced from the Global Burden of Disease 2010 with modification to the New Zealand distribution of cancers. Cancer survivors were modeled in the cancer state for five cycles accumulating QALYs with morbidity adjustment via the DWs, then returned to the healthy state.
Health System Cost Parameters
Just as QALYs were awarded to each individual as they traveled through states, so were health system costs. We used routine, linked administrative health data for the entire New Zealand population with costs attached, as described elsewhere. We assigned health system costs by sex and age to the healthy state (see Additional file 1: Table S2 http://www.biomedcentral.com/1471-2334/14/351/additional). The added cost for cancer patients at different stages of care (i.e., diagnosis, remission, terminal) were estimated using gamma regression, as per the previous research in the BODE Program for girls-only vaccination. The added cost for other disease states (CIN I-III and warts) were simple averages. All health system costs were measured in 2011 New Zealand dollars.
Interventions
Effectiveness: Vaccination Coverage and Future Reduction in HPV Prevalence. The vaccination coverage levels were assumed to be the same for both boys and girls for two interventions: (1G and 1G + B) the current girls' 3-dose coverage level in 12–13 year olds in New Zealand (45–56%); and (2G and 2G + B) Australia's coverage level of 12–13 year old girls (73%), with no catch-up vaccination modeled. We suspect that similar coverage is a reasonable assumption for school-based vaccination programs, provided that sufficient information is communicated to parents and providers about the direct benefits to males of HPV vaccination. We note that a recent review found a preference to vaccinate females over males among both health care providers and parents, which could lead to lower coverage for boys than girls. However, the review authors noted that in many studies the direct benefits to males were not communicated and the main reason given for refusal was the lack of perceived direct benefit for males.
The reduction in HPV infections for varying programs of girls-only and girls and boys vaccination was estimated by meta-regressions on outputs from Brisson et al's (2011) dynamic Canadian model, which allowed for changing likelihood of acquiring HPV over time and herd immunity effects. This model used gender and age-specific sexual behavior characteristics (e.g., partner acquisition rates, mixing between age groups) as risk factors for HPV infection. Thus, our Markov macro-simulation models assumed similar sexual behaviors to that in the Canadian model. We considered this generally reasonable on the basis of available comparable data on age of sexual debut and the epidemiology of genital warts. While data on adolescent sexual behavior are sparse, age of sexual debut are similar between Canada and New Zealand. Likewise, incidence of genital warts peak in the same age groups (<25 years old) in both countries.
Briefly, we fitted regression models to their output for the median, 10 and 90 percentile reductions in HPV prevalence with vaccine coverage as an independent variable for two types of HPV (6/11 and 16/18) for girls-only vaccination, doubled the uncertainty range on the logistic scale to account for 'structural' uncertainty when using the results in New Zealand, and then used these median and widened uncertainty intervals to generate Beta distribution parameters to sample in the model (method details described in an Appendix to the parallel study on girls-only vaccination). For the marginal impact of boys, we assumed a correlation of 0.5 between the HPV prevalence impacts of girls-only vaccination and the marginal impact of adding boys, and calibrated Beta distributions to achieve the uncertainty reported in for the marginal impact of adding boys' vaccination. Examples of future long-term HPV reduction for specified vaccine coverage levels are shown in Table 1. We assumed that the vaccine efficacy was 99% and had a duration of 20 years, that the effect of HPV vaccination was in a 'steady state' and that vaccine immunogenicity was the same in both boys and girls.
Intervention Costs
The vaccination costs were calculated per fully vaccinated girl/boy, based on the annual vaccine cost paid by the Ministry of Health (in 2011), resulting in the vaccine cost per dose of NZ$113 (Table 1). The delivery/administration costs of three main interventions were NZ$141 or NZ$126 per dose depending on the method of delivery (i.e., in schools and primary care settings or in schools only). These costing data were based on official Ministry of Health data which include funding to cover program management (a component which is often omitted from 'administration costs' of other studies). The vaccination costs were multiplied by three doses, and then assigned for each fully vaccinated member of the cohort (allowing for vaccine coverage); incomplete vaccine courses were not considered. We used a discount rate of 3% and as for health system costs, all intervention costs were measured in 2011 New Zealand dollars.
In scenario analyses, we explored reductions in vaccine cost-per-dose at several levels of reduction down to $1 and lower administrative costs (NZ$19 per dose). We also conducted a scenario analysis which removed the herd immunity cancer reduction benefits for males in the girls-only vaccination program (1G). We compared these results with the benefits of adding boys to the vaccination program (1G + B) and included MSM-attributable warts and cancers in the disease incidence data that populates the model. We also performed a threshold analysis to estimate the maximum cost per delivered vaccine dose (including vaccine and administration costs) which would allow vaccination of boys in addition to girls to be cost-effective compared to girls only vaccination.
Considering Uncertainty and Performing Cost-effectiveness Analysis
Monte Carlo simulation was used, with 2,000 draws from input parameter distributions. Incremental QALYs, costs and incremental cost-effectiveness ratios (ICERs) were calculated for each draw of Monte Carlo simulation. All modeling and uncertainty analyses were undertaken in TreeAge Pro 2012. ICERs were calculated for each intervention including boys (1G + B and 2G + B), compared to the equivalent intervention for girls-only (1G and 2G) and compared to no vaccination program. In line with contextualizing ICER results with GDP per capita comparisons, we made such comparisons for the New Zealand setting. Because there is no universally accepted threshold in New Zealand for describing interventions as being "cost-effective" or not, we relied on the WHO definition and used a nominal GDP per capita of NZ$45,000 in 2011 (US$29,600) as being such a threshold. Using the GDP per capita level is based on the WHO description of interventions below this value as being 'very cost-effective' (for the WPROA Region to which New Zealand belongs).
Source...