## Main

More than 25% of cervical cancer cases worldwide occur in India (Ferlay et al, 2004). Although cervical cancer is the most frequent cancer diagnosis in Indian women, age-adjusted incidence rates vary between regions, ranging from 10.9 in Trivandrum to 47.2 per 100 000 person-years at risk in Chennai (Parkin et al, 2005).

Persistent infection with high risk types of human papillomavirus (HPV) is established as the key causal agent for cervical cancer (IARC, 2005). In India, overall HPV prevalence in women with normal cytology has been estimated to be 7.5% (de Sanjosé et al, 2007). Similar to cancer incidence, region-specific prevalence of HPV varies considerably (Sankaranarayanan et al, 2004; Franceschi et al, 2005; Laikangbam et al, 2007), likely attributable to genetic and cultural diversity, as well as heterogeneity between studies (Basu et al, 2003; Duttagupta et al, 2004).

New options for cervical cancer prevention motivate important questions in low-income countries. Traditional cytology screening, conducted at frequent intervals and requiring multiple clinic visits for women with abnormal cytology (e.g., screening, diagnostic testing, and treatment), has been difficult to implement in India (Sankaranarayanan et al, 2001). Screening with HPV DNA testing and/or visual inspection with acetic acid (VIA) have been demonstrated to be acceptable and promising alternatives when embedded in a screening protocol that requires fewer visits (Bhatla et al, 2007; Ramachandran et al, 2007) and utilises cryotherapy conducted by nurses to treat the majority of early cancer precursors (Sankaranarayanan et al, 2001, 2006, 2007a, 2007b). These strategies, less dependent on existing health system infrastructure and associated with greater rates of follow-up, are also expected to be cost-effective (Goldie et al, 2005).

Most recently, two new vaccines have been shown to be highly effective in preventing infection with HPV 16 and 18 in women without prior exposure to these types (FDA, 2006). Contemplating an HPV vaccination programme adds complexity to decision-making about India's national approach to cervical cancer control. The country will want to consider the burden of cervical cancer, the comparative effectiveness – and potential synergies – of vaccination and screening, the financial costs required to initiate and sustain programs, their cost-effectiveness, and the programmatic capacity and infrastructure necessary to effectively deliver a three-dose pre-adolescent vaccine. Although a single study would be unable to include all of these factors, decision-analytic models can be used to synthesise the best available epidemiologic, clinical and economic data, and project long-term health and economic outcomes expected with different cancer prevention strategies. To provide insight to decision makers and stakeholders invested in reducing mortality from cervical cancer in India, we assessed the potential avertable burden of disease and cost-effectiveness associated with various vaccination and screening strategies.

## Materials and methods

### Sensitivity analysis

Assuming a cost of I$20 per vaccinated girl, if VIA screening was 30% less sensitive and costs were increased to I$2, vaccination alone dominated VIA screening, although VIA combined with vaccination approximated I$975 per YLS (see Supplementary Appendix). The choice between specific screening modalities was sensitive to assumptions about their relative costs, test performance, and ability to be delivered within one or two visits. For example, if training costs associated with VIA raised the total screening cost such that it would approach that of HPV DNA testing, vaccination plus screening three times per lifetime with two-visit HPV DNA testing was favored, at I$1780 per YLS – if a rapid HPV DNA test was available, allowing for a single-visit screen-and-treat strategy, this result would become even more favourable. All strategies requiring multiple visits became more attractive when loss to follow-up was reduced.

We have previously reported how the comparative performance of different cervical cancer prevention strategies depends on several factors (Goldie et al, 2005, 2007; Garnett et al, 2006; Kim et al, 2007a). For screening, as noted above, these include test performance, cost, and loss to follow-up. For vaccination, these include vaccine efficacy, coverage, and duration of protection. Figure 3 shows how different assumptions about vaccination and screening coverage will influence the level of cancer reduction achievable with a strategy of pre-adolescent vaccination, assuming 100% efficacy, combined with screening three times per lifetime using two-visit HPV DNA testing (shown by the bars). Similarly, Figure 3 shows how vaccine efficacy and coverage will influence the level of cancer reduction with a strategy of vaccination alone (shown by the lines). A combined strategy of vaccination and screening always provided greater cancer reduction than vaccination alone; however, the incremental benefits achieved by adding screening three times per lifetime were greater at lower vaccination coverage rates.

We can also obtain information about the coverage levels of vaccination and screening required to achieve a specific threshold reduction in cervical cancer from Figure 3. For example, without screening, a threshold of 50% cancer reduction (dashed red line) was expected to be attainable with vaccination alone only at vaccination coverage rates exceeding 80% and vaccine efficacy of 100%. When screening and vaccination were combined, however, several strategies would potentially achieve this threshold. At vaccination coverage rates of 40–50%, screening coverage would need to exceed 80%, but at vaccination coverage of 70%, screening coverage of 40–50% would suffice. Assuming a cost per vaccinated girl of I$10, the vast majority of these combination strategies had cost-effectiveness ratios less than the per capita GDP when compared to the next best strategy. ## Discussion The vaccine-preventable cervical cancer burden in India is a product of several factors, including the underlying cervical cancer incidence, the proportion of cancer attributable to HPV 16 and 18, the long-term vaccine efficacy, and the ability to achieve high coverage in adolescent girls before sexual activity. Our results showed that with 70% coverage, the expected mean reduction in the lifetime risk of cervical cancer with pre-adolescent vaccination alone was 44%. The effectiveness of the vaccine could be lower than we projected if older girls, who may have been infected previously with type 16 or 18, are vaccinated; if there is less robust vaccine-induced immunity in girls with other diseases, such as severe anaemia, chronic illness or HIV; or if vaccine-induced immunity wanes while individuals are still at risk for new HPV infections. On the other hand, the impact of the vaccine could be higher if there is long-term cross-protection against non-16,18 type infection; if there are herd immunity benefits to unvaccinated individuals; and if non-cervical HPV 16,18-related cancers and diseases are prevented. For most of these factors, a lack of data prohibit us from estimating the magnitude of individual potential effects; that being said, these are uncertain issues for which empiric data are needed. We found that a combined approach of pre-adolescent vaccination and screening three times per lifetime after age 30, both at 70% coverage, provided a mean cancer reduction of 56–63%, depending on the specific screening strategy. Because the mechanisms of effectiveness for vaccination and screening differ, with vaccination preventing infections with HPV types 16 and 18, and screening allowing for treatment of precancerous lesions (caused by any high-risk HPV type) before progression to invasive cancer, they are synergistic, and provide substantially greater benefits than either alone. Although screening three time per lifetime prevents additional deaths from non-16 and 18-type associated cancers, as well as 16,18-associated cancers in the proportion of the population not vaccinated, it also provides some insurance of cancer risk protection in the context of the uncertainty around long-term vaccine performance in a non-clinical trial situation. For example, across the good-fitting parameter sets, the mean cancer reduction with vaccination alone ranged from 28 to 57%, whereas with combined vaccination and screening three times per lifetime using two-visit HPV DNA testing ranged from 53 to 76%. The cost-effectiveness of HPV vaccination will depend mostly on the incremental programmatic costs associated with adding a pre-adolescent vaccine to India's vaccination programme, and also on the ultimate negotiated vaccine price for India. We found that provided the cost per vaccinated girl was I$10 or less (per dose cost of approximately $2), vaccination alone was more effective and cost-effective than screening alone. A combined approach of pre-adolescent vaccination and screening three times per lifetime (at ages 35, 40, and 45) using VIA cost I$290 per YLS; this ratio is a fraction of India's per capita GDP, and would be considered very cost-effective according to suggested benchmarks for developing countries (WHO, 2001). As the cost per vaccinated girl exceeded I$10, vaccination alone was no longer more efficient than screening alone. The incremental cost-effectiveness ratio for pre-adolescent vaccination followed by screening in adulthood three times per lifetime varied from I$340 per YLS at I$20 per vaccinated girl, to I$1920 per YLS at I$75 per vaccinated girl. At a cost per dose of approximately$100 (I$360 per vaccinated girl), vaccination followed by screening exceeded I$7000 per YLS.

Although the rank-ordering of screening strategies from most effective to least effective, was two-visit HPV DNA testing, single-visit VIA, and three-visit cytology, differences between these screening approaches were somewhat attenuated at higher vaccination coverage rates. Vaccination plus screening three times per lifetime with two-visit HPV DNA screening consistently exceeded I$80 000 per YLS due to a VIA strategy's lower cost and ability to deliver screening and treatment in a single visit. However, with plausible changes in relative costs and test performance, and the ability to obtain a same-day test result, the use of HPV DNA testing was equally attractive. If a cytology-based strategy could be conducted in two visits without the use of colposcopy for diagnostic confirmation, it could be more attractive than predicted by our analysis due to reduced loss to follow-up. The results of this sensitivity analysis indicate that regions in India should consider their own situations and infrastructure, and identify which of these screening test options will be most feasible for them. Most importantly, regardless of specific test choice, our results provide strong support for screening adult women two to three times per lifetime in addition to pre-adolescent vaccination. Other influential assumptions on the cost-effectiveness ratio include the values used for vaccine efficacy and the annual discount rate. If vaccine efficacy is low, girls who are not protected still accrue the costs of vaccination, thereby making it less cost-effective. Cost-effectiveness analyses should be revisited as future data become available on efficacy, necessity for and cost of boosters, and ability to reach young girls who are not yet sexually active. Since the cancer prevention benefits to vaccinated girls occur years after the costs are paid, the effect of discounting both costs and benefits equally, as recommended by guidelines for economic evaluations, (DCPP; Gold et al, 1996; Drummond et al, 2005; WHO CHOICE) is substantial. We acknowledge the complexity of the decision faced by countries on whether to adopt the HPV vaccine, given the irrefutable reality that women who will benefit from vaccination are not in the same birth cohorts as the women who benefit from screening in the short-run. Even so, as we have documented in other work, the benefits of HPV 16,18 vaccination, when presented in formats that are not influenced by discounting, are comparable to those of other new vaccines. For example, we found that per 1000 girls vaccinated, 15 deaths would be averted (Goldie et al. 2008 in press), which compares favourably to 3 deaths prevented per 1000 children vaccinated for rotavirus (IAVI/PATH, 2007). We recognise that because vaccination and screening are applied to such different age groups, rely to different degrees on existing infrastructure, and require the mobilisation of financial resources that are likely to come from different ‘pots’, the feasibility of achieving wide coverage with screening vs vaccination could vary greatly within India, providing a challenge in designing a national prevention programme. It is possible that in regions where screening is not likely to be feasible, focused efforts to achieve high coverage rates for pre-adolescent vaccination would be the most worthwhile investment. In contrast, in regions where screening is successfully being conducted within demonstration projects or clinical studies, it may be most cost-effective to continue to expand those efforts and combine them with pre-adolescent vaccination – at vaccination coverage rates of even 50%, a combined approach could substantially reduce the incidence of cancer. In India, where out-of-pocket costs of health care can topple a household into extreme poverty, if the opportunity for vaccination is restricted to those willing to pay higher costs in the private sector, most women will not benefit. From a population perspective, since 80% of Indian women live on less than I$2 per day (World Bank), even at a drastically reduced price, the vast majority could not afford this vaccine and provision by the public health sector will be necessary. Although a cost-effectiveness analysis provides information on value for money to a decision maker with a long-term perspective on investing in health, it is not equivalent to providing information on affordability to the payer for whom the short-term perspective is more relevant. Both the financial costs (i.e., affordability) and the cost-effectiveness profile (i.e., value for money) of an HPV vaccine will need to be favourable as this vaccine will compete for dollars earmarked for existing immunisation programs.

Our analysis has several limitations, and thus, we emphasise that our results are intended to provide quantitative approximations of the potential benefits of HPV vaccination, and to provide qualitative insight into the relative value of primary and secondary prevention. We have previously discussed the limitations inherent in our modelling approach (Goldhaber-Fiebert et al, 2007; Goldie et al, 2007; Kim et al, 2007a), but we briefly mention key points here. In addition to model parameter uncertainty, there are uncertainties with respect to the natural history of HPV, especially in older women, the nature of type-specific immunity following natural infection, and the relationships between different HPV types in the case of multiple infections. Similarly, a lack of high quality data on temporal trends of HPV-related disease limited our ability to calibrate the model to different time points, and contributed to the choice to simulate a single birth cohort.

We made a tradeoff in choosing to use our individual-based microsimulation model for this analysis and not our HPV 16 and 18 transmission model (Kim et al, 2007b). This was a purposeful decision as the former can simulate detailed screening strategies as well as vaccination, and includes other HPV types not targeted by the vaccine, enabling us to take advantage of all available data by calibrating to many epidemiological targets (Garnett et al, 2006). As the model includes vaccine-targeted and non-targeted HPV types, we were able to explore the potential effect of an increase in non-vaccine-targeted HPV types. We have previously documented a small expected increase in non-16,18 cancers (Goldie et al, 2003, 2007), although we again found that the impact on the main cost-effectiveness results was small. While our independent dynamic model (Kim et al, 2007b) has been previously used to include the herd immunity effects in an HPV vaccine policy analysis, sensitivity analyses that reduced incidence based on these findings, found the main policy results would be unlikely to change. Data on type-specific transmission by age and sex will be useful to include in future analyses once available.

Given the range of uncertainties and the limitations of this analysis, our findings should be considered exploratory and our estimates of cost-effectiveness approximate, and should be interpreted in the context of the analytic purpose of this work, that is, to leverage the best available data to provide insight into decisions that policy makers in India are discussing right now. A country as large and as heterogeneous as India will need to conduct its own financial analyses, assess the effectiveness and feasibility of alternative modes of delivery, and identify potential economies of scale with other programmes that might target adolescents. The strategies that we have identified as cost-effective in this analysis may still be prohibitively expensive in India, and information on affordability, potential financing mechanisms, and likelihood of uptake and acceptability will need to be considered by decision makers as well.

Prevention and treatment of cervical cancer, the leading cause of cancer in Indian women, is a priority according to the country's National Cancer Control Programme (Government of India, 2005). On the basis of age-specific incidence and projected demographic changes, the expected new cases of cervical cancer in India will increase from 132 082 in year 2002, to more than 330 000 in 2050 (Ferlay et al, 2004). The vast majority of women will lack curative treatment, and thus, approximately two-thirds will die from this preventable disease. The implications for these women's families and communities are profound, as this cancer affects women at an age when they are vital to social and economic stability. The opportunity to prevent these deaths is now inarguable with the availability of a vaccine to prevent HPV 16,18 infection, new diagnostics for HPV DNA testing, and promising secondary prevention strategies that target women after age 30 and perform screening and treatment in as few visits as possible (Goldie et al, 2005). If the cost per vaccinated girl is less than I$10, implying a per dose cost of approximately$2, vaccination is likely to be extremely cost-effective in India. The most effective strategy, within a framework that would still be potentially cost-effective in India, would be pre-adolescent vaccination, followed by screening three times per lifetime between ages 35 and 45; assuming a vaccination coverage rate of 70%, this strategy would be expected to prevent more than 1.25 million cervical cancer deaths over the lifetimes of 10 consecutive birth cohorts.