Modeling optimal cervical cancer prevention strategies in Nigeria

A previously developed Markov cohort model built in Microsoft Excel was adapted to the Nigerian setting and was used to estimate the clinical and cost outcomes associated with each specified prevention strategy analyzed separately among a female population [31‐33]. Screening was assumed to be cytology-based. Eight strategies were analyzed using the Markov model: one lifetime screening at age 35 years; two lifetime screenings at ages 35 and 40 years; three lifetime screenings at ages 35, 40, and 45 years; vaccination of girls at age 12 years; vaccination and one lifetime screening; vaccination and two lifetime screenings; vaccination and three lifetime screenings; and no prevention. The screening strategies (one, two or three lifetime screenings) were selected to reflect the screening programs that could be implemented in Nigeria and other Sub-Saharan African countries.The Markov model consisted of 12 health states, reflecting the natural history of the disease and the effect of screening and vaccination: no infection with an oncogenic HPV virus; infection with an oncogenic HPV virus without precancerous or cancerous lesion; cervical intraepithelial neoplasia (CIN) grade 1; CIN grade 2 or 3; persistent CIN grade 2 or 3; CC; diagnosed CIN grade 1 through screening; diagnosed CIN grade 2 or 3 through screening; diagnosed persistent CIN grade 2 or 3 through screening; CC cured; death from CC; and death from other causes (see Figure 1).

The natural history transition rates between the different model health states were assumed to be the same as those used in the original Markov model [32] and were based on published natural history studies (Table 1). The other input values were adapted to reflect the epidemiology and costs of CC in Nigeria whenever available, or in the continent of Africa if country-specific data were not available [30‐33]. In particular, the incidence of HPV infection was taken from a study of the prevalence of HPV infection in Nigerian women, converted to incidence data based on natural mortality in Nigeria, HPV regression and HPV progression rate [34].

The validity of the model was assessed by comparing the estimated age-dependent CC incidence without any prevention strategy to the CC incidence reported in GLOBOCAN [1]. Calibration to the reported CC incidence was done by adjusting the persistent CIN2/3 to CC transition probability.

Health care services used for treating CIN grade 1, CIN grades 2 and 3, and CC were taken from a retrospective chart review performed at the university college hospital at Ibadan where patients’ charts are archived. The chart review collected the medical resources used (outpatient health care professional visits, outpatient diagnostic procedures, outpatient treatment procedures, medications, and hospitalizations) to treat a patient with CIN1, CIN2/3 or CC. This study received approval from the University of Ibadan/University College Hospital Ethics Committee. For precancerous lesions, resources used over a 1-year period from the date of diagnosis were collected; for CC, lifetime resources were collected (from diagnosis until either death or cure). For each condition, 10 patients with the required information at the time of data collection (2010) were identified (using consecutive medical records) and the medical resources used were extracted from their hospital medical records. The associated costs were estimated after assigning unit costs from the hospital record to each of the medical resources used. The costs were adjusted to 2012 values based on the Nigerian consumer price index for the health care sector [40]. The average costs from all patients per condition were used as input to the model (Table 1). Health care services for screening were based on expert opinion, and unit costs were estimated based on the average unit costs for each procedure reported in the hospital records from the university college hospital at Ibadan. The cost of the vaccine program was assumed to be $15 per dose (based on the Pan American Health Organization (PAHO) price).

Vaccine efficacy was estimated as the weighted average vaccine efficacy of 98% for HPV types 16 and 18 and 68.4% for the 10 most frequent non-vaccine HPV types (HPV-31/33/35/39/45/51/52/56/58/59) related to CC, based on the clinical trial results of the AS04-adjuvanted HPV-16/18 vaccine, with weights reflecting the relative frequency of the different HPV types in Nigerian women (Table 1). Matching efficacy was assumed for both the 3-dose and the 2-dose vaccination schedule.

For each prevention strategy, the Markov model estimated the lifetime costs for prevention and treatment of CIN and CC and the lifetime incident CC cases for a cohort of women. The lifetime outcomes from the Markov model were divided by the total life-years lived by the cohort and multiplied by 100,000 to provide an estimate of 1-year values at steady state for a population of 100,000 women, assuming that the age distribution for the population remained constant over time. These outcomes were then used as inputs to the optimization model. Because the estimated Markov model results were used to estimate the steady-state, cross-sectional, 1-year values for the whole population of interest, no discount rate was applied.

Objective function:

Subject to the following constraints:

With x _i  ∈ ℝ

The pre-vaccination budget was estimated assuming that there is no national screening or vaccination program in Nigeria, the associated expenditure per woman per year being the cost of treatment for those with CC, corresponding to $0.25 per woman per year across the entire female population. In the base-case analyses, the constraint on annual expenditure per woman was increased gradually from $0.25 to $2.0, and the annual incident number of CC cases was estimated for each level of annual expenditure when the optimal combination of prevention strategies is implemented. For the two vaccination schedules 3 and 2 doses were considered. Three scenarios were estimated for the full range of budget constraints: (1) maximum screening coverage of 20% and maximum vaccination coverage of 95%; (2) maximum screening coverage of 40% and maximum vaccination coverage of 95%; and (3) maximum screening coverage of 20% and maximum vaccination coverage of 50%. These ranges were selected based on expected or targeted ranges for screening and vaccination coverage within the Nigerian setting. An additional constraint required a minimum number of people to receive no prevention. This constraint equals the lower of one minus the upper-bound coverage constraint for either screening or vaccination and hence is directly linked to the screening and vaccination coverage constraint.

Univariate sensitivity analyses were conducted to estimate the effects on the incident number of CC cases for two different budget constraints, set at $1 and $2 per woman per year, when changing the costs associated with screening, vaccination, and treatment of CIN and CC. Ranges of plus or minus 20% were used for the screening costs, and ranges of plus or minus one standard deviation from the means were used for the costs of treatment of CIN and CC, using the data from the retrospective chart review. In addition, the impact on CC cases of adding strategies with more frequent screening, two lifetime screenings (at ages 35 and 40 years) or three lifetime screenings (at ages 35, 40 and 45 years), was tested. The sensitivity analysis on treatment costs accounted for setting with higher or lower costs than the one estimated in the retrospective cost evaluation that may not be representative of all settings in Nigeria. An additional scenario included the possibility of implementing an HPV test screening instead of the Pap test. In this scenario the cost of the screening test was set to 50% of the Pap test costs and CIN sensitivity was increased by an absolute value of 10%. Finally, the duration of vaccine efficacy was reduced from lifetime to 25 years, and also the vaccine efficacy was reduced by an absolute value of 20%.

Table 2 presents the outcomes, total cost, and annual incident number of CC cases for the prevention strategies used to generate inputs for the base case and alternative optimization model. These results indicate that screening all women once in a lifetime ($75,418 per 100,000 women) or providing no prevention ($26,201 per 100,000 women) is less expensive than vaccinating all women ($191,415 and $130,603 per 100,000 women for a 3- and 2-dose vaccine, respectively). One lifetime screening is less effective than vaccination at reducing incident CC cases (12.15 per 100,000 women per year with one lifetime screening, and 6.01 per 100,000 women per year with vaccination), but more effective than no prevention (17.45 CC cases per 100,000 women per year). The most effective and expensive strategy is vaccination combined with three lifetime screenings for all women (2.74 CC cases per 100,000 women for a cost of $303,324 for a 3-dose vaccine and $242,523 for a 2-dose vaccine).

Figure 2A presents the optimal allocation of resources for screening and vaccination in Nigeria at different budget constraints (i.e., the maximum levels of expenditure per woman per year for the prevention and treatment of CC) with a 20% coverage limit for one lifetime screening and a 95% coverage limit for vaccination with a 3-dose vaccination schedule. The stacked columns represent the estimated optimal proportion of women in the population receiving each intervention in order to reach the maximum CC reduction compared with pre-vaccination. Figure 2B presents the percentage reduction in CC cases from the pre-vaccination value of 17.45 per 100,000 women when the optimal allocation of resources to screening and vaccination is achieved at different levels of budget constraint. Figure 2C presents the budget associated with the optimal strategies for each set of constraints included in the optimization model under the base-case. At maximum constraint values of a budget of $1.00 per woman and coverage of 95% for vaccination and 20% for one lifetime screening, the optimal strategy would be 39% with vaccination alone, 20% with one lifetime screening, 0% with vaccination and one lifetime screening, and 41% with no prevention strategy. This would result in a 31% reduction in the number of CC cases. With a 2-dose vaccination schedule, as presented in Figure 3, the optimal mix of prevention strategies at the $1 per woman budget constraint would be 71% with vaccination alone, 0% with one lifetime screening, 0% with vaccination plus one lifetime screening, and 29% with no prevention strategy. This would result in a 46% reduction in the number of CC cases.

With a budget constraint of $2.00 per woman per year, the maximum prevention coverage could be reached and the optimal mix of prevention strategies to minimize CC incidence would be (with both a 2- and a 3-dose vaccination schedule), 75% with vaccination alone, 20% with vaccination and one lifetime screening, and 5% with no prevention strategy. This would result in a CC reduction of 64% with both a 3-dose and a 2-dose vaccination schedule. The resulting expenditure would be $1.93 (3-dose vaccination schedule) and $1.35 (2-dose vaccination schedule) per woman per year.

Although the most effective of the four strategies included in the base case is vaccination plus one lifetime screening, the optimal mix of prevention strategies does not include this combination until the budget constraint per woman is set to $2.00 with a 3-dose vaccination schedule and $1.50 with a 2-dose vaccination schedule.Figures 4, 5, 6 and 7 show similar impacts of relaxing the budget constraint within different constraints on one lifetime screening and vaccination coverage. In Figures 4 and 5, representing a scenario with a vaccination coverage constraint of 95% and once in a lifetime screening coverage constraint of 40%, the maximum coverage of prevention strategies would result in an expenditure of $2.02 (3-dose vaccination schedule) to $1.44 (2-dose vaccination schedule) per woman, with an associated 66% reduction of incident CC cases. In Figures 6 and 7, which present a scenario with a vaccination coverage constraint of 50% and a screening coverage constraint of 20%, the maximum prevention coverage would result in an expenditure of $1.18 (3-dose vaccination schedule) to $0.88 (2-dose vaccination schedule) per woman per year, and an associated 35% reduction in CC cases.

With a 3-dose vaccination schedule (Figures 2, 4 and 6) the optimal mix of prevention strategies includes screening alone for the lowest budget constraint. With a 2-dose vaccination schedule (Figures 3, 5 and 7), the lowest budget constraints do not include screening but include vaccination alone. Screening is only part of the optimal mix of prevention strategies for budget constraints of at least $1.50 per woman per year with a vaccination coverage constraint of 95% or at least $1.00 per woman per year with a vaccination coverage constraint of 50%.

One way sensitivity analyses were performed for a budget constraint of $1 and $2 per woman per year. The costs of screening and treating CIN grade 1, CIN grades 2 and 3, and CC were varied, as was the frequency of lifetime screenings and the duration and level of protection resulting from vaccination. An additional scenario investigated the use of a HPV test as the screening method instead of the Pap test. The results of the sensitivity analyses, measured as the percentage of CC cases prevented compared with the pre-vaccination incidence of CC cases (17.45 per 100,000 women) are shown in Table 3 for a budget constraint of $1 per woman per year and Table 4 for a budget constraint of $2 per woman per year. The results indicate that the maximum reachable CC reduction was higher with a 2-dose than with a 3-dose vaccination schedule under all sensitivity analyses conducted. The optimal mix of strategies under the different sensitivity analyses are presented in the Additional file 1: Figure S1 and Additional file 2: Figure S2. The costs of CIN and CC treatment, as well as the vaccine characteristics, had the largest impact on the optimal CC reduction. A low cost led to a higher coverage of the population by a prevention strategy resulting in more CC cases prevented, while a high cost led to a lower coverage and hence a lower CC reduction. Interestingly, the optimal strategy with a high cost for treating precancerous lesion would imply vaccination alone or no prevention with either a 3- or a 2-dose vaccination schedule. However, the optimal strategy with a high cost for treating cancer would combine screening alone, vaccination alone and no prevention with a 3-dose vaccination schedule, and vaccination alone, vaccination combined with screening or no prevention with a 2-dose vaccination schedule. The use a 2-dose vaccine with a reduction in the vaccine efficacy also led to a lower CC reduction under the optimal mix of strategies and a combination of screening and vaccination with however a larger vaccination coverage than with a 3-dose vaccination. The use of a HPV test for the screening assuming a lower costs and higher sensitivity led to a higher CC reduction while the optimal mix would combine both screening and vaccination.

With a $2 per woman budget constraint, the maximum vaccination and screening coverage is reached under the optimal mix of strategies for all sensitivity analyses performed, and hence the maximum CC reduction is reached with both the 3-dose and the 2-dose vaccination schedule scenarios and under all sensitivity analyses.

The results of the Markov evaluation models indicated that the number of CC cases expected from a 100% coverage of each prevention strategy was lowest for vaccination (6.01 per 100,000 women per year) compared with one lifetime screening (12.15 per 100,000 women per year), two lifetime screenings (9.63 per 100,000 women per year), three lifetime screenings (7.85 per 100,000 women per year), or no prevention (17.45 per 100,000 women per year).

In the base-case optimization model analyses, with upper-bound coverage constraints of 20% for screening and of 95% for vaccination and a budget constraint at $1 per woman, the optimal mix of prevention strategies would result in a 31% CC reduction compared with today’s CC incidence with a 3-dose vaccination schedule, and in a 46% CC reduction with a 2-dose vaccination schedule. With a 3-dose vaccination schedule, the optimal combination would be 20% with screening alone, 39% with vaccination alone and 41% without any prevention, while with a 2-dose vaccination schedule the optimal combination would be 0% screened, 71% vaccinated, and 29% without any prevention. Under the lower budget constraints, the optimal strategy with a 3-dose vaccination schedule would always be a combination of screening alone, vaccination alone and no prevention, while with a 2-dose vaccination schedule the optimal combination would only include vaccination and no prevention by screening. Using an increment in budget constraint of $0.25 per women per year going from $0.25 to $2.00, any budget constraint equal to or higher than $2.00 per woman with a 3-dose vaccination schedule, or $1.50 per woman with a 2-dose vaccination schedule, would result in the optimal strategy with a maximum CC reduction using a strategy consisting of 75% with vaccination alone, 20% with vaccination and screening and 5% without prevention. The associated CC reduction would be 64%. These strategies would be specifically associated with a budget of $1.93 per woman with a 3-dose vaccination schedule and $1.35 per woman with a 2-dose vaccination schedule.

The results were similar for different screening and vaccination coverages, except that in the case of an upper-bound vaccination coverage of 50%, the percentage reduction in CC cases compared with the pre-vaccination situation had a maximum value of 35% at an expenditure of $1.18 per woman with a 3-dose vaccination schedule and $0.88 per woman with a 2-dose vaccination schedule. Sensitivity analyses indicated that the results were sensitive to the CIN and CC treatment cost. Higher treatment costs resulted in lower achievable coverage with vaccination and screening within the $1 budget constraint, and hence a lower reduction in the number of CC cases. With a 2-dose vaccination schedule, all sensitivity analyses tended to first maximize the vaccination coverage. With a 3-dose vaccination schedule, most sensitivity analyses (except a high CIN treatment cost) tended to maximize both vaccination and screening and in all cases resulted in a lower reduction in CC than the reduction obtained with a 2-dose vaccination schedule.

The results from the optimization model indicate that when introducing a prevention program for CC, in a country like Nigeria where none is currently in place, with a 2-dose vaccination schedule it would be optimal to invest in vaccination alone to the extent feasible, followed by vaccination plus one lifetime screening. With a 3-dose vaccination schedule, a combination of both vaccination and screening would be optimal. This result depends therefore on the relative costs of vaccination and screening as well as the relative efficacy between the two interventions. The optimization would favor a strategy with a low costs and a high efficacy (sensitivity for the screening).

A similar model was recently used to estimate the optimal mix of CC screening and HPV vaccination in the UK and Brazil [30]. These two countries already had a screening program in place, with coverage levels of 65% in the UK and 50% in Brazil, each with a screening frequency of 3 years applied to all screened women. The optimization model indicated that keeping the existing screening program while extending the screening interval to 6 years and adding vaccination with 80% coverage would result in a reduction in CC cases of 41% in the UK and 54% in Brazil, with no change in expenditure per woman compared with the pre-vaccination situation. This is very different from the situation observed in Nigeria, where no nationwide screening programs are currently in place and one lifetime screening is standard for the few women who are screened. As a result, reducing CC in Sub-Saharan Africa will require that additional resources be spent per woman on either screening or HPV vaccination programs.

The current evaluation estimated that an optimal allocation of a CC prevention budget of $1 per woman per year could lead to an important reduction in CC burden by 31% to 46%, depending on the scenario chosen. In Nigeria, three tiers of Government (Federal, State and Local) have responsibility for health and decide about budget allocation. The current evaluation suggests that a combination of 2-dose vaccination with or without one lifetime screening is expected to lead to an important reduction in CC.

Nationwide deployment of HPV vaccines to reduce the CC burden in Africa is more realistic than implementing a screening program, given the current failure of the national screening program due to logistic and funding challenges. Most African countries have efficient and very well accepted national childhood immunization programs, as in Nigeria. HPV vaccination could be integrated into these programs, provided some operational challenges linked to the adaptation of the program to a different target population (adolescent for the HPV vaccination) are addressed. The cost of implementation would also be drastically reduced compared to that expected from the implementation of a screening program. However, in each country the age at which the vaccination is implemented should be adapted based on their national average age of girls’ sexual debut.

The optimization model used for this analysis has its limitations. First, the model outcome is obtained when an optimal mix of different prevention strategies is applied in the population and a steady-state incidence of CC is achieved across the population. Many years will elapse before reaching that state. The results therefore represent an optimal “ideal” situation that provides a direction to be followed but not a result that will occur immediately, and nor does it address specific implementation issues. Vaccination will affect the girls who will become the women of the future. Today’s women would not benefit from vaccination, but could directly benefit from a single lifetime screening. The vaccinated girls would not need any screening in the years post vaccination. This cost saving could be used to implement a screening program benefiting older women. This would accelerate the overall CC reduction. No discounting was applied in this analysis, as the model presents a steady-state evaluation over a 1-year period. Given the time lag needed to reach steady state, discounting could be considered, but the means of implementation and especially the value of including a discount rate in such an optimization exercise is debatable. A discounting applied to the input of the model down to a today value would not impact the optimization results while ways to implement the optimization on a population overtime still need to be investigated.

Second, a static Markov model was used to evaluate the cost and outcomes of each potential strategy at steady state across a population over a 1-year period. As the optimization model takes results from the evaluation model, the robustness of the optimization results therefore also depends upon the validity of the evaluation model. The Markov model used for this analysis did not account for dynamic effects of vaccination on infection, such as herd protection. It also assumed stability in the population demographic structure over time. However, using a dynamic model as the evaluation model instead of the static model would be more difficult to adapt to countries with limited data. It would also lead to nonlinearity between coverage and the outcomes, and thus would require the use of a nonlinear programming optimization model.

Third, the models presented here are deterministic: costs and effects in all combinations are known. It would be possible to characterize the allocation problem under conditions of uncertainty by reformulating the model as a stochastic analysis. However, existing approaches for such stochastic analyses are limited [49]. This is an area for future research, although the principal outcome of evaluating uncertainty is to indicate the need and the potential value of searching for more detailed and specific information [49].

Fourth, the results depend on the strategies initially investigated using the evaluation model. Therefore, including a vaccine with a different profile or a screening program with different characteristics (e.g., HPV testing instead of cytology-based screening) may lead to other optimal scenarios, as presented in the sensitivity analysis conducted in the current evaluation. Furthermore, the analysis does not take into account the possibility that the introduction of vaccination may affect the sensitivity of the screening test [50, 51]. It also does not account for the cost associated with the implementation of a screening or vaccination program in Nigeria. Finally, our model assumes that the interventions evaluated are completely divisible in terms of intervals between screening tests and coverage. It ignores possible fixed set-up and implementation costs associated with interventions, which could constrain their divisibility. The fixed costs could be large for screening coverage, such as setting up the necessary infrastructure to collect and analyze samples, or implementing health education programs to increase awareness and overcome psychological and cultural barriers to uptake. Conversely, they may be relatively low for vaccine implementation in situations where other vaccination programs in the relevant age group are already established. In addition, the incremental implementation cost by coverage is unlikely to be linear. Disregarding such indivisible or nonlinear costs could be important in practice. Given the importance of the relative costs of vaccination and screening to the selection of the optimal program, the inclusion of these costs could have a large impact on the results.