The impact of a sexually transmitted infection (STI) prevention intervention is influenced by the epidemic phase of the STI.^{1,2,3,4,5,6,7,8,9,10} For example, prevention efforts that focus on the general population will be relatively less effective in phases of an epidemic when the disease is more concentrated in the core population than in phases when the disease is spreading beyond the core population.^1,4

Epidemic phase would also be expected to influence the cost effectiveness of STI prevention. For example, some prevention efforts will be more cost effective when initiated in the early phases of an epidemic by preventing the spread of the STI into the general population.^4,11 Conversely, because many STI prevention activities (such as screening and treatment) are more cost effective when prevalence is higher,^12–14 such prevention activities could be less cost effective if initiated in early phases on an epidemic when prevalence is low. So, in terms of cost effectiveness of prevention, there are advantages and disadvantages of initiation of prevention efforts in early stages of the epidemic. The present study explored how the cost effectiveness of a hypothetical, behaviour-change STI prevention programme varies over the course of an STI epidemic.

METHODS

We used a deterministic, compartmental model of STI transmission (see appendix) based on previously published work.^15–17 The model population was divided into two groups, susceptible and infected, and was stratified by sex and into low-activity and high-activity groups with characteristic rates of sex partner change. The low-activity and high-activity groups comprised 90% and 10% of the population, respectively, and had rates of sex partner change 1 per year and 10 per year, respectively.^8,17 Mixing between the low-activity and high-activity groups was determined by a mixing parameter (ε), set to 0.5 in the base case analysis, where ε = 1 corresponds to assortative matching (no mixing between the two sexual activity groups) and ε = 0 corresponds to random selection of sex partners without regard to sexual activity group.

The modelled population consisted of 200 000 exclusively heterosexual persons (100 000 women, 100 000 men) aged 16–35 years. We assumed that all sexual contacts of this population came from within this population. Each year, 10 000 individuals (5% of the population) exited due to age and were replaced with non-infected individuals. The per-partnership probability of transmission was 0.5 and the average duration of infection was six months. These values of the probability of transmission and the duration of infection are consistent with bacterial STIs such as syphilis, gonorrhoea and chlamydia.^18–23 At the start of the epidemic (year 0), 100 men and 100 women in the high-activity group were infected, and no individuals in the low-activity group were infected.

We examined the impact of a hypothetical public awareness campaign (delivered through the media) to increase the use of condoms. The cost of such a prevention effort, unlike the cost of prevention strategies such as screening and treatment, would not depend on the STI prevalence in the population. We assumed that the prevention intervention would cost $10 million a year. This estimate is probably unrealistically high for a population of 200 000 but was selected so that the cost-effectiveness ratios in the sensitivity analyses would be positive, as comparisons of negative cost-effectiveness ratios are difficult.²⁴ We assumed that the increase in condom usage attributable to the intervention would reduce the per-partnership probability of transmission by 20% while the intervention was in place. We also assumed that the average lifetime cost per case of treatment of the STI and its sequelae was US$50 and US$350 in men and women, respectively. These values are consistent with estimates of the discounted, lifetime cost per case (including possible costs of sequelae) of bacterial STI such as gonorrhoea and chlamydia.²⁵

The average cost-effectiveness ratio is the net cost of the intervention (programme costs minus the treatment cost averted by the intervention) divided by the number of cases averted.²⁶ Specifically, we calculated the average cost-effectiveness ratio as:

(P−A×L)/A

where P is the programme cost, A is the number of STI cases averted by the intervention (compared with no intervention), and L is the average lifetime cost per case of the STI.

We examined how the cost effectiveness of the intervention changed when we varied the onset of initiation of the intervention from year 0 (at the start of the epidemic) to year 20 (by which time the incidence had reached equilibrium). The intervention, once implemented in a given year, was assumed to be in effect and equally effective in all subsequent years as well.

Programme costs and averted STI cases were discounted at an annual rate of 3% to present values.²⁷ That is, programme costs and averted STI cases on the first day of the second year were divided by 1.03, programme costs and averted cases on the first day of the third year were divided by the squared value of 1.03, and so on. Programme costs were incurred only in years when the intervention was offered. Averted STI cases were calculated over a 40-year horizon. The 40-year horizon was applied to ensure that all of the benefits of the delayed intervention would be included in the cost-effectiveness analysis. We repeated the analysis, focusing on a five-year horizon rather than a 40-year horizon. In doing so, we calculated cost-effectiveness ratios based only on costs and benefits in the first five years of the intervention.

Sensitivity analyses

We conducted one-way sensitivity analyses to see how the estimated cost-effectiveness ratios (in the 40-year horizon) would change when varying one parameter at a time, while holding other parameters at their base case values. Specifically, we calculated the cost-effectiveness ratios of interventions initiated in year 0 and year 20 when varying the discount rate, the per-partnership probability of transmission, the mixing parameter (ε), the proportion of the population with a low rate of sex partner change, the impact of the intervention, the annual cost of the intervention, and the direct medical costs per case of STI.

RESULTS

The steady-state STI prevalence rate was 8.0% in the absence of the intervention and 5.2% with the intervention, regardless of whether the intervention was implemented in year 0 or in year 20 (fig 1). The cumulative incidence of cases of STI was substantially lower when the intervention was implemented in year 0 than when the intervention was implemented in year 20 or when there was no intervention (fig 2). In the base case analysis, the intervention was more cost effective when implemented in year 0 than in year 20 for both the 40-year and 5-year time horizons (table 1). When implemented in year 0, the intervention averted 268 004 discounted cases of STI at a cost of US$675 per case averted over the 40-year time horizon (compared with no intervention). When implemented in year 20, the intervention averted 91 828 discounted cases of STI at a cost of US$710 per case averted over the 40-year time horizon (compared with no intervention). When the costs and benefits were calculated for only the first five years of the interventions, the earlier intervention was again more cost effective than the later intervention, and the difference was slightly more pronounced than in the 40-year horizon.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

 Prevalence of sexually transmitted infection with no intervention, earlier intervention and later intervention.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

 Cumulative incidence of sexually transmitted infection (discounted number of cases) with no intervention, earlier intervention and later intervention.

In the 40-year horizon, the cost per case averted increased as the onset of implementation was increased from year 0 to year 20, but this relationship was not linear (fig 3). That is, delaying the onset of the intervention from year 0 to year 1, or from year 1 to year 2 increased the cost-effectiveness ratio more substantially than delaying the onset of the intervention from year 10 to year 11, or year 11 to year 12.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

 Cost per case of STI averted (compared with no intervention) when intervention onset is varied from year 0 to year 20.

Sensitivity analyses

Earlier intervention was more cost effective than later intervention, over a range of parameter values (table 2). The cost-effectiveness ratios were particularly sensitive to changes in the mixing parameter, the proportion of the population in the group with low sexual activity, the impact of the intervention on the per-act probability of STI transmission, the annual cost of the intervention and the costs of STI treatment (table 2). However, the absolute difference between the cost effectiveness ratios of the earlier and later interventions did not exceed US$75 over the range of the sensitivity analyses, except when the mixing parameter (ε) was set to 1.

DISCUSSION

The model results suggest that a sustained, behaviour-change STI prevention intervention will generally be more cost effective the earlier in the epidemic it is implemented—that is, the cost per case averted increases over time as the onset of the intervention is delayed. These findings are in general agreement with two recent, model-based cost-effectiveness analyses of HIV prevention interventions.^28,29 In our model, the cost per case averted was higher for the later intervention (implemented in year 20) than for the earlier intervention (implemented in year 0) over a range of assumptions about the per-partnership probability of transmission, the partner mixing patterns, the intervention’s impact and cost, and the costs of STI treatment.

Although the later implementation of the intervention led to a rapid decline from the peak rates of STI prevalence of 8% (fig 1), the earlier implementation of the intervention prevented the occurrence of this 8% peak from the outset. Earlier intervention is more cost effective because the effective reproductive number—the average number of new infections arising per existing case—is higher earlier in the epidemic, so each case averted in turn averts more future cases. As the epidemic progresses the effective reproductive number declines for two reasons. First, the epidemic is initially concentrated in the high-risk group in which infection is transmitted more effectively but as the epidemic progresses the proportion of all infections that is in the low-risk group increases, thus reducing the average rate of onward transmission. Second, as prevalence increases, the proportion of the contacts that an infected person has with susceptible individuals declines.

In the scenario that we examined, the cost of the intervention did not depend on STI prevalence, as it was a behaviour-change intervention delivered through the media. The cost of other STI prevention activities (such as screening and treatment) will vary with the prevalence of the STI.^12–14 For such activities, it is possible that earlier interventions could be less cost effective than later interventions, if the number needed to screen to find an infected person is sufficiently high. Rapid spreading of infection results in a limited window of opportunity to capture the benefits of earlier initiation of the intervention. Although we found the cost per case of STI averted to increase as the onset of the intervention is delayed, this increase was most notable over the first two years and became much less pronounced after year 4 (fig 3). Thus, as initial prevalence increases, the cost per case of STI averted by early intervention also increases.

Although the model results predicted that earlier interventions are more cost effective than later interventions, the influence of the timing of the intervention on cost effectiveness was less important than other factors such as the cost of the intervention, the cost of STI treatment and the impact of the intervention (table 2). Nevertheless, over the range of parameter values we examined, earlier intervention was more cost effective than later intervention. When applying more extreme parameter values than reported in the sensitivity analyses, we found scenarios in which earlier intervention was less cost effective than later intervention. For example, delaying the onset of the intervention improved the cost-effectiveness ratio when both a low probability of STI transmission (β = 0.025) and a lifetime duration of infection were applied simultaneously. In this example, STI prevalence increased slowly in the population, but took over 50 years to reach 10%. However, note that these parameter values correspond more closely to a viral infection such as due to herpes simplex or HIV, which have higher lifetime treatment costs than we assumed for bacterial sexually transmitted diseases in this analysis.

Our analysis is subject to several limitations. First, our study is quite hypothetical in its assumptions about the cost and impact of the intervention. As noted earlier, an unrealistically high intervention cost ($10 million per year) was selected to ensure that the cost-effectiveness ratios would be positive (ie, the intervention would not be cost saving). The assumed impact of the intervention (a 20% reduction in the per-partnership transmission probability due to increased condom usage) is probably unrealistically high for long-term partnerships, although it is more reasonable for short-term partnerships in the high-activity group, which are the ones contributing to most of the transmission of infection. Assuming a high impact of the intervention helped to clarify the differences in the cost effectiveness of interventions over the phase of the epidemic. Finally, we made a range of additional, simplifying assumptions. For example, our model included only two sexual activity classes (and was not stratified by other potentially important factors such as age), and we did not examine all the potential combinations of possible parameter values and time horizons.

In summary, we found that STI prevention interventions were generally more cost effective when initiated earlier, at least for interventions whose cost does not depend on the prevalence of the STI in the population. However, this “earlier is better” finding must be considered with two important caveats. First, this finding does not always hold, as it is possible to devise plausible scenarios in which earlier interventions are less cost effective than later interventions. For example, when we simultaneously applied a low probability of STI transmission and assumed a lifelong duration of infection, the delayed intervention was more cost effective than the earlier intervention. Second, the relative impact of the timing of the STI prevention intervention on the intervention’s cost effectiveness can be quite small compared with other important factors, such as the cost and impact of the intervention and the lifetime medical cost of the STI prevented by the intervention. Therefore, it is important to obtain reliable estimates of these factors to guide decision making.

APPENDIX 1

DESCRIPTION OF MODEL

λ_k,l = βc_l[ρ_lω_k′,l+(1−ρ_l)ω_k′,l′],

where S and I are the numbers of susceptible and infected people, respectively; the subscripts k and l denote sex and sexual activity class, respectively; the subscript k′ refers to the opposite sex to k and the subscript l′ refers to the opposite sexual activity class to l; N_k,l = S_k,l+I_k,l; the force of infection (λ_k,l) is the product of the per-partner transmission probability (β), the rate of sex partner change (c_l), and the prevalence of STI in the pools of sex partners (determined by ω_k,l, the prevalence of infection in sex k and sexual activity class l, and ρ_l, the probability that the partner of a person in sexual activity class l is in the same sexual activity class); and other symbols are defined as follows: in the base case, the rate of entry and exit from the population (μ) was 0.05, the recovery rate (ν) was 2, the per-partnership transmission probability (β) was 0.5, the rate of sex partner change (c_l) in the low-activity and high-activity groups was 1 and 10, respectively, the proportion of the population in the low-activity and high-activity groups (χ_l) was 0.9 and 0.1, respectively, and the mixing parameter (ε) was 0.5.