The potential impact of the demographic transition in the Senegal-Gambia region of sub-Saharan Africa on the burden of infectious disease and its potential synergies with control programmes: the case of hepatitis B

Focus on Senegal and The Gambia as a unique demographic entity (denoted from now on as SG) was motivated by the fact that The Gambia, apart from its coastline (< 50 km), is entirely surrounded by Senegal, and also that there is a strong ethnic and cultural overlap between the two countries, which for a period in the 1980s together formed a loose confederation, Senegambia [29]. The Gambia itself is an exemplar of a country in SSA with good-quality data on the prevalence of pre-vaccination HBV infection by age [25]. Moreover, the two countries supplied a substantial portion of the data used previously to estimate a key function relating age and likelihood of becoming a carrier [25]. In addition, although patterns of age-specific growth rates of the small population in The Gambia are somewhat erratic, this is not the case for the much larger Senegal, which has exhibited, according to United Nations (UN) data, fairly regular patterns of DT and mortality decline since 1950, and a well-separated late onset of fertility transition (by about 1990).

Data on previous experience of HBV infection by age were drawn from published work on The Gambia [30]. Plausible ranges for HBV epidemiological parameters relating to infectivity, duration of stages of infection, etc. (Additional file 1: Table S2.1) were gleaned from the literature [23]. Demographic data on fertility and mortality rates in Senegal and The Gambia were drawn from the UN 1950–2015 estimates and the subsequent 2015–2150 projections, as well as data on the population age distribution by 5-year age groups which were used for model validation [31]. In particular, data from all three main UN projections variants, namely, the “medium” one, which is the UN key variant taken as our baseline here, as well as the “high” and “low” variants, were considered. Moving averages of estimated or projected UN age-specific fertility and mortality rates in successive 5-year periods from 1950 to 1955 to 2095–2100 provided time-varying vital rates for the model over a 150-year period.

The population model was age-sex structured with time-changing age-specific fertility and age- and gender-specific mortality rates to depict a realistic time course for the DT. The part of the model simulating the transmission dynamics of HBV (see the flow diagram in Fig. 1) included age dependencies in vertical, horizontal, and heterosexual modes of transmission, as well as in the probability of developing HBV carriage once infected (see Additional file 1: Figure S3.1 and Text S2 for details including the full partial differential equations (PDEs) describing the model). A computer program designed for the purpose and coded using Fortran 77 provided for numerical solution of the model PDEs.

In the model all births to acutely and chronically HBV infected mothers were at risk of vertically transmitted HBV, acquired at a rate depending on the relative proportions of females with acute or persistent (i.e. chronic) infection in each fertile age group. A proportion of vertically infected offspring was assumed to develop chronic infection [25]. Children and adults at all ages were at risk of horizontal HBV transmission at an age-specific rate determined by an appropriately specified WAIFW (who acquires infection from whom) matrix [6] obtained by multiplying the WAIFW matrix estimated for The Gambia by Edmunds and colleagues [30] by a single scaling parameter for purposes of data fit. The influence of the evolving age distribution of the population on social contact patterns relevant for horizontal HBV transmission, and therefore on the WAIFW, was modelled in accord with previous work [10‐12, 16]. Sexual transmission started at the age of sexual debut, assumed to be 15 years, at a rate dependent on (1) the rate of acquisition of new partners for each gender and age group, (2) sexual mixing matrices determining the proportion of sexual contacts to be drawn from each age group, (3) the proportion of acutely and chronically infected partners in each gender age group, and (4) the risk of transmission per acutely and chronically infected partner. The sexual mixing matrix adopted obeyed a preferred mixing rule [32] with a parameter determining the degree of age assortativeness of sexual partnerships. The underlying age-gender specific rates of sexual behaviour are kept constant during population evolution, but as male and female populations evolve separately, at each time point an adjustment is made to partner acquisition rates to ensure that male and female partnerships always balance. Note that the preferred mixing rule, in which the force of infection acting on each age-gender group explicitly depends on the proportions of the population in the various age groups, straightforwardly incorporates the feedback of the evolving age distribution of the population on sexual contact patterns.

Upon infection by any route of transmission, susceptible people moved to a latently infected but noninfectious stage and subsequently to an acutely infected and infectious stage, a small proportion of the latter being assumed to die as a result of fulminant liver disease. Following acute infection, a strongly age-dependent proportion [25] moved to a state of chronic carriage having an average duration of several decades prior to recovery, while the complementary proportion resolved the infection, moving directly into the recovered stage; for simplicity, it was assumed that there was no additional disease-specific mortality during chronic carriage and that no screening and treatment programme was in place which might hasten recovery or reduce infectivity.

The model could also be set to include HBV vaccination at birth and/or at any subsequent age point, starting from a pre-specified time point, with a vaccine assumed to have 100% efficacy and lifelong duration. Those vaccinated in the model are moved to a vaccinated and immune compartment.

Estimated (period 1950–2015) and projected (2015–2100) mortality and fertility rates from the UN for Senegal and Gambia were pooled to obtain a unique set of rates for SG as a single entity. In order to depict the course of an entire DT for SG, from its ancien régime stationary demographic equilibrium, we hypothesised that SG fertility at 1950 reflected true SG fertility through its entire previous history back to when the ancien-demographic régime prevailed. We then projected mortality rates into the past to reconstruct a reasonable ancien régime mortality table and the related stationary population (as expected for the ancien régime), i.e. one characterised by a constant total population size and an invariant age distribution, (see Additional file 1: Text S1 for details). The stationarity of the ancien régime age distribution was ensured by requiring that the corresponding female life table once combined with SG 1950 female fertility rates promoted a net reproductive rate of the population equal to one. This equilibrium population was taken as the initial condition for the age distribution of the SG population, i.e. the one prevailing before destabilisation due to the onset of the mortality transition (Fig. 2a). Departing from the ancien régime equilibrium, the demographic model was then run forward in time to reproduce the DT in SG using the pooled set of SG mortality and fertility rates. Model predictions were compared, by a least squares criterion, to the 1950 SG male-female age distribution, assuming smooth adjustment over time of ancien régime mortality to the 1950 UN estimate.

We assumed also that HBV was, in the demographic ancien régime equilibrium, at an epidemiologic equilibrium which itself was destabilised by mortality decline at onset of the DT. Based on the HBV literature, uncertainty ranges were specified for HBV epidemiological parameters in the model, and Latin hypercube sampling (LHS) was used to explore these parameter ranges (see Additional file 1: Table S2.1 for full details of HBV parameters included in the LHS procedure). By LHS we generated 24,000 sampled parameter sets. Departing from the ancien régime demo-epidemiologic equilibrium, the full HBV model was first run forward for each sampled parameter set until 1984 using smoothed UN-based vital rate estimates 1950–1985 [31], under the hypothesis that no interventions (vaccination, use of protection for sexual activity, etc.) were in place. Model predictions at 1984 were compared, by a least squares criterion, to Gambian early 1980s’ HBV pre-vaccination age-prevalence data on both past exposure and current infection. As best fit we took the parameter combination generating the lowest value of the least squares function over the entire set of 24,000 LHS samples. In particular, the horizontal transmission WAIFW matrix was fitted to the data by adjusting the scaling factor applied to all the elements of the matrix within the overall fitting procedure. During each simulation the HBV parameters were kept constant over time, ensuring that any changes over time in results for disease prevalence arose solely from demographic change. Uncertainty bands about the best-fit set of parameters were generated by retaining the 2400 (i.e. 10% of the total) parameter sets generating the largest values of the least squares function. These uncertainty bands about parameter values were then used to generate uncertainty bands about the predicted HBV trajectories over time.

Our baseline vaccination scenario was designed to approximate the Senegal situation in view of its possible larger relevance as: (1) The Gambia could be considered a special case due to the work of The Gambia Hepatitis Intervention Study [33], (2) the Senegal population is many times the size of its neighbour, (3) in common with many other countries in SSA, in Senegal the lack of availability of monovalent HBV vaccine prevented the implementation of universal vaccination at birth [34]. Thus, in our baseline scenario, vaccination is administered to children aged 3.5 months, with effective coverage of 46% [34] and 2005 as year of introduction of the vaccine into the Expanded Programme on Immunisation [34]. Alternative scenarios were considered including higher coverage levels (55% and 75%) as well as the possibility of vaccination at birth to prevent vertical transmission.

As a preliminary step, using the best demographic model, predictions for the age-sex distribution of the population for 1955–2100 were systematically compared with the corresponding UN estimates/projections, in order to ensure that the model-based prediction of the evolution of the age-sex structure depicted a course of the DT in SG always close to UN projected figures.

The model identified 1840–1850 as the epoch when the ancien regíme equilibrium ended due to initiation of the mortality transition (Fig. 2a). Forward projections of the demographic model satisfactorily fitted the 1950 UN estimate of population age distribution for both sexes (Fig. 2b). Subsequent model-based forward projections of the age-sex distribution remained quite close to UN population estimates and projections for the entire horizon 1950–2100, as reported in Fig. 2b–f for the case of the UN “medium” projection variant. The model also reproduced quite satisfactorily the 1980s’ pre-vaccination age-prevalence data for exposure to HBV (Fig. 3a) and current HBV infection (Fig. 3b).

Figures 4 and 5 report predictions by the best model for the evolution of the natural history of HBV that would be observed in the absence of immunisation during the whole course of the DT in SG, from around 1850 (corresponding to the ancien régime) to 2100 according to the UN medium variant. As well summarised by the transitional trend of prevalence of exposure to HBV infection (Fig. 4a), starting from the HBV pre-transitional equilibrium profile at 1850, where prevalence by age 15 was in the region of 70%, HBV was predicted to gradually expand, reaching prevalence levels as high as 85% by age 15. Thereafter HBV entered a phase of continuing decline, returning to levels comparable to, and subsequently below, the pre-transition phase when projected fertility approached and stabilised near to replacement levels (between 2100 and 2150). At 2150, HBV prevalence (in the absence of any control measures) is predicted to settle at 25–30% by age 15. This same story is told by the other graphs of Fig. 4: in particular, the population age distribution of disease burden post-equilibrium (Fig. 4d) remained heavily skewed towards younger ages until 1950–2000 before becoming substantially more evenly distributed in later years, reflecting the changed demography.

Much of the explanation for the predicted trends lies in the dramatic changes of horizontal transmission during the DT. This is made clear by the temporal trend of the corresponding age-specific force of infection (Fig. 5a) among young individuals (aged below 15 years). Indeed, the horizontal force of infection among the youngest systematically increased during the HBV expansion phase up to levels of around 15% per year, approaching levels commonly seen for typical childhood infections such as varicella [35]. This resulted from the continued rejuvenation of the population during the major epoch of mortality decline among young individuals and persistent high fertility (roughly up to 2000). Subsequently, the horizontal force of infection among young individuals experienced continued decline, falling to very low levels — well below those of the ancien régime – during the major phase of fertility decline (after 2050) and corresponding progressive population ageing. This pattern arises as a consequence of the dramatic changes in the proportions of social contacts between younger and older individuals during the various stages of the DT [11, 12, 16, 17]. Notably, vertical and sexual transmission routes also showed complex patterns of evolution during the course of the DT, as signalled by the changes in route-specific incidences (Fig. 5b–d), even under the assumption of constant sexual behaviour rates. In the ancien régime, horizontal vs sexual transmission accounted for 85% and 8% of overall HBV incidence, respectively. During the HBV expansion phase, the two routes are predicted to diverge, with horizontal incidence growing to a maximum in the 1980s and sexual incidence declining to a minimum not long after the 1990s, when the horizontal route is predicted to account for almost 92% of total incidence. In the 1990s, after horizontal transmission has begun its long decline, sexual transmission inverts its trend, returning to ancien régime levels by around 2060; this pattern of incidence due to horizontal transmission is mirrored, at an order of magnitude lower, by that for vertical transmission, results well reflecting observations from The Gambia [36].

To separate the effects of the continuation of the DT from those of vaccination, we focus on the proportion of the overall population with persistent infection, which serves as a useful summary measure of burden of disease. Under the best model in the absence of any immunisation, this increased monotonically along the DT (in the medium variant scenario) to a peak of 17% in ~ 1992, i.e. a decade prior to the introduction of universal HBV vaccination in Senegal in 2005, before decreasing steadily thereafter (Fig. 6a, uppermost line), reaching levels as low as 5% in 2150.

Figure 6 also investigates the effects of possible vaccination programmes (all initiated in 2005) comparing our baseline DT scenario in SG (Fig. 6a) with the alternative scenario where SG total fertility is frozen at its maximal level observed in 1990–1995, namely before the fertility transition initiated (Fig. 6b), the scenario representing the worst case in terms of infection control given that it completely prevents the phase of HBV retreat. The differences are stark. If fertility remains constant at its highest level, the prevalence of chronic HBV in the absence of immunisation would remain nearly constant over time at its maximum level (Fig. 6b, uppermost line). At baseline coverage (46%), chronic HBV would cease its decline at a level of around 3% prevalence, whereas coverage in excess of 75% would be effective in bringing the infection under full control.

In contrast, the onset and completion of the DT (along the medium scenario) is predicted to “sustain” immunisation by yielding a substantial decline in the prevalence of chronic HBV (down to 1% by 2150) even with the baseline coverage, and bringing it to near elimination level with a coverage just in excess of 55%. This suggests the possibility of a strong synergy between vaccination and the HBV retreat phase triggered by the DT. The impact of a hypothetical alternative programme of vaccination at birth at the same rate of 46% shows an even steeper decrease in the prevalence of chronic infection (see Additional file 1: Text S4).

Finally, the impact of different patterns of DT in the future, as embedded in the “low” and “high” variants of UN projections (in contrast to the “medium” one), and the interplay of this with vaccination are also reported (Fig. 7). The persistent fall into below replacement fertility, as hypothesised in the low variant, would made control conditions easier but only after 2050.