Geographical hotspots and correlates of early sexual debut among women in Ghana

The data for the analysis was secondary data derived from the 2017 Ghana Maternal Health Survey (GMHS) [28]. The GMHS adopted a two-stage stratified cluster (Census enumeration areas) sampling design. A total of 900 (466 urban and 434 rural) clusters (census enumeration areas) were selected, with 30 households selected from each cluster, resulting in a total sample of 27,000 households. Within the selected households, 25,062 women aged 15–49 years were interviewed. This study covered 21,392 women for whom complete data was available. The study considered all women aged 15–49 years to examine generational changes in early sexual debut among women in Ghana. The 2017 GMHS collected detailed demographic and health information from respondents including marriage and sexual activity. The survey also collected information on the deaths of women aged 12–49 years, who died on or after 1 January 2012. The GMHS reported a median age at first sexual intercourse of 18.1 years for women aged 20–49 years, with 12.0% of these women having their first sexual intercourse by age 15 years.

Following Ghana’s Criminal Code (amendment) Act, 2003 (Act 646) [4], the outcome variable for the analysis was binary coded 1 if a woman aged between 15 and 49 years had her first sexual intercourse before attaining 16 years of age and 0 otherwise. The covariates for the analysis were selected based on evidence from the literature [2, 7, 12, 29, 30] and their availability in the GMHS data. The covariates were grouped into three categories: demographic, socioeconomic and pregnancy outcome factors to examine their associations with the spatial patterns of early sexual debut at the district level. The selected covariates, their classification and coding are shown in Table 1.

The Analysis of Variance (ANOVA) technique was used to examine the mean distribution of the continuous covariates (current age of respondent and parity) aggregated by respondents who had early sexual debut and those who did not have early sexual debut. The percentage distribution of respondents who had early sexual intercourse by the categorical covariates was examined, using the Chi-squared test to assess statistically significant differences. The Bayesian Geoadditive Semiparametric (BGS) regression technique was used to examine geospatial clustering of early sexual debut at the district level and the covariates associated with the observed clustering [31]. The BGS approach allows for simultaneous estimation of non-linear effects of the continuous covariates and the fixed effects of the categorical covariates in addition to the unobserved spatial effects, both spatially structured and unstructured [31]. The analysis covered the 216 districts created in 2012 and adopted for the 2017 GMHS.

The dependent variable \({y}_{ij}\) was coded 1 if woman i in district j reported having early sexual intercourse and 0 otherwise. Thus, the dependent variable follows a binomial distribution with the expected probability \({\pi }_{ij}\) of a woman having early sexual intercourse. Thus, the logistic model linking the probability \({\pi }_{ij}\) of having early sexual intercourse is of the formwhere \({\eta }_{ij}\) are the covariates of interest. If we have a vector \({x}_{ij}^{^{\prime}}={({x}_{ij1},\dots ,{x}_{ijk})}^{^{\prime}}\) of k continuous covariates and \({\lambda }_{ij}^{^{\prime}}={({\lambda }_{ij1},\dots {\lambda }_{ijd})}^{^{\prime}}\) a vector of d categorical covariates, then the predictor \({\eta }_{ij}\) can be specified aswhere \(\alpha\) is a vector of unknown regression coefficients for the categorical covariates, \({\lambda }_{ij}^{^{\prime}}\),\(\beta\) is a vector of unknown regression coefficients for the continuous covariates \({x}_{ij}^{^{\prime}}\).

The BGS framework, which replaces the strictly linear predictors with flexible semiparametric predictors was adopted to account for the non-linear effects of the continuous covariates and their spatial correlation with the outcome variable. The model is thus specified aswhere \({f}_{k}(x)\) are the non-linear smoothing function of the continuous variables \({x}_{ijk}\), and \({f}^{spat({S}_{i})}\) accounts for unobserved spatial heterogeneity at district j (j = 1, …, S), of which some may be spatially structured (correlated) and others unstructured (uncorrelated). The spatially structured effects show the effect of location by assuming that geographically close areas are more similar than distant areas, whilst the unstructured spatial effect accounts for spatial randomness in the model [32]. Equation 5, in this regard, is specified as.where \({f}^{str}\) is the structured spatial effects, and \({f}^{unstr}\) is the unstructured spatial effects and \({f}^{spat({S}_{i})}={f}^{str}+{f}^{unstr}\). The spatially structured effects depict the extent of clustering of early sexual intercourse at the district level and the associative effects of unaccounted covariates, which may be spatially clustered or random [32]. The smooth effects of continuous factors are modelled with P-spline priors, whilst the spatial effects are modelled using Markov random field priors.

The posterior modes of the structured spatial effects and their corresponding probabilities at 95% nominal level were used to examine spatial correlates of the outcome variable at the district level. The posterior probabilities at the 95% nominal level show districts where early sexual debut was statistically significantly high. That is, high positive and statistically significant estimates of the posterior mode show clustering of districts with high early sexual debut.

The estimated posterior mode of the spatial effects characterises unexplained spatially correlated covariate information. Thus, a sequential modelling approach was used to identify districts where the demographic, socioeconomic and pregnancy outcome covariates were spatially associated with the observed clustering of high early sexual debut. Model 0 was a null (constant) model. Model 1 accounted for only the spatial effects. Model 2 included the demographic covariates; Model 3 added the socioeconomic factors and Model 4 further added the pregnancy outcome predictors. Only covariates significant at p < 0.05 were retained in the model. The statistical software R was used for the analysis [33].

Table 2 shows the weighted percentage distribution of respondents who had first sexual intercourse before the legal age of 16 years by the demographic, socioeconomic and pregnancy outcomes covariates. Overall, 26.7% of women reported having their first sexual intercourse before attaining the age of 16 years. With regards to the demographic covariates, a significantly (p = 0.000) higher percentage of co-habiting women (32.5%, 95% CI = 31.0–33.9) and women not in union (27.6%, 95% CI = 26.5–28.7) reported having early sexual debut when compared to those currently married (24.0%, 95% CI = 23.2–24.8). The mean current age for women who had their first sexual intercourse before attaining age 16 years is 30.5 years (95% CI = 30.3–30.8), whilst that for those who did not have early sexual debut is 31.8 years (95% CI = 31.7–32.0). This suggests that early sexual initiation is more common among younger women compared to their older counterparts. On parity, the estimates show that women who initiate sex early are more likely to have had a higher number of children (mean = 3.2, 95% CI = 3.1–3.3) when compared to those who initiate sex after attaining the legal age (mean = 2.4, 95% CI = 2.4–2.4).

Considering the socioeconomic covariates, generally, the proportion who had early sexual debut decreases with increasing levels of educational attainment (Table 2). The estimates also shows that a significantly (p = 0.000) higher proportion of women of traditional believe initiated sex before age 16 years when compared to women of other beliefs. Further, women from poorer households were significantly (p = 0.000) more likely to have initiated sex early compared to those from richer households. Table 2 shows that women who are grandchildren to the head of the household were more likely to have initiated sex early compared to other relations. With regards to place of residence, women from rural areas (31.9%, 95% CI = 31.0–32.7) were significantly (p = 0.000) more likely to have had early sexual intercourse compared to their urban (21.5%, 95% CI = 20.7–22.3) counterparts. On source of information, Table 2 show that a higher proportion of women who do not read newspapers and those who cannot read, those who do not listen to radio and those who do not watch television, initiated sex early when compared to those who read newspapers, listen to radio or watch television at least once a week. With respect to pregnancy outcomes, the results show that women who initiated sex early were more likely to have had a miscarriage, abortion or stillbirth (Table 2).

The estimated posterior variance of the continuous covariates and the posterior odds ratios of the categorical covariates for a woman having early sexual intercourse and their corresponding 95% credible intervals are shown in Table 3, along with their model summary statistics. Interpretation of the model coefficients is based on the final model (Table 3, Model 4). The estimated deviance and AIC for Model 0 (null model) were 24,838.3 and 24,840, respectively (Table 3). When the spatial effects were included in the model (Model 1), the deviance and AIC reduced by 1039.9 and 755.0, respectively. The high reduction in the deviance and AIC after the spatial effects were included in the model indicates spatial clustering of early sexual intercourse at the district level. Figure 1a show districts where the posterior mode of the structured spatial effects was high, positive and statistically significant (clusters of high early sexual debut), at the 95% nominal level. The figure shows clustering of high early sexual debut in 19 districts. The observed hotspot districts are predominantly in the Oti region (Biakoye, Jasikan, Krachi East, Krachi West, Nkwanta South, Nkwanta North and Kadjebi districts) and Volta region (Kpando and Hohoe Municipals, Afadzato South and Krachi Nchumuru districts). Also, districts in the Bono East region (Sene West and Sene East), the Ashanti region (Amansie Central and Sekyere Afram Plains North districts), Eastern (Kwahu Afram Plains North and Kwahu Afram Plains South), Northern region (Kpandai district) and Central region (Upper Denkyira West district). An interesting spatial pattern observed was that, apart from the Upper Denkyira West district in the Central region and the Amansie Central district in the Ashanti region, all other districts with high early sexual debut were clustered along the mainstream of the Volta Lake (Fig. 1a).

The demographic covariates were included in Model 2 (Table 3), leading to a reduction of 2363.9 and 2311.0 in the AIC and BIC, respectively. The considerable decline in the AIC and BIC suggest that the demographic factors are spatially associated with the odds of a woman having an early sexual debut. The current age of the respondent and parity were observed to have a non-linear association with the observed spatial clustering of early sexual debut (Fig. 2a and b). Figure 2a shows higher odds of having early sexual intercourse for younger women which decreases exponentially with increasing age. This indicates that the younger generation have higher odds of engaging in early sexual intercourse when compared with the older generation. With regards to parity, Fig. 2b shows that women who had early sexual debut have higher odds of being of higher parities. The figure shows an increasing exponential relationship between the odds of having early sexual debut and women’s parity, indicating that women who had early sexual debut are more likely to have higher number of children. Further, marital status was observed to be statistically significantly associated with the probability of a woman having an early sexual intercourse. The estimates show that women who had early sexual intercourse have increased posterior odds ratio of 1.43 of being in co-habiting relationships and 1.69 of not being in union compared to those who were married. This suggest that women who had their first sexual intercourse after attaining the legal age are more likely to be married (Table 3), whilst those who had early sexual intercourse are likely to be co-habiting or not in union. The demographic factors were observed to be associated with the spatial clustering of early sexual debut in the Upper Denkyira West district in the Central region, the Amansie Central and Sekyere Afram Plains North in the Ashanti region, Nkwanta North in the Oti region, Sene West in the Bono East region and the Kpandai district in the Northern region (Fig. 1b).

The socioeconomic factors were added in in Model 3 (Table 3), leading to a decline of 381.5 and 328.6 in the AIC and BIC, respectively. The socioeconomic characteristics identified to be statistically significantly (p < 0.05) associated with the observed spatial clustering of early sexual debut were educational attainment and reading of newspapers/literacy (Table 3, Model 3). The estimates show that the higher the level of educational attainment, the low the odds of having early sexual debut. The estimated posterior odds ratios show that women with middle and primary education are not statistically significantly different from women with no formal education, however, women with junior high school (POR = 0.66, 95% CI = 0.59–0.74), senior high school (POR = 0.46, 95% CI = 0.39–0.55) and higher-level (POR = 0.30, 95% CI = 0.23–0.39) education have significantly lower odds of having early sexual debut when compared with those with no formal education. Considering frequency of reading newspapers/literacy, the results show that women who cannot read have increased odds of 27.0% of having early sexual debut compared to those who read newspapers at least once a week. This indicates that women who were illiterate were more likely to engage in early sexual intercourse. The socioeconomic factors were observed to be associated with the spatial clustering of early sexual debut in the Biakoye, Krachi East, Krachi West, Nkwanta South and Kadjebi districts in the Oti region and the Krachi Nchumuru district in the Volta region (Fig. 1b).

With regards to the pregnancy outcome covariates, the results show that women who engaged in early sexual intercourse have higher odds of having miscarriage, abortion and stillbirth. The estimated posterior odds ratios show that women who had early sexual debut are 1.36, 2.10 and 1.18 times more likely to have miscarriage, abortion and stillbirth, respectively, when compared to their counterpart who did not engage in early sexual intercourse. The pregnancy outcome covariates were associated with the observed spatial clustering of early sexual debut in the Kwahu Afram Plains North and Kwahu Afram Plains South districts in the Eastern region and Sene East in the Bono East region (Fig. 1b).

Although the Kpando and Hohoe Municipalities and the Afadzato South districts in the Volta region and the Jasikan district in the Oti region were observed among the spatial clusters of high early sexual debut, neither the demographic, socioeconomic nor pregnancy outcome covariates were significantly associated with the observed spatial clustering. Thus, the observed spatial clustering of high early sexual debut in these districts were unexplained by the selected covariates (Fig. 1b).