Background characteristics of study population (Additional File 1)
Of the 5,588 household surveyed, 21% were from the SWZ, 14.4% from SEZ, 13.5% from SSZ, 17% from NWZ, 13.2% from NEZ and 20.9% from NCZ. 36% were urban and 64% rural (ratio 1:2). About two thirds of the households were sampled from communities with health facilities. Overall, distribution of wealth was unequal with slightly higher proportions in the upper two quartiles of the combined data, and the lowest two quartiles of the urban and rural data respectively.
In the combined data, wealth index also varied within and between the zones. In the southern zones (SWZ, SEZ and SSZ), a majority of the population fell within the upper two quartiles (>50%) while the northern zones have more households in the lower two quartiles (54% in NWZ and 76% in NEZ), except NCZ where more than 57% are also in 3rd and highest quartiles, the largest proportion (33.5%) being in the 3rd quartile.
Of the 4,625 households with valid data on household religion, 3,044 (65.8%) were Christian, 1,529 (33.1%) Muslim and 52 (1.1%) practiced other forms of religion. A majority of the households in the southern zones (SWZ, SEZ and SSZ) were Christians; the NEZ was largely Muslim (73%), while the ratio of Muslim to Christian households in NWZ and NCZ was about 1.2:1.
Males headed 93% of the households while females headed 7% of them. Generally, female family heads are commoner in the south (ranging from about 9% to 16%) than the north (1% – 3%). Overall mean family size was 5.4 (SD 3.0) ranging from 4.7 (SD 2.4) in SWZ to 6.5 (SD 3.4) in NWZ (Additional File
1).
15.3% of household members were children under-five and 14.5% were 5–14 years old. Age group 50 and above constituted 5.6%, typical for a developing country.
Univariate analysis
Household net ownership
Overall household ownership of any net was 23.9% (95% CI, 22.8%–25.1%) and 10.1% (95% CI, 9.2%–10.9%) for ITN (Additional file
2). A significantly (p < 0.0001) larger percentage of rural dwellers (22.8%) owned any nets compared to urban dwellers (18.3%), and were more likely (p < 0.0001) to own more than one net (11%) than urban dwellers (0.6%). Net ownership varied very significantly (<0.001) by region. Households in SSZ consistently own the least nets in all categories while NEZ households own more nets than other regions, except for more than one net, which is commoner in the NWZ (35.4%). However, information about net ownership was only available from 58% of the households in the NEZ, which could have resulted in an overestimate if the valid cases were more likely to own nets than the missing cases. Also, the SEZ has the 2
nd largest proportion of households with any net and the 3
rd largest for more than one net. Since the population in the sample from SEZ was mostly rural (98%, possibly attributable to an error in data collection) and rural households were more likely to own nets than urban households, this could have caused an overestimate of the true proportion for the zone.
When household net ownership by region (dichotomized N/S) was stratified by residence (urban or rural), residence was found to modify the association between region and ownership of any net. Rural households in the north were one-and-a-half times more likely to own nets than urban households in the same region (59.5% vs. 40.5%, RR, 1.47, 95% CI, 1.25–1.74), and in the south, about four times more likely than urban households (79.8% vs. 20.2%, RR, 3.94, 95% CI, 3.17–4.91) and this was statistically significant (p > 0.0001). There was no synergism according to the additive model.
These findings were similar for ITN ownership, with the rural households in the north almost twice as likely to own ITNs as the urban households (62.7% vs. 37.3%, RR, 1.67, 95% CI, 1.29–2.19) while rural households in the south were twice more likely to own ITN than their urban counterparts (67% vs. 33%, RR, 2.01, 95% CI, 1.52–2.73). No evidence for synergism was found on the additive model.
Using the wealth index generated for all households irrespective of their urban or rural status (combined), households that fell in the highest quartile were hardly more likely to own any net (26.3%) compared to the lowest quartile (24.1%) with rate ratio, RR of 0.91 (CI, 0.79–1.07); while the poorest households were more likely to have more than one net (13.3%) compared to the richest (10.0%) with a significant RR of 1.3. There was no significant difference in ownership of at least one ITN between the rich and the poor (p = 0.05) and though poorest households were more likely to own more than one ITN than those in the other quartiles, this was significant only for the 3rd quartile (RR 1.5, CI, 1.03–2.21).
Using wealth index generated for urban and rural areas separately, the scenario was similar for urban households and that of the combined data. More households in the lowest group owned any compared to the 2nd and 3rd, but those in the highest owned more nets than any other group (p = 0.003). When urban wealth index was dichotomised, the rich households (upper two quartiles) owned significantly (p = 0.045) more nets (22%) than the poorer households (lower two quartiles; 18%).
By rural wealth index, the 2nd quartile households had the largest proportion of nets in all categories. In the category for any net, this was followed by the highest quartile and the lowest had the least proportion of households with nets (<0.0001). When dichotomized, the richer households (upper two quartiles) were more likely to have nets (28.2%) than the poorer (lower two quartiles) households (23.2%).
Multivariable analysis
Ownership of any net
Variables that were significantly associated with net ownership at the univariate level; region, residence, and household wealth index; and other covariates; presence of at least one under-five child in household, religion, gender and age of household head, family size and presence of health facility in the community; were entered in an unconditional logistic regression model (combined Model A) to determine predictors of household net ownership using the stepwise forward likelihood ratio.
In the initial model, combined wealth index, residence, region, religion, household head's gender, presence of at least one under-five child in the household were significantly associated with net ownership; however, when 'presence of an educated eligible woman in the household' as a proxy for household education status, and interaction terms for region by residence, religion by region, and combined wealth index by household head's gender, were entered into the model, the final model did not include religion, however, education and family size became significant.
The odds of a household owning a net if there was an educated eligible woman in the household were 30% higher than a household without an educated eligible woman (p = 0.005). A unit increase in family size increased odds of ownership of any net more than twice (p < 0.0001) while controlling for all other variables; and if a household had at least one under-five child the odds of owning any net was about 60% higher than households with no under-five child (adjusted OR, 1.60, CI, 1.40–1.90). Living in a rural area raised ownership odds by 26% compared to urban residence (p = 0.021), and there was interaction between residence and region with a more than 50% increase in the odds of owning a net if the household was in the north and rural compared to urban households in the south (p = 0.024). Every unit rise in wealth index was found to increase ownership odds 1.24 times (CI, 1.15–1.34).
Data was split by residence and separate models were developed for urban and rural areas (Models B and C respectively) and outputs were generated for the north and south regions separately to identify predictors for the regions. Additional File
3 shows the adjusted odds ratios for the final variables in the different models.
In the final model for the urban region (Model B), the presence of an educated woman in the household raised the odds of owning a net by 42% in the north compared to those without (p < 0.0001), while this was not predictive in the south controlling for other variables in the model. The odds of owning a net were almost three times higher for households with a health facility in their community in the south than those without (OR, 2.88, CI, 1.54–5.37); while they were more than one-and-a-half times lower in the north (OR, 0.59, CI, 0.41–0.81), controlling for other variables in the model. Also, in the north, a unit increase in urban wealth index independently increased the odds of net ownership by 28% (OR, 1.28, CI, 1.09–1.50), a difference not found in the South. An under-five child in the household was very significantly predictive of net ownership in the south with an OR of 3.24 (CI, 1.99–5.27) but not in the north.
For rural areas (Model C), presence of under-five child in the household and health in the community, family size, presence of health facility in the community, wealth index, education and religion were predictors of ownership of any net in the south while only family size and religion were important for net ownership in the north.
Households with at least one under-five child was 1.7 times (p < 0.0001) as likely to own any net as those without, and those with a health facility in their community were 1.6 times (p = 0.001) more likely to have nets than those without health facilities, controlling for other variables in the model. The odds of net ownership increased by more than three-and-a-half in the south, and more than two-and-a-half in the north, for every additional family member (p > 0.0001 and 0.002 respectively).
Christians were more than twice more likely to own nets in the south (OR, 2.35, CI, 1.32–4.20) and twice less likely to in the north than Muslims (OR, 0.23, CI, 0.03–0.91). A unit rise in wealth index and having an educated woman in the house independently raised the odds of net ownership in the south by 33% and 55% respectively, with no effects in the north.
Ownership of ITN
Similar logistic regression models were developed in which the dependent variable was ownership of at least one ITN by the household. Variables analysed on the univariate level were adjusted for potential confounders in a combined model (Model I) and the data was then split by residence (urban, Model II/rural, Model III) and outputs generated for the north and the south (Additional File
4).
In the combined model, although region and combined wealth index were significant on the univariate level, when entered into the model with other covariates, including presence of an under-five child in the house, presence of an educated eligible respondent in the household (as a proxy for education status of the household), health facility status, gender of household head, residence and interaction terms for region by residence, education by combined wealth index and combined wealth index by household head's gender; these two variables were not significantly associated with household ITN ownership. The single variables in the final model were under-five child in the household; health facility status, household head's gender, education status and family size and these were found to predict ITN ownership. Additional File
5 shows the adjusted odds ratio, 95% confidence intervals and p-values for these predictors.
A household with an under-five child was more than on-and-a-half times as likely to own an ITN as those without (OR, 1.55, CI, 1.53–1.94), while presence of health facility in the community and male gender of household head independently increased the odds of ITN ownership by about 100% (p < 0.0001 and 0.023 respectively). A unit increase in family size raised the odds of ownership by about 90% (OR, 1.88, CI, 1.13–3.13) and the presence of an educated woman in the household increased the odds by 36% (P = 0.012).
Interestingly, there was evidence for interaction between region and residence in relation to ITN ownership; being a rural household in the north significantly raised the odds of owning ITN about twice (p < 0.0001), even though residence in itself was not predictive. There was also evidence for synergism on a multiplicative scale between education and household wealth index on ITN ownership. A unit rise in wealth index when there was an educated woman in the household led to a 15% rise in the odds for owning a net (p = 0.025). There was, however, no evidence for interaction between combined wealth index and gender of household head.
For urban households, under-five child in the household and household religion were significant predictors of ITN ownership in the south after controlling for family size, education (using presence of an educated woman in the household as proxy), urban wealth index, and urban wealth index by household head's gender, household head's gender and health facility status. Only education was significantly predictive in the north and strongly so (OR, 2.05, CI, 1.18–3.55); however when an interaction term for education by urban wealth index was introduced into the model, while it did not change the picture in the south, it was significant in the north (OR, 1.33, CI, 1.02–1.75), such that in the final model, the effect of education alone on ownership of nets was slightly reduced, but was still significant (OR, 1.93, CI, 1.01–3.38).
Christianity increased the odds of ITN ownership about seven fold compared to Islam in the south (OR, 6.78, CI, 1.64–29.12) while the presence of an under-five child in the household increased the odds 3.4 times (OR, 3.42, CI, 1.92–6.40).
Similarly, ITN ownership in the south was also dependent on under-five child in the household and household religion, and rural wealth index in addition. Model III showed that ownership odds significantly increased 1.5 times with under-five in the household (p < 0.0001) and six-fold when household religion was Christianity compared to Islam, each variable adjusted for others in the model.
Religion was predictive of ITN ownership in the rural north and south. Just like it was for urban households in the south, in the rural south, the odds of ownership was significantly increased if the household was Christian (OR, 5.95, CI, 1.45–24.4) compared to if they were Muslim. However in the north, the relationship was in the opposite direction; Christian households had twice lowered odds of ITN ownership in this region compared to Muslim households, even after controlling for possible interaction between religion and wealth index. There was evidence for interaction between wealth index and household head's gender in the north; the odds of net ownership reduced by 30% with a unit rise in wealth index when the household head was female. Other religions were unimportant in determining net ownership in both regions.
Number of nets in the household
Overall, household ownership of any net was 23.9%, with 10.9% households owning more than one net. ITN ownership was however 10.1%, with 4.7% reporting ownership of more than one ITN. This means that of all households with nets, 42% had ITN while 58% had untreated bed nets; and of those who have more than one net, 43% were treated nets while 57% were untreated (Additional File
5). The mean number of nets per household reporting any net ownership was 1.82 (SD, 1.11) and 1.17 (SD, 1.25) for the total sample population.
Utilization of mosquito nets by under-five children
Overall, 11.5% (95% CI, 10.4%–12.6%) and 1.7% (95% CI, 1.3%–2.2%) of all eligible children slept under any net or ITN respectively, the night before the survey. Younger children (<2 years old) were more likely to be put under any net than older children although this was not significant for ITN. There was no association between the gender of the child and the use of nets (p = 0.36) however, region was significantly associated with utilization (p = < 0.0001); southerners were more likely to keep their children under nets than northerners, who were more likely to own nets. Utilization was commoner among rural children than urban children for any net but did not differ for ITNs (Additional File
6).
Education was very significantly associated with net utilization among under-five children (p = < 0.0001). The rate of utilization increased monotonously with level of education; those with higher education than secondary were about thrice as likely to put their under-five children under a net as the uneducated, and twice as much secondary school leavers.
In the combined data, household wealth index was significantly associated with utilization of any net and ITN by children under-5 (p = 0.004 and 0.003 respectively). Children who fell in the upper 2 quartiles are more likely to have slept under a net the night before the survey than the lower two quartiles. Those in the highest quartile are 1.5 times and 1.3 times as likely to use any net or ITN than those in the lowest quartiles respectively. However, when rural and urban dwellers were separated according to their wealth indices, utilization of nets was independent of the household wealth, although utilization still varied with residence by caregiver's level of education.
Additional File
7 shows utilization rates of ITN and any net for rural and urban under-five children by caregiver's level of education. Reported use of any net was significantly (p < 0.0001) higher among rural children of the less educated and uneducated caregivers (56%) than those whose caregivers were more educated (44%). However, the proportion of under-five children who used ITN among them was significantly (p < 0.0001) higher for caregivers with higher levels of education (secondary and higher, 56.7%).
In urban households, caregivers with secondary and higher levels of education were significantly (p < 0.0001) more likely to put their children under any net and were more likely to protect them with ITN, although this was not statistically significant (p = 0.146).
Utilization of any net
Multivariate analysis of utilization of any net by under-five children using the combined data showed that fever/convulsion in the previous two weeks, availability of health facility, residence and caregiver's level of education were predictors of utilization of any net, controlling for other variables in the model (Model 1, Additional File
8). Combined wealth index (CWI), child's age, family size, religion, region and region by residence, were not predictive of utilization of any net by under-five children.
If a child had fever/convulsion in the last two weeks, the odds of using a net the night before the survey was about 1.3 times higher than if fever/convulsion was not reported (p < 0.0001). This finding was similar for presence of a health facility in the community (OR, 1.29, CI, 1.01–1.63). In addition, the odds of an under-five child sleeping under any net the night before the survey was 40% higher if the caregiver was educated compared to uneducated caregivers (p = 0.016). There was also evidence for interaction between CWI and caregiver's level of education; for an educated caregiver, a unit increase in wealth index resulted in a 30% rise in the odds of utilization of any net (OR, 1.29, CI, 1.14–1.45) independent of other variables.
Under-five children who live in rural areas were about one-and-a-half times as likely to use any net as their counterparts in urban areas. When the data was split by urban or rural status of residence, using the combined wealth index as a measure of household wealth in both residences, (Models 2 and 3 respectively), fever/convulsion episode in the child and wealth index were found to be important predictors in the rural areas. The odds of utilization of any net was 1.5 times higher in children who had fever two weeks before the survey than under-five children who did not have fever (OR, 1.49, CI, 1.11–1.99); while wealth index independently increased the odds by 17% for every unit rise (OR, 1.17, CI, 1.01–1.37).
The presence of health facility in the community, caregiver's level of education and age of the child were found to predict utilization of any net in urban children. The odds of an under-five child using a net the previous night was 2.3 times higher for children who live in communities with health facilities than those without heath facilities (p = 0.001). An educated caregiver had a 2.16 higher odds of putting her child under a net than the uneducated, controlling for other factors (p = 0.008).
Caregiver's education was also found to interact with wealth index; the odds of net use increased by 30% when caregiver was educated (OR, 1.3, CI, 1.05–1.57) per unit rise in wealth index compared to an uneducated caregiver. Children less than two years of age were almost twice as likely to be put under a net as children between two and five years (OR, 0.56, CI, 0.36–0.85). When the effect of urban and rural wealth indices were examined with separate models developed for urban and rural children, the predictors were overall similar to those found using the combined wealth index. However, the age of the child was not predictive in urban region and fever was not significant for rural children.
Utilization of ITN
Additional File
9 shows the variables in the final logistic regression models developed to predict the use of ITN by under-five children. Only valid cases (2009) were included in the analysis. In the combined data (Model X), presence of health facility in the community where a child lived strongly predicted the use of ITN the night before the survey, with an odds three times higher than where health facilities were absent (OR, 2.93, CI, 1.29–6.78). Among Christian caregivers, the odds of an under-five child sleeping under an ITN the night before the survey was more than three times higher than for Muslim caregivers. When a caregiver was educated, a unit rise in wealth index increased the odds of utilization of ITN by 43%, other factors controlled for.
Age, family size, combined wealth index, residence, region, fever/convulsion episodes and caregiver's level of education were not found to independently predict utilization of ITN by under-five children in the combined data. However, when spilt by residence and output for the combined model generated for urban and rural communities, region was predictive in urban areas (Model Y) while there was no variable in the final model for rural communities. Children living in the north had five-fold lower odds of sleeping under an ITN than children living in the south (OR, 0.18, CI, 0.06–0.54).
Using rural and urban wealth indices, separate models were used to check for predictors. These models showed the same result for urban communities with region predicting use of ITN; while an interaction was found between rural wealth index and caregiver's level of education. When a caregiver in a rural area was educated, a unit increase in rural wealth index raised the odds of an under-five child sleeping under an ITN by 57% compared to an uneducated caregiver (OR, 1.57, CI, 1.06–2.32).
Utilization of nets and fever and/or convulsion prevalence
Overall fever/convulsion prevalence among under-five-children who slept under any net was 1.2 times greater than those who did not use any net; and more than one-and-a-half times greater in those who used ITN the night before the survey (Additional File
10).
There was a positive association between fever episodes in the last two weeks and the use of nets in under-five-children. The odds of using any net the night before the survey were about one-and-a-half times higher in children who had fever and/or convulsion in the last two weeks than children who had no fever and/or convulsion; and this was significant (p = 0.013). The difference was explained by the use of ITN; children with history of fever were twice as likely to have been put under an ITN the previous night as those with no such history, p = 0.006, while there is no significant difference among children who used other nets.
Among all children who used any net, those with history of fever and/or convulsion were 1.7 times more likely to use an ITN than an ordinary net; this was however not statistically significant. Usage of other nets was independent of fever history. Users and non-users had a prevalence of roughly 30%.