Insufficient Ratio of Long-Lasting Insecticidal Nets to Household Members Limited Universal Usage in Western Kenya: A 2015 Cross-Sectional Study

Study site.

Villages in the highlands (Kapkangani) and lowlands (Miwani) of western Kenya were selected for inclusion in this study, following a cross-sectional survey performed by our team in these sites in August 2012. Kapkangani is in Nandi district in the western Kenyan highlands, where the altitude ranges from approximately 1,600 to 2,100 m. Malaria transmission in this area is low and unstable with acute seasonal peaks, generally after the rainy season in April–May. The population comprises primarily indigenous Kalenjin people and Luhya settlers from neighboring highland areas. Most people participate in small-scale agriculture, whereas some work as casual laborers on local tea estates. By contrast, Miwani is located on the Kano Plain in western Kenya, 30 km east of Kisumu, where the altitude is approximately 1,200 m above sea level. Malaria transmission in this area is holoendemic and occurs year-round. People are primarily of the Luo tribe and are small-scale farmers with some casual labor on nearby corporate sugarcane and rice farms. This research group had been working in the proposed study sites for approximately 2 years. Few previous studies on malaria had been conducted in these sites. The sites were chosen specifically for their differences in malaria prevalence: low and seasonal in the highlands of Kapkangani versus high year-round in the lowlands of Miwani.

Study personnel training.

A 4-day classroom training for all field personnel was conducted to provide an overview of the ethical conduct of research. The curriculum included modules on recruitment, informed consent, confidentiality, survey administration, and quality control for data collection. Two weeks of field training was conducted following classroom sessions to implement the data collection tools, identify gaps in the process, and give teams an opportunity to receive guidance from the principal investigator and other senior personnel before implementation.

Ethics statement.

This study was approved by the Kenya Medical Research Institute (KEMRI) as minimal risk under protocol SSC 2810. The University of Arizona Institutional Review Board deferred review of this study to KEMRI. Permission was sought from local leaders and community members through individual and community meetings before initiation of the study. Individuals aged 18 years and older provided informed consent for the study. Youths aged 7–17 years were asked for their signature or mark of assent for malaria testing.

Study site census enumeration.

A full census of the population was undertaken in 2013 and updated in July 2014 for five sublocations, two in the lowland area of Miwani and three in the highland area of Kapkangani. Household members were listed by age, gender, and relationship to household head. Latitude, longitude, and elevation were recorded using handheld Garmin global positioning system units.

Study design and sampling.

The present analysis involved a cross-sectional survey implemented shortly after the onset of the rainy season in 2015, with data collection occurring between June and August. Household sampling was linked to a previous 12-month cohort study in the study site.

The cohort study involved 250 households, 50 randomly selected from each of the five sublocations from the census list enumerated in 2013. To sample additional households for the cross-sectional survey, a list was created of all households from the updated 2014 census that fell within a 1-km buffer of any cohort house in ArcGIS (Esri, Redlands, CA). In the more densely populated lowlands, this included nearly all households in the study area. Additional households were sampled proportional to size for each village in the highlands and in the lowlands. Randomly ordered lists were generated for each village that included an oversample of households by 20% to supplement village sample sizes in case household members were all absent or declined to participate. Households were recruited until the desired sample size for each village was obtained.

Recruitment and data collection.

Male or female heads of households were approached in person by the field team staff for recruitment. A household was defined as individuals who regularly eat meals together. Households were included if they had resided in the study area for a period of at least 1 month. Households of all sizes and configurations were included.

The primary female caretaker was interviewed to list all household members and bed nets and complete associated household-level, individual-level, and bed net–level survey forms. In the absence of a female caretaker, the male head of the household was surveyed to obtain information. The surveys collected data about the socioeconomic status, bed net ownership, perceptions of malaria risk, barriers to use of various malaria control strategies, and net care/misuse. Individual data were collected on demographics, net use, and recent malaria illness. The bed net forms collected data on net age, quality, and use on the previous night.

Data management and analysis.

All field teams had a team leader who reviewed and initialed all forms before putting them in envelopes sorted by household. On returning the forms to the research office, a second level of quality control was conducted by the site lead. If discrepancies were identified, the forms were returned to the field team for correction the following day.

Socioeconomic status of the households was estimated using an asset index, with contributing items weighted based on the inverse frequency of their ownership in the community. These index scores were then grouped into quartiles to estimate relative wealth across the entire study population, defined at the household level. A household’s perceived malaria risk was calculated as a composite score from two questions which asked the respondent to rank the seriousness of malaria as a problem: 1) for the family and 2) for the community on scales of 1–5, with 1 representing “not at all serious” and 5 representing “extremely serious.” The sum of the two scores was then grouped as “low” (sum of 2–4), “moderate” (sum of 5–7), or “high” (sum of 8–10).

Analysis was performed with Statistical Analysis Systems version 9.4 (SAS Institute, Cary, NC) to assess gaps in LLIN access and use. Several common LLIN indicators were calculated, most predicated on the assumption that all household members have access to a net if there is at least one available for every two household members.^{10,15,21–23}

• 1. Proportion of households with at least one LLIN (household level)

• 2. Proportion of households with at least one LLIN for every two people (household level): that is, sufficient LLIN ownership to achieve the WHO recommendation for “universal coverage”

• 3. Proportion of population with access to an LLIN in their household (individual level): the proportion of each household with access to a net was calculated by multiplying the number of LLINs owned by two and dividing that by the number of reported household members, capped at the maximum coverage of 1.0. This value was assigned to all individuals in the household, and population means of these values were calculated.

• 4. Proportion of population that reported sleeping under an LLIN the previous night (individual level): the number of individuals reporting the use of an LLIN on the previous night divided by the number for whom the behavior was recorded.

• 5. Proportion of existing LLINs used the previous night (net level)

We assessed distributions and predictors for gaps in both access to nets at a household level and for household- and individual-level choices to use the LLINs at different levels of household access (ratio of nets to household members). Most outcomes were analyzed dichotomously: households with versus without any LLINs, households with sufficient versus insufficient access among those with at least one net, and individuals who reported using versus not using LLINs. We also performed an exploratory qualitative assessment of households underusing nets (i.e., those where fewer people used nets than had access, assuming two people per net) and households that had unused LLINs.

All potential predictors were categorical, except the total number of household members, which was evaluated both as a linear variable and categorically to allow for nonlinearity. Wilcoxon–Mann–Whitney and chi-squared tests of association were used for the continuous predictors and categorical predictors in univariate analyses, respectively. Fisher’s exact test was used when cells had expected values less than 5. Logistic regression was also used to evaluate the strength of each relationship. Given a priori knowledge about the major differences in demographics and malaria risks between the highlands and lowlands, all models were stratified by site.

Population characteristics.

The survey included 643 households in the highlands and 574 households from the lowlands, comprising 2,742 and 1,677 people, respectively. Households from the highlands and lowlands had significantly different distributions of all demographic, socioeconomic, and net-related variables (Table 1).

Men were underrepresented in the study. Approximately 72.5% of respondents for the household form were female. The household listings included more women (55.6%) than men (44.4%) in both the highlands and lowlands, and nearly half of the participants (47.9%) were adults aged > 15 years. Of the 4,419 reported household members, 3,834 (86.8%) were present and filled out individual forms reporting on bed net usage. Men were disproportionately likely to be absent at the time of the survey, with data on their net use missing; they accounted for 64.1% of the people who were missing net use data.

Progress on net indicators.

Distributions of LLIN indicators and characteristics by site (highlands versus lowlands) and sublocations from the 2015 survey are provided in Table 2. There was evidence of considerable improvement on all LLIN indicators in 2015 compared with a similar cross-sectional survey our team undertook in these same sites in August 2012,¹² although the reported discrepancies between the highlands and lowlands persisted. We found that the spatial heterogeneity was pronounced even between sublocations within the highlands and lowlands in 2015, with better LLIN indicators in Chepsonoi villages than Kiborgok or Tindinyo in the highlands.

Household ownership of at least 1 bed net essentially doubled from 2012 to 2015 in both the highlands (from 37% to 68%) and lowlands (from 53% to 98%), seemingly attributable to mass distribution campaigns. In 2012, only 8% of nets in the highlands and 36% in the lowlands were reportedly obtained from mass distributions (free distribution at the clinic was the primary source).¹² Only 3 years later, the vast majority (85% and 90%, respectively) were reported to have come from mass distributions. Progress on the proportion of households owning sufficient nets for a ratio of 1:2 people was even more striking from 2012 to 2015, increasing from only 5% to 37.5% in the highlands and 10% to 72% in the lowlands, but there remained a significant gap on this indicator. In the lowlands, where the prevalence of malaria is higher, an LLIN distribution had taken place in September 2014, 8 months before the survey, aiming to provide at least one net for every two individuals enumerated in the households without a per household cap/ceiling. Although 32% of households in the highlands did not own any nets in mid-2015, an LLIN distribution effort that coincided with this cross-sectional survey was increasing coverage levels. Of the ∼40% of highland households that reported knowledge of a community distribution campaign within the year of the survey, only 15.9% owned no nets and 28.3% owned insufficient numbers of nets compared with 42.3% and 33.1%, respectively, among the households that were unaware of the recent distribution campaign.

At the individual level, the proportion of the population with access to an LLIN was not calculated in 2012, but the proportion of people who reported using a bed net approximately doubled by 2015, from 22% to 54% in the highlands and from 48% to 94% in the lowlands. In the highlands, this was approximately equal to the estimated 53% of people with access to an LLIN in 2015 (metric not calculated in the 2012 study). In the lowlands, however, a greater proportion of people used nets than the 85% of the population estimated to have access to one. This pattern was reflected in the household use to access ratios: in the lowlands, only 4% of households underused nets (i.e., fewer people used nets than had access, assuming two potential users per net), compared with 21.5% of households in the highlands. In 2012, it was uncommon for nets to go unused, but it was even rarer in 2015 (0.9% of nets in the highlands and 1.8% in the lowlands).

Predictors of household-level net ownership.

Only 11 households in the lowlands (1.9%) did not own any bed nets in 2015. Eight of these 11 households only had one or two members, and only one of them had a child younger than 5 years living in the house. Eight of the households were in the lowest wealth quartile. Comparatively, in the highlands, nearly a third of households did not own any bed nets in 2015 (31.6%). Households in the lowest wealth quartile were significantly less likely to own any nets than those in higher wealth quartiles, those that could not afford to buy a net were significantly less likely to own a net than those that could, and households with poor construction quality were significantly less likely to own any nets than households with better construction quality. Details are provided in Supplemental Table 1.

Although mass distribution efforts had made great strides since 2012, and most households owned at least one LLIN, many households still had insufficient numbers of nets in both sites, particularly among large households (Figure 1). Of households with six or more members, only 12.3% and 8.7% had sufficient nets in the highlands and lowlands, respectively. Table 3 (highlands) and Table 4 (lowlands) present the distribution of insufficient versus sufficient bed net ownership ratios among households that owned nets and had at least three household members. Households with only one or two members were excluded from tables because ownership of one net inherently qualified the household as having a sufficient number. As indicated in Figure 1, increasing household size was the strongest predictor of having insufficient bed nets in both sites. Households in the lowlands tended to be smaller than those in the highlands, potentially contributing to the discrepancy by site for this indicator.

View Full Size

Long-lasting insecticidal net (LLIN) ownership by household size in highland and lowland sites, western Kenya, 2015. Note: It was impossible by definition for households with only one to two members to qualify as having an insufficient number of nets.

Citation: The American Journal of Tropical Medicine and Hygiene 102, 6; 10.4269/ajtmh.19-0119

• Download Figure
• Download figure as PowerPoint slide

Given that previous bed net distribution tended to target pregnant women and children younger than 5 years, we hypothesized that households in which there was at least one child younger than 5 years would be more likely to have sufficient nets, but this was not the case. Even after adjusting for household size, households with a child younger than 5 years had 1.65 times the odds (95% CI: 1.01–2.68) of having an insufficient number of bed nets than households that had no children younger than 5 years in the highlands, and 2.65 times the odds (95% CI: 1.60–4.40) in the lowlands. As expected based on previous data¹² and the results from the net ownership data presented earlier, low socioeconomic status was associated with insufficient net ownership in the highlands; however, there was weak evidence of a relationship between net ownership and socioeconomic status variables in the lowlands.

Given the concurrent mass distribution campaign in the highlands, we ran a sensitivity analysis on the predictors of insufficient net ownership. Despite a loss of precision due to the smaller sample sizes, there were no notable differences in predictor patterns among the ∼40% of households in which the interviewee was aware of the recent community distribution campaign compared with those that were not.

Predictors of individual bed net usage.

Figure 2 displays the relationship between number of nets owned, number of household members, and individual-level net usage for the highlands (Panel A) and the lowlands (Panel B), with the dotted line dividing households with access to at least one net for every two members (sufficient nets). In the highlands, the percentage of individuals who used nets was highest for the households with sufficient nets, and usage was poorer in households with insufficient nets. Alternatively, in the lowlands, the reported personal net use was high even in households where access was considered insufficient, with > 90% of people sleeping under bed nets in all households that owned at least one net, regardless of the number of members.

View Full Size

Long-lasting insecticidal net use and ratio of household size to the number of nets owned in (A) the highlands and (B) the lowlands of western Kenya.

Citation: The American Journal of Tropical Medicine and Hygiene 102, 6; 10.4269/ajtmh.19-0119

• Download Figure
• Download figure as PowerPoint slide

Predictors of individual-level net use were evaluated among people in households that owned at least one net, where nonuse of a net may reflect either individual preference or household prioritization of limited nets among members (Table 5). Nearly all (1,404 of 1,469 [95.6%]) lowland residents in houses with at least one net reported using a bed net on the previous night, compared with only 78.1% (1,273 of 1,630) in the highlands. Sufficient household-level net ownership was significantly associated with members using nets in both sites, although the relationship was stronger in the highlands than the lowlands, corresponding to observations from Figure 2. In the highlands, 95.1% of people used a net in households with sufficient nets, compared with only 66.2% of people in households with insufficient net ownership (odds ratio [OR] = 9.86, 95% CI: 6.78–14.35). In the lowlands, the corresponding numbers were 98.4% and 91.6% (OR = 5.56, 95% CI: 3.05–10.13). Furthermore, although few people reported having problems hanging a net over their sleep space overall in either site, those in the lowlands who reported having a problem were much less likely to have used a net (OR = 0.10, 95% CI: 0.05–0.18) than those who did not report problems, and many stated that the problem was “not having enough nets.” In the highlands, the opposite relationship was observed. People who reported having problems hanging a net were more likely to have used a net than those who did not report any problems (OR = 2.17, 95% CI: 1.15–4.13). Only one person said that “not having enough nets” was a problem; the most common problems reported were not having a bed or a place to hang the net.

Variables related to malaria risk were also associated with reported net use (Table 5). Although perceived severity of malaria was assessed at the household and not at the individual level, people in the lowlands who lived in households with high perceived risk were significantly more likely to use a net than those in households reporting low or moderate perceived risk. The relationship was inconsistent in the highlands, where households generally reported lower perceived risk than in the lowlands. Individual perceptions of risk may be influenced by the timing of last malaria diagnosis. People in the highlands were most likely to report using a net if they had been diagnosed with malaria in the past month, with use becoming increasingly less prevalent among people whose last malaria diagnosis was older. However, there was no significant association between the timing of last malaria diagnosis and net use in the lowlands.

Finally, individual demographic characteristics were also associated with reported net use. People who belonged to the nuclear family of the head of the household were significantly more likely to report using a net than secondary family members (aunts, uncles, nieces, and nephews) in both sites. Significantly fewer men reported sleeping under a net than women in all age categories. School-aged children (aged 5–15 years) made up more than 60% of all nonusers of nets and had roughly 0.3 times the odds of using a net as children younger than 5 years in both sites. Young adults (aged > 15–30 years) were also less likely to use nets than children younger than 5 years, particularly in the highlands. Notably, the number of nets owned by the household (i.e., access) modified the relationships between net use and age and gender (Figure 3, Table 6). The differences in net use by age or gender were not statistically significant in households that owned a sufficient number of nets in either the lowlands or highlands. However, relationship to the head of the household remained strongly predictive of net use in both households with insufficient and sufficient nets, with people who were not members of the nuclear family being significantly less likely to use nets in both settings.

View Full Size

Use of long-lasting insecticidal nets by (A) age, (B) gender, and (C) relationship to the head of the household, stratified by site and household LLIN ownership ratio. Note: Bars represent the 95% CI of each estimate using Wald for those with counts > 5 and the Clopper–Pearson exact method for those with counts ≤ 5.

Citation: The American Journal of Tropical Medicine and Hygiene 102, 6; 10.4269/ajtmh.19-0119

• Download Figure
• Download figure as PowerPoint slide

Unused and underused nets.

Although it was uncommon for nets to go unused by any household members in the 2012 survey (4.5% of nets in the lowlands and 7.3% in the highlands), essentially all nets were used in the 2015 survey. In 2015, there were only 24 nets in 17 households that had not been used on the previous night—eight of 870 nets in the highlands (0.92%) and 16 of 864 nets in the lowlands (1.85%). Of these 24 nets, 15 (62.5%) belonged to nine households where all members already reported sleeping under other nets on the previous night. Another four nets were reported not to have been used because the typical user had not slept in that space the previous night. No explanation was given by the four households that owned the remaining five nets as to why they had not been used the previous night, despite there being 10 people in the households who reported not having used a net. These five nets represented only 0.3% of all recorded nets in the study sample.

We further examined the composition of households reporting underused nets, that is, those that reported low net usage relative to their predicted access when assuming two users per net (Figure 4). There were 29 households in the highlands and four in the lowlands that averaged one or fewer users per net, despite there being household members who had not used a net the previous night. The composition was variable, but often consisted of a single head of household living with his/her school-aged or adult children, nonnuclear family members, or both, where either the head of the household (16 households) or a child (three households) used a net alone. There were another seven households in which both the head of the household and a spouse were present and used separate nets, whereas one or more of their school-aged/adult children or nonnuclear family members did not use a net.

View Full Size

Percent of long-lasting insecticidal net (LLIN) users vs. the percent of the household estimated to have access to an LLIN in the (A) highlands and (B) lowlands. Note: The percent with access was calculated using all declared household members as the denominator; the percent using a net was calculated among only those members whose net use on the previous night was known.

Citation: The American Journal of Tropical Medicine and Hygiene 102, 6; 10.4269/ajtmh.19-0119

• Download Figure
• Download figure as PowerPoint slide