Baseline characteristics of respondents
This study was designed to evaluate the impact of family structure and malaria intervention strategies on episodes of malaria in northern Uganda. A total of 193 households were surveyed for the study. From responses to survey questions, the average number of individuals per household was 3 (±2). This average was determined by extrapolating the declared number of individuals who spent the night before the survey to one (1) household. Of the 193 households surveyed, 111 with 388 individuals were headed by women and 82 with 217 individuals by men. Additionally, of the 605 total number of individuals captured in our data (∑ 388 and 217), 255 had malaria in the three months prior to this survey (42% of samples), 171 of which were children (67%).
Among the malaria intervention strategies included in our survey questionnaire were the availability of bed nets (coded as Number.BedNet for analysis), the number of individuals who used the nets at night (Use.BedNet) and extent of usage as measured by a count of individuals that spent the previous night to the interview day in a household (O.N.LstNight). Layered over these strategies were the following qualitative predictors: indoor residual insecticide spraying by respondents themselves (IRS) or by others (In.Res.Spr), testing bed nets – whether treated or not – before use (Test.BedNet), and impregnating or treating bed nets with insecticides (Treat.BedNet). Furthermore, household were asked the question: how do you control malaria vectors in the family to which several intervention strategies were offered in addition to, or as the only means of, controlling malaria. These included the followings: using only bed nets, cutting grass around homestead, draining stagnant water pools, relying on mosquito-repelling incense (from burning dried paste of pyrethrum powder in coils at night), keeping household doors closed, maintaining general cleanliness, applying shea butter on the skin, and not using solar lighting at night. Overall, obtaining questionnaire with all questions answered was possible for only 159 of the 193 households. That being said, to have had 80% of an indigent community in a sub-Saharan Africa country to respond to all questions asked here was remarkable, and has no bearing on management or data quality.
In Fig. 1 are provided brief descriptions of the data collected in this study. It can be seen that each of the 159 households which responded to all questions on the questionnaire had at least two bed nets (317/159) under which a total of 460 individuals (out of 535; 86%) slept the night before this survey (Fig. 1a). The observed 86% usage is empirical large from experience and perhaps a reflection a general acceptance of bed nets usage in the study area. This statement is supported by the fact that a correlation was observed between the number of individuals that spent the night under bed nets within each household and the total number of individuals counted per household (Pearson r = 0.654, p = 8.96e−21).
In computing the correlation above, (i) 10 households were excluded because more individuals were reported as sleeping under bed nets than those declared to have spent the previous night in the households, and (ii) 21 responses were excluded because they were other than a Yes or a No required to the question: Who slept under treated mosquito bed nets? The 31 excluded households owned at least one mosquito bed net (impregnated or treated with insecticides), applied IRS in their houses, and reported having someone in their midst suffer from malaria in the previous three months prior to interview.
Overall we believe the data collected here shows an active inclination towards taking measures against the spread of malaria in the study area, particularly the use of insecticides in some form. As can be seen from Fig. 1b, the number of households whose malaria controlling mechanism involve insecticides was 433 (i.e. sum: 137+123+173) out of a total of 516 (84%). The number 516 (vis-a- vis 193) is due to the fact that some households practiced up to three of the intervention measures shown.
As mentioned earlier, additional malaria controlling mechanism adopted for controlling mosquitoes – the carrier of malaria pathogens – were offered by respondents. They range from cutting grass around homestead and draining stagnant water pools to not having any strategy. By the number of times these additional control mechanisms were provided (Fig. 1c), the primary one appeared to be the use of a bed net (122 out of 177). This was followed unfortunately, albeit to a minor extent, by not having any form of preventive measure against malaria whatsoever – not even receiving treatment after falling sick (16/177).
Seven households kept grass around houses low ostensibly to deny mosquitoes hiding places during the day, or relied on medical treatment only after becoming sick. A few households reported combining two-to-three strategies to control malaria. Four households for example combined keeping grass low around houses with using bed nets, or keeping doors closed with draining stagnant water pool where mosquitoes might breed.
To assess the relative effectiveness of these different strategies, a simple ratio of the number of malaria episodes s and number of times a control mechanism was offered by respondents n was computed for each category for comparison (column s/n in Table 1). The smaller the value of this ratio in comparative terms, the more successful the intervention strategy was. In that context, relying on bed nets for controlling malaria, or combining bed net usage with (i) IRS, (ii) draining stagnant water pool, (iii) clearing grass around the house, or (iv) receiving medical treatment after falling sick had the smallest value at 1.33 and therefore the best strategies among the 22 studied. Note that the ratio was higher for households (i) that relied exclusively on treatment alone or (ii) that had no strategy (1.50 and 2.00, respectively) which clearly show that the use of bed nets is not only important in controlling malaria in the study area, but that prevention is better than getting treated after falling sick. Furthermore, the worse a household can do by this analysis is to have no strategy at all for controlling malaria – including receiving no medical treatment after falling sick.
.
Table 1
Comparing the effectiveness of malaria control strategies offered by respondents beyond indoor spraying.
| Malaria control | | Malaria episodes | | Ratio |
| Count (n) | | Children | Adults | Sum (s) | | s/n |
Bed nets only | 27 | | 21 | 15 | 36 | | 1.33 |
Bed nets + others | 9 | | 7 | 5 | 12 | | 1.33 |
Treatment only | 6 | | 9 | 0 | 9 | | 1.50 |
No strategy | 5 | | 6 | 4 | 10 | | 2.00 |
Overall, children bore the brunt of malaria in northern Uganda in the three months prior to this survey, irrespective of actions taken to control malaria at the household level.
Selecting Number Of Mca Dimension To Retain
The MCA technique was the method of choice here because (i) our questionnaire provided a 22-level categorical dataset containing a mixture of integers and factors, and (ii) non-linear associations were found among the observables by regression (not shown), meaning regression was not appropriate for analysis here. MCA did not only reveal the internal structure of the non-linear correlations observed (as will be shown below), but allowed us to assess relationships which would not have been obvious by regression.
Of the 22 levels of the categorical dataset collected here, MCA yielded 17 important dimensions by decomposing the total MCA inertia (variance in a multivariate dataset) into a possible maximum of 22. Five of these dimensions were identified to have little information for inclusion into the final data matrix used in further analysis. Classically, identifying these five dimensions would require factoring into the decomposed values the individual household contributions to the observed inertia. In normal practice though, these steps are assumed and all but the first 2 to 5 MCA dimensions are excluded from further analysis since many studies have shown that the five dimensions have most of the variations contained in data [23]. In line with that practice, (i) we further reduced the number of MCA dimensions for inclusion into further analysis from 12 (i.e., 17-5) to the first five dimensions (accounting for 14.41%, 9.70%, 7.92%, 6.35% and 5.88% of uncorrected inertia, respectively), and (ii) uncorrected inertia was used since the kind of information we expected from MCA, which was the locations of points in the 'cloud of individuals', are unaffected – corrected or not [23, 29].
Identifying The Most Important Malaria Intervention Strategies By Mca
Out of the 22 intervention strategies tested, we identified eight as the most significantly related to malaria in both children and adults in northern Uganda using the MCA technique. Identification was based on the quality of their representation along the five chosen dimensions which was high (see next paragraph below). These eight strategies were (i) the number of individuals staying overnight in a household (O.N.LstNight), (ii) the number of bed nets in the households (Number.BedNet), (iii) the number of individuals who actually used the bed nets (Use.BedNet), (iv) and (v) the number of households that do practice indoor residual spraying (In.Res.Spr and IRS), (vi) the number of households that test bed nets before use (Test.BedNet)), (vii) the number of households that impregnate or treat bed nets with insecticides before use (Treat.BedNet) and (viii) a collection of additional strategies offered by respondents assembled under malaria.control for analysis.
The overall qualities of how well the original 22 categories of predictors are represented by the first five MCA dimensions are displayed graphically in Fig. 2. This assessment of quality is needed here to support our contentions later on here that not only were the first three dimensions the most important in understanding strategies for controlling malaria in the study region, but that the eight listed predictors of malaria were the most significant. The coordinates of the predictors along the first five MCA dimensions (Dim 1, Dim 2, Dim 3, Dim 4 and Dim 5 - color coded for clarity) are displayed in Fig. 2a. Here, the further the coordinate of a predictor is from 0 along a dimension (negative or positive direction), the better the predictor is represented along the dimension.
The quality of the representations as measured by the values of Cos2 (squared cosine) for each predictor, and their contributions along the dimensions are also displayed in Fig. 2b and Fig. 2c, respectively. Cos2 measures the degree of association between predictor categories and each of the selected five dimensions. A predictor category is perfectly represented by two of more dimensions if (the sum of) Cos2 along those dimensions is closed to one (i.e. a simple 2-D scatter plot using coordinates of the predictor category and the identified pair of axes would adequately explain variance in the category data). Among multiple predictors, a higher relative contribution value for a predictor along any dimension means the dimension would be quite different without the predictor category (i.e., the predictor is important). In that context, results show that although some predictors appeared to be well represented (e.g., E/G along Dim 1 and Dim 3; D/G along Dim 3 and Dim 4), the qualities of their representation were low (Fig. 2b) and hence the predictors did not contribute much to those dimensions (Fig. 2c). On the other hand, variables such as those associated with In.Res.Spr., Test.BedNets and Treat.BedNets which are lowly represented in relative terms (Fig. 2a), their presentations were actually of high quality (Fig. 2b), hence higher contributions along the dimension representing them.
To definitively decide on the wellness of predictor representations along a pair of dimensions require such plots as shown in Fig. 3 in which the further a predictor is from 0, the better the representation along that dimension. One can then see that most information in this study are represented by the first two dimensions (Fig. 3a). That is, Dim 1 and 2 together represent adequately Test.BedNet, Treat.BedNet, malaria.control, IRS and I.Res.Spr, the five important predictors of malaria identified by MCA, and that Dim 3 and Dim 4 represent only the collection of malaria.control strategies offered by respondents.
When the information displayed in Fig. 2 and Fig. 3 are factored into (i) determining the most important predictors of malaria, and (ii) the number of dimensions to retain for further MCA and HCPC analyses, the following is clear: that attempts to understand malaria in the study area ought to focus on the representation of bed nets, their treatment and indoor residual spraying along the first two MCA dimensions, and perhaps the third as well (i.e., Dim 1, Dim 2 and Dim 3). With this in mind, below are the MCA results with potential interpretations.
Linking identified strategies to households by MCA factor map
In Fig. 4 and Fig. 5 are displayed MCA results as factor maps showing clusters of correlated households in groups. The selection of the first three MCA dimensions to retain (Dim 1, Dim 2 and Dim 3) comes directly from Fig. 3. The first dimension (Dim 1) is tied to the state of indoor residual spraying of insecticides by households (denoted by IRS and In.Res.Spr in Fig. 4). The second dimension (Dim 2) correlates with testing and treating bed nets; the third (Dim 3) with whether some measure of malaria control is used by household or not.
In details, Dim 1 (Fig. 4a) represents steady shift from indoor residual spraying (In.Res.Spr_yes: positive loading, I) to none (In.Res.Spr_no: negative loading, II). Thirty-four (34) of the 193 households surveyed scored strongly along Dim 1, 29 of which – by relating to the original data – used bed nets (85%). Concurrently, households that answered no to indoor residual spraying – 24 out of the 34 which score strongly along Dim 1 (In.Res.Spr_no and IRS_no: loading negatively) – do test bed nets before use (Test.BedNet_yes) and/or used bed nets impregnated or treated with insecticides (Treat.BedNet_yes). These households are split 1:1 between non-bed nets users (who score strongly along the dimension) and bed nets users (who score less strongly along Dim 1). This trend could perhaps be explained speculatively on the basis of perception that (i) insecticides are not acceptable to 100% of the 24 households if given a choice between spraying residual insecticides indoors and using treated bed nets, or to a minor extent that (ii) bed nets are not a requirement once indoor spraying has taken place. Lack of a bed net therefore would be the driving force to spraying insecticides indoor. Further study is required to explain this observation.
The locations of the households that scored strongly (positively or negatively) along Dim 1 are Parabongo in Agago District, Minakulu in Oyam District, Layamo in Kitgum District, and Awach and Unyama in Gulu District. The village-level distributions are as follows: 11 in Pacer Parish in Parabongo sub county, Agago District (10 in Jinja Village and one in Olwor Nguu), six in Adel Parish, Minakulu sub county, Oyam (all in Obapo village), two in Abanya Parish, Oyam District (Mot-mot Atwero and Bar Owor, Acaba,), two in Pakwelo Parish, Unyama sub county, Gulu District (Akonyibedo village) and two in Pagen Parish, Layamo sub county, Kitgum (Lelamur village), and one household in Gweng Diya Parish in Awach sub county, Gulu district (Pageya).
Dimension 2 (Dim 2) represented a gradual increase from those households that use treated bed nets (Treat.BedNet_yes) without testing them (Test.BedNet_no) to those households that test bed nets (Test.BedNet_yes) in addition to taking other malaria control measures such as draining water pool (B) and keeping grass short around the homestead (D). Dim 2 also separates households with bed nets (scoring strongly and negatively) from those without bed nets (scoring strongly and positively). For intervention, the households with bed nets rely mostly in treatment after contracting malaria while the households with no bed nets relies primarily on clearing grasses around the house to remove mosquito hiding places. All households represented by Dim 2 reported at least one household member with malaria in the three months prior to this survey.
To provide actionable items on the basis of alignment along Dim 1 and Dim 2, the vectorial relationship among each of those most important supplementary quantitative predictors are displayed in Fig. 4b. The alignment of the vectors suggests the following: that more than any other intervention strategies studied, incidences of malarial (Malaria.Adults and Malaria.Children) were strongly related to lack of bed nets or lack of use thereof (i.e., Number.BedNets and Use.BedNets vectors point in opposite directions to vectors representing malaria incidences in children and adults), and directly associated with the number of individuals spending a night together in a household (i.e., O.N.LstNight vector points approximately in the same directions as vectors representing malaria incidences). The results also show that more children are affected in northern Uganda than adults (vectors representing children is longer than that representing adults). These two findings provide ground for concerted effort to encourage usage of bed nets at night in northern Uganda, especially where children are involved.
Dimensions 3 and 4 (Dim 3 and Dim 4) represent differences combinations of the additional control strategies offered by respondents under the malaria.control category (Fig. 5). (That is, the coordinates of predictors (A-Z) are furthers from 0 along these two dimensions in; Figs. 2a and 3, meaning they are well represented along these two dimensions.) The results suggest that for households which use bed nets in combination with IRS, there is a difference between a households accept treatment after getting sick (A,H,G; blue) and those that do not include treatment after the onset of malaria (A, H; pink). We offer no plausible explanation for this difference at this time.
Hcpc Results
HCPC was used here to assess similarities among the 193 households surveyed with the hope of informing a uniform public outreach on malarial control in northern Uganda, or of detecting potentially localized misinformation or misperception of activities that households are being asked to engage in regarding malaria control in the region. The data source for HCPC was the first five MCA dimensions.
The results, displayed in Fig. 6, shows that our survey households do group naturally into four clusters (Fig. 6a). Cluster 1, the most distinct of the four clusters (i.e., the longest ‘arm’ separating it from the others in Fig. 6b), comprises households that use bed nets but sleep in houses not sprayed with insecticides (Fig. 6c). In cluster 2 are households with no bed nets but do spray residual insecticides indoors. Clusters 3 and 4 possess similar intrinsic attributes (IRS, I.Res.Spr and Treat.BedNet), but have enough of a difference to be apart (Test.BedNet) (Fig. 6c). The conclusion from HCPC is that, at the minimum, four categories of households exist in northern Uganda in relation to malaria intervention (Fig. 6c): one that relies on bed nets only to control malaria, a second which relies mostly on residual insecticides indoor spraying, a third which relies on indoor residual spraying along with bed nets insecticides pretreatment, and a forth category of households comprising very cautious members who implement all the four important malaria control strategies identified by HCPC from our data.
For quality assurance, the links between the household clusters and the four major categorical malaria predictors, as determined by chi-square test reported by p-values (and degrees of freedom in parenthesis), were Treat.BedNet: 3.60e−13 (3); Test.BedNet: 3.95e−10 (3); IRS: 2.52e−09 (3); and Malaria.Control 3.85e−05 (39). This means the clustering in Fig. 6 are highly significant on the basis of the major predictors identified by MCA.