Household perception and infestation dynamics of bedbugs among residential communities and its potential distribution in Africa
Sample collectionA survey was conducted among the residents of nine counties in Kenya (Mombasa, Kisumu, Machakos, Nairobi, Makueni, Bomet, Kericho, Kiambu, and Narok) and GPS location coordinates were recorded and later used to build the predictive model (“Infestation dynamics of bedbugs in residential communities” section). These counties represent diversity in cultural practices, livelihood strategies (such as fishing, tourism, farming), and infrastructure development. Also, they comprise different altitudes above sea level, temperatures, and differing in average annual rainfall.Samples identification using morphological identification keysIn each county where the survey was conducted, bedbug samples was taken and preserved in ethanol 70% for morphological identification. Cimex belonging to Cimicidae family is the common genus adapted to human environment and reported throughout the world and comprising species such as Cimex lectularius and C. hemipterus that are hematophagous mainly feeding on human blood5. The key morphological features used in identifying bedbugs include: (1) the head has a labrum that appears as a free sclerite at the extreme anterior margin, ecdysial lines form a broad V, eyes project from the sides composed of several facets and the antennae are 4-segmented, (2) thorax is subdivided into prothorax, mesothorax and metathorax, (3) legs have all other normal parts except pulvilli and arolia, tarsus is 3-segmented with 2 simple claws, (4) the abdomen has 11 more-or-less segmented recognizable segments, 7 pairs of spiracles borne on the second to eighth segments, hosts the genital structures, paramere in males and mesospermalege in females45. Bedbug specimen morphological features were examined using Leica EZ24 HD dissecting microscope (Leica Microsystems, UK) and photos documented using the associated software.Survey for household’s knowledge and perceptions on bedbugsThis study was a community-based cross-sectional survey conducted from November–December 2020 with respect of the rules/guidlines introduced by the Ministry of Health to contain the COVID-19 pandemic in Kenya (wearing mask, social distance, washing hand, etc.). It was based on a stratified, systematic random sampling where 100 respondents were selected from each county.A total number of 900 respondents were randomly selected and the household head or the representative showing willingness and consent was interviewed face-to-face. The interview was conducted using a semi-structured questionnaire prepared in the English language (Appendix A). The questionnaire was translated into the local native language (Kiswahili) to avoid biasness and improve the understanding between the enumerator and the respondent. Prior to the commencement of the survey and authentic data collection, a pre-testing exercise was performed by training enumerators on a similar socio-demographic pattern. This was useful for improving the quality of data, ensuring validity, familiarizing the enumerators with the questionnaire, and data handling.The information collected using the semi-structured questionnaire included residents’ socio-economic profiles, knowledge, and perceptions on the pest, bedbug incidence, and management practices. The socio-economic profile factors addressed in the survey comprised gender, age, education, access to basic social amenities, and household size. The study also prioritized the financial consequences, the severity of the bites, perceptions of respondents on the pest, and management practices for its control.Survey data were checked for errors, completeness, summarized, and entered in Microsoft-Excel. It was then cleaned and transferred to Statistical Package for Social Science (SPSS) version 25 software (IBM Corp., Armonk, NY) for purposes of descriptive statistics (means and percentages).In contrast, in instances where more than one reason was given for a single question, percentages were calculated based on each group of similar responses. Chi-square was performed to determine the differences regarding socio-demographic characteristics, knowledge, and perceptions on bedbugs and control practices. Additionally, data were disaggregated by gender and age categories to understand the existing differences among the various respondent categories. Besides, F-test statistics was performed on the ages of respondents to determine the mean, standard deviation and statistical significance. The level of significance was considered when the p-value was below 5%.Infestation dynamics model of bedbugModel simulation assumptionsHouses infestation dynamics was studied following Susceptible-Infested-Treatment (SIT) model46. Therefore, houses in the community are classified into three groups: susceptible, infested or treated. Within a house, bedbug population dynamics was ignored, while it was considered from one house to another where infested houses have some potential to spread the infestation to other houses in the community. A population of bedbugs in an infested house has some probability per unit of time of becoming extinct either naturally or after treatment. In the infestation dynamics, the rate of house infestation depends on the number of infested houses, the movement of people from one house to another and the proportion of treated houses in the community. We assume that infested houses (I) spread the infestation at the rate β and only a fraction S/N of the houses is susceptible (S) to infestation. Infested houses become extinct at a certain rate known as rate γ. Infested houses are treated at the rate τ and the protection conferred is lost at the rate α. Ordinary differential equation developed to study SIT model were used in this study46. All the models used have the generic formulations displayed below:$$frac{dS}{dt}=frac{beta }{N}SI+gamma I+alpha T$$
(1)
$$frac{dI}{dt}=frac{beta }{N}SI-(gamma +tau )I$$
(2)
$$frac{dT}{dt}=tau I-alpha T$$
(3)
where β > 0, τ > 0, α ≥ 0 and γ > 0. The total population size is N = S(t) + I(t) + t(t). The initial conditions satisfy at S(0) > 0, I(0) > 0, T(0) ≥ 0 and S(0) + I(0) = N, where N is the constant total population size, dN/dt = 0.Infestation dynamics models implementationThe method used to implement the infestation dynamics model of the pest is based on the system thinking approach with its archetypes [Causal Loop Diagram (CLD), Reinforcing (R) and Balancing (B)] by a mental and holistic conceptual framework. This is important for mapping how the variables, issues, and processes influence each other in the complex interactions of bedbugs within and between houses and their impacts. Despite these archetypes being qualitative, they are necessary for elucidating and disclosing the basic feedback configurations that occur in houses and their environs when infested with pests like bedbugs. A dynamic model was generated by converting the causal loop diagram (CLD) obtained using stocks, flows, auxiliary links, and clouds. Consequently, these in turn were translated into coupled differential equations for simulations.The SIT model was translated into causal loop diagram where arrows show the cause-effect relations where positive sign indicates direct proportionality of cause and effect while negative sign shows inverse proportionality relations, and two different scenarios have been assessed: (1) homogeneous houses where there is a single community of houses of the same quality, and (2) heterogeneous houses where there is a community of good and bad houses. Ancient houses presenting slits/fissures with less cleanliness and filled with old or secondhand furniture at low grade are considered bad houses as they may sustain high level of bedbug infestation; and new houses don’t provide well enough conditions for bedbug population to survive, and they are called in the model good houses47. Bad houses are considered to act as sources while good houses act as sinks, but all together are randomly distributed where each house has the same probability to contact good or bad houses.In the scenarios of homogeneous houses, the causal loop diagram (Fig. 7) has two feedback loops: (a) one positive, as the number of infested houses increases, the probability to get susceptible houses infested also increases resulting in infested houses increase; (b) one negative, as the infested houses increases, the treated houses increase resulting in susceptible houses decrease. The causal loop diagram is displayed in Fig. 7A while Fig. 7B showed the stocked and flows diagram and axillary variables obtained from causal loop diagram.Figure 7Susceptible-Infested-Treatment (SIT) model translated into causal loop diagram (A) and stock and flow diagram (B) for homogeneous houses and causal loop diagram (C) and stock and flow diagram (D) for heterogeneous houses in the community.Full size imageSusceptible, infested, and treated houses are stocks in the system, representing the number of houses susceptible, infested, and treated, respectively at a given point of time. The rates represent in and out-flows of the diagram. Auxiliary and constants that drive the behavior of the system were connected using information arrows within them and flows and stocks to represent the relations among variables in terms of equations.In the scenarios of heterogeneous houses, the causal loop diagram (Fig. 7C) comes with the two previous feedback loops but for each category of house. In addition, there is a fifth feedback loop that connect bad house to good house and vice versa.Therefore, as the infested bad houses increase, the probability to infest good houses increases. The more they are exposed the more they get infested. In turn, as the infested good houses increase, the chance to infest susceptible bad houses increases and the more they are exposed, the more they get infested, resulting in the increase of infested bad houses. The stocks and flows diagram of each of the two categories of houses occurred with interconnexion relationships between the two categories (Fig. 7D).Models’ simulationsThe survey data (“Bedbug Genus identification” section) on prevalence, knowledge, perceptions and self-reported; in addition, the respondents’ reported control mechanisms and their average time of effectiveness (Appendix B, Table S1) were used for model simulations. The different control methods reported were reclassified in three control approaches: chemical control, other control methods (including exposure to direct sunlight, use of hot water, painting, application of diesel, paraffin and wood ash, use of Aloe Vera extract and Herbs), and combination of chemical and other control methods. All the models commodities and units were checked before performing the simulations. Simulation and implementation of the models were done using Vensim PLP 8.1 platform (Ventana systems, Harvard, USA). It consists of a graphical environment that usually permits drawing of Causal Loop Diagram (CLD), stocks, flow diagrams and to carry out simulations. After we simulated the infestation dynamics under the two scenarios, we explored the effect of the different control methods.Spatial distribution analysis of bedbugs using MaxEnt modelEnvironmental data for MaxEntThe environmental variables used as the other maxent input were obtained by deriving bioclimatic, land cover, and elevation data. Bioclimatic variables and elevation (Digital Elevation Model; DEM) data were obtained from the Global Climate Data official website, Worldclim (http://www.worldclim.org/bioclim.htm)48 including 19 bioclimatic variables (Appendix B, Table S2). The land cover data were downloaded from the Global Land Cover Facility (GLCF).In order to reduce collinearity between predictors, a collinearity test was performed on all the variables by filtering them according to the following steps36: firstly, the MaxEnt model was run using the distribution data of bedbugs and 19 bioclimatic variables to obtain the percent contribution of each variable to the preliminary prediction results. Secondly, following the generation of the percentage contribution of all the variables, we then imported all distribution points in Arc-GIS and extracted the attribute values of the 19 variables. Furthermore, the “virtual species” package49 in R-software (R Foundation for Statistical Computing, Vienna, Australia) was used to explore the extracted variables’ clusters spatial correlation using Pearson’s correlation coefficient and the cluster tree (Fig. 8). Thus, the final number of predictor variables after screening was 5 establishing the potential geographical distribution of bedbug, which includes Temperature Seasonality (bio4), Precipitation of Driest Month (bio14), Temperature Annual Range (bio7), Precipitation of Driest Quarter (bio17) and Precipitation of Warmest Quarter (bio18) (Appendix B, Table S2). The land cover was considered because studies have shown its importance on insect spatial distribution50,51,52 and it was setled as a categorical variable53. Elevation was selected as variable because it greatly influences species’ occurrence and dispersal by affecting the temperature, precipitation, vegetation, and sun characteristics (direction, intensity, etc.) on the earth’s surface54,55,56. The study variables had different resolutions and were therefore, resampled to 1 km. The variables were clipped to Kenya and Africa boundaries and converted to ASCII (Stands for “American Standard Code for Information Interchange”) format using the ‘raster’ package49 in R statistical software (R Foundation for Statistical Computing, Vienna, Australia).Figure 8Key model predictor variables.Full size imageDistribution modelling in Kenya and AfricaIn our study, we used the maximum entropy distribution modelling method. This is because it has been recommended to have the ability to perform best and remain effective despite the use of small sample size relative to the other modelling methods57.Our selected bioclimatic variables (5) and occurrence/prevalence data for bedbugs were then imported into MaxEnt model and the options of ‘Create response curves’ and ‘Do jackknife’ were selected to measure variable importance’ options. The model output file was selected as ‘Logistic’, the commonly used approach is the random portioning of distribution datasets into ‘training’, and ‘test’ sets57,58. MaxEnt model was run with a total number of 5000 iterations and five replicates for better convergence of the model and rescaled within the range of 0–1000 suitability scores using ‘raster’ package49 in R statistical software (R Foundation for Statistical Computing, Vienna, Australia).The modelling performance/MaxEnt accuracy was evaluated by choosing the area under the receiver operating characteristics (ROC) curve (AUC) as the estimation index. This was important for the calibration and validation of the robustness of MaxEnt model evaluation. Furthermore, the area under the ROC curve (AUC) was necessary as an additional precision analysis59. The range of AUC values greater than 0.7 was considered a fair model performance, while those greater than 0.9 indicated that the model was considered an excellent model performance. Therefore, by considering the AUC values, the excellently performing model was selected to analyze the suitability of bedbugs in Kenya and Africa59,60,61,62.The ASCII format output was then imported into QGIS 3.10.2 (using the QGIS 3.10.2 software, https://qgis.org/downloads/), following its conversion into a raster format file using R software. This was useful for the classification and visualization of the distribution area63,64. The potential suitable distribution of bedbugs was extracted using the Kenyan and African maps. At the same time, Jenks’ natural breaks were also used to reclassify and classify the suitability into five categories, namely: unsuitable (P More
