Study design and area

The general design was a prospective study with the single net as the unit of observation and multiple cross-sectional surveys for evaluation of the primary outcomes, bio-assay and chemical residue analysis. The study protocol followed the WHO guidelines for phase III field trials [6] with minor modifications and compared the performance of the LLIN with that of a comparable mosquito net conventionally treated with the same insecticide as used in the LLIN. Study site was five villages in Kirongo Parish, Kyenjojo District, which have been described in detail previously [12]. In short, this is an rural area in Western Uganda with moderate climate at altitudes of 1350-1550 m and a meso- to hyperendemic malaria situation.

Nets and net treatment

The LLIN Interceptor^® was provided by BASF Corporation (Research Triangle Park, NC, USA). They were white, rectangular multifilament polyester nets of 75 denier and medium size (160 × 150 × 180 cm W × H × L). Long-lasting treatment was applied at production with FENDOZIN^®, a mixture of the insecticide alpha-cypermethrin with a binding polymer at a target dose for the insecticide of 6.7 g/kg or 200 mg/m²[13].

Nets for the conventional treatment were also white, rectangular multifilament polyester nets of 75 denier (Siamdutch Mosquito Netting Co., Bangkok, Thailand) but of size 190 × 150 × 180 cm (W × H × L). Net treatment was done by a team of trained dippers and supervised by one of the authors (AK). Nets were treated individually in basins using one sachet of 6 ml alpha-cypermethrin 6% (FENDONA^®, BASF, Midrand, South Africa) and a standard amount of water adequate for the size of net. Based on the content of the sachet of 360 mg of alpha-cypermethrin and the average size of the nets of 14.1 m² the target dose was 24.9 mg/m². Nets were dried lying flat on the ground without direct exposure to sunlight. The dipping team was provided with adequate protective gear.

Net distribution

A total of 200 LLIN and 100 conventionally treated nets were provided or prepared.

After treatment 10 of the conventionally treated nets were randomly selected for the baseline assessment. Similarly, 10 of the LLIN were also selected as baseline nets at this time. All nets for distribution to households (190 LLIN, 90 conventionally treated) were identified with a unique ID number written with wash resistant. The allocation of numbers to nets was random and only the principal investigator had the allocation list. In addition, each net was also marked with a water-soluble ink as a quality control for the assessment of washing.

From previous and on-going studies a complete household list for all 5 villages was available indicating the number of beds in the household and any study net already allocated. Based on these lists study nets were randomly allocated to households by the village health workers. The net allocation list was computerized and served as a net master list. Net allocation took place in May 2006.

Surveys

A survey assessing the demographic and socio-economic characteristics of all households participating in the LLIN studies was undertaken in May 2006 before net distribution.

Net follow-up surveys were then undertaken every six months in September or October and April or May with a total of eight surveys, the last being in April 2010. During the net surveys all remaining nets were assessed for usage, dirtiness (clean, slightly dirty, dirty or very dirty as subjectively judged by the interviewer), washing frequency during the past six months, method of washing and drying as well as number and size of any holes in the net. Holes were categorized in three groups:

Size 1 (finger size): Any hole not larger than 2 cm in maximum diameter

Size 2 (hand size): Any hole larger than 2 cm and less than 10 cm in maximum diameter

Size 3 (head size): Any hole larger than 10 cm in maximum diameter

Any loss of nets was also recorded and the net master list updated accordingly.

Net collections and sample preparation

From the master list nets were randomly selected for outcome evaluation with 2-3 replacement numbers drawn in case the selected nets could not be traced on the day of sampling. These lists were communicated to the field team and nets were collected after the net follow-up surveys in order to ensure that washing information for the collected net was obtained. Households received a new LLIN as replacement but these nets were not part of the study. LLIN collections were done after six, 12, 24, 36 and 42 months of follow-up with a target sample size of 40 nets each except for the sample after 42 months. As the study was originally only planned for 36 months this collection comprised of all remaining LLIN, which was 21. Conventionally treated nets were sampled at six and 12 months with a target of 40 nets each and then all remaining nets at month 24.

Sampled nets were prepared in the laboratory in the following way: each net was carefully inspected and the general condition, presence of the wash-control mark and number and sizes of holes recorded. Using specially prepared templates netting material was cut always from the same location on the net, i.e. on the long side of the net below the label and half way between roof and lower border. The size of the sample was 30 × 30 cm for bio-assay samples and 10 × 10 cm for chemical residue samples. The chemical residue sample was cut directly next to the bio-assay sample. The labelled samples were packed in aluminium foil and kept in a fridge at 4-8°C before transport to the laboratory.

For each net one sample was taken for each laboratory test except for the baseline nets for which one bio-assay and three chemical residue samples were obtained. The chemical residue samples were taken from different sides of the net to allow assessment of within-net variability of insecticide distribution.

Bio-assays

Bio-assays were carried out by the Centers of Disease Control, Atlanta, USA using WHO standardized procedures. For the tests 2-4 day old, unfed female Anopheles gambiae s.s. (Kisumu strain) were used. This species has been well established in culture for a long time and is known to be pyrethroid sensitive. Five mosquitoes were introduced into WHO cones at a time and four cones applied simultaneously onto the net sample with a three-minute exposure of the vectors. Tests were made at 25 ± 2°C under subdued light. After exposure, females were grouped into batches of 10 or 20 in 200 mL plastic cups and maintained at 28°C ± 2°C and 80% ± 10% relative humidity with honey solution provided. For each sample, a total of 40 mosquitoes were used. For each series a control was run with no exposure and results were only used if control mortality was less than 5%. Numbers of mosquitoes knocked down were recorded at 30 and 60 minutes and knock down rate at 60 minutes (KD60) calculated. Percentage mortalities were recorded after 24 hours using immediate and delayed mortality as defined by WHO guidelines [6], i.e. mosquitoes were scored as dead if they could not fly or stand upright on either the side or the bottom of the paper cups. Those that had lost one or more legs and could fly and stand upright without collapsing were considered to be alive.

In May 2009, the testing was shifted to the CDC partner Kenya Medical Research Institute in Kisumu using the same mosquito strain and methodology. However, due to the development of a resistance problem of the Kisumu strain the last bio-assays (42 months) were done at the Vector Control Reference Unit, National Institute for Communicable Diseases, Johannesburg, South Africa using the An. gambiae s.s. SUA strain. The methodology differed in that the WHO tubes for vector sensitivity testing were used as exposure device introducing the netting instead of the paper. Otherwise conditions were the same as in the CDC tests.

Chemical residue

Chemical residue analysis was done at the Wallon Agricultural Research Centre (CRA-W), Gembloux, Belgium (WHO Collaborating Centre for Quality Control of Pesticides) using the ISO 17025 accredited analytical method RESMM002. The samples were measured and weighed and then introduced into a 100 mL Erlenmeyer flask. Alpha-cypermethrin was extracted from the sample by heating under reflux for 60 minutes with 40 mL xylene. After cooling to ambient temperature the extract was quantitatively transferred into a 50 mL volumetric flask. The flask was filled up to volume with xylene. A 10 times dilution was achieved in xylene. The final extract was then analysed for determination of alpha-cypermethrin by Capillary Gas Chromatography with ⁶³Ni Electron Capture Detection (GC-ECD) using an external standard calibration. For each sample two chromatographic injections were performed and the mean reported as g/kg and mg/m² of alpha-cypermethin. Before the analysis of samples, the analytical method was successfully validated on its specificity, linearity of detector response, repeatability, accuracy and limit of quantification. During the analysis of samples, the performance of the analytical method was checked in order to validate the analytical results.

Data entry and analysis

All data was entered in an EpiData 3.1 data and then transferred to Stata 11.0 statistical software (Stata Corp., College Station, USA) for management and analysis. For proportions (rates) exact binomial 95% confidence intervals were used. For continuous variables either the arithmetic or geometric mean or median was used depending on the distribution of values compared to a normal distribution. For the core outcomes multivariate analysis was applied using a logit or linear regression model with all potential co-variates. For analysis of net survey results with repeated observations on the same net generalized estimation equations (gee) were used.

The primary outcome of net effectiveness was based on the bio-assay results using the following criteria:

Optimal effectiveness: KD60 ≥ 95% or functional mortality ≥ 80%

Minimal effectiveness: KD60 ≥ 75% or functional mortality ≥ 50%

While the optimal effectiveness is equivalent to the WHOPES evaluation criteria [6], the minimal effectiveness criteria were taken from an unpublished recommendation by WHO (Pierre Guillet).

The physical integrity of the nets was evaluated by applying a proportionate holes index (pHI) based on the number of holes per category and weighted as follows:

pHI = size 1 holes + (size 2 holes × 9) + (size 3 holes × 56) and then a mean hole index calculated for the sample or sub-sample. The multiplication factors were chosen to reflect the approximate surface areas of the hole sizes (4, 36 and 225 cm² respectively) resulting in one unit of the pHI being equivalent to 4 cm² of hole surface. Data were then grouped into three categories of pHI 0-24 (or maximum 100 cm² total hole surface) representing a net in good condition, pHI 25-299 (or maximum 0.1 m² total hole surface) for a torn net and pHI 300 or above for a severely torn net.

Attrition rate for the net cohort was estimated by adjusting for the potential loss from sampled nets. For each group of sampled nets the number of nets that would have been lost was calculated by applying the observed attrition rates from the remaining nets. These were then added to the actually observed attrition and divided by the total of nets distributed.

The study was conducted according to the principles of the Declaration of Helsinki and the International Guidelines for Ethical Review of Epidemiological Studies and approved by the Uganda Council of Science and Technology.

Household and net surveys

Of the 382 households involved in the multi-brand LLIN study in the five villages 211 (55%) had received an Interceptor^® LLIN or alpha-cypermethrin conventionally treated net. The proportion of households with only LLIN was 71.6%, those with conventionally treated nets (ITN) only 28.9% and with both 5.2%. The mean number of study nets per household was 1.3 with 25.1% having two and 3.3% three nets. There was no difference between number of nets received between households with only LLIN or only ITN.

The demographic characteristics of the population was not significantly different from the assessment six years earlier [12]. Heads of households were predominantly male with 21.8% female lead households. Average age was 45 years (male 42, female 52), family size 6.1 persons with 1.4 children under five years of age and 1.9 persons per bed or sleeping place. Educational level was better for male than female heads of households with 10.9% and 47.8% being illiterate respectively.

Houses had mainly tin roofs (88.6%), mudded walls (73.0%) and closable windows (91.5%). Fuel for cooking was firewood for 96.2% of households and for almost all families (98.6%) cooking was exclusively outside the main house, usually in a small kitchen hut (assessment during rains in May). A radio was owned by 86.3% and any means of transport by 60.2%, predominantly bicycles (58.3%). The mean wealth score based on assets, animals and land ownership was 27.1 (95% CI 25.9, 28.3) which indicated a statistically significant increase compared to 2000 when it had been 24.4 (23.1, 25.6) using the same methodology.

Demographic and socio-economic variables did not differ statistically between households that had received only LLIN or only ITN (p > 0.2) demonstrating the effectiveness of the random distribution of the study-nets.

During the study period a total of 175 of the 190 LLIN and 88 of the 90 ITN were sampled with details shown in Figure 1. In addition, 16 nets (14 LLIN) were lost to follow-up and five remained unsampled. Using the number of lost nets that would have been observed if the sampled nets had been exposed to the same loss rate for each time interval (see methods) the retention rate was calculated and is presented in Figure 1. Looking at the loss rather than retention the attrition rate after one year then was 1.4% (95% CI 0.2, 5.1), after two years 5.6% (2.5, 10.9), three years 12.1% (7.2, 18.7) and after three and a half years 20.1% (13.7, 27.8). The reason for loss was not recorded for all nets but reports from the villages suggest that reasons include nets stolen, burnt and taken to other locations (e.g. boarding school) in addition to being discarded because of damage. The proportion of remaining nets that were assessed in each of the eight net surveys varied between 94.2% and 100%.

Washing and aspect

Nets were generally washed in cold water (96.7%) in a basin (99.5%) and with local soap (95.1%) rather than a detergent and none of the nets was rubbed with or on rocks during washing. Drying occurred outside (98.6%) with about one third (37.3%) lying flat and 62.5% hanging. Washing and drying patterns did not change during the follow-up period and did not differ between the types of net.

Physical condition

Net use

After 6-12 months of use 93.3% of nets were reported to have been used every night. This proportion decreased to 87.1% after 18-24 months, 85.9% after 30-36 months and 81.2% after 36-42 months (p = 0.006). Among the not regularly used nets the proportion of those used not at all during the past six months also increased over time from 20% after 6-12 months to 33%, 56% and 66% in the following observation periods resulting in 12% of all assessed nets not being used at all after 36-42 months. The correlation between non-regular net use and physical condition of the net was even stronger increasing more than three-fold between nets in good condition (6.1% non-regular use) to severely damaged nets (19.0%). This was confirmed by a logit model of non-regular net use which showed a strong gradient for the physical condition with an Odds-ratio of 6.2 (95% CI 2.2, 16.4) for severely damaged compared to nets in good condition after controlling for age of net. In contrast, in the presence of the variable on physical condition the time factor was only relevant in the first 12 months but non-regular use did not increase with age of net thereafter. There was no difference in use between the LLIN and conventionally treated nets controlling for the varying time of observation between them.

Sampled nets

While at baseline arithmetic mean and median were very close (see Table 3) indicating an approximate normal distribution, the distribution of values from chemical residue tends to deviate from normal with increasing time of follow-up. Therefore, the median alpha-cypermethrin concentration was used for the assessment of the performance of the nets over time (Table 4). After three years of field use the median insecticide concentration for the LLIN was 2.12 g/kg or 73.8 mg/m² and after three and a half years 1.61 g/kg or 56.2 mg/m² implying a loss of 62% after three and 71% after 3.5 years respective the initial values. There was some fluctuation of the rate of decline but as shown in Figure 2 the loss rate was very similar to the other polyester LLIN tested in the same villages [12] with approximately 20% per year irrespective of the differing initial doses and deltamethrin being used in the other LLIN brands. After three years 94% (95% CI 81.3, 99.3) of the LLIN samples still had more than 15 mg/m² alpha-cypermethrin. After three and a half years of field use this figure was 81.0% (58.1, 94.6) and 95.2% (76.2, 99.9) of the LLIN had more than 3 mg/m². In contrast, the median alpha-cypermethrin concentration for the conventional ITN decreased by 69% after 12 months to a median of 7.5 mg/m² and by 93% after two years with 1.6 mg/m². At the end of year two none of the net samples for the ITN had more than 15 mg/m² and only 37% more than 3 mg/m² although the sample was only 8 nets and the confidence interval accordingly very wide (9, 75).

Net type	Time of follow-up
	6 months	12 months	24 months	36 months	42 months
LLIN
Number tested	38	37	37	36	21
Median mg/m2	193.0	143.0	135.7	73.8	56.2
IQR	166, 210	95, 177	81, 184	51, 105	17, 116
% > 15 mg/m2	100%	100%	97%	94%	81%
% > 3 mg/m2	100%	100%	97%	94%	95%
Conventional ITN
Number tested	40	40	8
Median mg/m2	13.9	7.5	1.6	Not done	Not done
IQR	2.3, 19.7	0.7, 13, 3	0.8, 7.4
% > 15 mg/m2	40%	18%	0%
% > 3 mg/m2	70%	53%	37%

Correlation between bio-assay results and active ingredient content was explored by using optimal effectiveness as "gold standard". Using chemical residue of more than 15 mg/m² as a test for optimal effectiveness had a sensitivity of 77.2% (95% CI 72, 83) and a specificity of 52.4% (30, 74). Decreasing the cutoff-level to > 3 mg/m² increased sensitivity to 89.0% (85, 93) but decreased specificity to 38.1% (18, 61). At the proportions of LLIN with optimal effectiveness observed (Table 4) the positive predictive value (ppv) of finding a LLIN sample to have > 3 mg/m² alpha-cypermethrin was 95.8% (95, 97) at three and 96.5% (96, 98) at three and a half years of follow-up. However, negative predictive values (npv) were only 18.1% (11, 36) and 15.4% (9, 32) respectively.

In order to assess potential factors that influence the performance of the nets a data set was prepared merging the results from chemical residue and bio-assay with information on use, washes, physical condition and aspect (cleanliness) from the net follow-up surveys and information on the household. From the 259 sampled nets complete information on these variables was available for 228 nets, 144 LLIN and 84 ITN. Regression analysis of insecticide content as a function of time of follow-up and the total number of washes the net had received before being sampled showed for the LLIN that each additional wash reduced insecticide content by 4.6 mg/m² (95% CI 1.8-7.5), i.e. given an average of 1.5 washes per year a loss of 6.9 mg/m² per year. The total average loss per year was 31.5 mg/m² (25.0, 37.9) meaning that washing was only responsible for approximately 22% of the annual loss. For the conventional ITN the loss per wash was very similar with 2.1 mg/m² (0.5, 3.6) or 3.2 mg/m² per year, but this represented 53% of the overall annual loss of 6.0 mg/m² (2.0, 10, 0).

A joint analysis of insecticide content using all nets and controlling for net type confirmed the significant negative influence of time (p < 0.0005) and a modest negative impact of washing (p = 0.1). Another factor that could be identified to impact on insecticide content of the net were houses with plastered or brick walls (p = 0.001) which had 21 mg/m² less insecticide than nets in houses with mudded walls. A marginal influence was seen by nets not used the last 6 months before sampling (p = 0.05) which had 35 mg/m² higher insecticide levels than regularly used nets, being clean at time of sampling (p = 0.05) which had 12 mg/m² less than dirty nets but with no differences between level of dirt, the head of household being non-literate (p = 0.8) with 12.3 mg/m² less insecticide content, and the number of children under five in the household (p = 0.1) indicating that with each additional child the average insecticide content was 4 mg/m² less. Physical condition showed an association with declining insecticide content initially but this effect disappeared when the interaction term between pHI and time since distribution was introduced (p = 0.07 for interaction). Other interaction terms did not reach significance level. All variables in the model explained 75% of the variability in the data. Variables tested but not included due to lack of significance were wealth quintiles and number of people in the household.

Finally, the possible impact of dirt on the net with respect to bio-assay results was explored. Whether knock-down rate, mortality rate or optimal effectiveness was used as an outcome variable, in all cases no significant impact of the level of dirt on the net was detectable after controlling for insecticide content.