Sample
At baseline a total of 998 households (98% of target) were recruited and 1,988 campaign LLIN labelled for follow up, 726 in Inhambane (93% of target), 601 in Tete (77%) and 661 (85%) in Nampula. The slightly lower recruitment rates in Tete and Nampula were due to a lower household size than expected. Table 1 summarizes the follow-up status of households and cohort LLIN. The most important reasons for loss of households in Nampula and Inhambane was that they had lost all their labelled LLIN meaning no further follow-up was needed. In contrast, most common reason for loss in Tete was households not present as families there spent extended time at far away farms or in mining areas. In addition, during the last survey two clusters could not be reached by the field teams due to flooding of the roads. The second most common reason was households moving away from the community. This was particularly high in Nampula with 18% of households. Only 17 households refused participation (2%), eight in Tete and nine in Nampula. Due to the high absentee rate in Tete, the proportion of cohort LLIN with a definite outcome there was only 45% compared to 72% in Nampula and 80% in Inhambane. At 36 months, data from two clusters in Tete were excluded from final analysis due to concerns related to data quality.
Table 1 Follow-up status of households and campaign cohort LLIN at final survey
Variable
|
Inhambane
|
Tete
|
Nampula
|
Households
|
N=340
|
N=333
|
N=325
|
Still has any campaign LLIN
|
54.1 (48.3-59.9)
|
36.3 (27.4-46.4)
|
34.5 (24.7-45.8)
|
Lost all their campaign LLIN
|
28.8 (22.3-36.3)
|
16.5 (10.8-24.4)
|
39.1 (28.8-50.4)
|
Moved away
|
8.2 (6.3-10.7)
|
9.9 ( 6.6-14.7)
|
17.5 (14.4-21.3)
|
Refused
|
0.0
|
2.4 ( 1.2-4.7)
|
2.8 ( 1.2-6.2)
|
Nobody home at survey or not reached
|
8.8 (5.9-13.0)
|
34.8 (24.2-47.2)
|
6.2 (3.4-10.9)
|
|
|
|
|
Campaign cohort LLIN
|
N=726
|
N=601
|
N=661
|
Known outcome
|
79.8 (74.9-83.9)
|
44.9 (34.1-56.3)
|
72.0 (64.4-78.5)
|
Unknown outcome
|
20.3 (16.2-25.1)
|
55.1 (43.7-65.9)
|
28.0 (21.5-35.6)
|
Household moved away or refused
|
7.2 (5.5-9.3)
|
11.8 (7.8-17.5)
|
19.5 (15.2-24.7)
|
Net used elsewhere
|
0
|
3.5 (1.2-9.4)
|
0.2 (0.0-1.1)
|
Fate of net unknown
|
13.1 (9.3-18.1)
|
39.6 (28.0-52.6)
|
8.3 (5.4-12.6)
|
Socio-demographics
Comparing households that participated in both the baseline and 36-month surveys (N=550) the data was explored for differences between provinces (Additional File 1). Household size was significantly larger in Inhambane (5.0) than in Tete (4.0) or Nampula (4.1). Inhambane also had slightly older and more educated heads of household, and more female-headed households.
Population age structure was similar. There was a clear difference between the three sites across all indicators showing that Jangamo district in Inhambane was economically significantly better off than the other two sites.
Other indicators confirm the differences between sites. Access to safe water was 100% in Inhambane and 79% and 89% in Tete and Nampula respectively. Any form of latrine was available for 97% of households in Inhambane compared to 67-75% in the other two sites. Ownership of any phone was higher in Inhambane (71%) and Nampula (58%), but lower in Tete (29%) where in some parts of Changara District there is no coverage. Similarly, ownership of smartphones was 18% in Inhambane and only 4% in Tete and 1% in Nampula. A similar difference was seen for other “luxury” household assets such as television (40% vs. 7% and 15% respectively), refrigerator (17% vs. 5% and 5%, respectively) and fan (12% vs. 3% and 4%, respectively). There was no evidence of a significant change in the socio-economic status of households during the study period.
Quality of housing was more similar with mainly thatch or grass roofs, but the wealth difference can be seen in the floor materials. In Inhambane 78% of houses had floors made from improved materials compared to 15% in Tete and 18% in Nampula.
Determinants of durability
Factors that have previously been shown to be associated with LLIN durability were explored. These can be divided into factors of the net use environment in the household, knowledge and attitudes towards net care and repair of the household respondent, net handling and washing, and type of sleeping place. Household-level factors depended on the information provided by the respondents and these were in Nampula in 66% the head of household, 29% the spouse and 5% other adult household members. In Tete the respective rates were 57%, 37% and 7% and in Inhambane 42%, 43% and 15%, respectively.
Perceived presence of rodents was generally very high and highest in Nampula where at least 90% of household respondents were aware of the presence of rodents at all time points, followed by Inhambane with a consistent reporting of rodent presence by three quarters of respondents. Only in Tete was there some variation with lower values at baseline (48%) and at 12 months (29%) and increased reported rodent presence in the last two surveys (62% and 71%).
Other household level factors were calculated across all surveys and are presented in Table 2.
Table 2: Net use environment at household level across all survey rounds. Results were aggregated across all four surveys i.e. “never” = responded with "never" in all surveys the household participated; “at times” = household reported the behaviour as “sometimes” in at least one survey round or had conflicting statements; “always” = responded with "always" in all surveys the household participated in. Exposure and attitude were similarly aggregated, i.e. “once” = reported exposure or positive attitude score at one of the four survey rounds; “twice or more” = at two or more survey rounds.
Variable
|
Inhambane
|
Tete
|
Nampula
|
P-value
for site comparison
|
Households
|
N=245
% (95% CI)
|
N=132 % (95% CI)
|
N=173 % (95% CI)
|
Storing of food in sleeping rooms
Never
At times
Always
|
41.3 (33.7-49.3)
54.7 (47.6-61.6)
4.1 ( 2.2 - 7.5)
|
41.2 (29.7-53.7)
54.9 (44.3-65.2)
3.9 ( 2.0 - 7.2)
|
10.1 ( 4.7-20.2)
55.4 (43.7-66.5)
34.5 (23.1-48.1)
|
P=0.001
|
Cooking in sleeping room
Never
At times
Always
|
76.1 (62.9-85.7)
22.1 (13.7-33.7)
1.8 ( 0.6- 5.4)
|
45.6 (33.8-57.8)
43.6 (33.7-54.1)
10.8 (6.0-18.7)
|
18.2 (7.7-37.2)
58.7 (42.6-73.1)
23.1 (11.9-40.1)
|
P=0.001
|
Exposure to net use or care messages
Never
Once
Twice or more
|
9.2 (4.4-18.4)
17.4 (11.8-24.9)
73.4 (60.1-83.5)
|
36.4 (25.9-48.3)
36.5 (28.3-45.6)
27.1 (16.7-41.0)
|
20.9 (13.2-31.4)
39.1 (32.2-46.5)
40.0 (30.4-50.5)
|
P=0.001
|
Very positive net care attitude (score >1.0)
Never
Once
Twice or more
|
62.6 (42.1-79.3)
17.2 (11.5-34.2)
20.2 ( 9.6-37.6)
|
63.8 (53.5-73.0)
30.9 (23.6-39.2)
5.3 (2.9-9.6)
|
40.7 (30.3-52.1)
35.3 (29.8-41.1)
24.0 (14.4-37.2)
|
P=0.012
|
Storing food in the sleeping room is thought to attract rodents and thereby increases the potential damage of LLIN by rodents. Across the four surveys, this practice was less common in Inhambane or Tete with 40% of households never reporting doing this in any of the surveys they were interviewed and 4% always reported it, while in Nampula only 10% of households never stored food in sleeping rooms and 35% reported always doing so in all surveys they were interviewed.
Cooking in the same room where LLIN are hanging is a potential source of heat damage (melting of polyethylene yarn), especially if the cooking fuel is firewood or charcoal as was the case for 99% of all enrolled households. This practice was again very uncommon in Inhambane, moderately common in Tete, but was reported much more frequently in Nampula.
Recall of messages heard or seen in the last 6 months about net use or care and repair was low in Tete and Nampula, with only 27-40% of households recalling hearing net care messages at two or more surveys, but was better in Inhambane at 73%. The household care and repair attitudes were generally low with less than a quarter of households having a very positive attitude at any time during the three years.
Net-level factors are presented in Table 3. At least two-thirds of cohort LLINs were observed hanging at some point during the following up, and 60-70% were reported used with no difference between sites. The proportion of cohort LLIN that were hanging loose over the sleeping place and were not folded up or tied during the day was consistently high in Inhambane with nearly 90% and 75% of them never tied up over the three years in Inhambane and Nampula, respectively, but in Tete over a third of LLINs were always tied up and another 20% sometimes tied up (p<0.0001 for site comparison). The cohort LLINs were mostly used over reed mats in Tete (93%), and finished bed frames in Inhambane and Nampula (around 40%). Mattresses were rare in all sites.
Table 3: Net use environment and washing of cohort LLIN from campaign across all survey rounds
Variable
|
Inhambane
% (95% CI)
|
Tete
% (95% CI)
|
Nampula
% (95% CI)
|
P-value for site comparison
|
Cohort LLIN
|
N=737
|
N=619
|
N=675
|
Ever hung
|
65.7 (61.4-69.7)
|
75.0 (65.3-82.7)
|
70.2 (63.2-76.4)
|
0.18
|
Ever used
|
62.8 (58.5-67.0)
|
70.6 (63.0-77.2)
|
69.6 (62.7-75.8)
|
0.15
|
Cohort LLIN ever hung
|
N=484
|
N=464
|
N=474
|
|
Tied up or folded when hanging
Never
At times
Always
|
88.0 (83.3-91.6)
8.1 (5.3-12.0)
3.9 (2.2-7.0)
|
43.1 (35.0-51.6)
19.8 (14.7-26.3)
37.1 (30.8-43.8)
|
74.1 (61.9-83.4)
19.2 (12.3-28.7)
6.8 (3.0-14.8)
|
<0.0001
|
Type of sleeping place**
Bed frame (finished)
Bed frame (sticks)
Foam mattress
Reed mat
|
40.2 (32.8-48.1)
26.1 (20.2-32.9)
4.1 (2.3-7.0)
29.7 (23.8-36.4)
|
0.9 (0.4-2.2)
1.6 (0.5-4.8)
4.3 (0.8-19.9)
93.3 (81.2-97.8)
|
42.8 (30.9-55.6)
23.7 (17.2-31.7)
2.0 (0.7-5.6)
31.6 (20.6-45.1)
|
<0.0001
|
Cohort LLIN ever used
|
N=445
|
N=430
|
N=461
|
|
Net was used only by:
Children
Children with adults
Adults
|
17.5 (14.1-21.5)
15.5 (11.4-20.8)
67.0 (59.7-73.5)
|
18.1 (13.5-23.9)
52.1 (43.4-60.7)
29.8 (24.8-35.3)
|
15.4 (11.7-20.1)
20.4 (16.3-25.2)
64.2 (58.2-69.8)
|
<0.0001
|
Ever washed
|
87.3 (84.1-89.9)
|
76.2 (58.9-87.7)
|
68.1 (58.1-76.6)
|
0.03
|
Cohort LLIN ever washed
|
N=437
|
N=373
|
N=322
|
|
Washes last 6 months
Median (IQR)
|
1.5 (1-2)
|
2.6 (2-4)
|
2 (1.5-3.5)
|
<0.0001
|
Use of detergent
Never
At times
Always
|
77.6 (68.9-84.4)
15.6 (10.9-21.7)
6.9 (4.3-10.8)
|
36.5 (26.9-47.3)
24.9 (16.9-35.2)
38.6 (28.7-49.6)
|
74.5 (63.2-83.3)
16.2 (10.1-24.8)
9.3 (4.4,18.5)
|
<0.0001
|
Drying net outside
Never
At times
Always
|
3.2 (1.3-7.5)
7.6 (3.0-17.7)
89.2 (76.8-95.4)
|
2.1 (1.0-4.7)
6.2 (3.3-11.1)
91.7 (86.6-95.0)
|
0.9 (0.3-2.6)
3.1 (1.2-7.6)
96.0 (91.9-98.0)
|
0.31
|
Drying over bush or fence
Never
At times
Always
|
62.5 (50.0,73.5)
19.5 (12.8-28.5)
18.1 (12.6-25.2)
|
33.5 (23.0-45.9)
24.7 (18.3-32.4)
41.8 (30.9-53.6)
|
58.4 (42.5-72.7)
18.0 (11.5-27.1)
23.6 (13.1-38.8)
|
<0.0001
|
** most rudimentary type of sleeping place ever reported for net
Across the four survey rounds, the majority of LLINs were used only by adults for Inhambane and Nampula, but there was a higher proportion of LLINs shared by adults and children in Tete.
The proportion of cohort LLINs reported ever washed after 36 months was 87% in Inhambane, 76% in Tete and 68% in Nampula. The washing frequency showed some variations but was an average rate of about two washes every six months at all three sites. The proportion of households reporting washing with a detergent was overall low in Inhambane and Nampula and moderate in Tete with 38% always using detergent. Nearly all LLIN were reported to be dried outside and drying on fences or bushes was somewhat more common in Tete than the other sites.
At baseline only 13% of the cohort LLIN in Inhambane were hanging, 21% in Tete and 29% in Nampula and 86%, 68% and 67%, respectively, were still found in the package. Fig. 2 (left) illustrates that after 12 months the situation had significantly improved and hanging rates further increased at 24 months finally reaching between 66% in Inhambane and 75% in Tete. At the 36 months survey only between 2% and 4% of the cohort LLIN that were still present in the households were still in the package. Of the cohort LLINs not hanging, some were still being used the previous night, especially in Tete, where 88% of “taken down” LLINs were used and 30% in Nampula. This indicates that they might be removed during the day to gain space in the house. Of nets used the previous night, 81.3% were reported to have been used every night during the previous week. Of the household respondents 81% in Inhambane, 73% in Nampula and 65% in Tete said they used the LLIN equally in the rainy and dry season, but a significant proportion of 18% in Tete also said they used them only during the rains while only 4% in Inhambane and none in Nampula stated this.
In order to interpret the hanging and use of the cohort LLINs, the overall net ownership situation needs to be taken into account and this is shown in Fig. 2 (right). Initially a significant number of non-cohort LLINs were only found in Inhambane and this was due to the fact that during the 2015 campaign some Olyset LLIN had been distributed among the recruited households, but these had not been included in the durability monitoring cohort. At all sites the proportion of households with any non-cohort net and the proportion of these among all LLIN owned by the households declined sharply after baseline suggesting that older, non-campaign LLIN had been discarded. In Nampula new campaign LLIN came in both in 2016 and 2017 (in both cases DuraNet) resulting in an increase to near or above 50% of non-cohort LLIN. In Tete the situation was similar. A campaign that preceded the 36-month survey increased the proportion of non-cohort LLIN within participant households to around 60% (a mix of Olyset and MAGNet). In Inhambane there was a sharp increase of non-cohort LLIN in the 36-month survey (all MAGNet) with non-cohort LLIN reaching a share of 70% of the nets within participant households and these were clearly from the follow up campaign. However, a moderate increase was also seen at the 24-month survey and these LLIN (mostly Dawa Plus) were described in part being from “health facilities” and in part as “from NGO” which could represent the same source.
Attrition
The all-cause cohort net attrition rates and losses due to wear and tear (including LLINs that were reported to have been lost between the 2015 campaign and the baseline survey) are shown in Fig. 3. These include only those LLIN for which a definitive outcome could be determined (e.g. if no one was home to be interviewed, or cluster was inaccessible, net status could not be determined). The highest all-cause attrition was seen in Nampula with 74% followed by Inhambane at 56% and Tete with 50%. However, taking into account the different times of observation between Tete and the other two sites as shown in Fig. 3 reveals that all-cause attrition increased more or less linearly at all three sites and was highest in Nampula followed by Inhambane and was lowest in Tete. Attrition due to wear and tear increased in a more curvilinear fashion with very slow increase initially followed by near exponential gains. Attrition due to wear and tear was similar in Inhambane and Tete, but clearly higher in Nampula.
Reasons for loss among the discarded LLIN was similar across the three sites (p=0.3), with 54% thrown away, 36% destroyed and 10% used for other purposes. Overall there were only 28 cohort LLINs used for other purposes or 2% of all cohort LLINs with a known outcome. Protecting plants was the most commonly reported alternative use in Inhambane and Tete. In Nampula six of the 14 cohort LLINs used for other purposes (1.2% of all LLINs with known outcome) were reported as used for fishing; one was used for drying fish. Other uses were cutting the net up for various uses (two) and as window cover (one).
Physical condition of observed LLINs
As one would expect, the proportion of LLINs still present in the surveyed households with any sign of damage continued to increase significantly during the monitoring period (Table 4). In Inhambane and Nampula the increases continued up to the final survey while in Tete an equilibrium seems to have been reached where LLINs getting holes and LLINs being discarded occurred at similar rates so that the proportion with any hole no longer increased. The proportion of surviving LLINs that were no longer fit for use due to the level of damage (“torn”) significantly increased at the final survey compared to the modest increases seen previously. In Inhambane it reached 22%, while in Tete it was 36% and in Nampula 37%. This suggests that in Inhambane LLINs were discarded at lower levels of damage compared to Tete as both had similar attrition rates due to wear and tear, but LLIN were much less damaged Inhambane (median pHI among those LLINs with any damage was 269 in Inhambane but 1,745 in Tete). In Nampula the proportion of “torn” LLINs was highest and accordingly the proportion of surviving LLINs found in serviceable condition after three years was lowest with 63% followed closely by Tete with 64% and Inhambane with 78%.
Table 4: Integrity of campaign LLIN present in households
Variable
|
Baseline
|
12 months
|
24 months
|
36 months
|
|
% (95% CI)
|
% (95% CI)
|
% (95% CI)
|
% (95% CI)
|
Inhambane
|
N=726
|
N=589
|
N=423
|
N=257
|
Mean months since campaign
|
1.0
|
9.7
|
21.5
|
33.4
|
Net has any hole
|
2.3 (1.2-4.5)
|
20.0 (13.6-28.5)
|
46.8 (40.1-53.6)
|
58.4 (48.9-67.3)
|
Physical condition
Good (0-64)
Damaged (65-642)
Torn (643+)
Serviceable (0-642)
|
99.6 (98.2-99.9)
0.1 (0.0-1.0)
0.3 (0.04-2.1)
99.7 (97.9-99.9)
|
94.2 (91.2-96.2)
4.2 (2.7-6.7)
1.5 (0.7-3.3)
98.5 (96.7-99.3)
|
77.8 (71.9-82.7)
15.6 (12.4-19.4)
6.6 (3.9-10.9)
93.4 (89.1-96.1)
|
58.4 (49.7-66.6)
19.8 (15.2-25.6)
21.8 (15.2-30.2)
78.2 (69.8-84.8)
|
Median pHI if any hole (IQR)
|
23 (2-47)
|
23 (3-98)
|
60 (25-290)
|
269 (51-1193)
|
Has any repairs if any hole
|
0 (-.-)
|
0.8 (0.1-5.4)
|
12.1 (7.4-19.3)
|
4.7 (1.9-11.21)
|
Tete
|
N=601
|
N=464
|
N=306
|
N=112
|
Mean months since campaign
|
6.2
|
13.4
|
24.1
|
35.8
|
Net has any hole
|
7.7 (4.1-13.9)
|
17.5 (10.3-28.1)
|
62.8 (50.1-73.8)
|
58.9 (38.8-76.5)
|
Physical condition
Good (0-64)
Damaged (65-642)
Torn (643+)
Serviceable (0-642)
|
95.7 (90.8-98.1)
3.0 (1.4-6.2)
1.3 (0.5-3.9)
98.7 (96.1-99.6)
|
92.0 (82.9-96.5)
5.2 (2.6-10.0)
2.8 (0.9-8.4)
97.2 (91.6-99.1)
|
58.5 (48.3-68.0)
22.2 (15.9-30.2)
19.3 (11.9-29.7)
80.7 (70.3-88.1)
|
47.3 (27.2-68.4)
17.0 (10.3-26.7)
35.7 (18.2-58.1)
64.3 (41.9-81.8)
|
Median pHI if any hole (IQR)
|
137 (23-381)
|
54 (23-309)
|
162 (41-1125)
|
1745 (228-5780)
|
Has any repairs if any hole
|
4.3 (1.0-16.8)
|
7.4 (3.1-16.5)
|
21.4 (12.3-34.4)
|
27.3 (12.6-49.3)
|
Nampula
|
N=661
|
N=414
|
N=268
|
N=129
|
Mean months since campaign
|
1.2
|
9.9
|
21.9
|
33.3
|
Net has any hole
|
0.6 (0.2-1.6)
|
38.2 (32.3-44.4)
|
56.3 (43.7-68.2)
|
88.4 (79.4-93.7)
|
Physical condition
Good (0-64)
Damaged (65-642)
Torn (643+)
Serviceable (0-642)
|
99.9 (98.8-99.9)
0.2 (0.0-1.2)
100 (-.-)
|
82.6 (78.4-86.5)
13.8 (10.7-17.5)
3.4 (2.0-5.7)
96.6 (94.3-98.0)
|
68.3 (61.5-74.4)
23.5 (17.7-30.5)
8.2 (5.3-12.4)
91.8 (87.6-94.7)
|
23.4 (15.4-33.5)
39.5 (27.7-52.8)
37.2 (26.7-49.1)
62.8 (50.9-73.3)
|
Median pHI if any hole (IQR)
|
na
|
47 (6-226)
|
98 (29-336)
|
584 (201-1180)
|
Has any repairs if any hole
|
na
|
10.1 (5.4-18.2)
|
10.6 (4.6-22.6)
|
9.6 (4.0-21.5)
|
Survival
Overall physical survival of LLIN in serviceable condition after 36 months, i.e. the combination of attrition due to wear and tear and the integrity of the still existing LLIN, was 57% in Inhambane, 43% in Tete and 33% in Nampula (Table 5 and Fig. 4). Inhambane performed best and the result was significantly different compared to Nampula (p=0.0004), but not compared to Tete (p=0.15). This was due to the higher design effect in Tete of 6.9 (compared to 1.7 in Inhambane and 2.8 in Nampula) which resulted in a very wide confidence interval. In other words, there was a very high variation between communities in durability in Tete.
The time of follow-up differed slightly between Tete (35.8 months at the last survey) compared to Inhambane (33.4) and Nampula (33.3). In order to standardize the analysis of survival in serviceable condition, the results were plotted against the hypothetical survival curves with defined median survival (Fig. 4). The survival estimates roughly follow the hypothetical curves and that the relationship between the three sites was the same throughout the time of follow-up.
In addition to estimating median survival at each time point from the graph, it was also calculated from the final two data points (see methods) and results are shown in Table 5. Calculated median survival was 3.0 years in Inhambane (Royal Sentry LLIN), 2.8 years in Tete (MAGNet LLIN), and 2.4 years in Nampula (Royal Sentry LLIN). Estimates obtained from the graph were very similar to the calculated ones at 36 months, but also show that early on in the monitoring the results tend to overestimate the final outcome. Considering the confidence intervals around the median survival, LLINs in Inhambane performed according to the three-year expectation and also in Tete the result was still compatible with the “three-year durability” although given the huge variation between communities in that site this certainly was not true for all villages. In contrast, in Nampula median survival was below the three-year mark.
Table 5: Estimated proportion surviving and median survival in serviceable physical condition
Variable
|
12 months
|
24 months
|
36 months
|
Inhambane
|
|
|
|
% surviving in serviceable condition (95% CI)
|
98.0 (96.0-99.0)
|
85.3 (78.9-90.0)
|
57.3 (50.2-64.1)
|
Estimated from Fig. 4
|
4.9
|
3.7
|
3.1
|
Calculated from last two data points (95% CI)
|
-.-
|
-.-
|
3.0 (2.8-3.3)
|
Tete
|
|
|
|
% surviving in serviceable condition (95% CI)
|
95.8 (90.7-98.1)
|
74.2 (64.2-82.1)
|
43.4 (27.2-61.1)
|
Estimated from Fig. 4
|
4.2
|
3.1
|
2.7
|
Calculated from last two data points (95% CI)
|
-.-
|
-.-
|
2.8 (2.4-3.5)
|
Nampula
|
|
|
|
% surviving in serviceable condition (95% CI)
|
93.7 (90.6-95.8)
|
73.2 (62.2-81.9)
|
32.5 (23.5-43.1)
|
Estimated from Fig. 4
|
2.7
|
2.7
|
2.2
|
Calculated from last two data points (95% CI)
|
-.-
|
-.-
|
2.4 (2.1-2.6)
|
Kaplan Meier survival curves comparing the intention to treat and per protocol analysis are shown in Fig. 5 and show a similar pattern of survival curves only shifted to the left by 0.2 to 1.0 years when risk of damage is considered to start only when the net is hung for the first time.
Determinants of durability were explored with Cox proportional hazard models. Separate models were constructed for household factors and for net level factors, such that models with net-level factors included only LLIN that had been hung for use during the study. Factors were tested first in individual models then used to construct the final multivariate models. There was interaction between recall of net care messages and net care attitude scores, thus these two variables were combined into a composite variable measuring the intensity of recall and attitudes. Factors that were significantly associated with physical durability were province, message recall/attitudes, household size, cooking in the sleeping room, storing food in the sleeping room, folding up LLIN. Once these were combined in the final multivariate models (Table 6), increased household size was associated with an increased hazard ratio (HR 2.02, p<0.001 for households with 7 or more people vs those with 1-3 members). Higher net care attitudes and SBC exposure to social and behaviour change (SBC) messages were protective, with moderate attitudes and SBC exposure associated with a nearly 50% reduction in the hazard ratio (p=0.008) and high attitudes/exposure associated with a nearly two thirds reduction (p<0.001). Storage of food in sleeping rooms was not significant in the final model. Compared to Inhambane, Tete was not significantly different, but Nampula had a hazard ratio of 2.25 (p<0.001), aligning with the overall survival estimates.
Table 6: Determinants of physical durability from Cox proportional hazard model
Variable
|
Adjusted Hazard Ratio (HR)
|
95% CI
|
P-value
|
At household level; N=5857 obs / 2031 LLIN
|
|
|
|
Province Inhambane
|
Reference
|
|
|
Tete
|
0.83
|
0.52 - 1.31
|
0.41
|
Nampula
|
2.25
|
1.60 - 3.17
|
<0.001
|
Household Size 1-3
|
Reference
|
|
|
4-6
|
1.30
|
1.01 - 1.68
|
0.045
|
7+
|
2.02
|
1.52 - 2.69
|
<0.001
|
High net care attitude score and SBC exposure combination across surveys (ref=never (neither at any survey))
|
Reference
|
|
|
Moderate (high attitude never or once and SBC exposure at least once)
|
0.58
|
0.39 - 0.86
|
0.008
|
Higher (high attitude at least once and SBC exposure at least twice)
|
0.37
|
0.24 - 0.58
|
<0.001
|
Never stored food in sleeping rooms
|
0.84
|
0.66 - 1.06
|
0.14
|
|
|
|
|
At net level (LLIN ever hung) N=4419 obs / 1422 LLIN
|
|
|
|
Province Inhambane
|
Reference
|
|
|
Tete
|
0.94
|
0.58 - 1.53
|
0.81
|
Nampula
|
2.75
|
1.88 - 4.01
|
<0.0001
|
Household Size 1-3
|
Reference
|
|
|
4-6
|
1.22
|
0.93 - 1.61
|
0.15
|
7+
|
2.03
|
1.48 - 2.79
|
<0.0001
|
High net care attitude score and SBC exposure combination across surveys (ref=never (neither at any survey))
|
Reference
|
|
|
Moderate (high attitude never or once and SBC exposure at least once)
|
0.59
|
0.40 - 0.86
|
0.008
|
Higher (high attitude at least once and SBC exposure at least twice)
|
0.39
|
0.25 - 0.62
|
0.0001
|
Net folded up Always
|
Reference
|
|
|
At times
|
0.56
|
0.40 - 0.79
|
0.001
|
Never
|
0.95
|
0.69 - 1.31
|
0.76
|
Most rudimentary type of sleeping place ever reported for the net (ref=finished bed frame)
|
Reference
|
|
|
Unfinished bed frame
|
0.95
|
0.62 - 1.47
|
0.82
|
Foam mattress
|
0.72
|
0.26 - 2.01
|
0.53
|
Reed mat or ground
|
1.08
|
0.79 - 1.48
|
0.64
|
For net-level factors, the final model was similar to the household model but with the addition of folding behaviour and the type of bed frame used with the net. Interestingly, “always” folding up the net was not different from “never” folding up the net, whereas LLINs that had sometimes been folded up had a hazard ratio of 0.56 (p=0.001). As the vast majority of LLINs in Inhambane and Nampula were never folded up (>75%) this may be confounded with province. Ultimately the type of sleeping place used with the net was not a significant factor in the final net-level model. Factors that were not significant predictors of physical durability included household head’s age, sex, or education level, discussion of net care in the home, wealth quintile, or the dominant user (adults vs adults-children vs children only).
Insecticidal efficacy
The target of sampling 30 campaign LLIN at each site with bio-assay testing was achieved at 12 and 24 months, but at 36 months 30 LLIN were sampled from Inhambane while only 27 each could be obtained from Tete and Nampula. Results of the bioassay testing are shown in Table 7. There was a decline over time of 60-minute knock-down percentage at all three sites, with a median of 58% after 36 months in Inhambane and Nampula and 72% in Tete. Decline of vector mortality at the final survey was even more pronounced with a median of 55% in Inhambane, 59% in Tete and 57% in Nampula. As result, optimal insecticidal effectiveness, which was 100% at 12 and 24 months at all sites dropped to just 3% in Inhambane, 11% in Tete and 29% in Nampula at 36 months. However, most samples still achieved the minimal effectiveness threshold with 93% in Inhambane, 100% in Tete and 96% in Nampula meaning that overall only 4% of the 36-month samples must be considered as providing insufficient insecticidal protection. Net handling and use of the sampled LLIN, which were external from the cohort at 12 and 24 months, was comparable to that of the cohort LLIN. This implies that the bio-assay samples can be considered representative of the overall campaign LLIN at these sites.
Chemical testing of the 36-month samples found that median g/kg was 4.70 for Inhambane (81% of target dose), 2.36 for Tete (42% of target dose), and 1.85 for Nampula (32% of target dose).