Egg data validation in quantitative monitoring of Aedes albopictus in Emilia-Romagna region, Italy

ABSTRACT

The monitoring of Aedes urban species is usually conducted by ovitraps, which can provide a good estimate of population density at low cost and relatively easy management. When the monitoring network is managed by many operators, as in the case of the Emilia-Romagna regional plan, it becomes necessary to develop a quality control methodology capable of highlighting the possible data incoherency and ensuring that the monitoring system provides reliable data.

This paper presents the procedure applied in the egg counting phase as developed in Emilia-Romagna in order to check the data quality and validate the data before being included in the database.

Several steps have been identified and protocols developed to serve quality control such as training of technicians and intralaboratory quality check, daily random double counting of Masonite paddles and interlaboratory periodic comparison.

The qualitative test showed that all operators were able to distinguish Aedes albopictus eggs from other mosquito species eggs. The quantitative test showed significant differences between the 11 operators, with a relative error resulting in the range of 0.1–25.8%.

The daily internal double counting of randomly extracted Masonite slides was performed by the coordinator showing a relative error in the range 2.47–2.63% without statistical differences, confirming a good alignment of the operators with the coordinator throughout the monitoring period.

The interlaboratory comparison trial produced an average relative error of 7.20% showing a sufficient alignment between the three laboratories.

Moreover a new time-saving methods in egg counting was developed and tested in real condition.

1. Introduction

Aedes albopictus is an invasive mosquito species that has spread rapidly in many regions in recent decades [1–3]. This species is a major public health concern due to its vector capacity in relation to arboviruses, such as chikungunya (CHIKV), dengue (DENV) and zika (ZIKV) [4–6]. In southern Europe, Ae. albopictus shows the capacity of causing local CHIKV outbreaks as has been observed in northern Italy in 2007 [6] and in central-southern Italy in 2017 [7], as well as in southern France in 2014 [8]. Small foci of DENV have been observed in the south of France in 2010 [9], in 2013 [10] and in 2015 (http://www.invs.sante.fr/Dossiers-thematiques/Maladies-infectieuses/Maladies-a-transmission-vectorielle/Chikungunya/Donnees- epidemiologiques/France-metropolitaine/Chikungunya-et- dengue-Donnees-de-la-surveillance-renforcee-en-France- metropolitaine-en-2015), as well as in Croatia in 2010 [11] and Spain in 2018 [12].

In 2007, the Emilia-Romagna region experienced one of the largest outbreaks in Europe due to Ae. albopictus, which stimulates the implementation of the: ‘Regional Plan for the Monitoring and Control of the Tiger Mosquito and the Prevention of Chikungunya and Dengue’ which was started in 2008. This plan includes several actions, such as managing a quantitative monitoring network, larval control in public permanent breeding sites, community engagement campaigns, adult control around imported CHIKV, DENV and ZIKV cases to prevent local transmission, and education in primary schools and major ordinances (http://bur.regione.emilia-romagna.it/dettaglio-inserzione?i=31b143d82c884b00ac0b3d67f3ac43f6), as presented in 13.

The quantitative monitoring network has been organized to provide regular standardized data on the spatial and temporal distribution of population density, assisting in evaluating the possible impact of control plan actions on population density and supporting the efforts made by municipalities in mosquito control.

The monitoring network is based on a system of ovitraps, presented in detail in previous cited papers [14, 15]. We have shown that ovitrap data can provide a sufficiently good estimate of female biting density and can therefore be a useful tool in epidemiological risk assessment [16,17].

The data produced by the monitoring network is regularly uploaded to a dedicated website (www.zanzaratigreonline.it), which is organized in different levels of data accessibility and serves a global audience from citizens to the specialist operators.

Monitoring by ovitraps has certain advantages, such as a good level of sensitivity, easy management, and relatively low cost of the ovitrap itself that is about 4.0 euros each, as well as the management that is about 9–10 euros/ovitrap/turn-in inspection. However, the estimate of Ae. albopictus population density has become quite difficult, mainly due to the highly heterogeneous distribution under urban conditions. The monitoring data are strongly influenced by the microenvironmental conditions of the ovitrap site, such as vegetation type, degree of shade, breeding sites competing with ovitrap, by the changing laying behavior of females [18] and local adulticide treatments [19]. Furthermore, when the monitoring system is used to evaluate Ae. albopictus density, organizations responsible for mosquito control and local administrations may wish to demonstrate that they are doing well in controlling mosquitoes, so they may want the egg density to be lower than the actual case. Therefore, it is necessary to develop a quality control methodology capable of highlighting the possible data incoherency and ensuring that the monitoring system provides reliable data [20].

This paper presents the procedure applied in the egg counting phase as developed in Emilia-Romagna in order to check the data quality and validate the data before being included in the database.

Materials and methods

In the period 2008–2016, the Ae. albopictus monitoring system of the Emilia-Romagna region included a range of 2,600–2,700 ovitraps (the exact number depends on the number of municipalities participating in the network, which may vary slightly each year), positioned at fixed stations in urban areas and activated fortnightly during the summer months. For more details, see [14] and Carrieri et al. [15–17,20]. The ovitraps were managed by about 130 operators (volunteers, municipality or local public health employees, pest control operators], while the egg counting was conducted in three laboratories by about 10–12 operators, under the responsibility of the Regional Agency for Environmental Protection (ARPAE) sections of Modena and Forlì, and the Department of Life Sciences and Biotechnology, University of Ferrara. Both field and lab operators may be subjected to some change every year due to temporary work condition. Masonite paddles collected in the field were coded, inserted in a separate plastic envelope, and sent immediately to the laboratory for egg counting. Skilled operators conducted egg counting using the stereomicroscope.

In 2015, ARPAE’s laboratory developed and adopted procedures for the standardization of egg counts, for the training of laboratory technicians, and for the evaluation of data generated by means of regular qualitative and quantitative proficiency testing in ‘training of technicians and intralaboratory quality check,’ ‘daily random double counting’ and ‘interlaboratory periodic comparison,’ whereas, in 2016, a new method for counting mosquito eggs was developed.

Training of technicians and intralaboratory quality check

Two trials were conducted to test the performance of 10 operators: one is a qualitative analysis of egg discrimination capacity and the other is a quantitative analysis of precision counting capacity.

Both tests were conducted using 10 stereomicroscopes (10–50X) and field-exposed Masonite oviposition substrates. Each operator rotated on each stereomicroscope in which an oviposition substrate was positioned.

For the egg discrimination test, a series of Masonite strips with Ae. albopictus or with other mosquito species eggs were prepared. Operators were asked to indicate the strip with Ae. albopictus eggs.

The quantitative test was performed using Masonite strips with Ae. albopictus number of eggs in the range of 70–700, prepared and counted by an independent technician (control), which is considered to be representative of the common number of eggs found in standard ovitraps. Operators had a maximum time availability of 20 min for each paddle.

Daily random double counting

In each laboratory, a daily internal double counting of randomly extracted Masonite paddles directed to each operator was conducted. At the end of each daily work schedule, the coordinator randomly withdrew 1–2% of the Masonite paddles from the batch processing of the day, which was double counted by him or herself.

Interlaboratory comparison

During the 2015 operating season (May to September), two interlaboratory comparisons were realized to verify the reliability of operators working in the three laboratories.

The first intercomparison was realized in July, involving six operators working at the ARPAE laboratory, while the second one was realized in September, involving all nine operators working in three laboratories.

In each laboratory, 40 Masonite paddles previously exposed in the field were randomly selected after being read normally. The paddles were recoded with a progressive number from 1 to 40, separately placed in plastic bags and sent to another laboratory.

Each operator performed egg counting on 10 randomly drawn paddles from the 40 received.

Developing new time-saving methods in egg counting

Since the oviposition substrate remains for two weeks inside ovitraps in the field, the number of eggs on the strips, especially during the summer months, can be as high as 2,000–3,000 eggs/ovitrap, resulting in peaks of working time. Analysis of the deposition substrates under the microscope shows eggs of mosquitoes and other insects, as well as organic material, making it impossible to use automatic or semi-automatic counting software. To reduce egg counting time and errors, the study developed a new method for counting mosquito eggs. Similar to cell counting chambers, the current study made a transparent acetate mask of the same dimensions as the Masonite strip (150 x 25 mm), and divided into 16 counting cells of 37.5 × 6.25 mm dimension, as shown in Figure 1.

Figure 1.

Transparent mask for egg counting

The substrates were dried and the larger-sized organic material was eliminated by ensuring the non-removal of eggs. The mask was fixed on the dry substrate, so as to adhere perfectly to the surface.

The eggs present in the separation of the cell lines were calculated using the following methods:

1. Eggs touching the cell separation line, but not passing through it, were counted in clearly visible cells;

2. Eggs between lines visible in multiple cells were counted in the upper cell in case of horizontal line or in the left cell in case of vertical line.

In each cell the eggs present and the time required to count them were reported.

At the end of the count, the mask was removed and an accurate count was made of the total number of eggs present on the strip and the total time taken.

The study was carried out on 79 deposition substrates (15 in the pretest draw up, 16 at the University of Ferrara, 24 and 24 in ARPAE Modena and Forlì, respectively).

Statistical analysis

In the training of technicians and intralaboratory quantitative check the error respect of control were calculated.

In the intralaboratory quality control, a block ANOVA (operator paddle) was performed, while in the interlaboratory periodic comparison and in the daily double counting, the data were submitted to the ANOVA F-test.

In developing new methods for egg counting, forward stepwise multiple regressions (FSMR) was performed to evaluate the most predictive cells in the egg counting model. The choice of predictive variables was made by analyzing the Beta coefficient (the Beta value measures the contribution of each predictive variable).

Statistical analyses were performed using Statistica 8.0 (Statsoft).

Results

Training of technicians and intralaboratory quality check

The qualitative test showed that all operators (100% correctly identified eggs) were able to distinguish Ae. albopictus eggs from other mosquito species eggs.

The quantitative test showed significant differences between the operators (F_10,80 = 2.65 and P < 0.008). The relative error of the 11 operators included in the study were calculated using the mean value of the egg number, resulting in the range of 0.1–25.8%, with five operators showing an error below 10%, five operators showing an error between 10% and 20%, and two operators showing an error above 20% (Table 1).

Table 1.

The quantitative test of technician training

Technician	Number of paddles	Mean	SD
A	9	276.67	162.86
B	9	290.33	167.40
C	9	278.67	151.04
D	9	296.11	170.85
E	9	280.44	162.54
F	9	278.11	174.65
G	9	313.44	194.64
H	9	289.44	160.89
I	9	271.78	144.70
L	9	291.56	174.97
Control	9	278.78	162.20
All Grps	99	285.94	158.23

Daily random double counting

A total of 418 paddles were double counted by operators and expert coordinators in three laboratories. The relative error of the double count varied between laboratories from 2.47% (SD 1.46%) to 2.63% (SD 2.03%), and no statistical differences were found (ANOVA F _2,417 = 0.19 and P = 0.82).

The low error of the single laboratory (Figure 2) confirms a good alignment of the operators with the coordinator, demonstrating that the operators involved in the analysis have maintained a good performance throughout the monitoring period.

Figure 2.

Percentage relative error observed in double counting by operators and expert coordinators in three laboratories

Interlaboratory periodic comparison

The periodic interlaboratory comparison involves the three laboratories in the analysis of the substrates. A total of 95 Masonite strips were analyzed, with an average relative error of 7.20% (SD 11.105) and a confidence limit of 4.94–9.46%. The comparison between the performances of each laboratory shows a sufficient alignment between the laboratories, especially considering that repeated manipulation of the paddles for counting as well as transportation can result in the detachment of eggs from the paddles making perfect repetition very challenging.

Developing new time-saving methods in egg counting

A total of 79 Masonite strips with 81,145 eggs were counted. As shown in Table 2, few eggs were laid in the upper and lower cells, with over 90% of the eggs laid in the central cells (B-C) of the strips.

Table 2.

Average number of eggs counted in the cells

	Valid No.	Mean	Std. Dev.	%
A1	79	2.61	10.00	0.25
A2	79	3.99	16.24	0.39
A3	79	2.25	10.75	0.22
A4	79	3.67	11.47	0.36
B1	79	118.76	131.59	11.56
B2	79	110.56	89.40	10.76
B3	79	129.25	95.79	12.58
B4	79	149.14	111.90	14.52
C1	79	122.65	109.53	11.94
C2	79	116.97	98.93	11.39
C3	79	109.10	92.69	10.62
C4	79	107.54	105.39	10.47
D1	79	13.73	28.73	1.34
D2	79	12.89	31.58	1.25
D3	79	12.97	39.01	1.26
D4	79	11.06	39.89	1.08

In the first step of the analysis, all variables with Beta > 0.1 corresponding to cells A and D were eliminated (the Beta value measures the contribution of each predictive variable).

In the second step analysis, FSMR was realized considering the eight central cells B and C as the most predictive variables (R² = 0.92, F_6,72 = 140.96 and P < 0.0001).

The most important factor that can modify the vertical distribution of eggs on the deposition substrate is certainly the climatic factor (Figure 3).

Figure 3.

Monthly climatic trend in Emilia-Romagna region in 2016

Under the conditions of low temperature and high rainfall, the evaporation is low, so the distribution of eggs is more concentrated in the upper part of the strip; on the contrary, during periods of high temperature and drought, the evaporation is high and the eggs are distributed on the whole substrate. For this reason, it was decided to use dummy variables as a function of the sample month to introduce a correction parameter in the model (R² = 0.95, F_8,70 = 153.07 and P < 0.0001) (Table 3).

Table 3.

Forward stepwise multiple regressions considering B and C cells and the month dummy variables

Effect	Comment	Param.	Std. Err	t	p	Beta (ß)	St. Err. ß
Intercept		185.20	48.75	3.80	0.0003
June – α		−89.55	38.06	−2.35	0.0214	−0.09	0.04
July – ρ		129.20	35.88	3.60	0.0006	0.12	0.03
August	Pooled
September	Pooled
October	Pooled
‘B1’		1.20	0.15	7.92	0.0000	0.36	0.05
‘B2’	Pooled
‘B3’		1.18	0.26	4.50	0.0000	0.26	0.06
‘B4’		0.78	0.22	3.49	0.0008	0.20	0.06
‘C1’		0.56	0.24	2.35	0.0214	0.14	0.06
‘C2’		1.40	0.32	4.38	0.0000	0.31	0.07
‘C3’	Pooled
‘C4’		1.85	0.21	8.79	0.0000	0.44	0.05

The FSMR found that the dummy variable alpha in the month of June was negative, which indicates that considering only the central cells B and C, we overestimated the number of eggs; conversely, in the month of July (high temperature and drought), the correction factor ρ was positive, so we underestimated the number of eggs.

The equation considering six cells and the dummy variable for the month is:

$$\begin{array}{rl}& E6=184\hspace{0.17em}.20-89\hspace{0.17em}.55\alpha +129\hspace{0.17em}.20\rho +1\hspace{0.17em}.20\hspace{0.17em}B\hspace{0.17em}1\hspace{0.17em}\\ & +1\hspace{0.17em}.18\hspace{0.17em}B\hspace{0.17em}3\hspace{0.17em}+0\hspace{0.17em}.78\hspace{0.17em}B\hspace{0.17em}4\hspace{0.17em}+0\hspace{0.17em}.56\hspace{0.17em}C\hspace{0.17em}1\hspace{0.17em}\\ & +1\hspace{0.17em}.40\hspace{0.17em}{C}_{\hspace{0.17em}}2+1.85\hspace{0.17em}C\hspace{0.17em}4\end{array}$$ (1)

where E6 is the estimated number of eggs, α = 1 in June, ρ = 1 in July, and B1, B3, B4 and C1, C2, C4 are the number eggs counted in the corresponding cells.

In the third step of the FSMR analysis, the four cells with the major Beta coefficient were considered, in addition to the dummy variables for the months of June and July (R² = 0.93, F_6,72 = 163.58 and P < 0.0001), and the equation is:

$$\begin{array}{rl}& E4=224\hspace{0.17em}2.66-109\hspace{0.17em}2.66\alpha +91\hspace{0.17em}2.41\rho +1\hspace{0.17em}2.37\hspace{0.17em}2B\hspace{0.17em}21\hspace{0.17em}2\\ & +1\hspace{0.17em}2.72\hspace{0.17em}2B\hspace{0.17em}23\hspace{0.17em}2+2\hspace{0.17em}2.04\hspace{0.17em}2C\hspace{0.17em}22\hspace{0.17em}2+1\hspace{0.17em}2.79\hspace{0.17em}2C\hspace{0.17em}24\\ & \end{array}$$ (2)

In the fourth step of the FSMR analysis, only the two cells with the high prediction coefficient used B1-C2 (Eq. 2a-R² = 0.85, F_4,74 = 101.98 and P < 0.0001) and B3-C4 (Eq. 2b-R² = 0.84, F_3,75 = 127.67 and P < 0.0001).

$$\begin{array}{rl}& E2a=427\hspace{0.17em}.08-210\hspace{0.17em}.03\alpha +125\hspace{0.17em}.96\beta \\ & +2.13\hspace{0.17em}B\hspace{0.17em}1\hspace{0.17em}+3\hspace{0.17em}.28\hspace{0.17em}C\hspace{0.17em}2\end{array}$$ (3)

$$E2b=4072.71-213.91\alpha +2\hspace{0.17em}2.65B\hspace{0.17em}23\hspace{0.17em}2+3.15C24$$ (4)

For each model, the absolute error was calculated based on the total number of eggs counted on the oviposition substrate and using the sum of the eggs counted in all cells (Table 4).

Table 4.

Absolute error calculated for the different models

Egg density	N	Method E Standard		Model E6 6 cells		Model E4 4 cells		Model E2a 2 cells		Model E2b 2 cells
		Means	SD	Means	SD	Means	SD	Means	SD	Means	SD
<500	8	0.12	0.11	0.17	0.12	0.24	0.15	0.33	0.34	0.36	0.16
501–1000	29	0.05	0.08	0.09	0.06	0.12	0.08	0.16	0.14	0.15	0.12
1001–1500	35	0.03	0.04	0.06	0.06	0.07	0.06	0.12	0.07	0.11	0.08
>1500	7	0.07	0.12	0.08	0.09	0.08	0.08	0.13	0.07	0.11	0.08
All Grps	79	0.05	0.07	0.08	0.08	0.10	0.10	0.16	0.15	0.15	0.13

This study shows that the absolute error in analyzing the whole deposition substrate using the traditional method E varies from 3% to 12%, depending on the number of eggs present. For the E2 model, the error is high when the number of eggs is less than 500, while the error decreases as the number of eggs increases. For the E4 and E6 models, the error seems to be acceptable (<25%) across all egg densities tested.

The new method for counting mosquito egg allows a significant reduction in counting times based on the model used (Figure 4).

Figure 4.

Simulation of egg counting times in a regional monitoring system using the 4-cell model E4 and the standard method

The E2a and E2b model (two cells) allows a reduction in counting times compared to the whole egg counting of up to 76.9%, with a mean error of 0.15 (0.11–0.36). However, the error in the strips with a number of eggs less than 500 is very high (>30%), which is considered unacceptable by the monitoring system.

The E4 model seems to be the best model for reducing egg counting time (reduction of up to 54%) with an error of 0.10 (0.07–0.24), which can be considered acceptable by the monitoring system. Finally, the six-cell model E6 produces a very low error of 0.08 (0.06–0.17), but does not allow a significant reduction in egg counting time (up to 27.5%).

Conclusion

In large Aedes monitoring networks based on ovitrap management and involving many field and laboratory operators, the risk of unreliable data due to human error or inaccuracy (intentional or unintentional) is high, and measures should be adopted to ensure data quality.

In this study, we describe the experience conducted in the Emilia-Romagna Aedes albopictus monitoring network aimed at data quality control and working time optimization. Regular internal quality control procedures in egg counting are essential to guarantee data reliability. The technicians in charge of egg counting attended a training session held in the central laboratory, where their expertise in discriminating Ae. albopictus from other mosquito eggs or non-mosquito eggs was assessed.

The results showed that the technicians were well trained in the discrimination of the eggs present on the Masonite substrate, thus ensuring the counting of Ae. albopticus eggs only, so the three laboratories that regularly conduct regional surveillance can provide aligned, homogeneous and comparable data. In each of the three involved laboratories organized between technicians and laboratory coordinators, each laboratory performed daily random double counting to maintain a high quality of substrate counting.

We recommend this activity, although time consuming, to be essential in maintaining the necessary regular quality control over the egg data produced by the laboratory, thereby avoiding data quality decline due to routine.

The interlaboratory comparison shows that egg data produced by different laboratories are well aligned without significant differences. We suggest that this test be performed every two years considering the turnover of technicians.

The method tested to reduce the egg counting time showed possibility of application especially during period of highest egg density causing working load peak, with acceptable error while keeping the working time of technicians more constant during the season.

Funding Statement

This study has been conducted in the frame of the LIFE CONOPS project ‘Development & demonstration of management plans against – the climate change enhanced – invasive mosquitoes in S. Europe’ [LIFE12 ENV/GR/000466] co-funded by the EU Environmental Funding Program LIFE+ Environment Policy and Governance with the support of the Regional Health Authority of Emilia-Romagna.

Disclosure statement

No potential conflict of interest was reported by the authors.