Preventive planning against mercury over-exposure among Moroccan dentists using multidimensional statistical methods

Abstract

BACKGROUND:

Mercury used in dental amalgams constitutes a significant source of chronic exposure to this heavy metal among dentists. Thus, the safety of dental amalgam remains a controversial issue despite its long history of use. In Morocco, most studies about dental mercury were mainly focused on the environmental risk related to the management of mercury-contaminated waste.

OBJECTIVE:

In order to evaluate the occupational exposure to mercury among liberal dentists practicing in two Moroccan regions, a multidimensional statistical approach was used to analyze the collected data. The main objective was to help establishing a targeted prevention plan aiming to reduce the mercury exposure among Moroccan dentists.

METHODS:

Fifteen variables from 146 dentists were elected for a three-step classification procedure: a multiple correspondence analysis followed by a hierarchical ascendant clustering consolidated by the k-Means algorithm.

RESULTS:

Three homogenous clusters were identified. The most important one includes 57.5% of the population as well as the majority of the risky factors. The characterization of these clusters allows proposing concise guidelines for a targeted preventive plan.

CONCLUSIONS:

A real mercurial risk has been observed in the studied population. However, its impact on health as well as the efficiency of simple preventive recommendations remains to be unveiled.

Keywords

Unsupervised clustering missing data multiple correspondence analysis hierarchical ascendant clustering K-means

1 Introduction

Mercury used in dental amalgams constitutes a significant source of exposure to this heavy metal among dentists [1]. Thus, the safety of dental amalgam remains a controversial issue despite its long history of use. In this context, several studies and medical institutions support the use of dental amalgam as a safe material, and others do not [2, 3]. However, all of them are aware of the mercurial risk and recommend the application of prophylactic rules to reduce the occupational exposure and protect the environment [4].

In Morocco, most studies about dental mercury were mainly focused on the environmental risk related to the management of mercury-contaminated waste [5, 6]. Thus, in order to evaluate the occupational mercurial risk among liberal Moroccan dentists, we have previously carried out an exhaustive cross-sectional survey in two regions of central Morocco [7]. In this study, a univariate descriptive analysis revealed the presence of many factors increasing the occupational exposure to mercury within the studied population [7]. These factors were related to workplace characteristics, professional habits of preparing and handling amalgams, and individual means of protection [7]. Despite its importance, univariate analysis allows only a partial and one-dimensional view of the data. Indeed, using this approach, the studied population is characterized by a unique variable at each step of the analysis while each individual reflects, at the same time, all the values of the studied variables and it is often desired to study them simultaneously especially when they are linked [8].

In descriptive multivariate statistics, Exploratory Data Analysis refers to several methods used to describe and visualize datasets with many variables. Using a geometric approach based on the theory of linear algebra, individuals are assimilated to points in a high- dimensional Euclidean space. As far as the central objective of these methods is to study the resemblances between individuals from a multidimensional point of view, similarities between individuals are apprehended by studying the shape of the cloud made by these constructed points. In this sense, different strategies are available and the most common ones are the principal components methods. These latter allow approximating the points’ cloud into a subspace of lower dimensions while preserving as much as possible distances between individuals/points. Another strategy to study similarities between individuals while respecting all the variables is to perform a hierarchical clustering. This method requires defining a distance (Manhattan, Euclidean, etc.) and an agglomeration criterion (Ward, single, centroid, etc.). It is based on the idea of successively merging the pair of closest/most similar individuals together to form a new larger cluster and similarly merging these new clusters until all points are merged into a unique cluster. The result is a hierarchy represented by a tree-like diagram called dendrogram. A third strategy that can be used is the partitional clustering. Many partitioning algorithms are available and the most common one is the k-means method. This latter is based on the idea that each point of the cloud should be iteratively assigned to a cluster minimizing the overall distance between this point and the cluster centroid to which it has been assigned. The clusters are then described by the variables [8, 9].

From these points of view, the purpose of the present study is to use a combination of these three algorithmic tools to identify homogenous groups in a population of dentists differentially exposed to mercury according to the concomitant presence of the aforementioned over-exposure factors. To this aim, given the qualitative nature of the studied variables, a multiple correspondence analysis (MCA) followed by a hierarchical ascendant classification (HAC) consolidated by the k-Means algorithm were chosen to analyze the collected data. These three methods are using the same distance (Euclidean) and therefore can be combined. In addition, the Ward criterion should be used in the hierarchical clustering because it is based on the multidimensional variance like MCA [9, 10].

The described analysis strategy was performed after the resolution of three major problems: selection of variables to be integrated in the MCA and the subsequent clustering, regrouping categories with low frequencies in homogenous ones to avoid determining influences in MCA and finally handling missing data by making a judicious choice within three available strategies: listwise deletion, imputation or taking missing values as new categories for each concerned variable [11 –13].

The final objective of these multidimensional exploratory methods is to classify the studied population of dentists according to the intensity of mercury exposure and some other demographic characteristics. This may help decision-making processes in preventive health care against occupational mercurial risk within this population.

2 Materials and methods

2.1 Data collection

The present study uses a research database obtained from a previous exhaustive cross-sectional survey conducted to evaluate exposure to mercury among liberal dentists in two regions of central Morocco, Fez-Boulmane and Meknes-Tafilalet [7]. This survey took place over a three-month period, from the beginning of February to the end of April 2016. A written call for participation, explaining objectives of this study, and a self-administered questionnaire were distributed to all liberal dentists of the studied regions [7]. Moreover, verbal informed consents were obtained from all dentists who consented to participate in the survey. Participation was voluntary, and confidentiality and anonymity were ensured by coding data collection sheets [7].

Regarding the present study, as far as the studied variables are related to dental amalgam use, practitioners who no longer use it were excluded. Accordingly, data of only 146 dentists is analyzed in this study.

2.2 Data preparation

2.2.1 Variables selection

The intensity of mercury exposure within dentists using dental amalgam is related to several factors including the workplace characteristics and maintenance, professional habits of amalgam preparation and handling [14], and the personal means of protection [15]. In this context, several items in the questionnaire were devoted to explore these factors during the survey.

In the present study, twelve variables were selected for the analysis: the floor covering and its cleaning habits, ventilation mode, presence of fabric curtains in the workplace, amalgam mixing technique (the use of amalgam capsules or the use of bulk alloy and bulk elemental mercury), sterilization mode (dry or wet heat) and its place (in the treatment room or in a separate one), removal and polishing of amalgam with or without cooling water spray, handling of prepared amalgam (bare hands or with gloves), amalgam waste conditioning (open or closed container), amalgam use frequency and finally the former sensitization about the occupational exposure to mercury. This list was largely based on the dental mercury hygiene recommendations established by the World Dental Federation [16], the American Dental Association [17] and the French National Institute of Scientific Research [14]. All these variables were used as active in the MCA. Two other categorical variables were taken as illustrative: gender and exercise area (rural or urban). The number of years in the professional life (seniority) was taken as a supplementary quantitative variable.

Active variables with rare categories were recoded to make them dichotomous with a category that corresponds to the protective modality against another for the whole risky ones. Thus, better balanced frequencies and simplified interpretation were obtained. Regarding variables related to the individual means of protection (masks, surgical caps or professional shoes), they were not included in the analysis since they have to be systematically promoted against the biological risk independently of the use of dental amalgam [18].

2.2.2 Statistical software

Data analysis was carried out with the R statistical software version 3.6.1 (Foundation for Statistical Computing, Vienna, Austria; https://www.R-project.org). The packages, functions and references used for each process are indicated at each step of the analysis.

2.2.3 Missing data

Missing values are a very common problem affecting data analysis. When data are collected by a questionnaire, subjects may be unwilling or unable to respond to some items or sections generating by this fact missing values in the dataset. This may be problematic because most statistical software are not designed to handle these missing values.

Rubin [19] has described a widespread classification of missing data based on the randomness of missingness. Three mechanisms of missingness were defined:

Missingness completely at random (MCAR): A variable is missing completely at random if the probability of missingness is the same for all units and unrelated to any information in the dataset;

Missingness at random (MAR): The MAR mechanism is at work when the probability of missing data in a variable is related to some other variable(s) in the dataset;

Missingness not at random (MNAR): The MNAR mechanism happens when the probability of missing data in a variable is associated with this variable itself or on other variables not included in the dataset.

Another popular distinction that is often used with missingness is the distinction between ignorable and nonignorable missingness. Ignorable missingness usually refers to MCAR and MAR, whereas nonignorable missingness is often used synonymously with MNAR [20].

If these different mechanisms are relatively simple to be mathematically defined, their determination in practice is quite difficult. Visual exploration of missingness can help check these different assumptions. Certainly, this cannot prove that any randomness holds but visual checks can be used to reject MCAR assumptions. More else, this exploration can suggest what dependencies exist between variables to suspect MAR or MNAR mechanism. These investigations can help to define the method to deal with the missing values [21].

It is worthy to note that before exploring or handling missing data in the active variables, it was decided to not handle the variable “former sensitization” because of its high proportion of missing values (57.53%). Moreover, sensitization was considered as our further main recommendation and its former status should be reflected as faithfully as possible.

MCA is a very useful tool to visualize the missing data pattern especially when the number of variables is large. All variables should be recoded into two categories, observed and missing, and MCA will highlight associations between these two categories by searching the common dimension of variability between the corresponding variables. This will show if missing values occur simultaneously or if they occur when other variables are observed.

The MCA was run using the R package FactoMineR [22]. Figure 1 shows the result for the first two dimensions. This graph highlights three groups of missing categories; the first group has large positive coordinates on the first axis whereas the two others have opposite coordinates on the second axis. Consequently, MCA suggests that the missing data are not MCAR. As this assumption is necessary for a listwise deletion, this solution has to be discarded as an acceptable strategy to handle our missing data even though our objectives are purely descriptive. Moreover, some links between missing and observed categories are viewed on the other dimensions which can comfort the MAR assumption.

Fig. 1

Visualization of the missing data pattern: Graph of the first plane of MCA with “_m” for missing categories and “_o” for observed ones.

In order to run our study with the complete data, a first solution was the use of the “missing passive” approach where a new category is created for all variables with missing values. This solution was not appropriate in our case because many variables had missing data and relationships between observed categories were disturbed by missing ones which made interpretation hard to do in a meaningful way.

Another way consists in replacing missing values by plausible ones according to an imputation model. This is called single imputation. Thus, a unique and complete dataset is obtained. However, if a unique value is predicted for a missing entry, uncertainty is not reflected which constitute a major problem for predictive models [23]. As far as our objective is descriptive, this solution seems to be satisfactory. Indeed, single imputation can be used when a single dataset needs to be completed for a descriptive purpose when no inference is required [23]. From this point of view, this solution, under the ignorability assumption, responds largely to our objectives.

In 2012, Josse et al. proposed their regularized iterative MCA algorithm. This algorithm is implemented in the R package missMDA [12]. It allows either a single or a multiple imputation. To perform single imputation using this package, the first step is to estimate the number of dimensions. Then, the regularized iterative MCA algorithm is performed using the estimated number of dimensions as argument. The first output is a disjunctive table that corresponds to the completed indicator matrix resulting from the last step of the regularized iterative MCA algorithm. The imputed value in the indicator matrix can be seen as a degree of membership to the associated category. Consequently, each missing entry of the original dataset can be imputed with the most plausible category [12]. This completed data table is available on the second output. This latter was used for the analysis.

2.3 Classification procedure

First, the imputed data were pre-processed by a multiple correspondence analysis and reduced to their principal components. Then hierarchical ascendant clustering was performed to determine a consistent partition. This partition was consolidated by the k-means algorithm.

2.3.1 Multiple correspondence analysis

MCA is a descriptive analysis of multidimensional qualitative data. It allows the analysis of a matrix of I individuals depicted by K qualitative variables. Projections of these individuals in a K-dimensional space are used to calculate factorial axes, the first one is constructed in order to retain the maximum variance, and the following axes retaining similarly the residual variance and being perpendicular to each other [24]. In addition, MCA allows continuous quantitative coordinates to be attributed to individuals. Usually, only the most significant axes are selected for analysis in order to reduce dimensions of the studied space [8].

In this study, MCA was used only as a pre-processing step before the clustering in order to code the collected categorical variables into a set of continuous ones (the principal components) [24]. First, variables were classed into active and illustrative ones whereas all individuals were taken as active. Distinction between active and illustrative variables allowed reducing insignificant noise and maintaining homogeneity of variables used to calculate principal component. MCA was performed by the R package FactoMineR.

Multidimensional analysis is often supplemented by univariate analyses to characterize some specific variables. In this context, in order to explore relationships between the former sensitization status and the other variables before the clustering step, we have used the function “catdes” from FactoMineR. This function allows the description of the categories of one factor by categorical variables and/or by quantitative ones [8]. This step aimed to characterize the impact of the former sensitization on the application of the studied preventive measures.

2.3.2 Hierarchical ascendant clustering

This step was also performed by FactoMineR and its function “HCPC”. The hierarchical tree obtained by this function uses Ward’s criterion. It consists in aggregating two clusters so that the growth of the within-inertia is minimal at each step of the algorithm. The within-inertia characterizes the homogeneity of a cluster. The hierarchical clustering is performed onto the principal components of MCA. The output is represented by a dendrogram indexed by the plot of the absolute loss of inertia. The algorithm used by the “HCPC” function proposes in its output an optimal number of clusters. This suggested partition corresponds to the one with the highest relative loss of inertia [8].

2.3.3 Partitioning

As HAC clustering is not optimal due to the constraint of hierarchical grouping [24], individuals were partitioned into the defined clusters by the k-means method implemented in the “HCPC” function. The partition obtained from the cut of the hierarchical tree was introduced as the initial partition of the K-means algorithm and several iterations of this algorithm were done. The partition resulting from this algorithm was finally retained. The initial partition was not entirely replaced, but rather improved (consolidated). Indeed, the quality of the partition is better because clusters were, after this step, compact around their centroids and well separated from each other [10].

2.3.4 Clusters description

The classification assigns each individual, i.e. an MCA derived representation of him, to a cluster. In order to characterize the obtained clusters, the quantitative variable and the categories of the qualitative ones with significant links (p < 0.05) were used to describe them.

3 Results

To build a statistical classification, twelve variables were elected as active in MCA. Missing data in active variables (except for sensitization) were single-imputed using the R package “missMDA”. Two qualitative and one quantitative variables were used as illustrative. Table 1 summarizes the definitive data used for analysis.

Table 1
Distribution of the variables used in the multiple correspondence analysis and the subsequent clustering after a single imputation of their missing values

Variable Category Coding n %

Fabric Absence Curtain_0 34 23.29

Curtain Presence Curtain_1 112 76.71

Ventilation Natural Ventilation_0 120 82.19

[Air conditioner + Fan] Ventilation_1 26 17.81

Sterilization Wet heat SterMode_0 61 41.78

Mode Dry heat SterMode_1 85 58.22

Sterilization Seperate room SterRoom_0 40 27.40

Room Work room SterRoom_1 106 72.60

Removal and With water spray Removal_0 125 85.62

polishing of amalgam Without water spray Removal_1 21 14.38

Amalgam Capsule Capsule_0 78 53.42

Mixing Mercury bulk Capsule_1 68 46.58

Floor Tiling Floor_0 136 93.15

Covering [Mosaic+carpet] Floor_1 10 6.85

Floor Washing FloorClean_0 131 89.73

Cleaning [Manual broom+vacuum cleaner] FloorClean_1 15 10.27

Handling of Withgloves Handling_0 65 44.52

Amalgam Bare hands Handling_1 81 55.48

Waste Closed container Waste_0 35 23.97

Conditioning Open container Waste_1 111 76.03

Frequency of Frequent Freq 21 14.38

Amalgam use Medium Med 65 44.52

Rare Rare 60 41.10

Former Absence Sens_0 10 6.85

Sensitization Presence Sens_1 52 35.62

Missing Sens_NA 84 57.53

Illustrative qualitative variables

Variable Category Code n %

Exercise Rural Rural 26 17.81

Area Urban Urban 120 82.19

Gender Male M 84 57.53

Female F 62 42.47

Illustrative quantitative variables

Variable Range Mean SD

Seniority (years) [1, 27] 10.96 8.34

Variable	Category	Coding	n	%
Fabric	Absence	Curtain_0	34	23.29
Curtain	Presence	Curtain_1	112	76.71
Ventilation	Natural	Ventilation_0	120	82.19
[Air conditioner + Fan]	Ventilation_1	26	17.81
Sterilization	Wet heat	SterMode_0	61	41.78
Mode	Dry heat	SterMode_1	85	58.22
Sterilization	Seperate room	SterRoom_0	40	27.40
Room	Work room	SterRoom_1	106	72.60
Removal and	With water spray	Removal_0	125	85.62
polishing of amalgam	Without water spray	Removal_1	21	14.38
Amalgam	Capsule	Capsule_0	78	53.42
Mixing	Mercury bulk	Capsule_1	68	46.58
Floor	Tiling	Floor_0	136	93.15
Covering	[Mosaic+carpet]	Floor_1	10	6.85
Floor	Washing	FloorClean_0	131	89.73
Cleaning	[Manual broom+vacuum cleaner]	FloorClean_1	15	10.27
Handling of	Withgloves	Handling_0	65	44.52
Amalgam	Bare hands	Handling_1	81	55.48
Waste	Closed container	Waste_0	35	23.97
Conditioning	Open container	Waste_1	111	76.03
Frequency of	Frequent	Freq	21	14.38
Amalgam use	Medium	Med	65	44.52
Rare	Rare	60	41.10
Former	Absence	Sens_0	10	6.85
Sensitization	Presence	Sens_1	52	35.62
Missing	Sens_NA	84	57.53
Illustrative qualitative variables
Variable	Category	Code	n	%
Exercise	Rural	Rural	26	17.81
Area	Urban	Urban	120	82.19
Gender	Male	M	84	57.53
Female	F	62	42.47
Illustrative quantitative variables
Variable	Range	Mean	SD
Seniority (years)	[1, 27]	10.96	8.34

First, “catdes” function was run to catch links between the sensitization status and the other variables. Only seven of them had a p-value less than 5%. In a descending order, the concerned variables were: the sterilization room, waste management, presence of fabric curtains in the workplace, amalgam’s handling, sterilization mode, floor covering and finally the ventilation mode. Correlations between categories with positive v-test and the sensitization status are summarized in Table 2. Both positive and negative sensitization statuses were linked significantly (p-value < 0.05) with five protective categories and shared four of them. These common factors were related to sterilization, amalgam handling and waste conditioning. Moreover, 45.16% of female dentists and 44.61% of the population with a medium frequency of amalgam use have a positive status of former sensitization. The undefined (missing) sensitization status was usually linked with the risky categories except for the ventilation mode that was in a protective status.

Table 2

Correlations between former sensitization statuses and categories of variables related to mercury over-exposure

	% Cluster^*	% Global^†	v.test^‡
Negative: Sens_0
SterRoom_0	25.00	27.40	4.90
Waste_0	22.86	23.97	3.74
SterMode_0	13.11	41.78	2.42
Floor_0	30.00	6.85	2.26
Handling_0	12.31	44.52	2.25
Positive: Sens_1
SterRoom_0	59.02	41.78	4.96
Handling_0	55.38	44.52	4.44
Curtain_0	64.71	23.28	3.91
SterRoom_0	60.00	27.40	3.67
Waste_0	60.00	23.97	3.34
Gender = Female	45.16	42.47	2.04
AmalUse = Med	44.61	44.52	2.00
Undefined: Sens_NA
SterRoom_1	73.58	72.60	6.43
SterMode_1	78.82	58.22	6.17
Waste_1	70.27	76.03	5.54
Handling_1	77.77	55.50	5.53
Curtain_1	68.75	76.71	4.94
Capsule_1	67.64	46.57	2.28
Ventilation_0	61.67	82.19	2.11

^*Frequency of the modality in the cluster. ^†Frequency of the modality in the whole dataset. ^‡Only significant correlations (p-value less than 5%) with positive v-test are represented in a descending order.

MCA was then performed using the first ten axes that allowed loading almost 90% of the total variance (Fig. 2). The last axes were considered as just a noise. Variables contributing to axis definition differed from one axis to the other, and no rare categories had a determining influence. Figure 3 represents the first plane of MCA.

Fig. 2

Proportions of variance explained by the first ten axes of MCA.

Fig. 3

Multiple correspondence analysis: Modalities plot for dimensions 1 and 2. Active variables are in red and supplementary ones in green.

HAC was then performed on the individual-coordinates derived from the MCA. The shape of the hierarchical tree, much like the bar chart of the inertias associated with the nodes, suggested partitioning into three clusters (Fig. 4). Partition strengthening by the k-means method (10 iterations) was simultaneously used to attribute individuals to these three clusters. Three variables were not correlated with the cluster variable: amalgam mixing technique, cleaning mode and gender. The categories best describing the different clusters are presented in Table 3.

Fig. 4

Hierarchical ascending classification of dentists included in the study: Dendrogram. The dense horizontal line is the proposed partition by the algorithm (three clusters) based on the relative loss of inertia (the graph in the top right corner).

Table 3

Characterization of the obtained clusters by the qualitative variables modalities

Category	% Cluster^*	% Global^†	v.test^‡
Cluster 1: n = 84
SterMode_1	92.94	58.22	10.83
Handling_1	91.36	55.48	9.63
SterRoom_1	79.24	72.60	9.23
Waste_1	75.68	76.03	8.40
Sens_NA	85.71	57.53	8.20
Curtain_1	71.43	76.71	6.23
Ventilation_0	67.50	82.19	5.26
Floor_0	61.76	93.15	3.84
Removal_1	85.71	14.38	2.87
Rural	80.77	17.81	2.66
AmalUse = Freq	80.95	14.38	2.35
Area = Urban	52.50	82.19	-2.66
Cluster 2: n = 45
SterMode_0	67.21	41.78	8.28
Sens_1	71.15	35.62	7.82
Handling_0	63.08	44.52	7.79
SterRoom_0	67.50	27.40	5.68
Waste_0	65.71	23.97	4.90
Curtain_0	64.71	23.29	4.66
Ventilation_1	65.38	17.81	3.97
Removal_0	35.20	85.62	2.98
Floor_0	33.09	93.15	2.29
Cluster 3: n = 17
Sens_0	100.00	6.85	6.69
Floor_1	100.00	6.85	6.69
SterRoom_0	32.50	27.40	4.41
Waste_0	34.28	23.97	4.28
SterMode_0	24.59	41.78	4.09
Handling_0	21.53	44.52	3.29
Curtain_0	23.53	23.29	2.25
Area = Urban	14.17	82.19	2.19
Area = Rural	0.00	17.81	-2.19

^*Frequency of the modality in the cluster. ^†Frequency of the modality in the whole dataset. ^‡Only significant correlations (p-value less than 5%) are represented.

Cluster 1 comprised 57.5% of the population. It grouped the majority of the risky categories. These ones concerned especially the professional habits of handling mercury, the sterilization (mode and place) and the waste conditioning. The characteristics of the work place were rather protective except for the presence of fabric curtains. The majority of dentists in rural areas belonged to this cluster (80.77%). In the urban exercise, 53.33% of the practitioners were also present in this cluster. The sensitization status was undefined for the majority of this cluster’s dentists (85.71%). However, 23.08% of practitioners with a positive sensitization status belonged to this cluster.

Dentists in cluster 2 (30.8% of the population) were characterized by a positive sensitization status (71.15% of them). Except for the ventilation mode that was at risk, all the other associated categories were protective. The variable “exercise area” was not significantly correlated to this cluster.

Cluster 3 regrouped only 11.7% of the population. It was characterized by a negative sensitization status (100%). However, only the floor covering presented the risky category. No rural dentist belonged to this cluster which was composed in totality by urban practitioners (13.33% of them).

The quantitative variable (“Seniority”) was associated with Cluster 1 with a negative v-test (–3.58). Practitioners in this cluster were, on average, newer in the practice of the profession. Seniority was also associated with Cluster 2 with a positive v-test which suggested that dentists in this cluster were rather seniors. The association with Cluster 3 was not significant.

4 Discussion

The use of amalgam in dentistry results in the dental team’s exposure to mercury vapor. The exposure level depends on background pollution and on peak exposures during specific work tasks [25]. Several studies have revealed that working environment and professional habits played a crucial role in mercury exposure of dental personnel [26]. Indeed, the intensity of mercury exposure is strongly related to potential losses of this toxic, uncollected or unconditioned amalgam residues [27], the nature of the floor covering and its cleaning habits [14, 25], the presence of mercury traps such as carpets or fabric curtains [15] and the type of ventilation [14, 25]. Factors such as the frequency of amalgam use [28] and how it is prepared or removed [29], its mixing technique [14] and the place and the mode of sterilization [14, 30] have been pointed out as important in the severity of exposure. In this context, hygiene recommendations have been established by several dental organisms and official institutions to reduce mercury exposure related to these factors [14 , 17]. The application of these rules helps to reduce, in a significant way, the exposure to mercury among dental personnel [29].

Our study describes the use of a three-step clustering method to group dentists using dental amalgam in two Moroccan regions. This clustering was made according to the application of prophylactic rules against mercury exposure. The final purpose of this work was to identify characteristics of subgroups highly exposed to mercury in order to design efficiently a prevention plan. This statistical classification procedure showed its ability to handle data with large categorical variables to identify the parts of population that are at risk of occupational mercury overexposure. It merely indicated features that are characteristic and allowed distinguishing three homogenous clusters.

Cluster 1 is the most important. It includes 57.5% of the population and the majority of the risky factors related to the practitioners’ behavior. In fact, dentists in this cluster use amalgam frequently and in a risked way. They use dry heat for sterilization in the work room which generates over lasting peaks of exposure to mercury [14, 30]. Amalgam residues are conditioned in open containers which keep a high mercury background pollution [25, 27]. The presence of fabric curtains also participates in maintaining atmospheric pollution [15]. Amalgam removal or polishing is also done without cooling water spray. This provokes very high peaks of exposure [27, 29]. Handling mercury with bare hands is also a risky habit [15]. Fortunately, natural ventilation by regular windows opening could help to reduce the background pollution [14]. Tiling on the floor is also a protective factor that allows effective cleaning and traps few amalgam residues [14 , 25]. However, the interpretation of the presence of protective modalities related to the workplace should be taken with caution because it does not reflect necessarily awareness against mercurial risk and should be consolidated during a future sensitization campaign. This cluster includes 80.77% of rural dentists and 52.50% of urban ones. This shows that the majority of dentists, especially in rural, are at risk. Dentists in this cluster are on average newer in the profession. Characteristics of this cluster indicate that urgent sensitization campaign should be initiated and focused in priority on rural and newer dentists. Dentists in urban areas should not be overlooked, more than half of whom belong to cluster 1. Fortunately, all risky factors are related to practitioners’ habits. Thus, no heavy financial investments are required to modify workplace’ design for the majority of the concerned dentists. In this cluster 85.71% of dentists have undefined sensitization status.

Cluster 2 is characterized by a positive sensitization status and protective categories except for the ventilation mode. This shows the effectiveness of sensitization. Dentists in this cluster are on average seniors. This suggests that sensitization is not done periodically or newer dentists are not sufficiently reached by sensitization campaigns.

Cluster 3 includes the totality of dentists with negative sensitization. Only the floor covering is at risk and all the other characteristics are protective. Concomitant presence of protective categories and negative sensitization status seems contradictory. Nevertheless, item about sensitization in the distributed questionnaire concerned only the passive one and not the personal efforts of documentation which could be effective in adoption of prophylactic measures. These results showed that not to impute sensitization variable was a wise decision. Imputation would have biased interpretation. If a post hoc imputation is done, the positive and negative sensitization should be recoded logically as positive status and undefined one imputed as negative. However, univariate statistical exploration according to the sensitization status has showed that neither positive status (passive sensitization) nor negative one (self-documentation) was correlated with the whole protective modalities. This suggests a lack of exhaustiveness of recommendations or a laxness installed with the time. Moreover, despite the awareness against mercurial risk, characteristics of workplaces are in a risky situation. The heavy financial investment might be the explanation.

In summary, our study reveals that 57.5% of practitioners using amalgam are highly exposed to mercurial risk. Thus, a sensitization campaign is needed. It should be axed primarily on behavior. Priority should be given to newer and rural dentists. Exhaustive and clear recommendations should be formulated and periodically repeated. Sensitization is not sufficient to change risky workplaces’ characteristics; more efforts should be done to change them especially on the financial side.

The main weakness of our study is the limit of the regions covered by the study. However, despite the fact that no sampling technique has been done to guarantee the representativeness of the entire population of Moroccan dentists, our recommendations should be integrated into a national plan of sensitization. Indeed, the majority of Moroccan dentists has followed the same university training and is supervised by the same institutions concerned with the occupational health promotion.

The rules and recommendations governing the use of metallic mercury are the same in most jurisdictions. They are intended to protect human health and their working environment. Standards and best practices for handling this toxic and hazardous chemical are widely accepted [31, 32]. The Minamata convention on mercury, on January 2013, is a concrete manifestation of the international commitment against mercurial risk [33].

In dentistry, several studies have shown that work practices were associated with mercury exposure in dental personnel [34]. However, it is difficult to establish which recommendations related to mercury hygiene are followed by dental personnel in most of the countries highlighted in these studies [34]. Some investigations revealed that the well-known hygiene guidelines against mercury were not always observed by many dentists around the world [35 –39]. In this light, establishing recommendations is necessary but not sufficient to promote best practices against mercurial risk in dentistry. Active and periodic preventive campaigns should be made to consolidate the hygienic professional habits. Moreover, a periodically check of the dental operatory atmosphere for mercury vapor using dosimeters for routine monitoring of air-mercury levels may help dental personnel to assess objectively the encountered mercurial risk in their dental offices [32] and, if necessary, decontaminate the work place and reinforce the hygienic measures against mercury.

5 Conclusion

The present study reveals a real mercurial risk in a large proportion of the studied population of Moroccan dentists. Our approach allowed us to design a targeted preventive plan to reduce exposure to mercury. However, one question remains unanswered: the impact of this exposure on the health and whether the preventive measures are sufficient to curb the problem. An etiological approach is indicated to resolve this question.

Conflict of interest

The authors declare no conflicts of interest with respect to the research, authorship, and/or publication of this article.

References

Nagpal

, Bettiol

, Isham , Hoang

, Crocombe

. A review of mercury exposure and health of dental personnel, Saf Health Work. 2017;8:1–10.

Jirau-Colón

, González-Parrilla

, Martinez-Jiménez

, Adam

, Jiménez-Velez

. Rethinking the dental amalgam dilemma: An integrated toxicological approach, Int J Environ Res Public Health. 2019;16(1036):1–13.

Homme

, Kern

, Haley

, Geier

, King

, Sykes

, et al. New science challenges old notion that mercury dental amalgam is safe, Biometals. 2014;27:19–24.

Fisher

, Varenne

, Narvaez

, Vickers

. The Minamata Convention and the phase down of dental amalgam, Bull World Health Organ. 2018;96(6):436–8.

Manyani

, Rhalem

, Ennibi

, Soulaymani

, Hmimou

, Soulaymani-Bencheikh

. Gestion des déchets médicaux et pharmaceutiques produits au niveau des cabinets dentaires privés et publics des provinces de Rabat et Kénitra, Toxicol Anal Clin. 2019;31(2, Supplement):S68–9.

Chala

, Sawadogo

, Sakout

, Abdallaoui

. Gestion des déchets d’amalgame dentaire par les chirurgiens-dentistes au Burkina Faso et au Maroc, Odontostomatol Trop. 2012;35(140):31–7.

Attiya

, Fattahi

, El-Haidani

, Lahrach

, Amarouch

M-Y

, Filali-Zegzouti

. Exposition au mercure et état de santé des médecins dentistes de deux régions du centre du Maroc: Enquête transversale descriptive. Pan Afr Med J. 2020;36:110.

Husson

, Lê

, Pagès

. Exploratory multivariate analysis by example using R. 2nd ed. Boca Raton: CRC Press; 2017, p. 248.

Flynt

, Dean

. A survey of popular R packages for cluster analysis, Journal of Educational and Behavioral Statistics. 2016;41(2):205–25.

10.

Husson

, Josse

, Pagès

. Principal component methods – hierarchical clustering – partitional clustering: Why would we need choose for visualizing data? Appl Math Dep. 2010;1–17.

11.

Kang

. The prevention and handling of the missing data, Korean J Anesthesiol. 2013;64(5):402–6.

12.

Josse

, Husson

. missMDA: A package for handling missing values in multivariate data analysis, Journal of Statistical Software. 2016;70(1):1–31.

13.

Stavseth

, Thomas Clausen

, Røislien

. How handling missing data may impact conclusions: A comparison of six different imputation methods for categorical questionnaire data, SAGE Open Med. 2019;7:1–12.

14.

Schach

, Jahanbakht

, Livardjani

, Flesch

, Jaeger

, Haikel

. Le risque mercuriel dans les cabinets dentaires: Histoire ancienne ou futur proche? INRS: Documents Pour le Médecin du Travail (2003;93(1):7–23.

15.

Bouard

, Sawicki

, Choukroun

, Durand

, Baron

, Dewitte

J-D

. Exposition professionnelle au mercure des assistants dentaires, INRS: Documents Pour le Médecin du Travail. 2003;93(1):25–34.

16.

FDI World Dental Federation [homepage on the Internet]. Mercury Hygiene Guidance. [Adopted by the FDI General Assembly October, 1998 and revised October, 2007]. 2007 [cited 2020 Jun 25]. Available from: https://www.fdiworlddental.org/resources/policy-statements-and-resolutions/mercury-hygiene-guidance

17.

ADA Council on Scientific Affairs. Dental mercury hygiene recommendations, JADA. 2003;134:1498–9.

18.

Banaee

, Claiborne

, Akpinar-Elci

. Occupational health practices among dental care professionals before and during the COVID-19 pandemic, Work. 2021;68(4):993–1000. doi: 10.3233/WOR-205319.

19.

Rubin

. Inference and missing data, Biometrika. 1976;63:581–92.

20.

Graham

. Missing data analysis: Making it work in the real world, Annu Rev Psychol. 2009;60:549–76.

21.

Cheng

, Cook

, Hofmann

. Visually exploring missing values in multivariable data using a graphical user interface, Journal of Statistical Software. 2015;68(6):1–23.

22.

Lê

, Josse

, Husson

. FactoMineR: An r package for multivariate analysis, Journal of Statistical Software. 2008;25(1):1–18.

23.

Julie

. Contribution to missing values & principal component methods. Statistics [stat]. Université Paris Sud -Orsay. 2016. tel-01573493. Available from: https://hal.archives-ouvertes.fr/tel-01573493

24.

Warns-Petit

, Morignat

, Artois

, Calavas

. Unsupervised clustering of wildlife necropsy data for syndromic surveillance, BMC Vet Res. 2010;6(1):56.

25.

Langworth

, Sällsten

, Barregård

, Cynkier

, Lind

, Söderman

. Exposure to mercury vapor and impact on health in the dental profession in Sweden, J Dent Res. 1997;76(7):1397–404.

26.

Martin

, Naleway

, Chou

. Factors contributing to mercury exposure in dentists, J Am Assoc. 1995;126(11):1502–11.

27.

Paksoy

, Görgün

, Nalçaci

, Yagbasan

. Assessment of blood mercury levels in practicing Turkish clinicians, dental students, and dental nurses, Quintessence Int. 2008;39(4):173–8.

28.

Goodrich

, Chou

, Gruninger

, Franzblau

, Basu

. Exposure of dental professionals to elemental mercury and methylmercury, J Expo Sci Environ Eidemiol. 2016;26(1):78–85.

29.

Powell

, Johnson

, Bales

. Mercury release during insertion and removal of dental amalgam, Oper Dent. 1994;19(2):70–4.

30.

Parsell

, Karns

, Buchanan

, Johnson

. Mercury release during autoclave sterilization of amalgam, J Dent Educ. 1996;60(5):453–8.

31.

Bagheri Hosseinabadi

, Khanjani

, Mobarake

, Shirkhanloo

. Neuropsychological effects of long-term occupational exposure to mercury among chloralkali workers, Work. 2020;66(3):491–498. doi: 10.3233/WOR-203194.

32.

Ngim

, Ngim

. Health and safety in the dental clinic – Hygiene regulations for use of elemental mercury in the protection of rights, safety and well-being of the patients, workers and the environment, Singapore Dent J. 2013;34(1):19–24.

33.

Fisher

, Varenne

, Narvaez

, Vickers

. The Minamata Convention and the phase down of dental amalgam, Bull World Health Organ. 2018;96(6):436–8.

34.

Sawair

, Hassoneh

, Jamleh

, Al-Rabab’ah

. Observance of proper mercury hygiene practices by Jordanian general dental practitioners, Int J Occup Med Environ Health. 2010;23(1):47–54.

35.

Afify

, Arafa

, AllaaTaher

. Dentist’s attitude toward mercury hygiene practice and its correlation with urinary mercury levels among Egyptian dentists, Egypt J Occup Med. 2011;35(1):143–56.

36.

Bechir

, Bechir

, Sirbu

. The necessity to implement safety measures to prevent mercury pollution, determined by specific dental activities, Jepe. 2012;13(1):178–89.

37.

Raghavendra

, Ranadive

. Mercury hygiene practices followed in dental clinics in pune: A survey, World J Dent. 2013;4(2):92–5.

38.

Decharat

, Phethuayluk

, Maneelok

, Thepaksorn

. Determination of mercury exposure among dental health workers in nakhon si thammarat province, thailand, J Toxicol. 2014)401012.

39.

Makanjuola

, Ekowmenhenhen

, Enone

, Umesi

, Ogundana

, Arotiba

. Mercury hygiene and biomedical waste management practices among dental health-care personnel in public hospitals in Lagos State, Nigeria, Afr Health Sci. 2021;21(1):457–69.