Extent of Primary DNA Damage Measured by the Comet Assay in Health Professionals Exposed to Antineoplastic Drugs: A Systematic Review and Meta-Analysis

Background: Antineoplastic drugs (ANDs) are a broad group of chemicals showing, at the same time, carcinogenic effects. The potential, albeit true, risk of side effects cannot be accepted, especially if resulting from occupational exposure. The aim of this study was to evaluate the association between occupational exposure to ANDs and the extent of primary DNA damage in health professionals. Methods: A systematic review and meta-analysis was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. PubMed/Medline, Web of Science, and Scopus were used to perform the literature search. The databases were examined in July 2019. Sub-group, moderator, and cumulative analyses were conducted. The trim and fill method was used in the case of potential publication bias. Results: Twenty studies were included in the qualitative analysis, and 19 in quantitative evaluation. The pooled effect size was 1.27 [(95% confidence interval (CI) = 0.66–1.88), p = 0.000] based on 1569 subjects. The moderator analysis by duration of exposure showed a positive association between duration of exposure and primary DNA damage. Conclusions: This systematic review clearly shows a significant association between occupational exposure to ANDs and the extent of primary DNA damage in health professionals. Considering these results, health professionals should be warned against this potential occupational risk.


Introduction
Antineoplastic drugs (ANDs) are a broad group of chemicals, mainly used in chemotherapy; their anti-proliferative effect was accidentally discovered during the First and Second World Wars, when bone marrow and lymph nodes depletion was registered in soldiers exposed to sulfur mustard [1,2]. However, the anticancer applicability of sulfur mustard was discovered in the late 1940s, and it was first used to treat non-Hodgkin's lymphoma and leukaemia [3,4]. ANDs can be classified in several Table 1. Detailed inclusion/exclusion criteria, according to a population, intervention, comparison, outcomes, and study design (PICOS) statement extended with a language and time filter.

Search Strategy Details
Inclusion criteria P: health professionals occupationally exposed to antineoplastic drugs (ANDs) (female and male) I: comet assay measuring primary DNA damage C: subjects not exposed to ANDs O: mean and standard deviation of primary DNA damage S: primary studies (clinical trial, cohort, case-control, cross-sectional) Exclusion criteria P: workers not occupationally exposed to ANDs in health care settings I: no assessment of genotoxic effects (primary DNA damage) associated to occupational exposure O: other outcomes not related to primary DNA damage S: not original papers (opinion paper, review article, commentary, letter, article without quantitative data) Language filter English Time filter No filter (from inception) Database PubMed/Medline; Web of Science; Scopus

Data Extraction
Full text was downloaded only for eligible articles. Data extraction was conducted only for articles included according to the previously mentioned criteria. Two researchers independently performed data extraction using a pre-defined and pre-piloted spreadsheet elaborated in Microsoft Excel ® for Windows. Any potential disagreement was solved through discussion between the two researchers or, if disagreement persisted, by consulting a third researcher. Data extracted from the original studies were both quantitative and qualitative. Qualitative data recorded included the following: name of first author, year of publication, country where the study was conducted, and preventive measures adopted. Moreover, characteristics of the subjects were also recorded (e.g., age, gender, and job task). The quantitative data extracted were as follows: sample size, ward of the participants, duration of the exposure (expressed in years), and outcome (expressed as mean). In the case of missing details, the corresponding author of the original manuscript was contacted by e-mail. A crosscheck of the reference lists was performed in order to identify further potentially related articles. References were managed using the citation manager software EndNote.

Quality Evaluation
Two researchers independently performed the quality evaluation of the included articles. A validated tool comprising nine items assessing several methodological characteristics of primary studies (i.e., number of exposed and controls, matching for multiple risk factors, and number of cells scored per subject) was used for evaluation. For each item, the score could range from 1 to 3, while the total score could range from 9 to 27 points [7].

Statistical Analysis
The effect size (ES) was calculated based on the mean, standard deviation (SD), and sample size provided per each study. It was estimated by Cohen's d, reported with its 95% confidence interval (CI). Cohen's d is the preferred method of ES calculation, and it is calculated as a difference between two means, divided by the variability among the sample. In this way, it is possible to compare results expressed with different units of measures [22]. The comparison was performed between subjects exposed to the antineoplastic drugs and subjects without any type of exposure. In order to reduce the heterogeneity, a random-effect model was applied to conduct the meta-analysis. The heterogeneity among the included studies was evaluated through Chi 2 and I 2 tests. An I 2 value below 25% means no heterogeneity, a value ranging between 25% and 50% is considered as low, while values between 50% and 75% are considered as moderate, lastly a value >75% means high heterogeneity. Potential publication bias was evaluated through visual examination of the funnel plot and by the Egger's regression asymmetry test. The funnel plot graphically represents the distribution of the retrieved studies considering ES and standard error. It is used for visually detecting potential publication bias based on the symmetry of the plot. However, as the visual examination could be subjective, the Egger's regression asymmetry test was used as well, being able to provide an objective measure for publication bias. For the Egger's regression asymmetry test, the statistical significance was set at p < 0.10 [23]. If publication bias was detected, a trim and fill method was used in order to adjust for publication bias [24]. The trim and fill method is a statistical approach aimed to estimate potential missing studies, causing the asymmetry of the funnel plot. This method assumes that the studies with the most extreme ES have to be suppressed, adjusting the overall effect estimate [25]. To perform the meta-analysis, the software Prometa3 ® (Internovi, Italy) was used.

Sub-Group and Sensitivity Analysis
In order to reduce the heterogeneity, two sensitivity analyses were conducted, considering the following: (i) only studies with higher quality score (QS ≥ 16), and (ii) combining studies with the same unit of measure. The subgroups analyses are commonly performed because the effect of a measurement might vary among subgroups of subjects, defined by individual characteristics. In order to consolidate the validity of the results, several sub-group analyses were developed, taking into consideration the following: gender (male and female separately), protective equipment used (nothings or gloves and masks versus at least gloves, masks, and air-flow cabinet), and work task (excluding studies that considered nurses not exposed to ANDs as controls).

Cumulative Analysis
A cumulative analysis is a sequence of meta-analyses, aiming to estimate the changes of the ES starting with a single study and adding the other studies one at a time. The cumulative analysis accumulates the results from the first to the latest study, and each successive result includes a synthesis of all previous studies. This analysis expresses the potential consistency of the results [26]. We performed the cumulative analysis both chronologically (adding each study according to the year of publication) and considering the sample size (from smallest to biggest).
A total of 373 articles were identified-69 in Web of Science database, 149 in PubMed/Medline, and 155 in Scopus database. A total of 85 documents were immediately excluded because of duplicates. After the preliminary screening by title and abstract, 267 documents were excluded because they were either reviews (n = 53), conference papers, or letters to the editor (n = 13), or because of unrelated topics (n = 172) or different languages (n = 29). Out of the 29 articles excluded because of language restriction, we identified articles written in Chinese (n = 6), Italian (n = 4), French (n = 4), Hungarian (n = 3), Serbian (n =2), Czech (n =2), Finnish (n =1), German (n =1), Japanese (n =1), Korean (n = 1), Polish (n = 1), Portuguese (n = 1), Russian (n = 1), and Spanish (n = 1). Overall, 21 articles were eligible, but one reference was excluded because it was an in vitro study. Figure 1 shows the flow diagram, reporting the selection process. At the end of the screening process, 20 articles were included in the qualitative analysis [30][31][32][33][34][35][36][37][38][39][40][41][42][43][44][45][46][47][48][49], and 19 in the quantitative analysis. One study was removed from the quantitative analysis because, even if the authors had been contacted in order to collect further information, data were not available [49]. Moreover, because four papers reported the results both aggregated and divided by gender, we included the latter data in the sub-group analysis by gender [33,35,40,47]. In order to perform a robust sub-group analysis, when possible, data divided by gender were calculated from the original studies [34]. At the same time, the study conducted by Ursini et al., 2006 [46], showed the results divided by work task, and Kopjar et al., 2001 [37] showed the results divided by protective equipment used; for these reasons, these data were included in the sub-group analysis.

Characteristics of the Included Studies
Most of the included studies were conducted in Western countries, in particular, 10 were conducted in Europe [Italy (n = 5), Portugal (n = 2), Croatia (n = 2), and Austria (n = 1)] and 3 in the Americas [United States (n = 1) and Brazil (n = 2)]. The remaining studies were conducted in Asia, particularly China (n = 2), Turkey (n = 2), Japan (n = 2), and India (n = 1). The majority of the studies performed the comet assay using lymphocyte cells, while two studies incorporated the results of the comet assay of buccal cells [31,46]. Most of the studies were published before 2010 (n = 17); the oldest study was published in 1999 [45], while the newest retrieved study was published in 2015 [39]. All

Characteristics of the Included Studies
Most of the included studies were conducted in Western countries, in particular, 10 were conducted in Europe [Italy (n = 5), Portugal (n = 2), Croatia (n = 2), and Austria (n = 1)] and 3 in the Americas [United States (n = 1) and Brazil (n = 2)]. The remaining studies were conducted in Asia, particularly China (n = 2), Turkey (n = 2), Japan (n = 2), and India (n = 1). The majority of the studies performed the comet assay using lymphocyte cells, while two studies incorporated the results of the comet assay of buccal cells [31,46]. Most of the studies were published before 2010 (n = 17); the oldest study was published in 1999 [45], while the newest retrieved study was published in 2015 [39]. All the included studies except one [49] performed a cross-sectional analysis considering professionals exposed to ANDs compared to others professionals not occupationally exposed to ANDs. Only one study was followed by a follow-up [49]; in this study, the authors estimated the correlation between ANDs' exposure and comet assay by trend analysis, considering the individual risk with regard to the years of exposure. Concerning the quality evaluation, the score ranged between 11 and 19, with a mean of 16 (Table 2).

Characteristics of the Studied Populations
The sample size of the included studies ranged between 12 and 83 exposed subjects, and 12 and 74 controls. The mean age and SD of the exposed subjects were between 29 ± 5 and 44.17 ± 2.40 years, while for controls, they were 28.23 ± 6.30 and 44.33 ± 1.41 years. The exposed and control groups were perfectly matched for age (n = 3), gender (n = 9), and smoking (n = 6) in several studies. Considering the nutritional intake, a very large majority of the studies either did not match the two groups (n = 18), or only partially matched (n = 2) by taking into account this confounding variable. In all studies, the considered exposed subjects were nurses, whereas in two studies [34,35], they were the workers producing ANDs. Moreover, pharmacists or pharmacy technicians were also included in the exposed group. Only one study considered physicians as well [38]. Most of the time, the controls comprised workers not occupationally exposed to ANDs. In five studies, the controls were nurses not involved in oncological wards [30,32,40,44,48], while in two studies, no details regarding the characteristics of control subjects were provided [34,35]. Most of the time, exposure to ANDs was assessed through a self-reported questionnaire; environmental monitoring was included in the study design in four studies [39,[47][48][49], whereas only three studies [32,42,46] performed biological measurements. Regarding personal protective equipment, gloves and masks were the most frequently used; wearing lab coats, glasses, and caps was also reported. Among the environmental equipment, the use of an air-flow cabinet was reported in only seven studies ( Table 2).
Approximately half of the included studies (n = 9) scored 100 cells per subject, and eight studies scored a number of cells between 100 and 200; lastly, three studies scored more than 200 cells per subject. The shape of a comet tail represents the entity of DNA damage; the higher the migration of chromosomal DNA from the nucleus, the higher the DNA damage. Several units of measure are used to express the results. In the retrieved studies, tail intensity % was the most used (n = 7), followed by tail length (n = 6). The DNA damage index was used in three studies; the tail moment, the total comet score, and the log tail length were used once, respectively. Lastly, one study considered a tail factor without any other further explanation [49]. However, all the results were expressed as a mean and SD or standard error of mean (SEM). One study expressed the results as the median [39], and one reported both mean and median [48].

Results of the Meta-Analysis
Considering the comet assay performed on peripheral blood lymphocytes, a total of 19 studies were included in the quantitative evaluation. However, one study reported the data separately for pharmacists, nurses working in day hospital, and nurses working in the ward, so they were considered as three independent studies and, for this reason, the meta-analysis included 21 datasets. The pooled ES was 1.
(a) (b) Figure 4. (a) Forest plot and (b) funnel plot (after trim and fill method) of the meta-analysis assessing the comet assay on lymphocytes cells among professionals occupationally exposed to antineoplastic drugs (ANDs), only including studies with a quality score (QS) ≥16.

Sub-Group Analysis by Gender
In order to increase the robustness of the results, a sub-group analysis by gender was performed. The sub-group analysis considering only females included 10 datasets, and the pooled ES was 1.95 [ (95% CI = 0.88-3.02), p-value < 0.001] based on 849 subjects ( Figure 5). High statistical heterogeneity was found (Chi 2 = 355.84, df = 9, I 2 = 97.47, p-value < 0.001). The sub-group analysis considering only males included five studies and the pooled ES was 2.50 [ (95% CI = 0.57-4.44), p-value = 0.011] based on 95 subjects ( Figure 5). High statistical heterogeneity was found (Chi 2 = 42.31, df = 4, I 2 = 90.55, pvalue < 0.001). Potential publication bias was found by visual assessment of the funnel plot, and confirmed by Egger's linear regression test (intercept 7.28, t = 3.15, p-value = 0.008). However, after applying the trim and fill method, the estimated ES did not change.

Sub-Group Analysis by Gender
In order to increase the robustness of the results, a sub-group analysis by gender was performed. The sub-group analysis considering only females included 10 datasets, and the pooled ES was 1.95 [(95% CI = 0.88-3.02), p-value < 0.001] based on 849 subjects ( Figure 5). High statistical heterogeneity was found (Chi 2 = 355.84, df = 9, I 2 = 97.47, p-value < 0.001). The sub-group analysis considering only males included five studies and the pooled ES was 2.50 [ (95% CI = 0.57-4.44), p-value = 0.011] based on 95 subjects ( Figure 5). High statistical heterogeneity was found (Chi 2 = 42.31, df = 4, I 2 = 90.55, p-value < 0.001). Potential publication bias was found by visual assessment of the funnel plot, and confirmed by Egger's linear regression test (intercept 7.28, t = 3.15, p-value = 0.008). However, after applying the trim and fill method, the estimated ES did not change.

Sub-Group Analysis by Protective Equipment Used
The sub-group analysis including studies with a low level of protection (considered as no protection at all or use of only gloves and masks) counted six datasets. The pooled ES was 2.19 [(95% CI = 0.97-3.42), p-value < 0.001] based on 361 subjects (Figure 6a). High statistical heterogeneity was found (Chi 2 = 93.63, df = 5, I 2 = 94.66, p-value < 0.001). No publication bias was found by visual assessment of the funnel plot, which was confirmed by Egger's linear regression test (intercept 6.52, t = 1.30, p-value = 0.265). The sub-group analysis considering a high level of protection (at least gloves, masks, and air-flow cabinet), included six studies, and the pooled ES was 2.13 [ (95% CI = 0.85-3.40), p-value = 0.001] based on 429 subjects (Figure 6b). In this sub-group analysis, Kopjar 2009 was excluded because the exposed group was composed of nurses who used heterogeneous protective equipment (among the 50 subjects, only 3 of them used gloves, masks, and air-flow cabinet, but the results were only shown as aggregated data). High statistical heterogeneity was found (Chi 2 = 128.11, df = 5, I 2 = 96.10, p-value < 0.001). Potential publication bias was found by visual assessment of the funnel plot, and confirmed by Egger's linear regression test (intercept 6.97, t = 2.47, p-value = 0.069). However, after the trim and fill method was applied, the estimated ES did not change.
(a) Figure 5. Forest plot of the sub-group analysis by gender assessing the comet assay on lymphocytes cells among professionals occupationally exposed to antineoplastic drugs (ANDs). F, female, M, male.

Sub-Group Analysis by Protective Equipment Used
The sub-group analysis including studies with a low level of protection (considered as no protection at all or use of only gloves and masks) counted six datasets. The pooled ES was 2.19 [(95% CI = 0.97-3.42), p-value < 0.001] based on 361 subjects (Figure 6a). High statistical heterogeneity was found (Chi 2 = 93.63, df = 5, I 2 = 94.66, p-value < 0.001). No publication bias was found by visual assessment of the funnel plot, which was confirmed by Egger's linear regression test (intercept 6.52, t = 1.30, p-value = 0.265). The sub-group analysis considering a high level of protection (at least gloves, masks, and air-flow cabinet), included six studies, and the pooled ES was 2.13 [(95% CI = 0.85-3.40), p-value = 0.001] based on 429 subjects (Figure 6b). In this sub-group analysis, Kopjar 2009 was excluded because the exposed group was composed of nurses who used heterogeneous protective equipment (among the 50 subjects, only 3 of them used gloves, masks, and air-flow cabinet, but the results were only shown as aggregated data). High statistical heterogeneity was found (Chi 2 = 128.11, df = 5, I 2 = 96.10, p-value < 0.001). Potential publication bias was found by visual assessment of the funnel plot, and confirmed by Egger's linear regression test (intercept 6.97, t = 2.47, p-value = 0.069). However, after the trim and fill method was applied, the estimated ES did not change.

Sub-Group Analysis by Protective Equipment Used
The sub-group analysis including studies with a low level of protection (considered as no protection at all or use of only gloves and masks) counted six datasets. The pooled ES was 2.19 [(95% CI = 0.97-3.42), p-value < 0.001] based on 361 subjects (Figure 6a). High statistical heterogeneity was found (Chi 2 = 93.63, df = 5, I 2 = 94.66, p-value < 0.001). No publication bias was found by visual assessment of the funnel plot, which was confirmed by Egger's linear regression test (intercept 6.52, t = 1.30, p-value = 0.265). The sub-group analysis considering a high level of protection (at least gloves, masks, and air-flow cabinet), included six studies, and the pooled ES was 2.13 [ (95% CI = 0.85-3.40), p-value = 0.001] based on 429 subjects (Figure 6b). In this sub-group analysis, Kopjar 2009 was excluded because the exposed group was composed of nurses who used heterogeneous protective equipment (among the 50 subjects, only 3 of them used gloves, masks, and air-flow cabinet, but the results were only shown as aggregated data). High statistical heterogeneity was found (Chi 2 = 128.11, df = 5, I 2 = 96.10, p-value < 0.001). Potential publication bias was found by visual assessment of the funnel plot, and confirmed by Egger's linear regression test (intercept 6.97, t = 2.47, p-value = 0.069). However, after the trim and fill method was applied, the estimated ES did not change.
(a)  Figure 6. Forest plot of the sub-group analysis by protective equipment used assessing the comet assay on lymphocytes cells among professionals occupationally exposed to antineoplastic drugs (ANDs). (a) Without any equipment or at least gloves and masks; (b) using at least gloves, masks, and air-flow cabinet.

Sub-Group Analysis by Work Task
In the sub-group analysis by work task, we included only studies that considered, as controls, health professionals not exposed to ANDs different from nurses. In this analysis, 13 datasets were included for females and the pooled ES was 1.24 [ (95% CI = 0.58-1.89), p-value < 0.001] based on 924 subjects (Figure 7). High statistical heterogeneity was found (Chi 2 = 217.72, df = 12, I 2 = 94.49, p-value < 0.001). No publication bias was found by visual assessment of the funnel plot and confirmed by Egger's linear regression test (Intercept 4.68, t = 1.33, p-value = 0.209).

Figure 7.
Forest plot of the sub-group analysis by work task (including only studies where controls were other health professionals different from nurses) assessing the comet assay on lymphocytes cells among professionals occupationally exposed to antineoplastic drugs (ANDs).

Sub-Group Analysis by Continent
In the sub-group analysis by continent, the number of datasets of studies conducted in Europe was 11 (Figure 8). High statistical heterogeneity was found (Chi 2 = 27.64, df = 2, I 2 = 92.76, p-value < Figure 6. Forest plot of the sub-group analysis by protective equipment used assessing the comet assay on lymphocytes cells among professionals occupationally exposed to antineoplastic drugs (ANDs). (a) Without any equipment or at least gloves and masks; (b) using at least gloves, masks, and air-flow cabinet.

Sub-Group Analysis by Work Task
In the sub-group analysis by work task, we included only studies that considered, as controls, health professionals not exposed to ANDs different from nurses. In this analysis, 13 datasets were included for females and the pooled ES was 1.24 [ (95% CI = 0.58-1.89), p-value < 0.001] based on 924 subjects (Figure 7). High statistical heterogeneity was found (Chi 2 = 217.72, df = 12, I 2 = 94.49, p-value < 0.001). No publication bias was found by visual assessment of the funnel plot and confirmed by Egger's linear regression test (Intercept 4.68, t = 1.33, p-value = 0.209). (b) Figure 6. Forest plot of the sub-group analysis by protective equipment used assessing the comet assay on lymphocytes cells among professionals occupationally exposed to antineoplastic drugs (ANDs). (a) Without any equipment or at least gloves and masks; (b) using at least gloves, masks, and air-flow cabinet.

Sub-Group Analysis by Work Task
In the sub-group analysis by work task, we included only studies that considered, as controls, health professionals not exposed to ANDs different from nurses. In this analysis, 13 datasets were included for females and the pooled ES was 1.24 [ (95% CI = 0.58-1.89), p-value < 0.001] based on 924 subjects (Figure 7). High statistical heterogeneity was found (Chi 2 = 217.72, df = 12, I 2 = 94.49, p-value < 0.001). No publication bias was found by visual assessment of the funnel plot and confirmed by Egger's linear regression test (Intercept 4.68, t = 1.33, p-value = 0.209).

Figure 7.
Forest plot of the sub-group analysis by work task (including only studies where controls were other health professionals different from nurses) assessing the comet assay on lymphocytes cells among professionals occupationally exposed to antineoplastic drugs (ANDs).

Sub-Group Analysis by Continent
In the sub-group analysis by continent, the number of datasets of studies conducted in Europe was 11 and the pooled ES was 0.92 [ (95% CI = 0.16-1.68), p-value = 0.018] based on 890 subjects (Figure 8). High statistical heterogeneity was found (Chi 2 = 248.68, df = 10, I 2 = 95.98, p-value < 0.001). The number of datasets of studies conducted in Asia was seven and the pooled ES was 1.85 [ (95% CI = 0.32-3.38), p-value < 0.001] based on 494 subjects (Figure 8). High statistical heterogeneity was found (Chi 2 = 262.49, df = 10, I 2 = 97.42, p-value < 0.001). The number of datasets of studies conducted in Americas was three and the pooled ES was 1.15 [ (95% CI = −0.27-2.56), p-value < 0.001] based on 185 subjects (Figure 8). High statistical heterogeneity was found (Chi 2 = 27.64, df = 2, I 2 = 92.76, p-value < Figure 7. Forest plot of the sub-group analysis by work task (including only studies where controls were other health professionals different from nurses) assessing the comet assay on lymphocytes cells among professionals occupationally exposed to antineoplastic drugs (ANDs).

Sub-Group Analysis by Continent
In the sub-group analysis by continent, the number of datasets of studies conducted in Europe was 11 and the pooled ES was 0.92 [ (95% CI = 0.16-1.68), p-value = 0.018] based on 890 subjects (Figure 8). High statistical heterogeneity was found (Chi 2 = 248.68, df = 10, I 2 = 95.98, p-value < 0.001). The number of datasets of studies conducted in Asia was seven and the pooled ES was 1.85 [(95% CI = 0.32-3.38), p-value < 0.001] based on 494 subjects (Figure 8). High statistical heterogeneity was found (Chi 2 = 262.49, df = 10, I 2 = 97.42, p-value < 0.001). The number of datasets of studies conducted in Americas was three and the pooled ES was 1.15 [ (95% CI = −0.27-2.56), p-value < 0.001] based on 185 subjects (Figure 8). High statistical heterogeneity was found (Chi 2 = 27.64, df = 2, I 2 = 92.76, p-value < 0.001). Potential publication bias was found by visual assessment of the funnel plot and confirmed by Egger's linear regression test (intercept 8.92, t = 3.54, p-value = 0.002). However, after applying the trim and fill method, the estimated ES did not change. 0.001). Potential publication bias was found by visual assessment of the funnel plot and confirmed by Egger's linear regression test (intercept 8.92, t = 3.54, p-value = 0.002). However, after applying the trim and fill method, the estimated ES did not change. Figure 8. Forest plot of sub-group analysis by continent, assessing the comet assay on lymphocytes cells among professionals occupationally exposed to antineoplastic drugs (ANDs).

Cumulative Analysis
In order to estimate the evolution of the ES across the publications, a cumulative analysis was performed. The cumulative analysis by the year of publication showed a direct association between occupational exposure to ANDs and comet assay, and the results remained stable until the 2008. In 2009, the results started to lose significance, with a large 95% CI (Figure 9a). Referring to the cumulative analysis for sample size, datasets with small sample showed a larger 95% CI. On the contrary, the higher the sample size, the more stable the result (Figure 9b).
(a) Figure 8. Forest plot of sub-group analysis by continent, assessing the comet assay on lymphocytes cells among professionals occupationally exposed to antineoplastic drugs (ANDs).

Cumulative Analysis
In order to estimate the evolution of the ES across the publications, a cumulative analysis was performed. The cumulative analysis by the year of publication showed a direct association between occupational exposure to ANDs and comet assay, and the results remained stable until the 2008. In 2009, the results started to lose significance, with a large 95% CI (Figure 9a). Referring to the cumulative analysis for sample size, datasets with small sample showed a larger 95% CI. On the contrary, the higher the sample size, the more stable the result (Figure 9b). 0.001). Potential publication bias was found by visual assessment of the funnel plot and confirmed by Egger's linear regression test (intercept 8.92, t = 3.54, p-value = 0.002). However, after applying the trim and fill method, the estimated ES did not change. Figure 8. Forest plot of sub-group analysis by continent, assessing the comet assay on lymphocytes cells among professionals occupationally exposed to antineoplastic drugs (ANDs).

Cumulative Analysis
In order to estimate the evolution of the ES across the publications, a cumulative analysis was performed. The cumulative analysis by the year of publication showed a direct association between occupational exposure to ANDs and comet assay, and the results remained stable until the 2008. In 2009, the results started to lose significance, with a large 95% CI (Figure 9a). Referring to the cumulative analysis for sample size, datasets with small sample showed a larger 95% CI. On the contrary, the higher the sample size, the more stable the result (Figure 9b).
(a) (from the smallest to the biggest) assessing the comet assay on lymphocytes cells among professionals occupationally exposed to antineoplastic drugs (ANDs).

Moderator Analysis
Considering the year of publication as covariate in the moderator analysis, the results changed significantly between the three groups of years. Particularly, the early studies (before 2004) did not find a statistically significant association between occupational exposure to ANDs and comet assay. Considering the time period 2005-2010, a higher number of studies were conducted, increasing the sample size (n = 1082) and stabilizing the results. However, considering only studies conducted after 2010, only three datasets were included, reducing the sample size (n = 333) and increasing the uncertainty of the results, confirmed by the large 95% CI (Figure 10a). The results of the metaregression analysis by duration of the exposure showed a linear association; a higher duration of exposure was associated with a higher ES (Figure 10b). The association was less clear when the age of exposed groups was considered (Figure 10c). (from the smallest to the biggest) assessing the comet assay on lymphocytes cells among professionals occupationally exposed to antineoplastic drugs (ANDs).

Moderator Analysis
Considering the year of publication as covariate in the moderator analysis, the results changed significantly between the three groups of years. Particularly, the early studies (before 2004) did not find a statistically significant association between occupational exposure to ANDs and comet assay. Considering the time period 2005-2010, a higher number of studies were conducted, increasing the sample size (n = 1082) and stabilizing the results. However, considering only studies conducted after 2010, only three datasets were included, reducing the sample size (n = 333) and increasing the uncertainty of the results, confirmed by the large 95% CI (Figure 10a). The results of the meta-regression analysis by duration of the exposure showed a linear association; a higher duration of exposure was associated with a higher ES (Figure 10b). The association was less clear when the age of exposed groups was considered (Figure 10c and (c) meta-regression analysis by mean age of exposed groups, assessing the comet assay on lymphocytes cells among professionals occupationally exposed to antineoplastic drugs (ANDs).

Discussion
The current systematic review with meta-analysis-which included 20 studies in the qualitative evaluation and 19 studies in the quantitative analysis-provided data on the association between occupational exposure to ANDs and the extent of primary DNA damage, as evaluated by the comet assay. As some studies separately reported data for gender, work task, and protective equipment, and a total of 21 datasets were considered. The pooled ES based on 1569 subjects indicated a significantly higher extent of primary DNA damage in health professionals exposed to ANDs compared with controls, with an ES of 1.27 [(95% CI = 0.66-1.88), p-value < 0.001]. The ES did not change significantly when, in the sub-group analysis, other health professionals different from nurses were considered as controls, even if it was slightly lower [1.24 (95% CI = 0.58-1.89), p-value < 0.001].
In order to deeply understand the strength of the association between occupational exposure and primary DNA damage, a sub-group analysis by protective equipment was conducted. When studies that assessed the damage among health professionals who did not use any protective equipment or only masks and gloves were considered, the pooled ES was higher compared with the studies where health professionals used a combination of personal and environmental protective equipment (at least gloves, masks, and airflow cabinet). The importance of protective equipment could explain the differences obtained in the sub-group analysis by continent. Particularly, studies conducted in Asia showed the highest pooled ES compared with studies conducted in the Americas and even more in Europe. As suggested by Hon et al., different standard procedures, protocols, level of education, and legislations across the globe might influence the level of occupational risk, and consequently the conclusions drawn from one country might not be fully applicable to another [50]. Furthermore, in the cumulative analysis by year of publication, the results remained stable until 2008, while from 2009, they started to lose significance, with a large 95% CI. It should be noticed that a large number of guidelines on ANDs' manipulation, aimed to reduce the occupational risk, were published between 2000 and 2009 [27][28][29]51]. Probably, the reinforcement of protective occupational  ; and (c) meta-regression analysis by mean age of exposed groups, assessing the comet assay on lymphocytes cells among professionals occupationally exposed to antineoplastic drugs (ANDs).

Discussion
The current systematic review with meta-analysis-which included 20 studies in the qualitative evaluation and 19 studies in the quantitative analysis-provided data on the association between occupational exposure to ANDs and the extent of primary DNA damage, as evaluated by the comet assay. As some studies separately reported data for gender, work task, and protective equipment, and a total of 21 datasets were considered. The pooled ES based on 1569 subjects indicated a significantly higher extent of primary DNA damage in health professionals exposed to ANDs compared with controls, with an ES of 1.27 [(95% CI = 0.66-1.88), p-value < 0.001]. The ES did not change significantly when, in the sub-group analysis, other health professionals different from nurses were considered as controls, even if it was slightly lower [1.24 (95% CI = 0.58-1.89), p-value < 0.001].
In order to deeply understand the strength of the association between occupational exposure and primary DNA damage, a sub-group analysis by protective equipment was conducted. When studies that assessed the damage among health professionals who did not use any protective equipment or only masks and gloves were considered, the pooled ES was higher compared with the studies where health professionals used a combination of personal and environmental protective equipment (at least gloves, masks, and airflow cabinet). The importance of protective equipment could explain the differences obtained in the sub-group analysis by continent. Particularly, studies conducted in Asia showed the highest pooled ES compared with studies conducted in the Americas and even more in Europe. As suggested by Hon et al., different standard procedures, protocols, level of education, and legislations across the globe might influence the level of occupational risk, and consequently the conclusions drawn from one country might not be fully applicable to another [50]. Furthermore, in the cumulative analysis by year of publication, the results remained stable until 2008, while from 2009, they started to lose significance, with a large 95% CI. It should be noticed that a large number of guidelines on ANDs' manipulation, aimed to reduce the occupational risk, were published between 2000 and 2009 [27][28][29]51]. Probably, the reinforcement of protective occupational strategies, the higher attention reserved to the occupational health risk, and particularly the awareness of ANDs dangerousness even at low doses might have played an important role in reducing the occupational exposure risk. However, guidelines alone are not enough, because health professionals need to be empowered with regards to potential and accidental contaminations through tailored training and communication [52]. As a matter of fact, even if health professionals are exposed to a low or very low dose of ANDs, the exposure is nearly daily and prolonged at times. According to our results, the duration of exposure was an important moderator of the association between ANDs exposure and primary DNA damage. This result is in contrast with the data obtained by Mader et al.-the only study that stratified the results according to the duration of exposure (expressed in years) [49]. Considering the sub-group analysis by gender, the pooled ES was marginally higher among men compared with women. However, these data should be read with caution. Firstly, the total datasets assessing primary DNA damage among men were only 5, while 10 datasets performed the comet assay among women; secondly, the men's sample size was 10 times lower compared with that of the women; thirdly, the 95% CI was wider among men. An important aspect should be considered before the generalization of these results, that is, even though several sub-group analyses were conducted, the value of heterogeneity remained stably high. Indeed, although a sensitivity analysis including only datasets with a high quality score (QS ≥ 16) was conducted, the I 2 remained around 95%. Only in one case was the heterogeneity below 40%, when datasets performing the comet assay on buccal cells were considered. However, only four datasets provided this data, resulting in a small sample size (n = 180 subjects), wide 95% CI, and loss of statistical significance. The high value of heterogeneity found could be explained considering the methodological variation of the comet assay used in the considered primary studies. Indeed, primary papers expressed the results of the comet assay using different types of unit of measures. Moreover, an I 2 value higher than 90% means that heterogeneity is directly the result of heterogeneity among studies, rather than sampling errors [53]. Even if the pooled ES was estimated by Cohen's d, allowing comparability, this underlying heterogeneity might have affected the assessment of the I 2 . Another potential explanation of heterogeneity could be the different type of ANDs used or the type of exposure (not only duration or work task, but also doses and therapeutic scheme used). Moreover, even if the primary studies tried to match exposed and control subjects for several confounders, only a few studies perfectly matched for the most important elements, such as smoking, alcohol, and nutritional intake. Nevertheless, there are several strengths in this study: first, this is an extensive systematic review conducted according the PRISMA guidelines, which allowed retrieving a large number of studies; second, the pooled ES was based on 1569 subjects; and third, several sub-group analyses as well as sensitivity and moderator analyses were conducted. Particularly, the ES was estimated in both women and men, considering the work task and the protective equipment used. Moreover, in the meta-regression, the duration of exposure was considered as moderator. Furthermore, whenever we found potential publication bias, a trim and fill method was applied. Even if high heterogeneity was found, a random effect model was considered. Lastly, the pooled ES was estimated both in lymphocytes and buccal cells.

Conclusions
To conclude, the results of this systematic review with meta-analysis clearly show a statistically significant association between occupational exposure to ANDs and the extent of primary DNA damage. The comet assay method is able to identify reversible DNA damage; in this perspective, the identified damage is still potentially repairable. However, a prolonged exposure, a concomitant action of other risk factors, and the senescence of DNA repair systems might induce a risk of neoplasms or other side effects (such as, for instance, a teratogenic action). Having said that, health professionals should be made aware of this potential occupational risk.