Cancer risks of hematopoietic/lymphatic and urinary system in pharmaceutical plant workers: a systematic review and meta-analysis

Abstract

Introduction

In this study, we will investigate the risk of hematopoietic/lymphatic system and urinary system cancers in pharmaceutical plant workers.

Methods

Six pre-defined inclusion criteria were applied. The quality of all included studies was assessed based on the National Toxicology Program Office of Health Assessment and Translation Risk of Bias tool. Conventional meta-analysis was conducted for hematopoietic/lymphatic system and urinary system cancers. Between-study heterogeneity and publication bias were investigated.

Results

In this systematic review and meta-analysis, four cohort studies of a single pharmaceutical plant, one case-control study, and one cohort study with a nested case-control study were included. For publication bias, the p-value for Egger’s regression test was 0.51 and 0.38 for hematopoietic/lymphatic and urinary system cancer, respectively. Pharmaceutical plant workers showed a 3.19 (95% confidence interval, CI 1.53–6.64) times higher risk of hematopoietic/lymphatic system cancers than the general population. They also showed a 4.86 (95% CI 1.71–13.80) times higher risk of urinary system cancers compared to the general population. Higgins’ I-squared value for between-study heterogeneity was 61.77% (95% CI 0.00–99.06%) and 58.00% (95% CI 0.00–98.67%), respectively.

Conclusion

From this systematic review and meta-analysis, the authors concluded that potential hazardous occupational exposures of workers in pharmaceutical plants could pose an increased risk of hematopoietic/lymphatic system and urinary system cancer. Future studies should be conducted with a more thorough dose-response relationship distinction.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-025-24408-2.

Introduction

The pharmaceutical manufacturing industry is growing in the 21st century [1]. Many global and local pharmaceutical companies are manufacturing conventional and innovative new pharmaceuticals. However, because many chemicals are used during manufacturing processes, the pharmaceutical manufacturing industry could expose its workers to various hazardous chemicals [2].

The pharmaceutical industry is a vital sector that contributes significantly to public health by developing life-saving medications. However, the production processes and occupational exposures within this industry have raised concerns about the health risks faced by workers. Benzene may be found in pharmaceutical manufacturing environments as a volatile organic compound (VOC) [15]. Employees engaged in the production of antineoplastic medications (used in chemotherapy) may be exposed to carcinogenic hazards because of the genotoxic nature of these substances [18].

A systematic review by the National Toxicology Program (NTP) found inadequate evidence to conclusively link occupational exposure to cancer chemotherapy agents with increased cancer risk in pharmaceutical workers. However, moderate evidence exists for genetic toxicity and reproductive harms [3]. However, several individual studies have identified an increased risk of bladder and urothelial cancers among pharmaceutical workers. A retrospective cohort study in Sweden found a significant increase in the risk of urothelial tumors, with a standardized incidence ratio (SIR) of 3.5 (95% CI 1.5–6.8) among the highest exposed workers [4]. Breast cancer has also been a focus of several studies. A Danish cohort study found an elevated SIR of 1.5 for breast cancer among female pharmaceutical workers, particularly those who started working at the factory between the ages of 30 and 39 and had worked for 1–9 years (SIR = 2.8) [5]. Respiratory system cancers have been another area of concern. A mortality study of British pharmaceutical workers noted elevated risks for respiratory cancer among male maintenance workers and female production workers [6]. Acute leukemia has been identified as a significant risk in some studies. The Swedish cohort study mentioned earlier found a SIR of 4.5 (95% CI 1.2–12) for acute leukemia, although this risk decreased when a 10-year induction-latency period was applied [4].

In 1993, Notani et al. reported the increased odds ratio (OR) for bladder cancer among workers in chemical/pharmaceutical/rubber, etc. plants in a case-control study [7]. In this study, the OR for lung cancer among these workers was not increased with statistical significance. In 1995, Edling et al. reported an increased SIR for renal pelvis cancer and acute leukemia for male workers with no induction latency [4]. Except for the cancer of the peritoneum, no other organ showed an increased SIR with statistical significance. Considering only one case was reported as the peritoneal cancer, the authors concluded that hematopoietic/lymphatic system and urinary system cancer could be potential candidates. Youk et al. reported in 2009 that only hematopoietic/lymphatic system cancer showed an increased SMR with statistical significance [8]. Other systems did not show an increased SMR with statistical significance.

Based on these previous studies, the authors concluded that hematopoietic/lymphatic system and urinary system cancers would be the strong potential candidates for increased cancer incidence in pharmaceutical plant workers. Therefore, the authors will investigate the risk of hematopoietic/lymphatic system and urinary system cancers in pharmaceutical plant workers. We will conduct a systematic review and meta-analysis on the association between working in pharmaceutical plants and the risk of hematopoietic/lymphatic and urinary system cancers.

Methods

Literature search and inclusion criteria

Supplementary material A-1 provides search terms and syntax. The inclusion criteria were the following: (1) The study should address the incidence of/mortality due to cancers in pharmaceutical plant workers. (2) The study should report outcomes as quantitative measures, including risk ratios (RRs), ORs, hazard ratios (HRs), SIRs, or standardized mortality ratios (SMRs). (3) The article should be an original study: not review articles, conference abstracts, letters to the editor, or commentaries. (4) The article should be written in English. (5) The study subjects should be humans, not animals. (6) The studies that investigated workers dealing with oriental herbal medicine were excluded. Supplementary material A-2 provides the number of excluded studies and the main reason for exclusion (among the six pre-defined inclusion criteria).

Characteristics of the studies selected by the systematic review

For the included study for the systematic review, basic characteristics including publication year, study period, country of study, reported effect estimate statistic, comparison group, outcome (type of cancer), the specific exposure for the extracted effect estimate, the value of effect estimate, and whether or not included in the meta-analysis were summarized. The discussion of each paper regarding the association between the probable occupational exposure and the observed risk of cancer was summarized in a separate table.

Sources of heterogeneity

To investigate the sources of heterogeneity observed in meta-analyses, the authors summarized the selection of study population, exposure assessment, outcome reporting, and possible confounding (and adjustment) for each study. For readability, a concise summary of each aspect was provided just below the long explanation for each study.

Quality assessment of each included study

The quality of all included studies was assessed based on the National Toxicology Program (NTP) Office of Health Assessment and Translation (OHAT) risk of bias (RoB) tool. The specific questions in the NTP OHAT RoB tool are provided in Table 1. The first author, JM, rated each study.

Table 1.

Characteristics of the studies selected by the systematic review

Study (type)	Publication year	Study period	Country	Statistics	Comparison group	Outcome (for the meta-analysis)	Specific exposure (for the synthesized effect estimate)	Effect estimate (for the meta-analysis) (95% CI)	Inclusion in meta-analyses
Marsh et al. [12] (cohort of a single plant)	2005	1970–1996	US	SMR	local county population	Hematopoietic and lymphatic system cancer	Time since first employment > 20 (not included in the meta-analysis due to study population duplication with Youk et al. (2009) (13))	4.66 (1.71–10.14)	Not included
Dolan et al. [14] (cohort of a single plant)	2004	1950–1999	US	SMR	local county population	Hematopoietic and lymphatic system cancer (leukemia and aleukemia)	All employees, local county comparison	1.60 (0.64–3.29)	Included
						Urinary system cancer (kidney)	All employees, local county comparison	2.32 (0.85–5.06)	Included
Edling et al. [4] (cohort of a single plant)	1995	1960–1990	Sweden	SIR	national population	Hematopoietic and lymphatic system cancer (acute leukemia)	Men, no induction-latency	5.60 (1.20–16.00)	Included
						Urinary system cancer (renal pelvis)	Men, no induction-latency	12.00 (1.40–43.00)	Included
Hansen et al. [5] (cohort of a single plant)	1994	1964–1988	Denmark	SIR	national population	various cancers (not included in the meta-analysis)	Various, including overall, length of employment, and years since start of follow-up (not included in the meta-analysis)		Not included
Notani et al. [7] (a case-control study)	1993	1986–1990 (interview period)	India	OR	hospital control	Urinary system cancer (bladder)	Ever employed in the industry of interest vs. not exposed group	5.76 (1.20–26.90)	Included
Youk et al. [13] (cohort of a single plant with a nested case-control study)	2009	1970–1996	US	SMR (OR)	local county population	Hematopoietic and lymphatic system cancer	Time since first employment 20–29 years	4.15 (1.90–7.88)	Included

CI confidence interval, SMR standardized mortality ratio, SIR standardized incidence ratio, OR odds ratio

Examination of publication bias

The existence of publication bias was examined using Begg’s funnel plot (asymmetry) and Egger’s regression test. If Begg’s funnel plot shows an asymmetric shape, the existence of a publication bias is suspected. Egger’s regression test uses the precision and the standardized effect size of the effect estimate from a study as the independent variable and dependent variable, respectively [9]. If Egger’s regression test result shows a statistically significant result, the existence of a publication bias could be suspected. The statistically significant p-value for publication bias was set at 0.05.

Data extraction

Risk estimates from each study were extracted and stratified by the system of cancer, hematopoietic/lymphatic, and urinary systems, respectively.

Main meta-analyses

Each article reported a number of risk estimates. To synthesize evidence, we converted SIR and SMR into RR. The SIR and SMR can serve as estimates of RR under specific conditions. First, age-specific mortality/incidence ratios (MIRs) must be proportional between the study cohort and reference population. If MIRs vary non-uniformly across subgroups (e.g., age brackets), SIR/SMR becomes an oversimplified summary statistic. Second, the comparison population must share fundamental risk characteristics with the study cohort, except for the exposure under investigation. For occupational studies, national mortality rates often serve this purpose. Third, the multiplicative risk assumption should be satisfied. SIR/SMR inherently assumes risk operates multiplicatively (e.g., doubling baseline rates) rather than additively. This aligns with the relative risk’s multiplicative interpretation. Fourth, complete follow-up and accurate exposure data are needed. Missing outcome data or misclassified exposures could distort observed/expected ratios. Fifth, adequate adjustment for potential confounders is needed. While SIR/SMR standardizes for factors like age and sex, unadjusted confounders (e.g., socioeconomic status) may bias results. If these prerequisites are satisfied, SIR/SMR can be used as RR in occupational epidemiology studies. Based on these conditions, the authors determined that SIR and SMR reported in each study can be interpreted as RR.

In addition, we converted ORs into RRs. OR can be interpreted as RR under specific conditions, primarily when the rare disease assumption holds. This occurs when the outcome (e.g., disease) is uncommon in both exposed and unexposed groups. ORs approximate RRs only when the outcome is rare (< 10% prevalence). Because cancer is a rare disease, we concluded that the OR reported from each individual study can be used as the RR.

The results from the common-effect model and the random-effects model were provided together. Based on the calculated heterogeneity indices and the Cochran’s Q test results, the authors concluded that the random-effects model would be more appropriate for these meta-analyses. However, we provided both of these results so that readers could better understand the results of these meta-analyses. A pooled point estimate with a 95% confidence interval (CI) was provided for all meta-analyses.

Heterogeneity indices

For the between-study variance, τ², the restricted maximum-likelihood estimator was used. Higgins’ I-square statistic and Cochran’s Q-test statistic were also provided. The τ², τ, I², and H² were provided for heterogeneity indices. Tau-squared (τ²) represents the variance of the true effects across studies in a random-effects meta-analysis. Tau-squared quantifies the magnitude of heterogeneity in terms of variance. A larger τ² indicates greater heterogeneity. Tau (τ) is the square root of τ² and represents the standard deviation of the true effects across studies. I-squared (I²) quantifies the proportion of total variation in study estimates that is due to heterogeneity rather than sampling error. I² is calculated as the ratio of the excess variability (beyond chance) to the total variability, expressed as a percentage. I² values range from 0 to 100%, where 0%, 50%, and 100% indicate no heterogeneity (all variability is due to sampling error), moderate heterogeneity, and maximal heterogeneity, respectively. I-squared does not describe the magnitude of effect size variation; it only reflects the proportion of variability due to heterogeneity. For example, a high I² value does not necessarily mean significant differences in effect sizes, but rather that a large proportion of the observed differences are due to heterogeneity. While I² focuses on the proportion of variability due to heterogeneity, H² reflects the overall magnitude of heterogeneity. For example, an H² value greater than 1 indicates heterogeneity beyond sampling error, while a value close to 1 suggests minimal heterogeneity [10].

Cochran’s Q-test is a statistical method used in meta-analyses to assess heterogeneity. The null hypothesis of Cochran’s Q-test is that all studies estimate the same underlying effect with no heterogeneity. The alternative hypothesis is that at least one study’s effect size differs statistically significantly from others (in other words, heterogeneity exists). The test statistic Q follows a chi-square distribution with k − 1 degrees of freedom, where k is the number of studies. The p-value for Cochran’s Q-test indicates whether or not to reject the null hypothesis (no heterogeneity).

Sensitivity analysis (‘leave-one-out’ approach) and radial plots

To evaluate the effect of possible outliers in meta-analyses, the authors conducted separate meta-analyses, excluding one study at a time. For hematopoietic/lymphatic system cancer and urinary system cancer, this sensitivity analysis was conducted, respectively. We plotted a radial plot for each meta-analysis to assess the heterogeneity and detect possible outliers.

Statistical software

For all statistical analyses, R software version 4.4.2 was used. The R package ‘meta’ was used for meta-analyses. For the calculation of heterogeneity indices, the R package ‘metafor’ was used.

Results

Screening and selection processes

Supplementary material B provides search results. In total, 377 articles were retrieved from PubMed (n = 10), EMBASE (n = 347), Cochrane Library (n = 7), and Medline (n = 13), respectively. Among the total 377 articles, 24 were duplicate articles. Finally, 353 articles were included in the section process.

The PRISMA flow diagram is provided in Fig. 1. Four cohort studies with a single pharmaceutical plant, one case-control study, and one cohort study of a single pharmaceutical plant with a nested case-control study were included in this meta-analysis.

Fig. 1.

PRISMA flow diagram [11]

Characteristics of the studies selected by the systematic review

Table 1 provides the characteristics of the included studies. After careful examination of each study, we did not include Marsh et al. [12] in the meta-analyses because the study population was the same as that of Youk et al. [13]. We meticulously examined any remaining possibility of study population duplication during the process of checking the possibility of selection bias (making Table 2). However, we did not find any duplication among the remaining five studies. The outcome of the meta-analysis was hematopoietic/lymphatic system cancer in four studies and urinary system cancer in three studies. Because the reported outcomes do not include hematopoietic/lymphatic system cancer and urinary system cancer, the Hansen et al. (1994) study was excluded from the meta-analyses.

Table 2.

Sources of heterogeneity in meta-analyses

	Selection	Exposure	Outcome	Confounding
Dolan et al. (2004) [14]	“Study members included 1958 men and women who were identified as belonging to the cohort through an iterative series of computerized and manual searches of employment records, union rolls, historical documents, and interviews with retired employees. The accuracy of the cohort enumeration was verified through a series of internal data consistency checks.”	“Quantitative industrial hygiene data in an electronic format were readily available for only a fraction of the 50-year period of observation. Consequently, the plant industrial hygienist, a 22-year employee at the plant, was furnished with a computer listing of each job held by every employee, where available, and requested to classify each employee’s job type, in descending order of potential chemical exposure, as either: 1) production; 2) laboratory; 3) trade/craft; 4) service; or 5) administrative or clerical.” “As seen in Tables 1 and 43% of workers had their highest exposures in jobs held in the production category while around 14% of the workers had their highest exposures in jobs held in the unknown job category.”	“Using a two-stage vital status tracing protocol,13 we identified 384 deaths among this cohort as of December 31, 1999, and determined cause of death for 378 (98.4%) via the National Death Index or death certificates (Table 1). All deaths were coded to underlying cause according to the International Classification of Diseases and Causes of Death rules in effect at the time of death. No study members were lost to follow-up.”	“No formal adjustments were made for the many statistical comparisons performed.”
For hematopoietic and lymphatic system cancer	Cohort of a single plant	All employees vs. local county comparison	Leukemia and aleukemia, SMR	No confounder adjustment
For urinary system cancer	Cohort of a single plant	All employees vs. local county comparison	Kidney cancer, SMR	No confounder adjustment
Edling et al. (1995) [4]	“A retrospective entry cohort base was established, comprising all subjects who had worked for at least six months at the Pharmacia Company ally time during the period 1960–1990. The inclusion criterioil was, besides employment, a possible exposure to chemical, pharmacological, or biological agents. Thus not only laboratory workers, but also production personnel were included.”	“The information on work tasks was the base for the exposure classification. An exposure subgrouping was made by representatives from the company and the unions. Four different groups were created. Group I comprised subjects working at chemical laboratories or production units including high risk laboratories (ie, special and separate facilities for laboratory-scale safe handling of highly toxic chemicals) and workers exposed to ionizing radiation. Group I1 included those who had worked at pharmaceutical laboratories and production units. Group I11 was made up of worlters from biological laboratories, and group IV included those who could be indirectly exposed to chemicals and the like in, for example cleaning, washing, and storage jobs and administrative personnel working within the laboratory and production facilities. The exposure was considered low among people with indirect exposure and highest among those working in chemical laboratories and chemical production. Workers who had belonged to more than one exposure subgroup during employment were allocated to thc group where they had spent the longest exposure time. In cases of doubt, the highest exposure group in question was chosen. The potential exposures in the chemical laboratories have been extremely varied. With exception for common organic solvents, there had beell no long-term exposures to any specific chemical compound. Substances now classified as carcinogens according to the Swedish National Board of Occupational Safety and Health (4-aminodiphenyl, benzidine and beta-naphthylamine) had been used in the chemical aild pharmaceutical laboratories. Today these substances are banned, but there are still substances in use with carcinogenic potential (ie, diethyl sulfate, dimethyl sulfate, and ethylene dibromide). These substances were used with permission from the Factory Inspectorate.”	“Information about vital status and present address was collected from SPAR, a register of the entire Swedish population. Mortality data was collected from the National Death Register of Statistics Sweden. The information on cancer incidence was collected from the Swedish Cancer Register at the National Board of Health and Welfare.”	No confounder adjustment
For hematopoietic and lymphatic system cancer	Cohort of a single plant, Men with no induction latency	All male employees vs. national population	Acute leukemia, SIR	No confounder adjustment
For urinary system cancer	Cohort of a single plant, Men with no induction latency	All male employees vs. national population	Renal pelvis cancer, SIR	No confounder adjustment
Youk et al. (2009) [13]	“The cohort included male workers with some full-time work experience in the plant from 1970 to the end of 1996. Various computerized databases and hard copy employee files were used to identify all the employees who were eligible for inclusion in the cohort. Cohort completeness was verified by sampling a portion of the employee names from each source and checking the names against the other databases.”	“Individual worker-level, multiexposure profiles were developed for all exposure agents present in the plant for which air monitoring data on measurable exposure levels were available from the company (acetone, acetonitrile, dimethyl formamide, ethyl acetate, ethylene dichloride, isopropyl alcohol, methlyene chloride, trichloroethane, and toluene). Exposure estimates were linked to individual members of the cohort via detailed work histories. We computed three summary exposure indices for individual workers through the end of 1996: duration of exposure, average intensity of exposure, and cumulative exposure. Because many of the employees worked in administrative and staff positions with little or no potential for plant exposure, the cohort was also classified into those with and those without potential for exposure.”	“Vital status tracing was conducted through December 31, 2004. Underlying cause of death codes were obtained from the National Death Index-Plus system or from death certificates obtained from state health departments. Death certificates were coded to the underlying cause of death by a nosologist using the International Classification of Diseases rules in effect at time of death. Subjects with invalid Social Security Numbers were classified as lost to follow-up and person-years were stopped at the date of termination.”	Potential confounders were adjusted only in the nested case-control study. (The results were reported as ORs.)
For hematopoietic and lymphatic system cancer	Cohort of a signle plant, all male employees	time since first employment 20–29 years vs. local county comparison	Hematopoietic and lymphatic system cancer, SIR	No confounder adjustment
Notani et al. (1993) [7]	“Male cases of cancer of the lung (n = 246) and bladder (n = 153) who gave their residence as being within the State of Maharashtra were interviewed between June 1986 and May 1990. Most of these cases had histological/cytological confirmation (98% of lung cancers and 99% of bladder cancers) and the remaining few had radiological/clinical confirmation only. To avoid any biases which bring patients from farflung places, the selection of cases as well as the controls was restricted to those from Maharashtra State of which Bombay city is the capital. Sex-matched controls (n = 212) were obtained from patients with diagnosis of cancer of the mouth (n = 160), cancer of the oro- or hypo-pharynx (n = 27) and non-cancerous oral disease (n = 25). Cancers of the mouth and pharynx were selected as controls as no specific occupational exposure has been reported in their aetiology. The controls were selected such that the community distribution, by broad category, was similar to cases. These categories were: Hindus whose mother tongue was Marathi, i.e. the State language (44_52% of cases and controls), Hindus whose mother tongue was other than Marathi (20–25%), Muslims (15–20%), Christians (3–6%) and others (4–5%). The controls were also matched to cases in 5-year age groups. In addition to hospital controls, general population controls (n = 85) were obtained from areas in which individuals of socioeconomic strata similar to the cases lived. This group was only used for studying tobacco effects for the reasons given below.”	“The questionnaire obtained total lifetime occupational history and self-reported history on a few specific exposures. Furthermore, for each job held, information was obtained regarding job designation (title), name and address of place of work, name of specific department of the organization, description of work area and details of actual task performed and the period for which the job was held. The International Standard Classification of Occupation (Revised edition 1968)’ was used. Subjects were also asked whether the job involved handling of chemicals, whether there was dust/fume/smoke/gas in the work environment and whether any protective equipment was used. In addition to occupational history, other relevant information on demographic variables, and confounding variables like tobacco use, alcohol consumption and past medical history, was obtained.” “For analysis of occupational effects, the measure of employment experience was whether an individual had ‘ever’ been employed, for at least a year, in the occupation under consideration. The unexposed group was defined as those subjects who were ‘never’ employed in that occupation. Another assessment of risks was undertaken by using a second unexposed group comprising occupations in which there was little likelihood of exposure to any cancer-causing occupational agent (office workers, teachers, domestic servants etc.) and an odds ratio of about unity was found.”	“It was therefore decided to undertake a case-control study at Tata Memorial Hospital, Bombay, which is the largest facility in the region for treating cancer.”	“Stratified analysis involving eight strata for the confounding factors of age (four groups) and smoking status (two groups) was performed to assess the association between a particular occupation and cancer of the lung or bladder, and relative risks estimated by odds ratios (OR) were calculated by the Mantel-Haenszel method.”
For urinary system cancer	Male cases of bladder cancer (n = 153) in the Tata Memorial Hospital, Bombay (Workers in chemical/pharmaceutical etc. plants)	‘ever’ been employed, for at least a year, in chemical/pharmaceutical plants vs. unexposed group (occupations in which there was little likelihood of exposure to any cancer-causing occupational agent (office workers, teachers, domestic servants etc.) and an odds ratio of about unity was found.)	Bladder cancer, OR (case-control study)	Age and smoking adjusted

Supplementary material C provides the discussion of each paper regarding the association between occupational exposure and observed risk estimates. Among the six studies evaluated, only Marsh et al. [12] and Dolan et al. [14] expressed an opinion that the association between hazardous occupational exposures at the workplace and an increased risk of hematopoietic/lymphatic system or urinary system cancer could not be confirmed.

Sources of heterogeneity in meta-analyses

Table 2 provides the sources of heterogeneity in these meta-analyses. For hematopoietic and lymphatic system cancer in Dolan et al. [14], a cohort of a single plant was selected regarding the selection. Regarding the exposure, all employees who worked at the plant were compared to the local county population. Regarding the outcome, the SMR of leukemia and aleukemia was investigated. Regarding the confounding, no confounder adjustment was conducted. For urinary system cancer in the same study, the selection, exposure, and confounding aspects were the same as those of the hematopoietic and lymphatic system. Regarding the outcome, the SMR of kidney cancer was investigated. For hematopoietic and lymphatic system cancer in Edling et al. [4], men with no induction latency from a cohort of a single plant were selected. Regarding the exposure, all male employees were compared to the national population. Regarding the outcome, the SIR of acute leukemia was investigated. Regarding the confounding, no confounder adjustment was conducted. For urinary system cancer in the same study, the selection, exposure, and confounding aspects were the same as those of the hematopoietic and lymphatic system. Regarding the outcome, the SIR of renal pelvis cancer was investigated. For hematopoietic and lymphatic system cancer in Youk et al. (2009), all male employees from a cohort of a single plant were selected regarding the selection. Regarding the exposure, male employees with the time since first employment from 20 to 29 years were compared to the local county population. Regarding the outcome, the SIR of hematopoietic and lymphatic system cancer was investigated. Regarding the confounding, no confounder adjustment was conducted. For urinary system cancer in Notani et al. [7], bladder cancer cases of workers in the chemical/pharmaceutical plants and appropriate controls were selected from the Tata Memorial Hospital, Bombay, regarding the selection (case-control study). Regarding the exposure, workers who had ever been employed for at least a year in chemical/pharmaceutical plants were compared to ‘probable’ unexposed groups, such as occupations in which there was little likelihood of exposure to any cancer-causing occupational agent, and an odds ratio of about unity was found. Regarding the outcome, the OR of bladder cancer was investigated. Regarding the confounding, age and smoking were adjusted for in calculating the ORs.

Systematic review based on the OHAT RoB rating tool

Supplementary material D provides the results of a systematic review based on the OHAT RoB rating tool. Six original articles were assessed. The overall rating was four for four studies, including Marsh et al. [12], Dolan et al. [14], Edling et al. [4], and Hansen et al. [5], five for one study, Notani et al. [7], and six for one study, Youk et al. [13], respectively. The rationale for each rating is recorded in Supplementary material D.

The four studies with a score of four were rated yes for questions 7, 9, 10, and 11. The study with score five was rated yes for questions 3, 7, 9, 10, and 11. The study with score six was rated yes for questions 3, 7, 8, 9, 10, and 11.

The four studies with a score of four analyzed a cohort of all workers in a pharmaceutical plant. Then, they calculated SIRs or SMRs compared to the national/regional general population. Because of the features of this design, the questions related to outcome assessment (questions 7, 9, and 10) were rated as yes. Question 11 was also rated yes for each study.

The study with a score of five was a case-control study. This study selected lung and bladder cancer cases and appropriate controls matched with the cases. However, the matching processes were not complete. Only community distribution, age group (5-year age group), and socioeconomic strata were matched. Based on this process, the authors rated question 3 (selecting appropriate comparison groups) as yes.

The study with score six was a cohort study with a nested case-control study. As a cohort study, they analyzed a cohort of all male workers in a pharmaceutical plant. Then they calculated SMRs compared to the local county population. In the nested case-control study, respiratory system cancer and hematopoietic and lymphatic system cancer cases were selected as the cases. Each case was matched on the exact age of the case and year of birth (within 5 years) to 10 controls selected randomly from the remaining living and deceased members of the cohort. Based on these processes, the authors rated questions 7, 9, 10, and 3 as yes. In addition, individual worker-level, multiexposure profiles were developed for all exposure agents present in the plant for which air monitoring data on measurable exposure levels were available from the company (acetone, acetonitrile, dimethyl formamide, ethyl acetate, ethylene dichloride, isopropyl alcohol, methylene chloride, trichloroethane, and toluene). Exposure estimates were linked to individual cohort members via detailed work histories. Based on this aspect, question 8 (exposure characterization) was rated as yes.

Examination of publication bias and main meta-analyses

The p-value for Egger’s regression test was 0.51 for the cancer risk of the hematopoietic/lymphatic systems and 0.38 for the cancer risk of the urinary system, respectively. Therefore, the possibility of publication bias in both meta-analyses was scarce. The funnel plot for each analysis is provided in Supplementary material E. The x-axis and y-axis were the effect size (RR) and standard error of each risk estimate, respectively. Both funnel plots showed a generally symmetric distribution of risk estimates.

The authors determined that a random-effect model would be more appropriate than a common-effect model for these two meta-analyses based on Higgins’ I-squared values (64% and 59%) and Cochrane’s Q test results (p-value of 0.06 and 0.09).

Figure 2 provides the forest plot for the cancer risk of hematopoietic/lymphatic systems in pharmaceutical plant workers. Based on a random effects model, the pooled RR was 3.19 (95% CI 1.53–6.64). It means that pharmaceutical plant workers have 3.19 times higher risk of hematopoietic/lymphatic system cancer compared to the general population without exposure to hazardous agents in pharmaceutical plants. Figure 3 provides the forest plot for the cancer risk of the urinary system in pharmaceutical plant workers. Based on a random effects model, the pooled RR was 4.86 (95% CI 1.71–14.80). It means that pharmaceutical plant workers have 4.86 times higher risk of urinary system cancer compared to the general population without exposure to hazardous agents in pharmaceutical plants.

Fig. 2.

Forest plot for the cancer risk of the hematopoietic/lymphatic systems in pharmaceutical plant workers

Fig. 3.

Forest plot for the cancer risk of the urinary system in pharmaceutical plant workers

Heterogeneity indices

In Figs. 2 and 3, the heterogeneity indices for each meta-analysis are provided in a small separate table. The R codes for the calculation are provided in Supplementary material F. The I-squared value for the meta-analysis of hematopoietic and lymphatic system cancer and urinary system cancer was 61.77% with a 95% CI of 0.00–99.06% and 58.00% with a 95% CI of 0.00–98.67%, respectively.

Sensitivity analysis (‘leave-one-out’ approach)

Supplementary material G provides the results of the sensitivity analyses. For hematopoietic/lymphatic system cancer, when the Dolan et al. study was excluded, the pooled RR increased by the largest amount, from 3.19 (95% CI 1.53–6.64) to 4.50 (95% CI 2.60–7.78). The I-squared value decreased from 61.9 to 0%. When the Edling et al. study was excluded, the pooled RR decreased by the largest amount, to 2.62 (95% CI 1.03–6.65). The I-squared value increased from 61.9 to 73.3%. When the Youk et al. study was excluded, the pooled RR decreased to 2.82 (95% CI 0.83–9.56) without statistical significance. From this, we can infer that the Youk et al. study is paramount in making the pooled RR statistically significant. The I-squared value increased from 61.9 to 73.1%. For urinary system cancer, when the Dolan et al. study was excluded, the pooled RR increased by the largest amount, from 4.86 (95% CI 1.71–13.80) to 8.90 (95% CI 3.33–23.79). The I-squared value decreased from 59.4 to 0%. When the Edling et al. study was excluded, the pooled RR decreased by the largest amount, to 2.84 (95% CI 1.35–5.95). The I-squared value decreased from 59.4 to 6.1%. When the Notani et al. study was excluded, the pooled RR decreased to 4.87 (95% CI 0.98–24.17) without statistical significance. From this, we can infer that the Notani et al. study is paramount in making the pooled RR statistically significant. The I-squared value increased from 59.4 to 78.4%.

Supplementary material H provided the radial plots for each meta-analysis. For hematopoietic/lymphatic system and urinary system cancer, all three studies were included in the 95% confidence bands, respectively.

Discussion

This systematic review and meta-analysis included four cohort studies of a single pharmaceutical plant, one case-control study, and one cohort study of a single pharmaceutical plant with a nested case-control study. Pharmaceutical plant workers showed a 3.19 (95% CI 1.53–6.64) times higher risk of hematopoietic/lymphatic system cancers compared to the general population. They also showed a 4.86 (95% CI 1.71–13.80) times higher risk of urinary system cancers compared to the general population.

Observed increased risk: association with occupational exposure or not

As seen in Supplementary material C, only Marsh et al. [12] and Dolan et al. [14] stated that there was insufficient evidence to confirm a link between hazardous occupational exposures in the workplace and an increased risk of cancers of the hematopoietic/lymphatic or urinary systems. However, Youk et al. [13] used the same data as those used in Marsh et al. [12], and concluded that increased mortality risk due to hematopoietic/lymphatic system cancer might indicate a possible workplace association. In addition, the opinion of Dolan et al. [14] regarding this association was rather crude. They said the absence of clear exposure-disease mortality patterns emerging over half a century indicated that this association was not apparent.

Hazardous occupational exposures

Pharmaceutical workers may be exposed to known chemical carcinogens such as benzene (linked to leukemia) and formaldehyde (associated with nasal cancer and leukemia). These substances are common in industrial environments and drug production facilities. Benzene can be present in pharmaceutical production areas as a VOC. In dense pharmaceutical production areas, benzene was identified as one of the carcinogenic VOCs, posing a significant health risk [15]. Benzene derivatives are used in drug formulations, particularly in synthesizing active pharmaceutical ingredients and excipients. The presence of benzene in drug substances is regulated to be no more than two ppm, as per the International Council for Harmonization guidelines [16]. Formaldehyde is a reactive impurity that can affect drug stability and safety. Therefore, rigorous analytical methods, such as gas chromatography, are employed to detect and quantify formaldehyde impurities in pharmaceutical excipients, ensuring product quality and safety [17].

Workers involved in the production of antineoplastic drugs could face carcinogenic risks due to their genotoxic properties. These drugs could lead to secondary cancers in patients and pose similar risks to workers handling them during manufacturing [18]. Even though the following two studies did not deal with pharmaceutical plant workers, they reported the increased risk of cancer in healthcare workers dealing with antineoplastic drugs. A study of Italian healthcare workers found that biomonitoring using the buccal micronucleus cytome assay revealed higher DNA damage among workers handling antineoplastic drugs, particularly administrators [19].

Nitrosamines, classified as probable human carcinogens, have been detected as impurities in various pharmaceuticals, including valsartan and metformin. These compounds are formed during manufacturing processes involving certain solvents, catalysts, or raw materials. Even small exposures can damage DNA and increase cancer risks [20].

Chlorinated hydrocarbon solvents, such as perchloroethylene (PCE) and trichloroethylene (TCE), are commonly used in pharmaceutical production. These solvents are classified as carcinogens due to their association with various cancers, including leukemia, lymphoma, and urinary tract cancer [21]. The use of these solvents is regulated under the European Union’s Classification, Labelling, and Packaging Regulation, which classifies them as carcinogens, mutagens, or reprotoxic agents [22].

Aromatic solvents, such as benzene and toluene, are also used in pharmaceutical manufacturing. Benzene is a well-known carcinogen, classified by the International Agency for Research on Cancer (IARC) as a Group 1 carcinogen, meaning it is carcinogenic to humans. Occupational exposure to benzene has been associated with an increased risk of acute myeloid leukemia (AML) and other hematopoietic cancers [23].

Potent active pharmaceutical ingredients (APIs), such as those used in the production of cytotoxic drugs, are particularly hazardous. These compounds are designed to be biologically active at very low doses, which increases their potential to cause harm if inhaled, ingested, or absorbed through the skin. Workers handling these APIs are at risk of developing cancers, reproductive disorders, and other chronic diseases [24]. For example, antineoplastic drugs, which are used in cancer chemotherapy, are highly carcinogenic. Occupational exposure to these drugs has been linked to DNA damage, as evidenced by the presence of micronuclei in buccal cells of exposed workers [19]. Similarly, APIs such as methylene chloride and dimethyl formamide, used in the synthesis of various drugs, are classified as probable human carcinogens by IARC [23].

The primary routes of exposure to APIs in pharmaceutical plants include inhalation, dermal contact, and ingestion. Inhalation of airborne API particles is the most common route of exposure, particularly in processes such as weighing, dispensing, and charging. Dermal exposure can occur through skin contact with contaminated surfaces or equipment, while ingestion can result from poor hygiene practices, such as eating or drinking in contaminated areas [24, 25].

Disinfectants and cleaning agents used in pharmaceutical plants to maintain asepsis and cleanliness can also be carcinogenic. For example, formaldehyde, a common disinfectant, is classified as a human carcinogen by IARC and has been linked to nasopharyngeal cancer and leukemia [23]. Other cleaning agents, such as ethylene oxide, a sterilizing agent, are also classified as carcinogens and have been associated with an increased risk of breast cancer and lymphoma [23, 26].

According to a paper [2], the authors could verify the hazardous chemicals to which pharmaceutical plant workers could be exposed. By comparing the chemicals listed in this paper with the list of classifications by cancer sites with sufficient or limited evidence in humans, IARC Monographs Volumes 1-133 (https://monographs.iarc.who.int/list-of-classifications), the authors could identify several potential carcinogenic agents to which pharmaceutical plant workers could be exposed during their work. According to this comparison, benzene and dichloromethane (methylene chloride) could be associated with hematopoietic/lymphatic system cancers. Benzene is a carcinogenic agent with sufficient evidence in humans for acute myeloid leukemia and other acute non-lymphocytic leukemia. In addition, benzene is a carcinogenic agent with limited evidence in humans for chronic myeloid leukemia, chronic lymphocytic leukemia, non-Hodgkin lymphoma (all combined), and multiple myeloma. Dichloromethane (methylene chloride) is a carcinogenic agent with limited evidence in humans for non-Hodgkin lymphoma (all combined). Welding fumes and coal-tar pitch could be associated with urinary system cancers. Welding fumes are a carcinogenic agent with limited evidence in humans for kidney cancer. Coal-tar pitch is a carcinogenic agent with limited evidence in humans for bladder cancer.

Prevention strategies and safety measures

Safe handling procedures and engineering controls should be implemented to minimize exposure. For handlers of antineoplastic agents, the proper use of safety cabinets and closed-circuit transfer devices is a fundamental strategy for minimizing exposure [27].

Personal protective equipment remains the first line of defense against exposure. For workers handling antineoplastic drugs, medical gloves are particularly important, and careful evaluation of glove materials is necessary to prevent contamination. One study found that 33% of analyzed gloves from antineoplastic drug administration and preparation units were positive for at least one antineoplastic drug, highlighting the importance of proper protection [27].

Regular biological monitoring for workers potentially exposed to carcinogens can help identify exposures before health effects occur. Several studies have demonstrated that healthcare workers may have traces of antineoplastic substances or their metabolites in biological fluids, suggesting similar monitoring would be valuable for pharmaceutical production workers [28].

Limitations

This study has several limitations. First, the number of included studies was so small. Only six studies were included in the systematic review, and three studies were included in each meta-analysis, respectively. Second, the exposure assessment was relatively simple for each study. In future studies, a more detailed and diversely stratified exposure assessment is needed. In particular, if exposure to suspicious agents could be quantified through biological monitoring methods, the causal association between a hazardous occupational exposure and an increased risk of cancer could be confirmed. Third, no subgroup analyses (e.g., by gender, duration of exposure) were conducted in this study. In future studies, these stratified analyses could improve the quality of the study. Fourth, the exposure assessment for each risk estimate was so heterogeneous and not uniform. In addition, there could be a possibility of exposure misclassification. Future studies with a more detailed and precise exposure assessment could improve the interpretability of the research results. Fifth, five of the six included studies focus on a cohort from a single pharmaceutical plant. In addition, three of the five studies were conducted in the US. These factors could have caused a selection bias. Therefore, this could limit the generalizability of this meta-analysis. Sixth, six included studies were conducted and published from 1950 to 1999 and from 1993 to 2005, respectively. This potential obsolescence of findings could limit the generalizability of the study’s results. Seventh, of the six included studies, only the case-control part from one study (Youk et al. [13]) adjusted for potential confounders. The other five studies did not adjust for any potential confounders, such as smoking and socioeconomic status, at all. This could have introduced a bias in this study. Eighth, even though the p-values for Egger’s regression test were statistically insignificant, the power of this test is low if the number of studies is fewer than ten [29]. Therefore, the possibility of publication bias should be strictly examined in future meta-analyses.

Conclusion

In this systematic review and meta-analysis, pharmaceutical plant workers showed about three times higher risk of hematopoietic/lymphatic system cancer and about five times higher risk of urinary system cancer compared to the general population. These increased risks might be due to hazardous occupational exposures, including benzene, antineoplastic drugs, nitrosamine, chlorinated hydrocarbon solvents such as PCE and TCE, potent APIs, and possible disinfectants and cleaning agents. Even though the number of included individual studies was small and the exposure assessment in individual studies was rather crude, more future studies might add evidence supporting this association. Future research should focus on improved exposure assessment methods, better understanding of the mechanisms of carcinogenesis, and development of more effective protective measures to ensure the health and safety of pharmaceutical workers worldwide. In addition, the fact that most studies were conducted from a cohort of a single pharmaceutical plant and in the US could limit the generalizability of this study. Future studies from multiple pharmaceutical plants and various countries could add generalizability to the results of this study.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Jinyoung Moon: Conceptualization, Methodology, Investigation, Resources, Data Curation, Software, Validation, Formal analysis, Writing – Original Draft, Visualization, Investigation, Resources, Data CurationYongseok Mun: Writing –Review & Editing, Supervision, Project administration.

Funding

This study was supported by 2024 fund of the Korean Retina Foundation. This research was supported by the Hallym University Research Fund, 2024 (HURF-2024-61).

Data availability

This is a meta-analysis. All data used in this study were extracted from each original study. The extracted data are included in the main text of this article.

Declarations

Ethics approval and consent to participate

The authors confirm that all experiments were performed in accordance with the Declaration of Helsinki.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.