The health impacts of semiconductor production: an epidemiologic review

Abstract

Background:

Despite concerns over the harmful health effects of semiconductor production, epidemiological studies have shown mixed results.

Objectives:

We aim to critically appraise epidemiologic studies to date, and to suggest future research and actions to protect workers in semiconductor industry.

Methods:

Epidemiologic studies were identified through electronic database searches, review of reference lists of relevant published works, and expert consultations, and were narratively reviewed.

Results:

Most evidence suggests reproductive risks from fabrication jobs, including spontaneous abortion (SAB), congenital malformation, and reduced fertility. Although chemicals have been suspected as causal agents, knowledge of the likely contribution(s) from specific exposures is still limited. Evidence of cancer risk seems to be equivocal. However, the available studies had serious limitations including healthy worker effects (HWEs), information bias, and insufficient power, all of which are associated with underestimation. Nevertheless, excess risks for non-Hodgkin's lymphoma (NHL), leukemia, brain tumor, and breast cancer were observed.

Conclusions:

Monitoring and innovative research based on international collaboration with a focus on sentinel events are required.

Introduction

The semiconductor industry began as a parts production sector of the microelectronic industry in the 1960s. However, with the growth of the computer industry, semiconductor manufacturing has taken the linchpin of the entire microelectronics industry. Semiconductor global sales revenue reached US$307 billion as of 2011,¹ and the size of workforce is also huge; for instance, the number of employees in just two countries – South Korea (hereafter Korea)² and Taiwan³ – exceeds 290 000. This industry is typically characterized by highly specialized processes, rapid technological innovation, significant chemical use, and large-scale investment to support such innovation and mass production,⁴ all of which could influence workers' health and safety.

Contrary to its ‘clean’ image represented by ‘bunny suits’, semiconductor production includes a wide variety of hazards. Work processes mainly consist of three steps – wafer manufacturing, fabrication, and packaging/testing, and at each step, there are various sub-processes. In particular, the wafer-fabrication process includes material deposition, photolithography, etching, and doping, all of which include the use of hazardous materials.⁵ A variety of chemicals including organic solvents, acids, and metals are heavily used. Some individual chemicals are simply irritants to skin or respiratory organs, whereas others are carcinogens, reproductive toxins, and/or neurotoxins. Air recirculation systems designed to minimize dust make workers more vulnerable to effects from these chemicals both singly and in complex combinations. Physical hazards including extremely low-frequency electronic magnetic and radiofrequency radiation are not negligible, and are known to be associated with cancer development. In addition, there exist ergonomic hazards including awkward positions, repetitive work, shift work, and job stress, most of which are not unique but common to manufacturing industries in general. The detailed explanations of work process and potential hazards in semiconductor production are found elsewhere.^6–8

However, not enough is known regarding the full range of potential health effects of such exposures. Technical development is so fast that it is nearly impossible to examine all of them, and the widespread practices of trade secrecy within the semiconductor industry make the situation worse.⁴ In this context, social as well as scientific debates have become intensified regarding causality and liability for suspected occupational diseases.

In 1979, the US National Institute for Occupational Safety and Health (NIOSH) conducted a health hazard evaluation in the Signetics fabrication plant in Sunnyvale, CA, where workers claimed chronic harmful exposure to chemicals. National Institute for Occupational Safety and Health identified chemicals as ‘irritant’ and ‘narcotic’ in nature, although their concentrations were all well below the permissible exposure levels. National Institute for Occupational Safety and Health concluded that ‘a larger, more systematic study’ was required to characterize medical problems and determine the size of the affected workforce.⁹ The next year, the California Department of Industrial Relations in the U.S. reported that semiconductor workers were exposed to many hazardous chemicals including a variety of carcinogens and repro-toxic agents.¹⁰ Shortly thereafter, groundwater contamination near a Fairchild semiconductor factory in San Jose was revealed, which affected the drinking water in thousands of households.⁴ In 1985, a chemist at the IBM research facility in San Jose tried to alert top management after five of his 12 colleagues contracted cancer.¹¹ However, IBM had took no action until three more victims were identified, at which point they conducted an internal investigation.¹² In the late 1990s and 2000s, a series of suits were filed against IBM by its workers and their families who were struggling for recognition of work-relatedness of the workers' cancers and their children's birth defects.⁴ Across the Atlantic, some female workers employed by National Semiconductor, Scotland (hereafter, NSUK) claimed the U.S. head office was responsible for their exposure to carcinogens. In response, the UK Health and Safety Executive (HSE) launched an epidemiologic investigation.¹³

Health concerns, once confined to the US and UK, have spread to the Asian countries along with the expansion of the global supply chain in the microelectronics industry. In the early 1990s, four workers died and about 200 workers were diagnosed with lead poisoning in one Thailand HDD company, and similar incidents were repeated across India, China, Taiwan, and Malaysia.¹⁴ In the late 1990s, groundwater contamination similar to the San Jose case was identified in Taiwan.¹⁴ In addition, the Korean government has investigated cancer risk among semiconductor workers in response to the rising concern over cancer cluster in Samsung Electronics.¹⁵

In the meantime, scientific evaluation of health risks has evolved. Based on the Swedish cancer registry, Vagero and Olin¹⁶ first reported an excessive cancer incidence in electronic workers. Using the workers' compensation data in California, LaDou identified that the semiconductor sector had higher incidence of occupational illness, in particular, systemic poisoning, while the combined rates of occupational injuries and illnesses were lower compared to private industries in total.^6,17 The first cohort study result was reported in 1985 by the UK HSE.¹⁸ About that time, IBM began an internal investigation of the cancer cluster described above; the results were not published until 10 years later and the focus was limited to brain cancer.¹⁹ Since then, a series of epidemiological studies were launched, most of which were motivated by the struggles and lawsuits by workers and community residents.

Although several scientific papers have been published, there is still a dearth of knowledge on the health effects of semiconductor production.²⁰ In this article, we critically appraise the epidemiologic studies to date, and based on our review, suggest future research and actions to protect workers in the semiconductor industry. Epidemiologic studies were collected through electronic database searches, review of reference lists of relevant published works, and expert consultations, and narrative review of them was conducted. Compared to the previous reviews,^6,12 we extend the scope, both chronological and geographical; our review includes studies conducted in Asian countries, as well as those from the UK and the US, and covers more recent publications than were available at the time of the previous reviews.¹² In addition, we place more emphasis on epidemiologic methods, and cover health problems other than cancer and reproductive outcomes.

Reproductive Health

Concerns for reproductive health first received community attention in early 1980s, when residents' drinking water was contaminated by toxic solvents leaked from the nearby semiconductor factory in San Jose and abnormal pregnancy outcomes were reported. An epidemiologic study conducted in 1983 found a risk elevation for spontaneous abortion (hereafter, SAB) and congenital malformation (especially, cardiac anomaly) in the exposed neighborhoods.^21,22 The follow-up study covering the time up to 1985 failed to replicate this finding, although risk elevation for SAB was still significant when the analysis was limited to the contamination period of 1980–1981, which could not be explained by hydrogeologic modeling of contaminated groundwater.²³ Rightfully, the finding raised a question about adverse reproductive effects on workers who dealt with the chemicals directly inside the plant.

Reproductive toxicity is an important issue in semiconductor industry, because a majority of the workforce is females of childbearing age and hazardous effects could affect offspring. In addition, reproductive outcomes may act as sentinels for detecting occupational and environmental hazards with a long latency because of their relatively short latency from exposure.²⁴

Spontaneous abortion is a relatively well-documented outcome among many reproductive problems in the semiconductor industry. An early clue was presented by a Finnish study in 1980, which found particularly higher risks for SAB among electronics workers compared to other manufacturing workers as well as the general population.²⁵ Based on a questionnaire survey of female workers and spouses of male workers in semiconductor industry, Pastides et al.²⁶ reported that working in diffusion (RR = 2.18, 95% CI 1.11–3.60) and photolithography process (1.75, 0.77–3.25) was associated with higher risk for SAB, while through a community survey Lipscomb et al.²⁷ observed that women with a solvent exposure history in electronics production during the first trimester of pregnancy were more likely to experience SAB than controls (OR = 3.34, 95% CI 1.42–7.81). The Semiconductor Health Study launched by the Semiconductor Industry Association (hereafter, SIA) is the most comprehensive reproductive health study in this industry. It covered 14 US companies with silicon-based wafer-fabrication rooms, and examined SAB in both a retrospective^28–30 and a prospective way,^30,31 menstrual cycles³² and fertility³³ in a prospective design, and male fertility through a cross-sectional survey. In particular, its prospective component measured human chorionic gonadotropin through daily urine collection for about 6 months, in order to accurately assess menstrual cycles and early fetal loss.³⁰ They reported elevated risks for SAB among workers in fabrication processes, especially masking and photolithography.^29–31 More specifically, ethylene glycol ether (hereafter, EGE) and fluoride-containing compounds were noted to be associated with the excess risk.^28,34 Given that the methods of outcome ascertainment were diverse across studies and sample sizes were usually small because it was not easy to recruit enough number of eligible women from even several thousand workers, such consistent results strongly suggest a causal association. In addition, the risk estimates became greater by classifying exposures more specifically, from dichotomy (fabrication vs non-fabrication) to work group such as photolithography, and to specific agents like EGE. Furthermore, the fact that risk estimates were obtained through internal comparison between workers gives more credibility to the results by minimizing the ‘infertile worker effect’.³⁵ Two studies did not observe a significant risk elevation; of these, one case–control study was apparently subject to selection bias and insufficient statistical power³⁶ and another was hampered by misclassification bias (Tables 1 and A1).³⁷

Table 1. A summary of risk estimates for spontaneous abortion (SAB) in semiconductor industry.

Exposure	Design (measure)	Risk estimates (95% CI)	Source
Fab (vs non-fab)	Retrospective cohort (RR)	1.43 (0.95–2.09)	Schenker et al.30
	Prospective cohort (RR)	1.25 (0.63–1.76)	Eskenazi et al.31
	Case–control (OR)	0.58 (0.26–1.30)	Elliott et al.36
Work group
Photolithography (vs non-manufacturing)	Retrospective cohort (RR)	1.75 (0.77–3.25)	Pastides et al.26
Photolithography (vs non-fab)	Retrospective cohort (RR)	1.67 (1.04–2.55)	Beaumont et al.29
Diffusion (vs non-manufacturing)	Retrospective cohort (RR)	2.18 (1.11–3.60)	Pastides et al.26
Etching (vs non-fab)	Retrospective cohort (RR)	2.08 (1.27–3.19)	Beaumont et al.29
Masking (vs non-fab)	Retrospective cohort (RR)	1.78 (1.17–2.62)	Schenker et al.30
Masking (vs non-fab)	Prospective cohort (RR)	1.30 (0.59–1.84)	Eskenazi et al.31
Dope/film (vs non-fab)	Prospective cohort (RR)	1.39 (0.51–1.96)	Eskenazi et al.31
Agent-specific
EGE/fluoride high (vs non-fab)	Retrospective cohort (RR)	3.21 (1.29–5.96)	Swan et al.28
EGE high (vs no exposure)	Retrospective cohort (RR)	2.80 (1.40–5.60)	Correa et al.34

fab: fabrication; non-fab: non-fabrication; EGE: ethylene glycol ether; RR: relative risk; OR: odds ratio; CI: confidence interval.

Table A1. Summary of study findings on adverse developmental outcomes in semiconductor industry.

Authors	Year	Area/company	Subjects (design)	Out come	Main findings	Funding
Pastides et al.	1988	US (MA)/DEC	1984–1986, retrospective cohort of 471 female workers and 273 spouses of male workersPhoto and diffusion vs non-manufacturing	SAB	RR: photolithography 1.75 (0.77–3.25), diffusion 2.18 (1.11–3.60) – consistent findings even in multiple logistic regression analysis to adjust for potential confounders	n.a.
Dean et al.	1989	US (CA)	1980–1981, cross-sectional study through interview and medical records228 exposed vs 274 unexposed female residents145 live births in the exposed area vs 182 ones in the unexposed	SAB (congenital anomaly)	Prevalence of SAB: exposed 21.5% vs unexposed 11.0%OR for SAB: 2.3 (1.3–4.2) (consistently elevated risks across validation methods)RR for congenital malformation 3.1 (1.1–10.4)	Public
Swan et al.	1989	US (CA)	1981–1983, cross-sectional study through interview, vital statistics, and medical recordsExposed vs unexposed communities (by time-period)	Congenital anomaly	Prevalence of anomaly: exposed period – exposed area 0.56% vs unexposed 0.26%, unexposed period – exposed area 1.7% vs unexposed 3.2%RR: exposed period 2.2 (1.2–4.0), unexposed period 0.5 (0.2–1.7)Discrepancy between spatial distribution of anomaly and hydrogeologic modeling	Public
Wrensch et al.	1990	US (CA)	1980–1985, retrospective cohort study through interview and medical records of 1105 community living womenExposed vs non-exposed communities (by time-period)	SAB	OR: 1980–1981 original subjects 3.5 (1.2–10.3), new subjects 0.3 (0.1–1.1); 1982–1985 original 1.0 (0.5–1.9), new 1.3 (0.6–2.9)	Public
Lipscomb et al.	1991	US (CA)	1980–1985, cross-sectional study through interview of 1038 community living womenSolvent exposure vs non-exposure	SAB	OR: first trimester exposure 3.34 (1.42–7.81), regular exposure 4.44 (1.86–10.58)	n.a.
Shusterman et al.	1993	US (CA)	Case–control study based on questionnaire, pathology records, and birth certificatesElectronics (semiconductor fab, printed circuit board, assembly) vs community residents	SAB	OR: 0.94 (0.58–1.5)Risk elevation in diffusion, parts encapsulation, soldering, flux terminal (statistically not significant)	n.a.
Schenker et al.	1995	US/SIA	1986–1989, historical cohort study of 904 women through interview and medical records1989–1991, prospective cohort study of 402 women through 6-month daily urine collection, diaries, and interviewFab vs non-fab comparison	SAB	RR: retrospective fab 1.43 (0.95–2.09) – esp. masking 1.78 (1.17–2.62), prospective fab 1.25 (0.63–1.76)Dose–response relationship in photoresist and developer solvents	SIA, public
Eskenazi et al.	1995	US/SIA	1989–1981, prospective cohort study through 6-month daily urine collection, diaries, and interview152 fab vs 251 non-fab workers	SAB	RR: fab 1.25 (0.63–1.76), dopant and thin film 1.39 (0.51–1.96), and masking 1.30 (0.59–1.84)	SIA, public
Beaumont et al.	1995	US/SIA	1986–1989, historical cohort through interview and medical records447 fab vs 444 non-fab pregnancies	SAB	RR: fab 1.43 (0.95–2.09), masking 1.78 (1.17–2.62), etching 2.08 (1.27–3.19), and photo 1.67 (1.04–2.55)	SIA
Swan et al.	1995	US/SIA	1986–1989, historical cohort through interview and medical records: 447 fab vs 444 non-fab pregnancies	SAB	RR: EGE and fluoride high exposure 3.21 (1.29–5.96) – dose–response relation with EGEFab workers without exposure 0.98 (0.55–1.69)	SIA
Correa et al.	1996	US (NY, VT)/IBM	1980–1989, retrospective cohort through interview and plant records of 378 female workers and 375 spouses of male workersEGE exposure: no vs low vs med vs high	SAB	RR: female worker high exposure 2.8 (1.4–5.6) – dose–response relation, no association in spousesSuspected dermal contact of EGE (low level of air concentration)	IBM, public
Elliott et al.	1999	UK	1987–1993, nested case–control study of 36 cases and 80 controlsFab vs non-fab	SAB	OR: fab 0.58 (0.26–1.30)No evidence of elevated risk (even in EGE exposure)	Public
Lin et al.	2008	Taiwan	1980–1994, 5702 live births from 6834 male workersRetrospective cohort study based on national birth/death registry and labor insurance dataExposure: employed in semiconductor production during 2 months prior to conception vs not	Child death	OR: congenital anomaly 3.26 (1.12–9.44), heart anomalies 4.15 (1.08–15.95)Possible link between paternal preconception exposure of semiconductor manufacturing and an increased risk of congenital anomalies, especially of the heart	Public
Lin et al.	2008	Taiwan	1980–2000, 24 223 live births from 27 610 female workersRetrospective cohort study based on national birth/death registry and labor insurance dataExposure: employed in semiconductor production during pre/post 3 months of conception vs not	Child death	OR: congenital anomaly 0.56 (0.13–2.40) in 1980–1994 and 0.88 (0.37–2.11) in 1995–2000No significant risk elevation	Public
Sung et al.	2008	Taiwan	1978–2001, 40 647 first singleton babies from 40 647 female workersRetrospective cohort study based on cancer registry (1979–2001) – 639 051 person years (PYs)Exposure: employed in semiconductor production during pre/post 3 months of conception vs not	Child cancer	Exposed cases 11, unexposed cases 36RR for all cancer: total 2.26 (1.12–4.54), ≤6 years of education 3.05 (1.20–7.74), 7–9 years of education 2.49 (2.49–4.94), excluding manager group 2.32 (1.15–4.66)RR for leukemia: total 3.83 (1.17–12.55), excluding manager group 3.92 (1.20–12.86)Suspicious role of organic solvents exposure	Public
Lin et al.	2011	Taiwan	1997–2007, retrospective cohort study of 440 female workers with 158 live birthsRotating shift work: none vs intermittent vs persistent	Birth weight	Mean birth weight (g): none 3271.4±395.4>intermittent 3251.3±460.9>persistent 2998.5±381.2 (P<0.01)OR for lightest quintile of birth weight: persistent shift 4.3 (1.1–16.8)Significantly elevated risk for lighter birth weight in female workers	n.a.
Sung et al.	2009	Taiwan	1973–1992, 13 592 live births from 7202 male workersRetrospective cohort study based on national birth/death registry and labor insurance dataExposure: employed in semiconductor production during 3 months prior to conception vs not	Child death	RR: infant mortality 5.06 (2.33–11.0) (>10 years work) and 2.81 (1.44–5.51) (1–10 years)RR: congenital malformation 3.75 (129–10.94) and cardiac defect 5.06 (1.58–16.19)Suspicious role of organic solvents' exposure	Public

SAB: spontaneous abortion; OR: odds ratio; RR: relative risk; SIA: Semiconductor Industry Association; EGE: ethylene glycol ether.

As a causal agent(s), most studies suspected organic solvents, in particular EGE, a well-known repro-toxin. Some experts have cautioned, however, that it is a mistake to attribute reproductive problems exclusively to EGE, and have warned that merely removing EGE from the fabrication processes may not have eliminated the reproductive risks. The first report of excess risk for SAB in semiconductor production observed that the ambient EGE concentration was far below the federal standard,²⁶ and other studies reporting elevated risks also found that the companies fully complied with industrial exposure regulations for EGE.^30,34 Concentrations of glycol ethers in 400 air samples from seven US semiconductor factories were also far below the regulatory limit.³⁸ There could be alternative explanations for excess SABs; first, there might be a different mode of exposure to EGEs other than ambient air, such as skin contact, which is not routinely monitored; second, even ambient air levels of exposure well below ‘permissible exposure limits’ might cause harmful effects, which would imply the need for lower regulatory limits and/or safe alternative chemicals; third, chemicals or physical hazards other than (or in combination with) glycol ethers might be responsible for excess SABs. In fact, some scientists were suspicious about the industry attributing adverse reproductive outcomes solely to EGE, which had already been largely phased out.⁶ Little is known about the situation since the late 1990s, because no reports on SAB have been published further. Recently, Lin et al.³⁹ reported that persistent rotating shift work was associated with lighter birth weight based on a retrospective cohort study in a Taiwan semiconductor company. Further investigations are strongly warranted to identify all causal agents for SAB in semiconductor production.

Other Taiwanese studies examined hazardous effects on offspring. Two studies in retrospective cohort design reported that children born to fathers who were employed in semiconductor production during the preconception period were three times more likely to die of congenital malformation (RR = 3.75, 95% CI 1.29–10.94, OR = 3.26, 1.12–9.44), especially heart defects (RR = 5.06, 1.58–16.19, OR = 4.15, 1.08–15.95), compared to those born to non-exposed fathers.^40,41 Considering that the data lacked information on SAB and minor malformation, and we could not rule out the possibility of non-differential misclassification bias, such significant and large effects strongly suggest a causal association. Even though risk elevation for birth defects was not observed among children born to female workers employed in semiconductor production during the periconception period,⁴² a retrospective cohort study of the same female cohort found an elevated risk for cancer among their children. Relative risks for all cancer and leukemia in offspring were estimated to be 2.26 (1.12–4.54) and 3.83 (1.17–12.55), respectively, which further increased by excluding a manager group. From the past labor inspection reports, the authors presumed the observed carcinogenic effect to be attributable to in utero exposure to organic solvents, in which presumption could not be confirmed.⁴³ To understand seemingly conflicting results, underlying mechanisms (i.e. mutations in germ cells before conception, and teratogenic and carcinogenic effects in fetuses) should be considered. Mutagenesis and carcinogenesis are stochastic events without any threshold, and therefore even very low exposure could lead to DNA damage, whereas teratogenesis is a threshold phenomenon with a presumptive dose–response relationship.⁴⁴ Prior to conception, male germ cells are more prone to suffer mutagenic effects because toxicity depends on cell division; spermatogenesis that takes 3 months for completion of a cycle is a continuous process from puberty on, whereas female's oogenesis is complete early during the zygote period.⁴⁵ After conception, teratogenesis is assumed to be less likely at the lower level of maternal exposure, while even very low levels of maternal exposure to carcinogens could lead to an offspring's cancer because there is no threshold in carcinogenesis.

As for fertility, prospective cohort studies with daily urine collection^30,33 and retrospective cohort studies^34,46 reported that working in fabrication process, especially photolithography or dopant/film process and EGE exposure, was associated with lower fertility^30,33,46 or elevated risk for subfertility among female workers.³⁴ Although Samuels et al.⁴⁷ reported no significant risk elevation through a cross-sectional interview survey, it seems reasonable to give more credence to the evidence from cohort studies. Moreover, these cohort studies were based on internal comparisons with less likelihood of infertile worker effect.³⁵

In addition, Gold et al.³² reported that fabrication workers experienced more variable menstrual cycles, although mean cycle length was not different between fabrication and non-fabrication workers. A decade later, Hsieh et al.⁴⁸ observed that working in a fabrication process – in particular, photoresist and diffusion process – was associated with prolonged menstrual cycles through a cross-sectional survey of Taiwanese workers. They suggested that such prolonged cycles could be one explanation for prolonged waiting time to pregnancy of fabrication workers observed in their previous study.⁴⁶

In summary, most epidemiologic evidence to date indicates there are adverse impacts of semiconductor fabrication work on various developmental and reproductive outcomes (see Tables A1 and A2 in the appendix). Although chemical exposures – in particular to glycol ethers – were frequently suspected for causal agents, it seems clear that knowledge about the potential spectrum of specific agents contributing to and responsible for reproductive problems and their mechanisms is still very limited, making effective prevention efforts a major challenge.

Table A2. Summary of study findings on adverse reproductive outcomes in semiconductor industry .

Authors	Year	Area/company	Subjects (design)	Outcome	Main findings	Funding
Gold et al.	1995	US (CA, UT)/SIA	1989–1991, prospective cohort study through 6-month daily urine collection, diaries, and interview152 fab vs 250 non-fab workers	Menstrual cycle (days)	Mean cycle length: fab 34.8±1.7 vs non-fab 32.5±1.4 (P = 0.97)Mean standard deviation: ion implant 6.68±1.28, photo 5.72±1.24 vs non-fab 4.1±1.16 (P = 0.013 and 0.019)RR for cycle<24 days: supervisors 2.46 (1.19–3.63), photo 1.83 (0.94–2.88)	SIA, public
Hsieh et al.	2005	Taiwan	1997, cross-sectional study through questionnaire473 fab vs 133 non-fab workers	Menstrual cycle	OR for long menstrual cycle: fab 2.0 (1.0–3.9), etching 0.9 (0.4–2.2), thin film 2.0 (0.8–4.9), photo 4.4 (1.7–11.4), diffusion 3.8 (1.3–11.6) vs non-fabOR for short menstrual cycle: fab 1.5 (0.5–4.7), etching 1.4 (0.4–4.7), thin film 1.4 (0.3–5.8), photo 1.6 (0.6–9.4), diffusion 1.3 (0.2–8.6) vs non-fabSuspicious role of multiple chemical exposure in photo process	Public
Samuels et al.	1995	US/SIA	1984–1989, cross-sectional study of male workers through interview241 fab (165 births) vs 447 non-fab workers (300 births)	Fertility	RR for fertility: fab 0.98 (0.80–1.19)RR for previous conceiving difficulty: furnace/thin film/ion implant workers 1.79 (1.09–2.94)*Failure to control for maternal factors	SIA
Schenker et al.	1995	US/SIA	1989–1991, prospective cohort study of 402 women through 6-month daily urine collection, diaries, and interview1989–1990, cross-sectional survey of 1637 female and 1536 male workersFab vs non-fab comparison	Fecundability, infertility	FR: fab 0.51 (0.27–0.95), dope/film 0.32 (0.11–0.90), mask 0.75 (0.37–1.55)OR for infertility: male fab 1.17 (0.79–1.74), female fab 1.09 (0.72–1.46)	SIA, public
Eskenazi et al.	1995	US/SIA	1989–1991, prospective cohort study of 402 women through 6-month daily urine collection, diaries, and interview152 fab vs 251 non-fab workers	Fecundability	OR for fecundability: fab 0.59 (0.32–1.09)OR for clinical pregnancy: fab 0.43 (0.19–0.90)Lowest fecundability group: dopant and thin film, or exposed to glycol ether	SIA, public
Correa et al.	1996	US/IBM	1980–1989, retrospective cohort of 378 female workers and 375 spouses of male workers (561 and 589 pregnancies) through interview and plant recordsFocus on EGE: no vs low, med, and high exposure	Subfertility	RR: female worker's high exposure 4.6 (1.6–13.3) – dose–response relationRR: spouse's high exposure 1.7 (0.7–4.3)Suspected role of EGE exposure	IBM, public
Chen et al.	2002	Taiwan	1990–1997, retrospective cohort study of 173 female workersFab (EGE) vs non-fab comparison	Waiting time to pregnancy	RR for fecundability: photo 0.77 (0.45–1.32), EGE exposure 0.59 (0.37–0.94)RR for clinical pregnancy: fab 0.22 (0.05–0.96), EGE exposure 0.37 (0.11–1.19)Suspected role of EGE exposure on ovarian function	Public

RR: relative risk; SIA: Semiconductor Industry Association; OR: odds ratio; FR: fecundability ratio; EGE: ethylene glycol ether. *Confidence intervals and p-values are presented where present in the original studies. Presentation of CI and p-values are preferred to the use of arbitrary dichotomy - ‘significance vs. non-significance’. No information in these table means it was lacking in the original papers.

Cancer

In 1985, a cohort study to examine cancer risk in semiconductor production first appeared in a scientific journal. It estimated standardized mortality ratio (hereafter, SMR) and standardized incidence ratio (hereafter, SIR) of 1807 employees at one semiconductor factory located in the West Midland, UK with reference to the general population, and concluded that the observed number of cases was close to expectation.¹⁸ Since then, a series of epidemiologic papers have been published, all of which are from a few countries, including the UK, US, Taiwan, and Korea.

In the mid-1980s, IBM commissioned a research project in response to the rising concern over a suspected brain cancer cluster. Using the Corporate Mortality File containing information of tens of thousands of former and current workers, the case–control study concluded that VDT development work was unlikely to be associated with brain tumors. However, excess risk for brain tumor mortality was observed among engineers (OR = 1.7, 95% CI 1.3–3.0) and programmers (OR = 2.8, 1.1–7.0) with a tenure of 10 years or more.¹⁹ Another case–control study in a community setting reported that electromagnetic radiation exposure increased the risk for brain tumors, especially in design/manufacture/repair/installation jobs. The authors argued that the risk might be associated with combined exposure to chemicals and electromagnetic rays.⁴⁹ However, subsequent retrospective cohort studies commissioned by IBM repeatedly reported that there was no evidence to support the association between specific work exposures and cancer.^50,51 In addition, a retrospective cohort study sponsored by the SIA concluded that potential exposure to hazardous conditions (i.e. fabrication processes) was not associated with excess risk for cancer death.⁵² Meanwhile, two studies to estimate proportional mortality ratio (hereafter PMR) using the same IBM Corporate Mortality File, which became available to the plaintiffs through a court decision, found elevated risks for brain tumors in males (proportionate cancer mortality ratio (PCMR) 166, 95% CI 129–213) and hematologic malignancies in females (162, 121–218).^53,54

In response to workers' demands, the UK HSE conducted an epidemiologic investigation on cancer risk in the NSUK located in Greenock, Scotland,¹³ which was followed by successive studies to extend follow-up time and add a historical hygiene assessment.^55–57 Although the initial analysis found elevated risks for lung cancer incidence (SIR 273, 136–488) and mortality (SMR 241, 116–444) among female workers,¹³ they were not found to be significant in the follow-up studies. In addition, further analysis for breast cancer in a case–control design did not find any significant risk factors. Finally, the HSE concluded that there was little evidence to support excess cancer risk among semiconductor workers.^55,57 Meanwhile, another two follow-up reports (published in 1992 and 2005) of the above-mentioned West Midland cohort study also concluded that there was no evidence of excess cancer risk among semiconductor workers, although certain types of cancer were observed more frequently.^58,59

In Taiwan, a series of epidemiological analyses done after groundwater contamination by chlorinated solvents from the former RCA plant had been identified in 1998. The company had been operating from 1968 to 1992 in northern Taiwan, to produce remote controllers, main boards for color TVs, and IC products including semiconductor packaging. Lee et al.⁶⁰ reported that the mortality odds ratio for all cancer (2.07, 1.3–3.27) and for liver cancer (2.57, 1.21–5.46) were significantly higher in male residents living in the downstream community from the waste site, compared to the unexposed ones living in an upstream community.¹⁴ The risk estimates further increased during follow-up.⁶⁰ In case of workers inside, based on the industrial hygiene reports and interviews, it was assumed that almost all workers had been exposed to organic solvents during employment. A proportionate cancer morbidity study to compare the former RCA workers with textile and other electronics workers found that breast cancer risk was significantly elevated.¹⁴ Based on employment insurance and cancer registry data, Chang et al.⁶¹ estimated SMRs of the former RCA workers with reference to the general population. They found an elevated SMR for breast cancer among female workers (114, 85–151), in particular, those with longer employment (1–5 years: 125; 5 years or more: 132), while there was no risk elevation for other types of cancers.⁶¹ The following SIR study found no elevated risk except for breast cancer of female workers (119, 103–136).⁶² These two studies concluded that there was no evidence to support excessive cancer risk from the exposure to chlorinated solvents.^61,62 Meanwhile, Sung et al.⁶³ reported that breast cancer risk was higher for those who were hired before 1974 when regulations for organic solvents were introduced (SIR 138, 111–170) or worked for more than 10 years (168, 111–242).

The only Korean epidemiologic study was conducted by the government in 2008, in the context of determining eligibility for workers' compensation which several Samsung Electronics workers claimed for their cancer development. The study included all workers from all five semiconductor companies with eight factories involved in fabrication processes in Korea. Based on the national cancer registry and mortality database, SMRs and SIRs were estimated with reference to the general population. The study concluded that risks for overall mortality and most types of cancer were smaller than that of the general population. However, the incidence of non-Hodgkin's lymphoma (hereafter, NHL) in female workers (SIR 231, 123–395) and thyroid cancer in male workers (SIR 211, 149–289) was found to be significantly elevated, which was not explained by the subsequent sub-group analysis.¹⁵ The study team also published a case series on hematologic malignancies as a part of their investigation. It covered five cases of leukemia, one NHL case, and one aplastic anemia case, all of which involved claims for workers' compensation. Based on hygiene assessment of current work environment, review of the past monitoring records, and construction of detailed work history of the affected workers, they concluded that there was insufficient evidence of workers' exposure to occupational carcinogens.⁶⁴ However, a few months later, another case series was reported, which comprised 13 leukemia and 4 NHL cases, all of whom were fabrication workers in Samsung's Kiheung facility.⁶⁵ In this case, the authors selected cases from dozens of cases reported to a labor advocacy group, and obtained information solely from workers and their family members because they could not gain access to government and company data. The authors found that cancer developed in a relatively young worker with a short latency, and argued that the work environment was poor and chemical accidents were not uncommon, contrary to the official reports.

At the first glance, overall indications of excess cancer risk in the semiconductor industry seem to be at most equivocal. However, the results should be interpreted with caution because most studies have considerable methodological limitations, which should not be overlooked. In addition, by examining results in more detail, we could find some patterns in common.

First of all, most studies seeking to estimate SMR or SIR with reference to the general population were seriously impacted by the healthy worker effect (HWE). It is known that the HWE influences findings on cardiovascular disease while cancer findings are known to be less susceptible to this bias. The HWE is a mixed effect of different biases including (a) selection bias arising from selection process at employment or survival effect during follow-up, (b) information bias due to differential probability of case ascertainment, and (c) confounding bias such as lower smoking prevalence and higher socioeconomic status of the employed population.⁶⁶ Indeed, SMRs for mortality from all causes were significantly less than unity^{15,50–52,55} except for the cases of the UK West Midland male workers⁵⁸ and the Taiwanese female workers.⁶¹ As an extreme case, the SMR of Korean male workers was just a quarter of the expected one.¹⁵ Except for the West Midland studies,⁵⁸ SMRs of male workers were much smaller than those of female workers, which seemed to reflect gender difference in socioeconomic position; male workers were likely to be engineers with a relatively higher education, whereas most female workers were operators. It is well known that lower socioeconomic position is associated with higher mortality. Meanwhile, SMRs for cancer were consistently greater than those for total mortality, except for Taiwanese male and female workers, although many of them were still less than unity. Taken into account such HWEs, ‘near average cancer risk’ could be consistent with real risk elevation over what these specific workers otherwise have experienced.⁶⁷

In addition, selection bias was further exacerbated by the fact that vulnerable workers (i.e. migrant workers) or early retirees who quit their job before the start of follow-up were excluded from the analyses. The West Midlands studies^18,58,59 could not obtain information about the workers who quit their job before 1970, and the NSUK studies failed to follow-up non-UK nationals.^13,55,57 The US SIA studies did not include migrant workers and the workers who quit their job before 1983, and those excluded from the analysis amounted to 40.0% of total workforce.⁵² Also, the Taiwanese studies^14,42,63,68 failed to include workers exposed to more hazardous conditions of earlier times. Indeed, Sung et al.⁶³ reported that breast cancer risk was greater for the period before the adoption of government regulation in 1974. The Korean employment cohort covered only the period from 1998 to 2007,¹⁵ although mass production of semiconductors began in the early 1980s. By excluding those who worked in more dangerous or disadvantaged conditions, a bias toward the null is more likely to occur.

Another important limitation is the possibility of misclassification bias, in that most studies were based on historical data, originally constructed for administrative use, not for research. For example, the Korean employment cohort data designated 4295 out of 48 022 female workers (8.9%) in fabrication processes and 1669 out of 19 767 female workers (8.4%) in assembly processes as ‘engineers’, respectively.¹⁵ However, this classification is unreliable, because while usually men with diplomas from vocational schools or colleges are recruited as engineers, it is very rare for women to be trained as and then work as engineers in Korea. Furthermore, by treating all workers employed in semiconductor industry as one group (i.e. semiconductor workers vs general population), the risk of actually exposed workers could be diluted. Through simulation, Parodi et al.⁶⁹ showed that such dilution effect, in particular in case of dichotomous classification, could result in considerable underestimation of risk. In fact, the Semiconductor Health Study showed that risk estimates for SAB became greater by refining classification of the exposure status down to the level of the specific agent(s).⁷ This kind of non-differential misclassification bias can also lead to underestimation of risk.

Finally, false-negativity problem (type 2 error)⁷⁰ due to small sample size and short follow-up time should be considered; the UK West Midland and the NSUK studies covered only 1807 and 4339 workers, respectively,^57,58 whereas 86.9% of the Korean cohort members were born after 1970, relatively young for cancer development.¹⁵ In fact, when the NSUK study proposal was made public, experts and labor advocates pointed out the problem of small sample size and raised a call for a nationwide or international survey,^71,72 which was never carried out.

Despite these limitations, we could observe elevated risks for some types of cancer across studies – NHL, leukemia, brain tumor, and breast cancer. Given that most studies were subject to considerable biases toward the null, more attention should be paid to such findings, even if a statistical significance was not secured or effect sizes were small.

In case of NHL, the IBM workers with longer tenure showed elevated SMRs; those workers employed for 5 years and 10 years or more were 125 (95% CI 98–158) and 122 (88–165), respectively.⁵⁰ While the SMRs estimated from the US SIA were below unity, the point estimate was greater in fabrication workers than non-fabrication workers (86 vs 61).⁵² The Korean study found elevated SMRs, in particular, of female workers (250, 68–640),¹⁵ although they were not statistically significant. The PMR study based on IBM Corporate Mortality File found elevated risks among both males (150.4, 136.5–165.8) and females (139.5, 110.7–176).⁵³ Meanwhile, SIRs were not significantly elevated except for the case of Korean female workers (231, 123–395).¹⁵ However, we could still find a certain trend that longer employed or exposed groups were more likely to show larger point estimates of SIR; in the IBM East Fishkill plant, SIRs of total workers and those employed for 5 years or more were 94 and 98, respectively, whereas SIRs of workers unexposed, ever-exposed, and exposed for 5 years or more were 80, 100, and 109, respectively. Also, in the IBM San Jose plant, SIRs for total workers and those employed for 5 years or more were 91 and 110, respectively, although they failed to show statistical significance (Table 2).⁵¹

Table 2. A summary of risk estimates for non-Hodgkin's lymphoma (NHL) in semiconductor industry.

Population	Measure	Attributes	Risk estimates (95% CI)	Source
Mortality
US IBM	SMR	Total	99 (82–119)	Beall et al.50
		Employed for 5+ years	125 (98–158)
		Employed for 10+ years	122 (88–165)
	PMR	Males	150 (137–166)	Clapp53
		Females	140 (111–176)
	PCMR	Males	136 (123–149)
		Males, manufacturing	126 (98–163)
		Females	122 (97–154)
		Females, manufacturing	144 (86–240)
US SIA	SMR	Total	69 (48–97)	Boice et al.52
		Non-fab workers	61 (37–95)
		Fabrication workers	86 (47–144)
Korea	SMR	Males	133 (43–309)	Lee et al.15
		Females	250 (8–640)
Incidence
US IBM	SIR	East Fishkill, total	94 (74–118)	Bender et al.51
		East Fishkill, employed for 5+ years	98 (69–136)
		East Fishkill, unexposed	80 (43–137)
		East Fishkill, exposed	100 (76–128)
		East Fishkill, exposed for 5+ years	109 (72–157)
		San Jose, total	91 (69–117)
		San Jose, employed for 5+ years	110 (76–154)
		San Jose, unexposed	116 (77–168)
		San Jose, exposed	74 (50–106)
		San Jose, exposed for 5+ years	84 (46–141)
Korea	SIR	Males	93 (45–171)	Lee et al.15
		Females	231 (123–395)

SMR: standardized mortality ratio; SIR: standardized incidence ratio; PMR: proportionate mortality ratio; PCMR: proportionate cancer mortality ratio; CI: confidence interval.

As for leukemia, elevated risk was observed among the Korean females (SMR 137, 55–281)¹⁵ and the IBM workers (PCMR of males: 113.3, 102–116; females: 113.7, 89.6–144.4). Risk elevations were seen prominent in manufacturing jobs (PCMR of females: 165.8, 101.6–270.8).⁵³ In addition, a longer tenure was associated with a larger point estimate; in the IBM study, SMRs of total workers, those employed for 5 years, and those employed for 10 years or more were 85, 95, and 103, respectively,⁵⁰ while the SIA study reported smaller SMRs for fabrication workers.⁵² Although there were small increases of point estimates among IBM San Jose workers,⁵¹ we could not find any consistent pattern regarding SIRs, except that risk estimates of female workers were greater than those of male workers in gender-stratified analysis (Table 3).

Table 3. A summary of risk estimates for brain cancer in semiconductor industry.

Population	Measure	Attributes	Risk estimates (95% CI)	Source
Mortality
US IBM	SMR	Total	108 (87–132)	Beall et al.50
		Employed for 5+ years	114 (82–154)
		Employed for 10+ years	119 (77–174)
	PMR	Males	192 (173–212)	Clapp53
		Females	132 (101–174)
	PCMR	Males	162 (146–179)
		Males, manufacturing	166 (129–213)
		Females	109 (83–142)
		Females, manufacturing	119 (65–219)
US SIA	SMR	Total	111 (84–145)	Boice et al.52
		Non-fabrication workers	127 (90–172)
		Fabrication workers	83 (45–139)
UK West Midlands	SMR	Total	83 (17–143)	Nichols and Sorahan58
UK NSUK	SMR	Males	189 (39–552)	Darnton et al.55
		Females	n.a.
Taiwan RCA	SMR	Males	48 (1–266)	Chang et al.61
		Females	91 (33–199)
Korea	SMR	Males	92 (25–235)	Lee et al.15
		Females	34 (1–187)
Incidence
US IBM	SIR	East Fishkill, total	94 (65–132)	Bender et al.51
		East Fishkill, employed for 5+ years	94 (52–158)
		East Fishkill, unexposed	77 (28–168)
		East Fishkill, exposed	101 (67–147)
		East Fishkill, exposed for 5+ years	116 (60–203)
		San Jose, total	91 (56–139)
		San Jose, employed for 5+ years	126 (67–215)
		San Jose, unexposed	102 (47–193)
		San Jose, exposed	87 (45–152)
		San Jose, exposed for 5+ years	164 (75–311)
UK West Midlands	SIR	Total	50 (6–81)	Nichols and Sorahan58
UK NSUK	SIR	Males	209 (57–535)	Darnton et al.57
		Females	n.a.
Taiwan RCA	SIR	Males	n.a.	Sung et al.63
		Females	107 (59–180)
Korea	SIR	Males	137 (62–259)	Lee et al.15
		Females	22 (1–122)

SMR: standardized mortality ratio; SIR: standardized incidence ratio; PMR: proportionate mortality ratio; PCMR: proportionate cancer mortality ratio; CI: confidence interval.

Elevated SMRs for brain tumor were more frequently observed in the US studies – both the IBM (108, 87–132)⁵⁰ and the SIA (111, 84–145).⁵² In addition, the IBM PMR study showed statistically significant excess risk (PMR of males: 191.8, 173.2–212.3; females: 132, 100.8–173.8).⁵³ Although the SIR estimates seemed equivocal, longer employment or exposed status was associated with risk elevation in the IBM East Fishkill site; SIRs of workers unexposed, ever-exposed, and exposed for 5 years or more were 77, 101, and 116, while in the San Jose site, they were 102, 87, and 164, respectively (Table 4).⁵¹

Table 4. A summary of risk estimates for leukemia in semiconductor industry.

Population	Measure	Attributes	Risk estimates (95% CI)	Source
Mortality
US IBM	SMR	Total	85 (69–105)	Beall et al.50
		Employed for 5+ years	95 (69–127)
		Employed for 10+ years	103 (69–147)
	PCMR	Males	113 (102–126)	Clapp53
		Males, manufacturing	88 (63–121)
		Females	114 (90–144)
		Females, manufacturing	166 (102–271)
US SIA	SMR	Total	77 (54–107)	Boice et al.52
		Non-fabrication workers	82 (52–122)
		Fabrication workers	69 (34–123)
UK West Midlands	SMR	Total	96 (20–282)	Nichols and Sorahan58
Taiwan	SMR	Males	44 (5–159)	Chang et al.61
		Females	54 (23–107)
Korea	SMR	Males	39 (8–114)	Lee et al.15
		Females	137 (55–281)
Incidence
US IBM	SIR	East Fishkill, total	70 (49–98)	Bender et al.51
		East Fishkill, employed for 5+ years	62 (35–103)
		East Fishkill, unexposed	79 (34–157)
		East Fishkill, exposed	70 (46–102)
		East Fishkill, exposed for 5+ years	70 (36–123)
		San Jose, total	103 (73–142)
		San Jose, employed for 5+ years	115 (71–175)
		San Jose, unexposed	107 (59–180)
		San Jose, exposed	104 (66–156)
		San Jose, exposed for 5+ years	114 (56–200)
UK West Midlands	SIR	Total	121 (39–283)	Nichols and Sorahan58
Taiwan	SIR	Males	n.a.	Sung et al.63
		Females	78 (49–117)
Korea	SIR	Males	69 (30–137)	Lee et al.15
		Females	128 (61–236)

SMR: standardized mortality ratio; SIR: standardized incidence ratio; PMR: proportionate mortality ratio; PCMR: proportionate cancer mortality ratio; CI: confidence interval.

Excess risk for breast cancer was prominent among the Taiwanese workers (SMR 114, 85–151), and it further increased with longer tenure.⁶¹ Also, the IBM PMR study showed significantly greater PMR (137, 125–150) and PCMR (115, 106–125).⁵³ As for incidence, although the IBM East Fishkill site showed elevated SIR, there was scarcely a consistent pattern to imply a plausible association.⁵¹ In the UK, SIRs were elevated for the NSUK (123, 90–163), in particular, among fabrication workers (135, 96–184),⁵⁵ while there was no similar results in the West Midlands.⁵⁸ Similar to the mortality pattern, the Taiwanese female workers showed an elevated SIR (109, 96–122). The relative risk increased further among workers employed for 10 years or more (168, 111–242) and those hired before 1974 when the government regulations for chemicals were introduced (138, 111–170).⁶³ Such findings seem to be highly plausible, in accordance with expectation (Table 5).

Table 5. A summary of risk estimates for female breast cancer in semiconductor industry.

Population	Measure	Attributes	Risk estimates (95% CI)	Source
Mortality
US IBM	SMR	Total	95 (80–112)	Beall et al.50
		Employed for 5+ years	100 (73–134)
		Employed for 10+ years	102 (65–154)
	PMR	Total	137 (125–150)	Clapp53
	PCMR	Total	115 (106–125)
		Manufacturing	97 (79–120)
US SIA	SMR	Total	92 (75–112)	Boice et al.52
		Non-fabrication workers	109 (89–136)
		Fabrication workers	65 (42–95)
UK West Midlands	SMR	Total	47 (25–81)	Nichols and Sorahan58
UK NSUK	SMR	Total	104 (50–192)	Darnton et al.55
Taiwan RCA	SMR	Total	114 (85–151)	Chang et al.61
		Employed for 1–4 years	125 (n.a.)
		Employed for 5+ years	132 (n.a.)
Korea	SMR	Total	84 (10–302)	Lee et al.15
Incidence
US IBM	SIR	East Fishkill, total	104 (89–120)	Bender et al.51
		East Fishkill, employed for 5+ years	114 (88–145)
		East Fishkill, unexposed	115 (89–145)
		East Fishkill, exposed	98 (81–117)
		East Fishkill, exposed for 5+ years	94 (65–133)
		San Jose, total	102 (87–119)
		San Jose, employed for 5+ years	94 (69–125)
		San Jose, unexposed	109 (87–136)
		San Jose, exposed	97 (77–121)
		San Jose, exposed for 5+ years	80 (48–125)
UK West Midlands	SIR	Total	78 (59–102)	Nichols and Sorahan58
UK NSUK	SIR	Total	123 (90–163)	Darnton et al.55
		Fabrication workers	135 (96–184)
		Non-fabrication workers	81 (33–167)
Taiwan RCA	SIR	Total	109 (96–122)	Sung et al.63
		Employed for 10+ years	168 (111–242)
		Hired before 1974	138 (111–170)
		Hired after 1974	99 (85–114)
Korea	SIR	Total	77 (44–126)	Lee et al.15

SMR: standardized mortality ratio; SIR: standardized incidence ratio; PMR: proportionate mortality ratio; PCMR: proportionate cancer mortality ratio; CI: confidence interval.

In spite of ‘reasonable doubt’ about elevated cancer risks, we should be still cautious in determining causal association. As Boffetta et al.⁷³ commented, multiple comparisons by sub-group analyses could raise the likelihood of false-positivity by chance. In addition, lack of specific a priori hypotheses and insufficient adjustment for important confounders like tobacco use could contribute to false-positivity.⁷³ Many findings we mentioned are based on a series of sub-group analysis, usually without a priori hypotheses. Therefore, we could not rule out the possibility of false-positivity. However, considering that dilution effects are common in studies based on historical data, cases of substantial confounding (even smoking) are actually rare in occupational epidemiology,⁷⁴ potential biases in the studies we examined are consistently toward the null, and their statistical powers are limited, it is unlikely that the danger of false-positivity surpasses that of false-negativity.

To summarize, epidemiologic evidence to date does not seem to strongly support a causal association between semiconductor production work and overall cancer risk. However, most studies were subject to considerable limitations including HWEs, information bias, and insufficient statistical power, all of which are associated with underestimation of risk. In this context, data interpretation solely based on statistical significance could result in a faulty conclusion. At a minimum, the evidence to date suggests that excess risk may be present for NHL, leukemia, brain tumor, and breast cancer in the semiconductor industry, and further studies should be undertaken.

Other Health Outcomes

Although most attention has been paid to cancer and reproductive outcomes, other health problems have been also reported in semiconductor industry.

Respiratory abnormalities are one of them. Using pulmonary function test, Luo et al.⁷⁵ found that Taiwanese male fabrication workers, mostly process/maintenance/equipment engineers, were four times more likely to experience restrictive pulmonary abnormality compared to non-fabrication workers. Given that there were no definite abnormalities in chest x-rays and physical examinations, and pulmonary function abnormality was not observed among female operators, it is plausible to posit that the maintenance engineers had intermittent short-term peak exposures that could have resulted in mild pulmonary dysfunction.⁷⁵ Based on a cross-sectional questionnaire survey in the US, McCurdy et al.⁷⁶ reported that upper (prevalence 71%, RR = 1.08, 1.03–1.13) and lower respiratory symptoms (41.5%, RR = 1.12, 1.01–1.22) were more frequent among fabrication workers compared to non-fabrication workers. Moreover, there was a dose–response relationship between prevalence of upper respiratory symptoms and hours per day in the fabrication room, and photolithography or furnace jobs were significantly associated with symptoms. In addition, complaints of persistent wheezing were 1.9 times more likely among fabrication workers.⁷⁶

The same study also reported that dermatologic problems were more frequently observed among fabrication workers (RR = 1.19, 1.04–1.35) and specifically, methanol and isopropanol were suspected as causal agents. Alopecia was also more common among fabrication workers (RR = 1.73, 1.16–2.54), although the causative agents could not be identified.⁷⁶ Previously, Pastides et al.²⁶ reported that female workers in photolithography were more likely to experience skin rashes compared to the other workers (13.4 vs 5.7%, prevalence ratio = 2.38, 1.13–4.89). Various skin problems among electronics workers have been documented. Adams⁷⁷ and Koh et al.⁷⁸ presented a comprehensive review of dermatologic hazards potentially involved in specific electronics industry processes including semiconductor production – irritants such as solvents, soldering fluxes, hydrofluoric acid, and epoxy resin; allergens such as epoxy resin, colophony, metals, and anti-static agents; and others including fiberglass, use of protective equipment, and low humidity.

Musculoskeletal problems associated with poor ergonomic conditions including awkward posture and long working hours were also reported,^79–81 and are not unique to the semiconductor industry. Mental health problems have been reported in the microelectronics industry but are not specific to semiconductor production. Cases of mass psychogenic illness in electronics plants were found to be associated with stress from uncomfortable physical environments⁸² and repetitive work demanding a high degree of concentration⁸³ rather than specific chemical exposures. One review concluded that the so-called ‘mass psychogenic illness’ seemed to be triggered by objective stressors such as low level of chemical odors and insect bites, and transmitted by psychological factors within groups.⁸⁴ In fact, neuropsychological impairments were more frequently observed among former electronics workers who worked with organic solvent compounds compared to a control group,⁸⁵ and the clinical pattern derived from the MMPI questionnaire was similar to that observed in organic solvent toxicity.⁸⁶ On the other hand, a Japanese⁸⁷ and Taiwanese study⁸⁸ each reported a high level of job stress among workers in microelectronics. Recently, serial cases of suicide in the Foxconn plants in southern China raised the worldwide concern over the working conditions of microelectronics workers, mostly poor young workers who migrated from rural homes to industrial cities. Although the company attributed the tragedy to ‘immaturity and fragile’ state of mind of the young rural migrants, labor advocates argued that harsh working conditions such as intense work stress and labor rights abuses were associated with the incidents.⁸⁹

All of these reports suggest that exposure to chemical and physical hazards in semiconductor production could impair skin and respiratory organs, and that the scope of new research needs to be expanded to other types of hazards such as job stress, ergonomic conditions, and mental health issues.

DISCUSSION

Although considerable concerns over adverse health outcomes among semiconductor workers continue to be expressed, it is not yet possible to reach firm conclusions based on definitive evidence. In reality, we observed that many studies have varying degrees and types of limitations, most of which could bias the risk toward the null. In this context, we could summarize our findings as follows; first, there is a relatively clear association between reproductive abnormalities and fabrication work. Second, in spite of seemingly equivocal results, cancer studies provide reasonable grounds for us to suspect excess cancer risk in this industry, in particular, for hematologic malignancy, brain tumors, and breast cancer. Third, dermatologic and respiratory problems due to chemical exposures are relatively common, and ergonomic and psychosocial problems are shared with other industries. However, the specific causative agents or mechanisms are yet to be identified in this field.

Based on our review, we suggest some points for further research and actions regarding health and safety of workers in semiconductor production.

First of all, careful monitoring should be strengthened and expanded across countries. The problems we reviewed are not just a legacy of the past; technical developments have not caused them to disappear. To begin with, it should be noted that many of the study cohorts were so young that further follow-up is needed for health outcomes with long latency such as cancer. In addition, as the recent investigation in the contaminated neighborhoods in Taiwan showed, toxic chemicals persist in the ground at dangerous levels even now.^90,91 In addition, although considerable improvements have been achieved in health and safety measures, there are still many unknown hazards in the workplace, and protection and prevention measures are far from perfect. The UK HSE inspection of semiconductor-manufacturing plants in 2002 found that occupational health provisions varied across companies and many of them failed to meet legal compliance for health surveillance.⁹² Also, when the Korean government agency identified 424 chemical ingredients out of 509 chemicals used in one company, only 10% of them were monitored. Further, it should be noted that unexpected by-products could be generated under conditions of high energy plasma, high temperature, UV radiation, and/or ion emission; in fact, formaldehyde, arsenic, and ethylene oxide gas, which are supposedly not used in the processes themselves, are still generated from photolithography and/or ion implantation processes.⁵ In addition, the evidence reveals that seemingly complete controls cannot guarantee safe work places. For instance, risks for SAB were elevated even where the suspected agent – glycol ethers – was tightly controlled at far below the permissible exposure limits as evidenced by air sampling.^30,34 Decreased numbers of leukocytes were observed among male engineers working within a fully automated clean room.⁹³ In addition, genotoxic effects (increased frequency of micronuclei in leukocytes) were identified among exposed workers though their urine and airborne concentrations of chemicals yielded results consistent with air levels below regulatory standards.⁹⁴ Sometimes, alternative materials to replace toxic chemicals were proven to be also harmful; for example, propylene glycol ether introduced as a substitute for EGE was also found to be hazardous.⁹⁵ Furthermore, it should be noted that along with the expansion of the global supply chain, developing countries such as China, Eastern Europe, and South America are rising as a new production base, which have less resources, weak regulations, and insufficient surveillance systems.⁴ It is possible that materials that are banned or processes that are tightly controlled in developed countries are not strictly followed there. All of these facts indicate that strengthening monitoring and surveillance must be emphasized more than ever.

Regarding etiological studies, we believe that simple replication of current studies using the same design would not give much help and therefore alternative strategies should be taken.

First, for cancer epidemiology, international collaboration research is imperative. Based on larger samples with sufficient statistical power, detailed sub-group analysis would be available. In fact, we have little knowledge of whether dose–response relationships are linear or an approach assuming constant risk over time is appropriate. Estimates of specific rates according to age of diagnosis, exposure timing, and exposure duration would add an important clue for causal inference. Also, different timing of regulation or introduction of new chemicals from one country to another could provide valid counterfactual conditions for making internal comparison. As LaDou and Bailar¹² proposed, the International Agency for Research on Cancer needs to take the initiative.

In addition, information on unusual events in one country needs to be shared across countries even prior to publication of scientific papers. By doing so, further research could be organized if necessary, and warning signals could be acted upon in accord with the precautionary principle. For example, when some Korean workers afflicted with multiple sclerosis or amyotrophic lateral sclerosis claimed workers' compensation, no scientific papers were found to support the work-relatedness of semiconductor production, though a number of papers have reported strong associations between organic solvent exposure and multiple sclerosis.⁹⁶ Later, we learned that the IBM PMR study also found elevated risks for these conditions and compensation claims by some electronics workers with multiple sclerosis had been resolved, all of which were not mentioned in the original article with a focus on cancer.^53,97 As with individual case reports, this sort of information could be instructive for constructing health surveillance systems or launching new epidemiologic studies.

Special focus should be placed on sentinel events such as hematologic abnormalities, reproductive aberrations, and dermatologic and respiratory disorders. For example, leukocyte abnormalities among fabrication engineers⁹³ and excess risk for SAB^{26,28–30,34} or congenital malformations in offspring^40,41 could be relevant to risks of cancer development. Skin⁷⁸ or respiratory abnormalities^75,76 could be a clue to reveal the presence of workplace hazards. They are characterized by relatively short latency between exposure and case development, and events are relatively common compared to cancer. Accordingly, primary data collection is readily available and internal comparison based on exact exposure classification is possible, which could minimize information bias, HWEs, and insufficient statistical power. In the case of the Semiconductor Health Study, daily urine collection and diary was done for 6 months among 400 workers, which was quite expensive and labor-intensive method, but the compliance was fairly good.⁹⁸ Such studies seem to be readily feasible in Korea and Taiwan where a large number of young females are working.

In the meantime, barriers that hamper further epidemiological studies and surveillance should be identified. In general, semiconductor manufacturing needs huge capital investment and its impact on national economies is considerable, especially in some countries; for example, in Korea, semiconductor products generated 6.5% of real GDP and about 10% of total export revenues in 2010.² Therefore, it is not unusual that national economic strategies are deeply involved in promoting this industry, privileged conditions are provided for them,^2,3 and trade unions rarely exist.⁴ In this context, both domestic and transnational companies have overwhelming power to control workers, working conditions, and even research itself. For instance, under the name of trade secrets, Samsung rejected the call from the workers' advocates to disclose industrial hygiene assessment reports, which were eventually made public through parliamentary inspection.⁹⁹ IBM resisted the disclosure of its Corporate Mortality File until forced by a court decision to release it,⁴ and then tried to prevent the publication of a scientific paper derived from analyses of its contents.^11,100,101 Moreover, workers suffer from the widespread adoption of outsourcing, subcontracting, and temporary work arrangements, in which they are more likely to be denied labor rights and excluded from regulatory protection.¹⁰² In such conditions, the records of employment or work environment tend to be kept poorly, and therefore it is more difficult to identify work history or track exposure status for the purpose of assessing work-relatedness.¹⁰³ Gender inequality also needs to be considered. The majority of manufacturing workers in the semiconductor industry are females, with this pattern based on the gender stereotype; females' physical make-up is considered more suitable for such jobs and females, especially young and from poor families, are thought to be docile, motivated, and obedient to management.¹⁰⁴ The fact that chemicals that have been strictly controlled for their reproductive toxicity in the printing industry where skilled and unionized male workers dominate are still used in the microelectronics shows that unequal power relations based largely on gender affect the working environment differentially.¹⁰⁵ In such a context, it is highly probable that passive surveillance works poorly and identification of health problems among workers is difficult as well as incomplete. In order to overcome these barriers, concealment of health and safety information under the guise of trade secrets needs to be strictly restricted on a legal basis and workers' participation in health and safety programs and/or epidemiologic investigations should be substantially guaranteed.

In conclusion, it seems clear that workers in semiconductor manufacturing, and in particular within the fabrication process, have been exposed to hazardous conditions contrary to the misconception of ‘clean’ image of this industry. These hazardous conditions are developing to a different extent across countries. However, there is still a significant gap in knowledge about health problems with long latency and rare incidence, while the causal association is strongly supported in the case of reproductive abnormalities. Monitoring and innovative research based on international collaboration is indispensable, while protective measures based on the available evidence should be immediately exercised under the precautionary principle.

Disclosure Statement

There is no conflict of interest.