Stress, health, noise exposures, and injuries among electronic waste recycling workers in Ghana

Our previously described data collection methods [15] will be briefly summarized here. All research procedures were approved by the University of Michigan Institutional Review Board (HUM00084062) and the University of Ghana Institutional Review Board at the Noguchi Memorial Institute for Medical Research (NMIMR-IRB CPN 070/13-14). A research team composed of students, staff, and faculty members from the University of Michigan and the University of Ghana-Legon collected the data in May 2014.

E-waste recyclers 18 years of age and older who worked at Agbogbloshie during the study were recruited to participate in the study. We conducted recruitment with the approval and assistance of the Chairman of the Greater Accra Scrap Dealers’ Association. Recruiting was conducted with the assistance of hired translators, and scheduled for approximately 1 h in the mornings and 2–3 h in the afternoons of days in which we collected data. Our goal was to recruit a convenience sample of 60 e-waste recycling workers, recruited through word of mouth and by our translators actively approaching individuals onsite to assess their interest. Workers participated in this study over a single day each, but were not required to be actively working on their day of participation. Each participant was approached in person and read a recruitment script and the informed consent in their native language. Interested individuals signed or provided an ink thumbprint on the informed consent form for official enrollment in the study. Participants received 9 GHS (about 3 USD) and a 3 GHS (about 1 USD) snack as a thank-you for their participation.

All participants completed a comprehensive interview administered in the language of their preference by our hired interpreters. The interview consisted of questions on demographics (e.g., age, religion, marital status, education, time living in Agbogbloshie, and income); health-related behaviors and outcomes (e.g., smoking status, self-reported health status); and personal stress factors, as measured by a subscale of Cohen’s Perceived Stress Scale (PSS, [21]) with a total of 16 points possible and higher scores indicating greater stress. The PSS survey has been translated and validated in a number of countries, but not, to our knowledge, in Ghana. Occupational information was also collected through the survey, including work activities, work duration, use of personal protective equipment (PPE), information on job demands and working conditions, and the number of injuries received during recent e-waste recycling activities. Work activities were defined after making initial observations of ongoing work at the site and with the assistance of the Chairman of the Greater Accra Scrap Dealers’ Association. The final categories developed were: burning of e-waste, collecting e-waste from consumers or businesses, collecting e-waste after burning, dismantling e-waste, lead acid battery recycling, lead smelting, removing wire coverings, repairing e-waste, and sorting e-waste. Occupational stress scores (OSS) were calculated on 28-point scale based on occupational stress-related survey questions; as with the perceived stress scale, higher values indicated higher levels of occupational stress. The frequency of exposure to noise, intended to represent participants’ “typical” exposure, was subjectively assessed on a five-point scale, with categories of “Never,” “Almost never,” “Sometimes,” “Fairly often,” “Very often,” or “Don’t Know.”

While the perceived frequency of noise exposure question on the survey provided information about typical exposures, we also objectively measured personal noise exposures using ER-200D personal noise dosimeters (Etymotic Research Inc. Elk Grove Village, IL, USA). The dosimetry data were intended to compliment the subjective data by providing quantified estimates of noise exposure levels overall, as well as during particular work and non-work activities. These data were also collected to allow for assessment of the potential relationship between objective noise levels and injury risk. The devices were configured to measure the equivalent continuous average noise level (L_EQ) according to the exposure limit for community noise recommended by the World Health Organization: A-weighting, slow time response, 3 dB time-intensity exchange rate, 70 dB threshold, and 75 dB criterion level [13]. The measurement range of the noise dosimeter was 70–130 dBA. Average and maximum noise exposures were data logged every 3 min 45 s (the default and only data-logging interval length available for these dosimeters) for up to 24 h. Exposures during work activities were compared to the 85 dBA exposure limit used for occupational noise exposures in most countries around the world [49].

Study data were cleaned using methods we have described previously [15]. All statistical analyses were performed using SPSS Version 24.0 (IBM SPSS Statistics for Windows, IBM Corp., Armonk, NY, USA). For workers who completed personal noise dosimetry, we computed 8-h time-weighted average (TWA, in dBA) exposures using Eq. 1, where L_AEQ is a 3.75-min average equivalent continuous noise level, N is the total number of 3.75-min intervals i in the measured shift, and 128 is the number of 3.75-min intervals in a 480-min shift.

Descriptive univariate and bivariate analyses were conducted; for all inferential analyses, results were considered statistically significant where p < 0.05. We used multivariable Poisson regression [34] to model the relationship between the number of self-reported e-waste-related injuries and other occupational and non-occupational factors. Eq. 2 depicts the Poisson regression model, where the expected rate (E) of the number of injuries (Y) given the continuous and categorical predictor variables, x, equals the value of the effect on the predictors, α, added to the coefficient β^′, or the multiplicative effect on the mean of (Y) as the result of x.

We ran unadjusted (i.e., single predictor variable) and adjusted Poisson regression models on a number of potential injury risk factors. Variables tested in unadjusted models were chosen based on the participant responses to questions about work activities where they received the most injuries and the body sites reported where workers sustained injuries. Noise-related variables were selected due to the knowledge of the worksite conditions, worker responses to questions about noise exposure and a priori knowledge of noise related occupational injuries [4, 16, 17, 33, 38]. Each variable was tested individually in an unadjusted Poisson model to determine its effect on the outcome and on improvements in model fit (as assessed via the Akaike Information Criterion, AIC). A forward stepwise selection routine was used to select the final adjusted model. Two variables (age and self-reported health status) were forced into our adjusted model based on a priori assumptions derived from previous research [12, 22, 47], though it is important to note that these prior studies were conducted in formal (and not informal) work settings. However, recent studies on informal waste sites in Nigeria demonstrate a significant association of age to adverse outcomes of exposure and injury [41, 42].

We also performed a sensitivity analysis related to noise exposures. We repeated the final multivariable Poisson regression developed from all workers on the subset of workers who completed personal noise dosimetry, and replaced the self-reported noise exposure frequency variable with a variable representing measured noise exposure level in dBA.

Demographic characteristics are shown in Table 1 for all participants (N = 46 e-waste workers) and the subset of workers who completed personal noise dosimetry (N = 26). We did not track the total number of potential participants approached to participate, and so cannot report a participation rate. Among workers who declined to participate, the most common reason given was an insufficient financial incentive. No significant differences were noted between the total and subset samples (data not shown). Participants had an average age of about 25 ± 6 years old, and had been a resident at the site for an average of 5 ± 3 years. Slightly over half of had no formal education, and a similar fraction described their general health as fair or poor. The majority of participants were migrants from Northern Ghana who had traveled to Accra to pursue employment opportunities (data not shown). Dagbani was by far the most commonly-used interview language (65% of participants), and a slight majority of interviews were conducted by interviewer 1.

Participants had worked at the site an average of 5.7 ± 3.3 years, and worked an average of approximately 10 h/day (Table 2). Workers reported participating in a wide variety of activities, and generally did not specialize in any activity. The most commonly-reported activities were dismantling e-waste (reported by 83% of participants) collecting e-waste (80%) and sorting e-waste (78%). The least-commonly reported activities included lead smelting (reported by 39% of participants) and removing wire coverings from e-waste (28%).

The average score on the PSS and OSS was over 50% of the total possible points (PSS mean = 9.9 of 16 possible points, OSS mean = 19.7 of 32 possible points). Nineteen percent of participants reported impairments that limited their work abilities, and a minority of workers (43.5%) reported receiving any safety training prior to performing work that resulted in injury. Thirty-eight percent of the participants made less than 10 Ghanaian Cedis (about 2.20 USD) per day; for comparison, the average daily wage in Ghana during the time period of this study was 6 Ghanaian Cedis (about 1.30 USD, [1]). While seniority (i.e., more time on the job) is often associated with greater pay, the correlation between time at the site and pay was poor and insignificant.

The vast majority of workers (87%) reported frequent exposure to high noise on the job, and also reported being frequently bothered by this noise (78.3%, Table 2). Among the 26 workers with personal dosimetry, the average TWA for noise was 78.8 ± 5.9 dBA. Approximately 15% of TWA exposures exceeded the recommended 85 dBA exposure limit; all of these overexposures were associated with dismantling activities. Additionally, 73% of the workers reporting frequent exposure to high noise did dismantling. The Spearman correlation coefficient between measured noise level (in dBA) and self-reported noise exposure frequency among the participants who completed personal noise dosimetry was not significant (p = 0.21).

A total of 426 injuries in the prior 6 months were reported across all 46 participants. The average number of injuries was 9.9 ± 9.6 per participant, though the range of injuries was quite large (1–40 injuries per participant, Table 3). The total number of days of work missed due to occupational injuries was 193; the average number of missed workdays for injuries during this period was 4 ± 2.6 days, although few participants were hospitalized for their injuries (6.5%). Most of the reported injuries were lacerations (65.2%), and the most common injury locations was the hand (45.7%). The PPE categories most commonly used were pants and foot protection (defined here as shoes, as opposed to the open sandals commonly used by workers on the site). Few workers reported wearing gloves; only 26% reported glove use during dismantling or sorting of e-waste. Fisher’s exact tests of the associations between glove use and age, income, or seniority. No workers reported wearing hearing protection. The work activity associated with the most injuries was dismantling, an activity during which PPE use was rare (Fig. 1). PPE use was uncommon during all tasks, and at least 50% of injuries occurred when participants were not using any PPE.

When all workers were considered, the Spearman correlation coefficient between PSS and injuries (Fig. 2) was significant, where higher stress levels were associated with more injuries (r = 0.42, p = 0.001). Age, self-reported health status, perceived noise exposure, and measured noise exposure were not significantly correlated with number of injuries. Age and stress were also not significantly correlated (data not shown).

Our unadjusted Poisson regression results identified a number of variables that were significantly associated with number of injuries. These include: Age, education, perceived noise, impairment that limits work, measured noise, glove use, occupational stress score, perceived health status, perceived stress scale, and training prior to injury. Also, being interviewed in Dagbani and being interviewed by interviewer 2 were both significantly associated with an increased number of injuries. When all 46 participants were included in our adjusted multivariate Poisson regression model with number of injuries as the outcome (adjusted model, Table 4), higher PSS, not using gloves, and higher perceived noise exposure frequency were significantly associated with a higher number of injuries. Not performing dismantling work, greater age, and better self-reported health status were significantly associated with a reduced number of injuries. Interview language and interviewer were not selected in the forward stepwise selection process for the final adjusted model based on lack of improvement in the AIC when these variables were included in the model.

We completed a sensitivity analysis comparing the number of injuries associated with perceived noise exposure vs the measured noise levels from personal noise dosimetry (i.e., a multivariate Poisson model with the same variables in Table 4 but restricted to the subset of 26 participants with personal noise dosimetry). The subset of 26 workers did not differ significantly from those of the total sample for any of the variables assessed (data not shown). Our sensitivity analysis yielded model results which were generally very similar (data not shown); however, in this sensitivity analysis, higher measured noise levels were associated with a significant decrease in the number reported injuries (β = − 0.05 injuries per dBA, SE 0.01, p = 0.001).

This study has a number of limitations. First is the small sample size and cross-sectional nature of the study. While most participants had worked at the site for several years, and anecdotal reports indicated that conditions and work practices do not appear to have changed substantially over that period, it is nevertheless possible that injury risks could have changed in meaningful ways over our 6-month injury reporting window. This could introduce positive or negative bias into our estimates of injury frequency. Also, because exposures and injury experiences were evaluated simultaneously, we were unable to assess causality; we cannot determine, for example, whether workers who reported poor health status have more injuries as a result of that health status, or vice versa. Second, our assessment of injury risk may have been influenced by the healthy worker effect, and our injury estimates may have been biased negatively by the exclusion of less healthy workers who are no longer doing e-waste recycling work due to illness or injury. Third, our assessment of work activities conducted by the workers was completed using activities reported to us by participants, that we then collapsed into smaller post hoc categories. We used this approach to try to reduce the large number of specific work activities reported by individual workers into a manageable number of broader job categories. However, in doing so, we may have introduced misclassification of work activities, which could bias our understanding of injury risks associated with our defined work activities. Fourth, our participants came from Northern Ghana and other West African countries and spoke a number of languages; it is possible that errors or miscommunication by our translators, differences in the context and interpretation of some questions, or social response biases introduced by perceived or actual social class differences between our participants and translators, may have introduced additional misclassification. This possibility is reflected in the significant univariate Poisson regression results for interview language and interviewer. The cultural appropriateness of the PSS and OSS instruments we used has not been validated in this population. Fifth, the temporal misalignment between our objective and subjective measures of noise exposure may at least partially explain the poor correlation between objective or subjective noise exposures, as well as the inconsistencies in our injury risk models that included a noise exposure variable. Finally, although we emphasized to participants that their responses would be confidential and would never be shared with others, some workers may have purposely misrepresented the number of injuries experienced or health status to avoid perceived adverse impacts on their employment.