psychosocial effects of pcb contamination and remediation: the case of smithville, ontario

Pergamon

Sot. Sci. Med. Vol. 39, No. 8, 1093-l 104, 1994 pp.

0277-9536(94)EOO16-L Copyright 0 1994 Elsevier Science Ltd

Printed in Great Britain. All rights reserved 0277-9536/94 $7.00 + 0.00

PSYCHOSOCIAL EFFECTS OF PCB CONTAMINATION AND REMEDIATION: THE CASE OF SMITHVILLE,

ONTARIO

JAMES R. DUNN,’ S. MARTIN TAYLOR? SUSAN J. ELLIOTT~ and STEPHEN D. WALTERS

‘Department of Geography, Simon Fraser University, Burnaby, British Columbia, Canada VSA 1S6, *Department of Geography, McMaster University, Hamilton, Ontario, Canada L8S 4K1, )Department of Geography, University of Victoria, Victoria, British Columbia, Canada V8W 3P5 and 4Department of Clinical Epidemiology and Biostatistics, McMaster University, Hamilton, Ontario, Canada L8N 325

Abstract-A cross-sectional study design was used to investigate psychosocial effects in the population exposed to PCB contamination and remediation in Smithville, Ontario, Canada. Psychosocial effects were defined as the complex of distress, dysfunction and disability manifested in a wide range of psychological, social and behavioural outcomes as a consequence of actual or perceived contamination. This paper describes the results of logistic regression analyses conducted to investigate the determinants of psychosocial effects of exposure. The data come from an epidemiologic survey of a sample (N = 272) of Smithville households within 3 km of the site, and a sample from a matched comparison community (N = 263). Two types of outcome variables were used as indicators of psychosocial effects. Scores on the genera1 measures of psychosocial health and well-being (the GHQ-20 and the somatic complaints checklist of the SCL-90) for the Smithville sample did not differ from those expected in a ‘normal’ population, nor did they differ from those in the comparison community. Scores on these outcomes were associated with plausibly linked independent variables, but were not associated with exposure to the PCB site. Results for site-specific outcome measures (concern and health concern) showed that moderate levels of concern reported by Smithville respondents were explained by concern about another local hazardous waste issue and could not be explained by PCB site exposure. We conclude from these results that local community context exerts an important influence on psychosocial effects of environmental contamination. Further- more, the types of outcome measures employed and the timing of the research in the context of the site history were important factors in our ability to detect psychosocial effects of the PCB contamination and remediation in Smithville.

Key words-psychosocial effects, epidemiologic survey, PCBs, hazardous waste

INTRODUCTION

This paper reports findings from an epidemiologic survey conducted to investigate pyschosocial effects in a population exposed to PCB contamination and remediation at a site in Smithville, Ontario (see Fig. 1). Psychosocial effects have been investigated in the wake of other environmental contamination events including the Three Mile Island accident [l] and the Hagersville tire fire [2] and are gaining recognition as an important and legitimate concern for social and health scientists, though they remain poorly understood [3]. This investigation of the effects of contamination and remediation of a toxic waste site adds to the existing literature on psychosocial effects of environmental contamination.

Formerly a PCB transfer station, the site of interest was established in 1978. In 1985, pools of PCB- contaminated sludge created by leakage from storage containers were discovered. The contaminants had also leached into the groundwater, threatening the town’s main water supply, which at the time, was a large well located 550 m south of the PCB site. In response to a strong lobby from local residents, the

Ontario Ministry of the Environment made a com- mitment to be responsible for cleaning up the site. Smithville was connected to a water pipeline from Lake Ontario in 1988, and site clean-up began in 1989. Part of the clean-up involved a controversial mobile incinerator which operated continuously from January 1991 to December 1992. The facility’s emissions stack was very visible, and the site was a prominent feature of the local landscape. Throughout its history, the site has been the subject of sensation- alized media reporting, including headlines compar- ing it to Love Canal, and labelling Smithville ‘the PCB capital of Canada’. Later, local residents’ suspi- cions that numerous lengthy delays in the clean-up were part of a plan by the Ontario Ministry of the Environment to make the facility a permanent toxic waste incinerator were played out in the print media. In short, the PCB site has had a high public profile and has been an important aspect of the lifescape of Smithville residents [4].

Alongside the evolution of the issue, the Ontario Waste Management Corporation (OWMC) began a process to select a site for a permanent hazardous waste incinerator and landfill in the Niagara Region.

1093

JAMES R. DUNN et al.

Fig. 1. Location of the Smithville PCB site.

The proposed facility will provide toxic waste disposal and treatment services to the entire province. In the same year the contamination was discovered in Smithville, the OWMC selected a site for their facility only 12 km from Smithville, near Bismarck, Ontario (see Fig. 1). Environmental assessment hearings on the OWMC proposal began in 1990 and are ongoing. The importance of the proposed OWMC site as a factor influencing reactions to PCB contamination and clean-up in Smithville was confirmed in a preced- ing, qualitative stage of the research through depth interviews with Smithville residents [5].

Embedded in this history is a more complex social and community process involving the original discov- ery of the contamination, the community-based lobby for government response, and the subsequent negotiations resulting in the introduction of a controversial incineration technology to clean up the site, as well as responses to the proposed OWMC facility. The details of this contextual background and process are described elsewhere [5]. For the purposes of this paper, perhaps the most important implication from the site history is that at the time of the study Smithville residents were exposed to and were react-

ing primarily to PCB incineration and clean-up and not PCB contamination per se. This is an important distinction with implications for the type and magni- tude of psychosocial distress that could be anticipated a priori. This point is revisited in the conclusions in light of the study findings.

FOCUS AND CONTEXT

For the purposes of this study, psychosocial effects are defined as the complex of distress, dysfunction and disability manifested in a wide range of psychological, social and behavioural outcomes as a consequence of actual or perceived environmental contamination [6]. Psychosocial effects can occur at different levels of social organization: at the individual, the social network, and the community. At the individual level, they include emotional (e.g. worry, anger, loss of control, guilt), behavioural (e.g. task performance, help-seeking, information seeking) and somatic (e.g. headache, fatigue, depression) effects. Effects at the social network level include family disruption and social isolation, but effects may also be positive, for example when cooperative coping

Psychosocial effects of PCB contamination and remediation 1095

strategies are adopted or greater social cohesion occurs in response to the environmental events. Finally, at the community level the effects of environmental contamination often include stigmatization and dislocation, but again positive effects, including community enablement and empowerment, can occur. Effects at each level of social organization cannot be understood in isolation. For example, individual level effects cannot be understood in isolation from social network and community level variables. Con- sequently, the site history as reported in the local print media is drawn upon to contextualize the survey results. Also, a part of the survey instrument used in this study is devoted to gaining a greater understanding of individuals’ social networks and their community involvement. Although psychosocial effects do occur at different levels of social organization, the focus of this analysis is on effects at the individual level.

The definition also recognizes that psychosocial effects can occur as a consequence of actual or perceiued environmental contamination. The importance of perceived exposure to environmental contaminants in this definition derives from evidence that the impacts of perceived contamination are every bit as ‘real’ as the impacts of actual contamination. This is evident in other work by Elliott et al. [7], where high levels of concern, health concern and action were observed in the population proximate to a proposed solid waste facility near Milton, Ontario. In the Smithville case, there were no chemical exposures sufficient to cause physical health effects in the population proximate to the site. It follows that any effects observed can be attributed to perceived exposure.

The most useful theoretical framework for guiding this research comes from the environmental stress and coping literature [8-lo]. Environmental stress is defined by Baum et al. as “a process by which environmental events threaten, harm or challenge an organism’s existence or well-being and by which the organism responds to this threat” [l 1, p. 1861. Lazarus and Folkman propose that response to environmental stress is divided into two stages: primary appraisal whereby the individual appraises the stressor as a threat, harm or challenge; and secondary appraisal whereby one of two types of coping strategies is selected: (1) problem-focused (e.g. joining a citizen’s action group), or emotion-focused coping (e.g. adjusting one’s attitudes towards the stressor). Reappraisal occurs periodically as perception of the stressor or available coping resources change over time [lo].

The incidence of psychosocial effects, the experience of environmental stress and the choice of a particular coping response are shaped by a number of mediating factors which can be categorized into four groups [6]: characteristics of the environmental stressor [ 12-141; characteristics of the individual [13, 151; characteristics of the social network [4, 161; and characteristics of the wider community system

[4, 14, 171. These findings guided the design and analysis of the current study and the selection of environmental stress theory as the theoretical framework.

STUDY DESIGN AND METHODS

The objectives of this study were as follows:

(1) To determine the prevalence of psychosocial effects in individuals exposed to the Smithville PCB site;

(2) To investigate the determinants of psychosocial effects in individuals exposed to the Smithville PCB site.

A cross-sectional study design was chosen to investigate both objectives. A stratified random sample of Smithville households (N = 272) was selected across two zones using property tax data. A second sample (N = 263) was randomly selected from the population in Vineland/Jordan, a comparison community selected in order to assess the specific effects of the PCB site in Smithville. The communities were matched in two important respects: first, their simi- larity in socio-demographic profile, determined using 1986 Census of Canada data; second, their similar proximity to the proposed toxic waste disposal and treatment facility, known as the OWMC site. Resi- dents’ concerns about the proposed OWMC site were measured in both communities and included as an explanatory variable in the modelling of the determinants of psychosocial effects.

Stratification of the Smithville sample by distance zone allowed for internal comparison. The first zone included all Smithville households within a 1 km radius of the site and north of the railroad tracks, while Zone 2 included all other Smithville households within a 3 km radius of the site. Zones were defined in this manner for three reasons: (1) because within 1 km the incinerator emissions stack was visible and beyond 3 km the stack was not visible; (2) households in Zone 1 and 2 were qualitatively different: Zone 1 consisted mainly of low-density farm properties, while Zone 2 covered the town of Smithville and included mainly high-density, residential properties; and (3) for consistency with the qualitative stage of the study [5].

The sampling frame in both communities consisted of household phone numbers identified using tax assessment rolls and telephone directories. One adult member of each household was randomly selected. The comparison sample was randomly selected from within the boundaries of an agglomeration of census enumeration areas (EAs) in the area of Vineland and Jordan, Ontario. A brief summary of the socio- demographic characteristics of each sample appears in Table 1. With few exceptions, the samples are well-matched to each other and to the communities from which they were drawn. In Vineland/Jordan, a greater percentage of the sample was retired, the

1096 JAMES R. DUNN et al.

Table I. Sample characteristics

Characteristic SMVL 2.1 SMVL Z.2 SMVL Total VlNE/JRDN

(n = 20) (n = 252) (II = 272) (n = 263)

% Female Mean average % Total household income < $30,000 % Partner % High school+ Employment

% FULL TIME % PART TIME % RETIRED % HOMEMAKER % OTHER

Mean No. persons/ household % Household with children <5 years % English at home % Own dwelling Median No. years at:

Current address

55.0 60.0 55.9 58.2 38.7 43.5 43. I 49.2

15.0 12.7 12.9 I I.0 85.0 81.3 81.6 73.9 75.0 69.8 70.2 12.6

50.0 47.6 47.8 40.7 15.0 16.7 16.5 12.2 10.0 13.5 13.2 24.3 20.0 II.1 II.8 12.2

5.0 II.1 10.7 10.6

3.6 3.3 3.3 4.0

25.0 22.6 22.8 12.5 100.0 97.2 97.4 93.5 100.0 87.1 88.6 90.5

2.5 5.0 5.0 6.0 Area 10.0

Note: SMVL Z.1 = Smithville Zone I. SMVL 2.2 = Smithville Zone 2.

15.0 14.5 18.0

SMVL Total = Smithville total. VINE/JRDN = V&land/Jordan.

mean age was higher, and there was slightly greater residential stability than in Smithville. In the Smithville sample, a greater percentage of the sample had a partner, had children less than 5 years of age at home, and worked full-time than in Vineland/Jordan.

Although efforts were made to interview all 40 households in Zone 1, only 20 residents completed interviews. In Zone 2,252 respondents were surveyed. The overall response rate was 79%, with a total sample size of 272 individuals. In Vineland/Jordan, 263 households were surveyed, and the response rate was also 79%. The low response rate in Zone 1 may be attributable to the timing of the research. The survey was conducted 16 months after the incineration began and it is possible that the people closest to the site, who had been most involved in discussions with government and other agencies [5], may have tired of answering further questions about it. In addition, they may have been unwilling to complete interviews because they simply “wanted to get on with their lives”, a feeling expressed by residents affected by environmental contamination events elsewhere [2,4].

The survey instrument used in this study was adapted from the instrument used to collect data reported by Elliott et al. [6,7]. It was administered by telephone by the Institute for Social Research at York University between 13 June and 22 July 1992. It was introduced as a survey of residents’ attitudes towards quality of life in southern Ontario communities and consisted of eight sections, most of which were administered to both the Smithville and the Vineland/Jordan samples. Some sections were only relevant in one community or the other.

The first section of the instrument asked respondents to identify the factors they liked and disliked about the area where they lived and to rate their

satisfaction with their area as a place to live. It was expected that many respondents would volunteer the Smithville PCB site and/or the proposed OWMC site as a dislike. Section B determined the nature and strength of the individual’s social and community networks, so that their contribution to psychosocial effects could be ascertained. Section C contained general measures of psychosocial health and well- being. Two measures from the existing literature were included: the 20 item version of the General Health Questionnaire (GHQ-20) [18] and a modified version of the somatic complaints sub-scale from the Symp- tom Check List-90 [19,20]. These two scales were selected from a larger set of instruments [21] based on their appropriateness for use in a general population, their previous validation, and the existence of pub- lished population norms against which scores in the study could be compared [6]. The GHQ measures several aspects of emotional distress, including pre- disposition to depression, anxiety and social impair- ment. The SCL-90 sub-scale taps non-specific somatic complaints (e.g. headache, backache, fatigue) associated with psychosocial morbidity. The general health status section also contained items from the Critical Life Events Scale [22] to screen for major sources of psychosocial morbidity unrelated to environmental contamination in the respondent’s area.

Both closed- and open-ended site-specific questions about the Smithville PCB site were contained in Section D. Administered only to Smithville respondents, this section measured three major constructs: site awareness, site concerns, health-related concerns and actions taken or intended towards the site. Questions regarding awareness of and concerns about the proposed OWMC toxic waste facility appeared in Section E and were asked of both Smithville and Vineland/Jordan respondents. Section F, general


environmental concerns, was only administered to Vineland/Jordan residents to screen for major environmental perturbations unknown to the authors which could be sources of psychosocial morbidity. Questions about respondents’ actions or intentions to move, and their reasons for considering moving were the focus of Section G, asked of respondents in both communities. Finally, Section H included standard socio-demographic variables (i.e. gender, age, lifecycle status, socioeconomic status, length of resi- dence, etc.) as a check on the representativeness of each sample group and their comparability, and as potential mediators of psychosocial effects for inclusion in the analysis.

ANALYSIS

Two similar analytical models guided the analysis, depending upon the outcome variables under con- sideration. Two types of outcomes were examined: general measures of psychosocial health and well- being (Fig. 2); and site-specific measures of general and health-related concern about the PCB site (Fig. 3). The general measures of psychosocial health and well-being employed were the GHQ-20 and the SCL-90 somatic complaints sub-scale. For each item of the GHQ-20, respondents were asked first if they had experienced the condition in the past two weeks, and respondents who answered “yes” were asked if they had experienced it “more than usual” for them, “less than usual”, or “about the same as usual”. Goldberg [18] suggests two possible methods for scoring the GHQ. The first is the conventional Likert method, with O-l-2-3 values for the response cat-

egories. The more widely used and recommended scoring method [ 18,231, however, is the O-O-1-1 method, where the score for each item represents the presence or absence of individual problems over the last two weeks. Individual item scores are added to give the scale score. For this screening test, a score of 4 or more means the respondent is a “probable case” (i.e. probable case of emotional distress) [18]. In the analyses that follow, the O-O-1-1 scoring method was used and a cutpoint of 4 distinguished probable cases from probable non-cases. Cronbach’s 61, a measure of the internal consistency of a scale [24] was calculated to be 0.82 for the GHQ-20. An tl greater than 0.80, indicates the scale has good internal consistency.

An augmented version [6,7] of the somatic complaints sub-scale of the Symptom Check List-90 (SCL-90) was administered to all respondents, in both communities [19,20]. The SCL-90 sub-scale consisted of 16 items that measure a variety of non-specific somatic complaints (e.g. headache, backache, fatigue) hypothesized to be associated with psychosocial morbidity [6]. Items were scored on a five-point scale: respondents were asked if they had (for example), been bothered by faintness or dizziness over the past two weeks. Response categories were ‘not at all bothered’, ‘bothered a little bit’, ‘moder- ately bothered’, ‘bothered quite a bit’, or ‘extremely bothered’ and were scored O-l-2-3-4 respectively. Item scores were added and averaged to arrive at the scale score. A normalized cutpoint of 0.36 for the somatic sub-scale of the SCL-90 was generated by Derogatis [20], and was used in the subsequent analyses. Cronbach’s CI for the SCL-90 somatic

A

INDMDUAL

Age Gender Lifecyde SES

EXPOSURE

B

SOCIAL NETWORK

PRIMARY

Membership

Satisfaction

SUPpofi SECONDARY

Membership Sam- involvement

OUTCOMES

GHQ-20

SCL-SO

ENVlRONYENTALSlXBBDR

conoarn about OWMC

t-

Fig. 2. Analytical model: GHQ-20 and SCL-90.


I I

I SOCIAL NETWORK

-I PRIMARY

INDIVIDUAL

Age Gender Lifecycle SES Resi&ltial Status

Membership Satisfaction

SUPpofi SECONDARY

Membership

Satisfaction

t

1

Concern-PCB

Health Concern-PCB

Fig. 3. Analytical model: concern and health concern.

complaints sub-scale was 0.80, indicating the scale had good internal consistency.

Regarding site-specific outcome variables, at the beginning of the questionnaire, respondents were asked to identify the things they disliked about the area where they live, and were allowed up to three mentions. It was expected that Smithville respondents would identify the PCB site as a dislike, and that respondents in both communities would identify the proposed OWMC facility as a dislike. Later in the questionnaire, Smithville respondents were asked “Do you have any concerns about the PCB site in the Smithville Industrial Park?” and “Do you consider any of these concerns to be health related?“. Corre- sponding variables were called ‘concern’ and ‘health concern’. Note that these questions did not dis- tinguish concerns about the contamination from concerns about the remediation using the incinerator. Typically, respondents had difficulty separating the two issues [5].

Explanatory variables in both analytical models are categorized into blocks. Common to both models are the following blocks: individual (e.g. age, gender); exposure (respondent’s location (or zone)); social network (primary and secondary networks); individual stressors (e.g. death of spouse, divorce); and environmental stressor (concern re: proposed OWMC facility). There is an additional block of general health status variables (e.g. GHQ-20; perceived health status) in the analytical model for site-specific concern and health concern (Fig. 3) because these factors were hypothesized to be potential mediators between PCB site exposure and the out-

comes [7]. The structure of the analytical models is consistent with the conceptual framework [6]. The analytical models, however differ from the conceptual framework in two important respects. First, while the conceptual model clearly shows systemic, bidirec- tional relationships between variables, by necessity the analytical models could only accommodate uni- directional relationships between independent and dependent variables. This apparent inconsistency illustrates the difficulty of translating a complex conceptual framework into an operational model suitable for statistical analysis [7]. Second, community level variables are not explicitly incorporated into the analytical models. These were examined in a qualitative stage of the research project, which pre- ceded the epidemiologic survey. The results of the qualitative work provided contextual information which guided the study design, adaptation of the survey instrument and interpretation of the results of the statistical analysis [5].

Logistic regression was chosen as the method of analysis for three reasons: (1) the outcomes of interest were dichotomous; (2) the external and mediating factors were a mix of continuous and categorical variables; and (3) the relationships between the independent and outcome variables were sensibly described by a logistic function [7]. Logistic regressions were calculated for two outcome variables using SPSS/PC+. For each outcome, an hierarchical model was constructed by entering each block of variables shown in the analytical models (Figs 2 and 3, depending on the outcome of interest) in an additive fashion (i.e. first the ‘exposure’ block, then


Result

Table 2. Results from the GHQ-20 subscale SMVL Z.1 SMVL 2.2 SMVL Total (n = 20) (n = 252) (n = 272)

VINE/JRDN (n = 263)

% of respondents above cutpoint (4+)’ Sample group’s mean raw score (MRS)

10% 9% 9% 10%

1.31 0.96 0.99 0.97

‘Z.1 “S 2.2: x * = 0.037, P = 0.85. SMVL vs VINE/JRDN: I* = 0.178, P = 0.67. Note: SMVL Z.1 = Smithville Zone I.

SMVL 2.2 = Smithville Zone 2. SMVL Total = Smithville total. VINE/JRDN = Vineland/Jordan.

the individual block, then the social network block, and so on), keeping only those variables which contributed to the model, and removing those that did not, before moving on to the next block of variables. Models were run using a stepwise back- ward elimination algorithm within each block. Vari- ables were candidates to contribute to the model if the significance level (P ) for the Wald inclusion statistic was 0.10 or lower. Variables in the final model with a P-value less than 0.10 but greater than 0.05 were removed if the estimate of the relative odds was unstable (i.e. if the upper and lower limits of the confidence interval straddled unity, creating some doubt as to whether the relative odds for that variable was greater or less than one). Due to their a priori importance age, gender [18, 19,231, and respondent’s location (or zone) were forced into every model regardless of their significance level [6]. After all significant explanatory variables had been identified, their first-order interaction terms were added into the model using forward stepwise selection. Interaction terms were of two types: (1) age, gender and zone by each remaining independent variable, and (2) each remaining independent variable with all others. Once this step was complete, logistic regressions for each outcome were calculated including interaction terms that were shown to contribute significantly to the model, their lower order relatives and all single variables shown to contribute significantly to the model.

RESULTS

The results of the logistic regression analyses are presented in two parts: first, models of the two general psychosocial health and well-being outcome measures (the GHQ-20 and the SCL-90 somatic complaints sub-scale) are described. Data were col- lected for respondents from both communities on these two variables. Next, models for site- and situation-specific outcome variables are presented. Models are described for general and health-related concern about the Smithville PCB site. These two outcomes were only relevant to the Smithville population.

General measures of psychosocial health and well-being

The outcome variables investigated in this section are scores above or below the cutpoint for normal on

the GHQ-20 and on the somatic complaints sub-scale of the SCL-90. In each instance, descriptive results, including percentages of respondents scoring above the cutpoint for normal and the overall mean raw score are reported, followed by the results of logistic regression analyses.

The percentage of Smithville respondents scoring above the cutpoint on the GHQ-20 (Table 2) was relatively low (9%), when compared with other studies [25], and only small differences emerged between zones 1 (10%) and 2 (9%). A similarly small percentage of Vineland/Jordan residents (10%) scored above the cutpoint. Mean raw scores on the GHQ-20 also demonstrated little difference in scores between study sub-groups. In sum, the GHQ-20 scores for the Smithville sample do not differ from those expected in a ‘normal’ population.

The logistic regression model for the GHQ-20 (Table 3) had a p * of 0.35, where p * measures the overall goodness of fit of the model. It is defined as one minus the ratio of the maximum log likelihood values of the fitted and constant-only-term models [26]. Calculated values for p * range from zero to one, and values between 0.2 and 0.4 represent a very good model fit [26, p. 501. The positive predictive value of

Table 3. Results of logistic regression for outcome: above the cutpoint on the GHQ-20 (Smithville and Vineland/Jordan samples)

Relative odds Variable (95% confidence interval) Age*

Gender Zone Zone I vs V/J Zone 2 vs V/J Not satisfied w/social activities Had P serious illness in past 12 Inoaths*** Spouse/partner died in past 12 months*** Been worried about someone in Dust 12 months** had Anamial problems in household in past 12 months*** Employment status x presence of household financial problems*

Otber x no problems** Homemaker x no problems*

0.95 (0.91; 0.99) 1.88(0.49; 7.13)

0.83 (0.14; 5.03) 0.79 (0.34; 1.74) 4.93 (2.0; 12.2)

5.39 (2.12; 13.72)

26.2 (4.97; 138.4)

4.19 (1.717; 9.9)

12.3 (2.85; 53.4)

24.51(2.4; 248.7) 24.48 (2.0; 300.0)

% Correct: 91% Specificity: 93% Sensitivity: 55% 01: 0.35

++*p < 0.001; **P < 0.01; l P < 0.05.

1100 JAM= R. DUNN et al.

Table 4. Results from the (modified) SCL-90 somatic sub-scale

SMVL 2.1 SMVL. 2.2 SMVL Total Result

VINE/JRDN (n = 20) (n = 252) (n = 272) (n = 263)

% of resoondents above cut-point’(0.36)’ 35% 32% 32% 32% Sample group’s mean raw score (MRS) 0.29 0.32 0.32 0.32

‘Z.1 vs 2.2: * = P x 0.069, = 0.79. SMVL vs VINE/JRD: 2 y < 0.0001, P = 0.99. Note: SMVL Z.1 = Smiihville Zone I,

SMVL Z.2 = Smithville Zone 2. SMVL Total = Smithville total. VINE/JRDN = Vineland/Jordan.

this model (i.e. the percentage of those respondents who were predicted to be above the cutpoint who actually scored above the cutpoint) was low (24%) but the negative predictive value (i.e. the percentage of respondents predicted to be below the cutpoint who actually scored below it) was high (98%). The model had very good specificity (i.e. the percentage not above the cutpoint who were correctly predicted) at 93% but much lower sensitivity (i.e. the percentage above the cutpoint who were correctly predicted) at 55%. Overall, the model correctly classified 91% of respondents. Estimates of the percentages correctly predicted are based on a predicted probability of least 0.50 for the predicted outcome.

The significant explanatory variables in this model (shown in bold type in Table 3) included those from the individual, the social network and the individual stressor blocks (Fig. 2) with characteristics of the individual and individual stressor comprising the significant interaction effects. Note that respondent’s location, a proxy for exposure to the PCB site, did not make a significant contribution to the model. The relative odds (R.O.) and associated 95% confidence intervals also appear in the table. Relative odds (e 8) is the factor by which the odds of having the outcome variables changes when the independent variable increases by one unit (or, in the case of categorical variables, changes from one category to another) [27]. If /I is positive, relative odds are greater than one, which means that odds are increased. If /3 is negative, the relative odds are less than one meaning that the odds are decreased. Using satisfaction with social activities as an example, the relative odds was 4.93, indicating that an individual not satisfied with his/her social activities was more likely to score above the cutpoint for normal on the GHQ-20 than someone who was satisfied. The 95% confidence interval shows that the range of the relative odds for this variable was between 2.0 and 12.16.

Based on significant single effects in the model (Table 3), respondents were more likely to score above the cutpoint for normal on the GHQ-20 if they: were younger; were not satisfied with their social activities; in the previous 12 months had had a serious illness; had had a spouse or partner die; had been worried about someone; or had experienced financial problems in their household. In addition, the significant interaction effects indicated that respondents

were more likely to score above the cutpoint for normal on the GHQ-20 if they: had not experienced household financial problems and their employment status was ‘other’ as opposed to ‘employed full-time’, had not experienced financial problems and their employment status was ‘homemaker’ as opposed to ‘employed full-time’. Gender and zone did not make significant contributions to the model.

The percentage of Smithville respondents scoring above the cutpoint for normal on this scale (Table 4) was relatively low (32%) when compared with other studies [25], and only a small difference emerged between zones 1 (35%) and 2 (32%). A similarly small percentage of Vineland/Jordan residents scored above the cutpoint for normal (32%). Mean raw scores on the SCL-90 also demonstrated little difference in scores between any of the study sub-groups.

The model for the SCL-90 somatic sub-scale (Table 5) had a p * of 0.19. The positive predictive value was low (48%) while the negative predictive value was high (91%). Specificity (79%) and sensitivity (73%) were both quite high. Overall, 77% of respondents were classified correctly.

The significant explanatory variables that entered this model (bold type in Table 5) included those from

Table 5. Results of logistic regression for outcome: above the cutpoint on the SCL-90 somatic complaints sub-scale (Smithville and

Vineland/Jordan samples)

Relative odds Variable (95% confidence interval)

Age I .o (0.98; I .02) Gender** 2.76(1.48; 5.14) Zone

Zone I vs V/J 0.65 (0.2; 2.13) Zone 2 vs V/J 0.93 (0.6; 1.45)

Employment status** Other vs full-time* 2.19 (1.1; 4.39) Homemaker vs full-time’ 2.45 (1.z; 5.0) Part-time vs loll-time* 1.91 (1.0; 3.64)

Not satisfied w/social activities** 2.57 (1.29; 5.1) Low perceived involvement in local community events* 1.67 (1.08; 2.57) Had a serious illness in past 12 months*** 6.22 (3.13; 12.4) Had Baaacial pmblems in household in past 12 months** 2.13 (1.29; 3.52) Gender x been worried over someone close** 3.2 (1.35; 7.63)

% Correct: 77% Specificity: 79% Sensitivity 73% p+ 0.19

**+p < 0.001; **p < 0.01; ‘P < 0.05.

Psychosocial effects of PCB contamination and remediation

Table 6. Percentage of reswndents reoortinr! concern

1101

Type of concern

Concern re: Smithville PCB Site’ Health concern re: Smithville PCB site

SMVL Z.1 SMVL 2.2 SMVL Total (N = 20) (N = 252) (N = 272)

75% 49% 51% 50% 37% 38%

‘SMVL Z.1 vs SMVL 2.2: I’ < 0.05.

the individual, social network, and individual stressor blocks, with a characteristic of the individual and an individual stressor comprising the only significant interaction effect (Fig. 2). It is important to note that zone did not make a significant contribution to the model for the SCL-90 somatic sub-scale. Based on significant single effects in the model, survey respondents were more likely to score above the cutpoint for normal if: they were female; their employment status was ‘other’ as opposed to ‘employed full-time’; their employment status was ‘homemaker’ as opposed to ‘employed full-time’; they were not satisfied with their social activities; they perceived themselves to be ‘not too’ or ‘not at all involved’ in local community events; in the previous 12 months they had had a serious illness; or they had experienced financial problems in their household. In addition, the lone significant interaction effect indicated that respondents were more likely to score above the normalized cutpoint if they were male and had been worried about someone close in the 12 months prior to the survey administration. Age and zone were not significant.

There are three important observations to make about the results for the general measures of psychosocial health and well-being. The first pertains to the frequency of above normal scores, the second to the relative unimportance of respondent’s location (zone), and the third to the nature of the significant explanatory variables.

The descriptive results from the GHQ-20 and the somatic complaints sub-scale of the SCL-90 show no obvious difference between the Smithville and Vineland/Jordan samples, nor is there evidence of elevated scores amongst Smithville respondents closest to the site, such as we might expect if the Smithville population was experiencing distress associated with the PCB site remediation in their community. Furthermore, the frequency of above normal scores is low when compared with normal populations surveyed in other studies [25].

The relative unimportance of respondent’s location (zone) is confirmed by the results of the logistic regression analyses for the GHQ-20 and the somatic complaints sub-scale of the SCL-90. This measure of site exposure did not make a significant contribution to either of the models, further suggesting that there are no detectable psychosocial effects in the Smithville population associated with the PCB site.

Finally, it is important to note the explanatory variables that were significant in the models. Three variables from the individual block of the analytical model (Fig. 2), employment status, age and gender

made significant contributions. Social network variables appeared in both models, with respondent’s dissatisfaction with their social activities increasing the likelihood of a score above the cutpoint for normal on both outcome variables. Individual stressor variables appeared frequently in the models, and three such stressors made significant contributions to both models. Overall, the consistency of the models, particularly the number and nature of the significant variables suggests that scoring above the cutpoint for the GHQ-20 and the SCL-90 somatic complaints sub-scale is not random: it can be explained by plausibly linked independent variables, but is not associated with exposure to the PCB site.

Site - and situation -speciJic measures of psychosocial effects

In this section, results of site- and situation-specific outcome measures are described. The two outcome variables were concern and health-related concern about the Smithville PCB site.

Concern was reported by a very large percentage of Zone 1 respondents (75%), and approximately half of Zone 2 respondents (Table 6). There was a distance gradient of concern, and the difference was statistically significant. Overall, 51% of Smithville respondents reported concern about the PCB site.

Respondents who reported that they had concerns related to the site were asked to specify the nature of their major concern. The responses to this open- ended question were coded and grouped for descriptive purposes (Table 7). Of the 125 respondents who indicated the nature of their concerns, 28% expressed concerns about possible health effects of site exposure, 20% reported concerns about water quality, 13% were fearful that the site clean-up would be ineffective, 9% were concerned about site operation and delays, 8% were concerned about air pollution from stack emissions, and 6% reported fears that the site would be made a permanent hazardous waste treatment facility. Health concerns were very preva- lent, and in fact they may be underestimated by the way the open-ended responses were coded, since concerns about water pollution and air pollution are ostensibly health concerns.

Table 7. Maior site concern: Smithville PCB site

Rank Maior concern (N = 125)

I Health concern (28%) 2 Water pollution (20%) 3 Fear clean-up will be ineffective (13%) 4 Site operation/delays (9%) 5 Air pollution (8%) 6 Fear incinerator will be permanent (6%)

1102 JAMES R. DUNN el al.

Table 8. Results of logistic regression for outcome: concern about PCB site: Smithville only

Relative odds Variable (95% confidence interval)

Age 0.98 (0.95; I .02) Gender 1.29(0.77; 2.17) Zone 2.21 (0.71; 6.88) Concern about OWMC** 2.99 (1.76; 5.07)

% Correct: 65% Specificity: 64% Sensitivity: 66% p=: 0.10

*+*p < 0.001; **p < 0.01; ‘P < 0.05.

The model for concern about the PCB site (Table 8) had a p * of 0.10. The positive predictive value (66%) and negative predictive value (64%) were low. Specificity (64%) and sensitivity (66%) were also low. Overall, 65% of respondents were correctly classified.

Only one significant explanatory variable (bold type in Table 8), from the environmental stressor block, entered this model (Fig. 3). It suggests that Smithville respondents were more likely to be concerned about the PCB site in their community if they were concerned about the proposed OWMC facility. There were no significant interaction effects, and age, gender and zone were not significant explanatory variables in the model.

Smithville respondents reporting concerns about the PCB site were asked if they considered their concerns to be related to health. Overall, 38% of Smithville respondents reported health-related concerns (see Table 6). There was evidence of a distance gradient, with 50% of Zone 1 respondents and 37% of Zone 2 respondents reporting health concerns, but the zone difference was not statistically significant.

The logistic regression model for health concern (Table 9) had a p 2 of 0.12. The positive predictive value was low (51%) and the negative predictive value (83%) was high. Specificity (74%) and sensitivity (65%) were not high. Overall, 71% of respondents were correctly classified by the model.

Only one significant explanatory variable (bold type in Table 9), from the environmental stressor block, entered this model (Fig. 3). Based on the lone significant single effect in the model, Smithville respondents were more likely to have concerns related to health surrounding the PCB site in their community if they were concerned about the proposed OWMC facility. Like the model for concern, there were no significant interaction effects, and nor were age, gender or zone significant explanatory variables in the model.

The results in this section show that a high percentage of Smithville respondents was concerned about the PCB contamination and remediation in their community, and that a significantly greater percentage of Zone 1 respondents than Zone 2 respondents were concerned. This evidence of a distance gradient was not confirmed by the more powerful test pro-

vided by logistic regression analyses. The only important explanatory variable in the models was concern about the proposed OWMC facility. Smithville respondents were more likely to report general and health-related concern about the PCB site if they reported concern about the proposed OWMC facility.

DISCUSSION AND CONCLUSIONS

In the analysis of these data, we have examined the prevalence and determinants of psychosocial effects of PCB contamination and remediation in Smithville, Ontario, using two different types of outcome measures. First, the prevalence of elevated scores on the General Health Questionnaire (GHQ-20) and the Symptom Check List 90 (SCL-90) somatic complaints sub-scale were used as indicators of psychosocial distress. These outcomes could not be explained by proximity to the site (or zone) as we have defined it. These results do not, however, ex- clude the possibility of psychosocial effects associated with the Smithville PCB site. There may be psychosocial effects in the Smithville population attributable to the PCB site that could not be detected by these outcome measures, or explained by the exposure measurement.

Results from site- and situation-specific outcome measures suggest that there were psychosocial effects associated with the Smithville PCB site. Descriptive results for general and health-related concern about the Smithville PCB site showed a higher prevalence of concern amongst respondents living closer to the PCB site. This evidence of a distance gradient was not confirmed by the more powerful test provided by logistic regression analyses, but this result has to be qualified due to the small sample size in Zone 1. Logistic regression models for concern and health concern regarding the PCB site showed respondents were more likely to report general and health-related concern about the PCB site if they also reported concern about the proposed OWMC facility. These results have important implications for the study of psychosocial effects of environmental contamination.

First, it follows from the findings of the site- and situation-specific outcome measures that local community context exerts an important influence on

Table 9. Results of logistic regression for outcome: health concern re: PCB site_Smithville only

Variable Relative odds

(95% confidence interval)

Age 0.99 (0.97; 1.01) Gender I .56 (0.9; 2.7) Zone 0.83 (0.30; 2.26) Concern about OWMC*** 4.29 (2.40; 7.65)

% Correct: 71% Specificity: 65% Sensitivity: 74% p2: 0.12

***P < 0.001; l *P < 0.01; l f < 0.05.


psychosocial effects of environmental contamination. Second, the results illustrate the utility of the conceptual framework [6], which suggests psychosocial effects are influenced by characteristics of the wider community system. Respondent’s experience of the proposed OWMC facility is included as a characteristic of the wider community system. Finally, this study illustrates a practical reason for using a staged research strategy, employing a combination of quan- titative and qualitative methods, as recommended by Taylor et al. 1211. The influence of the OWMC facility was identified in the qualitative component of the larger study. In the media analysis, which was part of that component, the OWMC facility was a prominent local waste issue, with parallels to the PCB contamination and incineration in Smithville [5]. The inter- play of the two waste issues in Smithville residents’ perceptions was evident from many of the depth interviews [5]. These findings influenced the selection of the comparison community (proximity to OWMC was a criteria for selection) and the structure and content of the modified version of the survey instrument.

The findings of this study have other methodological implications. In previous work, considerable em- phasis has been attributed to the importance of using appropriate measures of psychosocial effects [6,21]. Typically, general measures of emotional distress, like the GHQ-20 and the SCL-90, are used as outcomes, primarily because the exposures are hazardous and acute, and a high prevalence of emotional distress would be expected as a consequence of such an exposure. In instances where these measures were used to investigate chronic, non-hazardous exposures [I, the GHQ-20 and XL-90 produced negative results, which was also unsurprising, given the nature of the exposure. The exposure investigated in this research was potentially hazardous, but chronic given the long site history. Moreover, the research was conducted near the end of PCB incineration, which was perceived by many Smithville residents as the solution to a problem, rather than a problem in and of itself. Consequently, the negative results from the GHQ-20 and SCL-90 in this instance, are not surpris- ing either. Therefore, the balance of evidence, from these and other studies, suggests that measures like the GHQ-20 and the SCL-90 may only be appropriate for investigating acute, hazardous exposures, when a significant level of emotional distress might be expected.

The other type of outcome measure used in this and other studies were specific to the exposure site and situation under investigation. The findings of this, and other research [‘II, have shown moderate to high levels of general and health-related concern across a wide range of exposures, none of which could be classified as acute and hazardous. For some of these exposures, a relationship was shown between residential proximity to the site (a proxy measure for exposure), and incidence of general and health-

related concern about that site. There was some evidence of such an association in the Smithville results. These findings suggest that concern and health concern may be more sensitive measures of the effects of environmental exposure, but whether concern is a valid indicator of adverse psychosocial health effects (as defined), with potential implications for public health interventions is uncertain. If so, there remains the problem of a lack of certainty about how to intervene effectively to reduce the adverse effects. This may point to an important role for qualitative methods in a staged research design: in an unobtrusive way, use of such methods could give residents the opportunity to voice their concerns, and could indicate ways to alleviate them. In the context of environmental stress theory, concern and health concern are indicators of an individual’s appraisal of threat, harm or challenge, and may be precursors to more profound psychosocial effects, which in acute exposure situations (e.g. Three Mile Island), may be detected by measures such as the GHQ and the SCL-90. For other situations of less acute exposure, there is a need to develop a generalizable measure of environmental stress as a standardized indicator of psychosocial effects. Consistent with the working definition, such a measure should include items to measure aspects of distress, dysfunction and disability. This methodological development is an important next step to advance our understanding of the psychosocial effects of environmental contamination on individuals, groups and communities.

Acknowledgements-The authors are part of a research team working on psychosocial effects in populations exposed to environmental contaminants. The team consists of John Eyles, John Frank, Murray Haight, David Streiner, Norman White, Dennis Willms, Jamie Baxter and Tobin Copley. The authors acknowledge financial support from the- Social Sciences and Humanities Research Council (SSHRC) for this oroiect. Jim Dunn also sratefullv . _ acknowledges support received from a Natural Sciences and Engineering Research Council/Social Sciences and Humanities Research Council (NSERC/SSHRC) Master’s Scholarship in Science Policy. Stephen Walter holds a National Health Scientist award from the National Health Research and Development Program, Health and Welfare Canada. Finally, the authors thank the people who graciously responded to the survey.

REFERENCES

1. Baum A., Gatchel R. J. and Schaeffer M. A. Emotional, behavioral and uhvsioloaical effects of chronic stress at Three Mile Island. J.-Consult. c/in. Psycho/. 519, 565-512, 1983. Eyles J., Taylor S. M., Baxter J., Sider D. and Willms D. The social construction of risk in a rural community: response of local residents to the 1990 Hagersville (Ontario) tire fire. Risk Analysis 13, 281-290, 1993. Lechat M. F. Update: the epidemiology of health effects of disasters. Epidemiol. Rev. 12, 192-198, 1990. Edelstein M. R. Contaminated Communities: The Social and Psychological Impacts of Residential Toxic Ex- posure. Westview Press, Boulder, CO, 1988.


5. Copley T. T. How a community heals: contamination, media reporting and residents’ concerns in Smithville, Ontario. McMaster University, Unpublished Master’s Thesis, October, 1992.

6. Taylor S. M., Elliott S. J., Eyles J., Frank J., Haight M., Streiner D. L., Walter S. D., White N. and Willms D. Psychosocial impacts in populations exposed to solid waste facilities. ioc. Sci. Med. 33, 441&7, 1991.

7. Elliott S. J., Taylor S. M., Walter S. D., Steib D., Frank J. and Eyles J. Modelling psychosocial effects of exposure to solid waste facilities. Sot. Sci. Med. 7, 791-804, 1993.

8. Evans G. W. (Ed.) Environmental Stress. Cambridge University Press, Cambridge, 1982.

9. Evans G. W. and Cohen S. Environmental stress. In Handbook of Environmental Psychology, Vol. I (Edited by Stokols D. and Altman I.), DD. 571-610. Wilev. New York, 1987.

I. __ _

10. Lazarus R. and Folkman S. Stress, Appraisal and Coping. Springer, New York, 1984.

Il. Baum A., Fleming R. and Singer J. E. Understanding environmental stress: strategies for conceptual and methodological integration. In Advances in Enviromen- Cal Psychot%gy, Vol.-S (Edited by Baum A. and Singer J. E.). Erlbaum Associates. New Jersev. 1985.

12. Vyndr H. M. Invisible Trauma: Psychosocial Effects of the Invisible Environmental Contaminants. D. C. Heath, Toronto, 1988.

13. Evans G. W. and Jacobs S. V. Air pollution and human behaviour. In Environmental Stress (Edited by Evans G. W.). Cambridge University Press, Cambridge, 1982.

14. Sims J. H. and Baumann D. D. Educational programs and human response to natural hazards. Environ. Behav. 2, 165-189, 1983.

15. Pearhn L. I. and Schooler C. The structure of coping. J. Hlth sot. Behav. 19, 2-21, 1978.

16.

sented to the U.S. Nuclear Regulatory Commission. Social Impacts Research Inc., Seattle, WA, 1978.

17. Butte1 F. H. New directions in environmental sociology. A. Rev. Social. 13, 465488, 1987.

18. Goldberg D. P. The Detection of Psychiatric Illness by Questionnaire: A Technique for the Identification and Assessment of Non-Psychotic Psychiatric Illness. Oxford University Press, London, 1972.

19. Derogatis L. R., Lipman R. S. and Covi L. SCL-90: an outpatient psychiatric rating scale, preliminary report. Psychopharmac. Bull. 9, 13-28, 1973.

20. Derogatis L. R. SCL-90: Administration, Scorina and Procedures Manual for the Revised Version. Chnical Psychometric Unit. Johns Hopkins Universitv School of Medicine, Baltimore, MD, 1977.

21. Taylor S. M., Frank J., Walter S. D., Haight M., White N. F., Streiner D., Willms D. G., Birnie S. and Elliott S. J. The Psychosocial Impacts of Exposure to Environ- mental Contaminants in Ontario: A Feasibility Study. Report to the Ontario Ministry of the Environment, May, 1989.

22. Holmes T. H. and Rahe R. H. The social readjust- ment rating scale. J. Psychosom. Res. 11, 213-218, 1967.

23. McDowell I. and Newell C. Measuring Health: A Guide to Rating Scales and Questionnaires. Oxford University Press, New York, 1987.

24. Last J. M. (Ed.) A Dictionary of Epidemiology. International Epidemiological Association, New York, 1988.

Flynn C. Three Mile Island Telephone Survey: Prelimi- nary Report on Procedure and Findings. Report pre-

27. Norusis M. J. SPSS/PC+ Advanced Statistics 4.0. ^_^^ _ -. . ̂ ^^

25. Elliott S. J. Psychosocial impacts in populations exposed to solid waste facilities. McMaster University, Ph.D. Dissertation, June, 1992.

26. Wrigley N. Cateogrical Data Analysis for Geographers and Environmental Scientisfs. Longman Group, New York, 1985.

SASS, Inc., Chicago, IYYU.

psychosocial effects of pcb contamination and remediation: the case of smithville, ontario

Documents