Industrial accidents may affect groups of workers exposed in the workplace as well as the population living around the plant where the accident takes place. When pollution caused by accident occurs, the size of the affected population is likely to be orders of magnitude greater than the workforce, posing complex logistic problems. The present article focuses on these problems, and applies to agricultural accidents as well.
Reasons for quantifying health effects of an accident include:
- the need to ensure that all exposed persons have received medical attention (regardless of whether or not treatment was actually needed by each of them). Medical attention may consist of the search for and alleviation of clinically recognizable adverse consequences (if any) as well as implementation of means for preventing possible delayed effects and complications. This is obligatory when an accident occurs within a plant; then all the people working there will be known and full follow-up is feasible
- the need to identify persons deserving of compensation as victims of the accident. This implies that individuals must be characterized as to the severity of disease and the credibility of a causal association between their condition and the disaster.
- the acquisition of new knowledge on disease pathogenesis in humans
- the scientific interest of unravelling mechanisms of toxicity in humans, including those aspects which may help in reassessing, for a given exposure, doses considered to be “safe” in humans.
Characterization of Accidents in Relation to Health Consequences
Environmental accidents include a broad range of events occurring under the most diverse of circumstances. They may be first noticed or suspected because of environmental changes or because of the occurrence of disease. In both situations, the evidence (or suggestion) that “something may have gone wrong” may appear suddenly (e.g., the fire in the Sandoz storehouse in Schweizerhalle, Switzerland, in 1986; the epidemic of the condition later labelled as “toxic oil syndrome” (TOS) in Spain in 1981) or insidiously (e.g., excesses of mesothelioma following environmental—non-occupational—exposure to asbestos in Wittenoom, Australia). In all circumstances, at any given moment, uncertainty and ignorance surround both key questions: “Which health consequences have occurred so far?” and “What can be predicted to occur?”
In assessing the impact of an accident on human health, three types of determinants may interact:
- the agent(s) being released, its hazardous properties and the risk created by its release
- the individual disaster experience
- the response measures (Bertazzi 1991).
The nature and quantity of the release might be difficult to determine, as well as the ability of the material to enter into the different compartments of the human environment, such as the food chain and water supply. Twenty years after the accident, the amount of 2,3,7,8-TCDD released in Seveso on July 10, 1976, remains a matter of dispute. In addition, with the limited knowledge about the toxicity of this compound, in the early days after the accident, any prediction of risk was necessarily questionable.
Individual disaster experience consists of fear, anxiety and distress (Ursano, McCaughey and Fullerton 1994) consequent to the accident, irrespective of the nature of the hazard and of the actual risk. This aspect covers both conscious—not necessarily justified—behavioural changes (e.g., the marked decrease in birth rates in many Western European Countries in 1987, following the Chernobyl accident) and psychogenic conditions (e.g., symptoms of distress in school children and Israeli soldiers following the escape of hydrogen sulphide from a faulty latrine in a school on the West Bank of Jordan in 1981). Attitudes towards the accident are also influenced by subjective factors: in Love Canal, for instance, young parents with little experience of contact with chemicals in the workplace were more prone to evacuate the area than were older people with grown-up children.
Finally, an accident may have an indirect impact on the health of those exposed, either creating additional hazards (e.g., distress associated with evacuation) or, paradoxically, leading to circumstances with some potential for benefit (such as people who stop smoking tobacco as a consequence of contact with the milieu of health workers).
Measuring the Impact of an Accident
There is no doubt that each accident requires an assessment of its measurable or potential consequences on the exposed human population (and animals, domestic and/or wild), and periodic updates of such assessment may be required. In fact, many factors influence the detail, extent and nature of the data which can be collected for such an assessment. The amount of available resources is critical. Accidents of the same severity may be granted different levels of attention in different countries, in relation to the ability to divert resources from other health and social issues. International cooperation may partly mitigate this discrepancy: in fact, it is limited to episodes which are particularly dramatic and/or present unusual scientific interest.
The overall impact of an accident upon health ranges from negligible to severe. Severity depends on the nature of the conditions which are produced by the accident (which may include death), on the size of the exposed population, and on the proportion that develop disease. Negligible effects are more difficult to demonstrate epidemiologically.
Sources of data to be used for evaluating health consequences of an accident include in the first place current statistics which exist already (attention to their potential use should always precede any suggestion of creating new population databases). Additional information can be derived from analytical, hypothesis-centred epidemiological studies for the purpose of which current statistics may or may not be useful. If in an occupational setting no health surveillance of the workers is present, the accident can provide the opportunity to establish a surveillance system which will eventually help to protect workers from other potential health hazards.
For the purposes of clinical surveillance (short or long term) and/or provision of compensation, the exhaustive enumeration of the exposed persons is a sine qua non. This is relatively simple in the case of intra-factory accidents. When the affected population can be defined by the place where they live, the list of residents in administrative municipalities (or smaller units, when available) provides a reasonable approach. The construction of a roster may be more problematic under other circumstances, particularly when the need is for a list of people showing symptoms possibly attributable to the accident. In the TOS episode in Spain, the roster of persons to be included in the long-term clinical follow-up was derived from the list of the 20,000 persons applying for financial compensation, subsequently corrected through a revision of the clinical records. Given the publicity of the episode, it is believed that this roster is reasonably complete.
A second requirement is that activities aiming at the measure of the impact of an accident be rational, clear-cut and easy to explain to the affected population. Latency may range between days and years. If some conditions are met, the nature of disease and probability of occurrence can be hypothesized a priori with a precision sufficient for the adequate design of a clinical surveillance programme and ad hoc studies aiming at one or more of the goals mentioned at the beginning of this article. These conditions include the rapid identification of the agent released by the accident, availability of adequate knowledge on its short- and long-term hazardous properties, a quantification of the release, and some information on inter-individual variation in susceptibility to the agent’s effects. In fact, these conditions are rarely met; a consequence of the underlying uncertainty and ignorance is that the pressure of public opinion and the media for prevention or definite medical intervention of doubtful usefulness is more difficult to resist.
Finally, as soon as possible after the occurrence of an accident has been established, a multidisciplinary team (including clinicians, chemists, industrial hygienists, epidemiologists, human and experimental toxicologists) needs to be established, which will be responsible to the political authority and the public. In the selection of experts, it must be borne in mind that the range of chemicals and technology which may underlie an accident is very large, so that different types of toxicity involving a variety of biochemical and physiological systems may result.
Measuring the Impact of Accidents through Current Statistics
Current health status indicators (such as mortality, natality, hospital admissions, sickness absence from work and physician visits) have the potential to provide early insight on the consequences of an accident, provided they are stratifiable for the affected region, which often will not be possible because affected areas can be small and not necessarily overlapping with administrative units. Statistical associations between the accident and an excess of early events (occurring within days or weeks) detected through existing health status indicators are likely to be causal, but do not necessarily reflect toxicity (e.g., an excess of physician visits may be caused by fear rather than by actual occurrence of disease). As always, care must be exercised when interpreting any change in health status indicators.
Although not all accidents produce death, mortality is an easily quantifiable endpoint, either by direct count (e.g., Bhopal) or through comparisons between observed and expected number of events (e.g., acute episodes of air pollution in urban areas). Ascertaining that an accident has not been associated with an early excess of mortality may help in assessing the severity of its impact and in addressing attention to non-lethal consequences. Further, the statistics needed in order to calculate expected numbers of deaths are available in most countries and allow for estimates in areas as small as those which are usually affected by an accident. Assessing mortality from specific conditions is more problematic, because of possible bias in certifying causes of death by health officers who are aware of the diseases expected to increase after the accident (diagnostic suspicion bias).
From the foregoing, the interpretation of health status indicators based on existing data sources requires a careful design of ad hoc analyses, including a detailed consideration of possible confounding factors.
On occasions, early after an accident, the question is posed whether the creation of a conventional population-based cancer registry or a registry of malformations is warranted. For these specific conditions, such registries may provide more reliable information than other current statistics (such as mortality or hospital admissions), particularly if newly created registries are run according to internationally acceptable standards. Nevertheless, their implementation requires the diversion of resources. In addition, if a population-based registry of malformations is established de novo after an accident, probably within nine months it will hardly be capable of producing data comparable to those produced by other registries and a series of inferential problems (particularly statistical error of the second type) will ensue. In the end, the decision largely relies on the evidence of carcinogenicity, embryotoxicity or teratogenicity of the hazard(s) which have been released, and on possible alternative uses of the available resources.
Ad Hoc Epidemiological Studies
Even in areas covered by the most accurate systems for monitoring the reasons for patients’ contacts with physicians and/or hospital admissions, indicators from these areas will not provide all the information needed in order to assess the health impact of an accident and the adequacy of the medical response to it. There are specific conditions or markers of individual response which either do not require contact with the medical establishment or do not correspond to the disease classifications conventionally used in current statistics (so that their occurrence would hardly be identifiable). There may be the need for counting as “victims” of the accident, subjects whose conditions are borderline between occurrence and non-occurrence of disease. It is often necessary to investigate (and evaluate the efficacy of) the range of therapeutical protocols which are used. The problems noted here are but a sampling and do not cover all those which might create the need for an ad hoc investigation. In any case, procedures should be established in order to receive additional complaints.
Investigations differ from the provision of care in that they are not directly related to the individual’s interest as a victim of the accident. An ad hoc investigation should be shaped in order to fulfil its purposes—to provide reliable information and/or demonstrate or disprove a hypothesis. Sampling may be reasonable for research purposes (if accepted by the affected population), but not in the provision of medical care. For instance, in the case of a spill of an agent suspected of damaging bone marrow, there are two totally different scenarios in order to respond to each of the two questions: (1) whether the chemical actually induces leukopenia, and (2) whether all exposed persons have been exhaustively screened for leukopenia. In an occupational setting both questions can be pursued. In a population, the decision also will depend on the possibilities for constructive intervention to treat those affected.
In principle, there is a need to have sufficient epidemiological skill locally to contribute to the decision on whether ad hoc studies ought to be carried out, to design them and to supervise their conduct. However, health authorities, media and/or the population may not consider the epidemiologists of the affected area to be neutral; thus, help from outside may be needed, even at a very early stage. The same epidemiologists should contribute to the interpretation of descriptive data based on the currently available statistics, and to the development of causal hypotheses when needed. If epidemiologists are not available locally, collaboration with other institutions (usually, National Institutes of Health, or WHO) is necessary. Episodes which are unravelled because of the lack of epidemiological skill are regrettable.
If an epidemiological study is believed to be necessary, however, attention should be addressed to some preliminary questions: To what use will predictable results be put? Might the desire for a more refined inference resulting from the planned study unduly delay clean-up procedures or other preventive measures? Must the proposed research programme first be fully documented and evaluated by the multidisciplinary scientific team (and perhaps by other epidemiologists)? Will there be adequate provision of details to the persons to be studied to ensure their fully informed, prior and voluntary consent? If a health effect is found, what treatment is available and how will it be delivered?
Finally, conventional prospective cohort mortality studies ought to be implemented when the accident has been severe and there are reasons to fear later consequences. Feasibility of these studies differs between countries. In Europe, they range between the possibility of nominal “flagging” of persons (e.g., rural populations in Shetland, UK, following the Braer Oil Spill) and the need for systematic contacts with the victims’ families in order to identify dying persons (e.g., TOS in Spain).
Screening for Prevalent Conditions
Offering affected people medical attention is a natural reaction to an accident which may have caused them harm. The attempt to identify all those in the exposed population who exhibit conditions related to the accident (and give them medical care if needed) corresponds to the conventional concept of screening. Basic principles, potentialities and limitations common to any screening programme (regardless of the population to which it is addressed, the condition to be identified and the tool used as a diagnostic test) are as valid after an environmental accident as in any other circumstance (Morrison 1985).
Estimating participation and understanding reasons for non-response are just as crucial as measuring sensitivity, specificity and predictive value of the diagnostic test(s), designing a protocol for subsequent diagnostic procedures (when needed) and the administration of therapy (if required). If these principles are neglected, short- and/or long-term screening programmes may produce more harm than benefit. Unnecessary medical examinations or laboratory analyses are a waste of resources and a diversion from providing necessary care to the population as a whole. Procedures for ensuring a high level of compliance have to be carefully planned and evaluated.
Emotional reactions and uncertainties surrounding environmental accidents may further complicate things: physicians tend to loose specificity when diagnosing borderline conditions, and some “victims” may consider themselves entitled to receive medical treatment regardless of whether or not it is actually needed or even useful. In spite of the chaos which often follows an environmental accident, some sine qua non for any screening programme should be borne in mind:
- Procedures should be laid down in a written protocol (including second level diagnostic tests and therapy to be provided to those who are found to be affected or sick).
- One person should be identified as responsible for the programme.
- There should be a preliminary estimate of specificity and sensitivity of the diagnostic test.
- There should be coordination between clinicians participating in the programme.
- Participation rates should be quantified and reviewed at regular intervals.
Some a priori estimates of efficacy of the whole programme would also help in deciding whether or not the programme is worth implementing (e.g., no programme for anticipating the diagnosis of a lung cancer should be encouraged). Also, a procedure should be established in order to recognize additional complaints.
At any stage, screening procedures may have a value of a different type—to estimate the prevalence of conditions, as a basis for an assessment of the consequences of the accident. A major source of bias in these estimates (which becomes more severe with time) is the representativeness of the exposed persons submitting themselves to the diagnostic procedures. Another problem is the identification of adequate control groups for comparing the prevalence estimates which are obtained. Controls drawn from the population may suffer from as much selection bias as the exposed person’s sample. Nevertheless, under some circumstances, prevalence studies are of the utmost importance (particularly when the natural history of the disease is not known, such as in TOS), and control groups external to the study, including those assembled elsewhere for other purposes, may be used when the problem is important and/or serious.
Use of Biological Materials for Epidemiological Purposes
For descriptive purposes, the collection of biological materials (urine, blood, tissues) from members of the exposed population can provide markers of internal dose, which by definition are more precise than (but do not replace totally) those obtainable through estimates of the concentration of the pollutant in the relevant compartments of the environment and/or through individual questionnaires. Any evaluation ought to take into account possible bias ensuing from the lack of representativeness of those members of the community from whom the biological samples were obtained.
Storing biological samples may prove useful, at a later stage, for the purpose of ad hoc epidemiological studies requiring estimates of internal dose (or early effects) at the individual level. Collecting (and properly preserving) the biological samples early after the accident is crucial, and this practice should be encouraged even in the absence of precise hypotheses for their use. The informed consent process must ensure that the patient understands that his or her biological material is to be stored for use in tests hitherto undefined. Here it is helpful to exclude the use of such specimens from certain tests (e.g., identification of personality disorders) to better protect the patient.
The rationale for medical intervention and epidemiological studies in the population affected by an accident ranges between two extremes—assessing the impact of agents which are proved to be potential hazards and to which the affected population is (or has been) definitely exposed, and exploring the possible effects of agents hypothesized to be potentially hazardous and suspected to be present in the area. Differences between experts (and between people in general) in their perception of the relevance of a problem are inherent to humanity. What matters is that any decision has a recorded rationale and a transparent plan of action, and is supported by the affected community.