Confianza Incertidunbre
Confianza Incertidunbre
Confianza Incertidunbre
Paper
WHEC-08 (2019)
WHEC Report
Confidence & Uncertainty
WHEC
Workplace Health
Expert Comittee
Foreword
The development of policy in HSE needs to be informed by the best available contemporary
scientific evidence. In 2015, HSE formed the Workplace Health Expert Committee (WHEC) to
provide independent expert advice to them on:
■ The quality and relevance of the evidence base on workplace health issues
Questions about workplace health issues come to WHEC from many sources, which include HSE,
trade unions, employers, interested individuals and members of WHEC. WHEC’s responses to
these questions are published online as reports to HSE, as position papers following investigation,
or as a briefer response where the current evidence is insufficient to warrant further investigation.
In cases where the evidence-base is limited WHEC will maintain a watching brief and undertake
further investigation if new and sufficient evidence emerges.
In its formal considerations, WHEC aims to provide answers to the questions asked based on
the available evidence. This will generally include review of the relevant scientific literature,
identifying the sources of evidence relied on in coming to its conclusions, and the quality and
limitations of these sources of evidence.
The purpose of WHEC reports is to analyse the relevant evidence to provide HSE with an informed
opinion on which to base policy. Where there are gaps in the evidence, which mean that this is
not possible, WHEC will identify these and, if appropriate, recommend how the gaps might be
filled.
Confidence and Uncertainty while duration, timing of exposures at work and interval
from first exposure can often be accurately assessed,
The primary role of WHEC is, based on the best
estimation of the level, or intensity, of exposure may in
contemporary scientific evidence, to provide advice to
many circumstances be informed guesswork, making
HSE on new and emerging health risks and on new and
accurate estimation of the relevant exposure difficult,
emerging evidence on recognised health risks to the
while information on potential confounding factors may
UK workforce. Two measures of risk are of particular
be absent or limited. All of these factors can serve to
relevance in this context: attributable risk – the additional
threaten the internal validity of a study and undermine
risk an individual incurs as a consequence of an exposure
confidence in the existence of an association, in
at work; and population attributable risk (PAR) – the
estimates of its magnitude or its causal relevance.
burden of illness or disease in the population that is a
consequence of the exposure. PAR will depend on both
Because of the many potential problems, the findings
the attributable risk and the prevalence of exposure in the
of a single observational study alone can only very
population.
rarely be considered sufficient to provide evidence of
cause and effect. Normally, it is necessary to consider
While the regulator’s policy is informed by such evidence,
a body of evidence drawn from all available studies,
it will be one of a number of considerations which
epidemiological, toxicological experiments and other
policy makers will take into account in decision-making.
forms of non-epidemiological research; if these are few,
Other factors may include the affordability and risks of
only cautious inferences can follow.
any proposed intervention in relation to the predicted
benefits, its deliverability, the values of policy makers and
Not uncommonly studies do not agree, sometimes
stakeholders and other constraints.
reaching opposite conclusions. More frequently they
disagree about the size and importance of effects and
The role of WHEC is the consideration and analysis of
the exposure levels that cause them. Circumstances
evidence: what do we know: how confident can we be in
can conspire to force different definitions of exposure
the inferences made and where do the uncertainties lie?
on different investigators. In this situation, an overview
In some circumstances, WHEC can also provide evidence
is needed, focussing on whether a causal relationship
in relation to different options for intervention.
between a putative hazard and adverse outcome is likely
to exist, at least in some circumstances; how big it is
Intervention in the workplace usually requires established
likely to be and in what circumstances; and how much
evidence on occupational causation. Deciding on this
confidence can be placed in the various judgements.
requires a judgement on the relevant scientific evidence.
Beyond this are judgements for the regulator about
Studies to investigate causation in working populations
the feasibility, expediency and effectiveness of control
are usually, necessarily, observational in nature (case
measures.
referent, cross-sectional or cohort studies), which are
more vulnerable to biases and confounding than well-
Bradford Hill proposed guidelines to assess the likelihood
executed blinded randomised controlled trials. Selection
of an association being causal. Although well-recognised
into, and for remaining in, the workforce is far from
exceptions exist to many of these (other than the absolute
random and blinding of participants to their exposures at
requirement that exposure must precede and not follow
work is not possible, each, potentially, depending on the
the effect), they offer a useful guide, especially given the
study question and design, leading to biased estimates
potential disparities to be found in a body of evidence.
of risk. In addition, an appropriate comparison group in
The most important of the criteria, from the perspective of
studies of working populations can often be difficult to
interpretation of occupational epidemiology, are probably
secure, making comparison of ‘like with like’ difficult.
1) the strength of the association, in terms of relative risk
In these circumstances internal comparisons based on
or risk ratio, the ratio of the risk in the exposed to the
different levels of exposure are often used. However,
risk in the unexposed: the higher the risk ratio, the less
likely it is to be explained by chance or by unrecognised silica and lung cancer, which could easily arise through
confounding; 2) evidence of exposure-response gradients: confounding and, in the case of lung cancer and silica, be
causal inference is supported if a bigger ‘dose’ causes a explicable wholly or in part by cigarette smoking and other
bigger effect or greater likelihood of disease; 3) repeatability agents such as ionising radiation and arsenic, to which
of the findings: a consistent observation, demonstrated exposure can occur in the workplace.
in different settings, different populations, different eras
by different study designs, is less likely to arise through In these circumstances, replication of the findings, with
bias, confounding or by chance; 4) biological plausibility: different studies generating similar risk estimates and
evidence of a plausible underlying biological mechanism consistent evidence of exposure-response relationships,
for the particular agent causing the disease: this is can provide important supporting evidence of causation.
inevitably limited by contemporary knowledge of disease Where undertaken, evidence for the effectiveness of an
mechanisms; and 5) the effectiveness of intervention: does intervention to reduce the level of exposure in reducing the
removal of the hazardous exposure makes disease less increased risk of disease, can provide the most powerful
likely? evidence of causation. It is important that replication of
findings comes from studies of high quality undertaken in
In a few circumstances the strength of an association can different populations in different circumstances of exposure.
be sufficiently great (say a risk ratio of more than 10) to Publication bias (the tendency for striking findings to be
have confidence that, (unless very strongly associated with published more readily than neutral, negative, or non-
another risk factor), it is probably causal. An example of significant ones), for instance, can ensure consistency
this would be the description by Percival Potts of scrotal of findings, often from small studies which are wrong or
cancer in boys climbing chimneys, in whom Richard Doll which exaggerate the risks. High quality studies often
estimated the risk ratio to be 200. Nonetheless, even here generate lower estimates of relative risk than low quality
confidence in causation is strengthened by the finding studies, as is illustrated by estimates of the association of
of an increased risk of scrotal cancer in those in direct residential exposure to electromagnetic fields and childhood
contact in their work with mineral oils also containing leukaemia (Fig 1. From Higgins et al, 2003). Rothman
polycyclic aromatic hydrocarbons. A risk ratio in excess of and Poole (1988) have suggested that weak associations
10 makes confounding by an unrecognised independent in epidemiological studies might be strengthened by a)
risk factor unlikely. Risk ratios of this level are found in restricting attention to populations otherwise at low risk and
relationships such as lung cancer with cigarette smoking, b) reducing non-differential misclassification of exposure
and with asbestos exposure in those exposed before the and of disease which tend to dilute associations.
1931 Regulations in the UK, and the risk of mesothelioma
in carpenters who worked in the construction industry
in the UK in the 1960’s and 1970’s. Another example
is adenocarcinoma of the nose or para-nasal sinuses in
workers heavily exposed to hardwood dust: relative risks in
the hundreds have sometimes been reported, with many
studies reporting values above 10.
A recent review by WHEC of the relationship between in 2001, of a pooled exposure-response analysis of 10
inhaled respirable crystalline silica (RCS) and lung cohorts of silica exposed populations of some 44,000
cancer identified several well-designed cohort studies, miners and some 22,000 non-miners found clear
undertaken in several different circumstances of exposure evidence of an exposure-response relationship (Fig 3).
(Workplace Health Expert Committee). The great majority
found an increased risk of lung cancer of the order of Interpretation of epidemiological findings is considerably
20% to 30% (RR 1.2-1.3), where the population at risk helped by evidence of biological plausibility, demonstrated
was followed up more than 15 years from initial exposure, in human and animal toxicological studies and in vitro
recognised confounders were taken into account, to investigations. Such studies often involve the use of
the extent possible, and contemporary measurements animal species (e.g. rodents from well-characterised
of exposure were sufficient to allow categorisation of strains), exposed in varying patterns, timeframes and
exposure (Fig 2). doses that simulate workplace exposures. Toxicological
studies involving workers or volunteers, although less
common, can also inform on subtle short-term pre-clinical
endpoints, such as irritation of the upper respiratory tract
or neurotransmitter enzyme perturbations. Experimental
control over confounders and effect modifiers can be
greater than in human field studies and this simplifies
interpretation. In addition, a wide range of doses can
be used to explore dose-response and dose-effect, and
substances can be tested in animals and in vitro for
which there is, as yet, no human experience. However,
since animals often have different metabolism and
pathophysiological responses from humans, there may
Figure 2. Risk ratio for lung cancer in cohorts of workers exposed to not be perfect read-across to the human experience.
crystalline silica Collectively, however, toxicological studies can offer
understanding of the molecular and pathophysiological
changes underlying disease-exposure associations
and strengthen confidence in the existence of a causal
relationship. Hazard and risk assessment is now usually
incomplete without the use of adjunctive toxicological
information.
capture the view of a majority of experts who have properly point when expert opinion would regard a causal association
considered the matter. This should provide sufficient guide as to be “probable”; but there would be less agreement between
to general confidence in the scientific evidence and therefore experts over where to set exposure limits. It may be helpful
the scientific grounds for intervention. therefore to contextualise a descriptor of certainty/uncertainty
by reference to defined exposure levels (and when these are
The framework we propose above is different from that used unclear to make this apparent to policy makers).
by the International Agency for Cancer (IARC) for the purpose
of identifying human carcinogens. Broadly speaking, the Figure 5 represents schematically the strengths of association
IARC classifies the available data for humans and animals (risk ratios) and repeatability of findings in epidemiological
separately and then employs a system for bringing the two studies for several hazard-outcome permutations of interest.
strands of evidence together. According to IARC’s system, The very high risk ratio of 150 reported in relation to sino
human data on carcinogenicity may fall into three categories nasal cancer and work in a nickel refinery prior to the 1950s
of evidence: ‘sufficient’ (meaning very strong); ‘limited’ makes occupational causation ”almost certain”, even though
(meaning plausible and with supporting evidence, but not the number of studies on the topic, and so the repeatability of
proven - “a causal interpretation is considered credible, evidence, is limited. For lung cancer and smoking, the subject
but chance, bias or confounding cannot be ruled out with of much investigation, the risk ratio is high, on average
reasonable confidence”); or ‘inadequate (weak or absent). 10, the repeatability of findings and the case for causation
Animal data on carcinogenicity are similarly classified into high – again, almost all experts would agree that a causal
three categories as ‘sufficient’, ‘limited’ or ‘inadequate’. relationship is “almost certain”.
What is WHEC?
https://webcommunities.hse.gov.uk/connect.ti/WHEC/
view?objectId=235408&exp=c1
WHEC membership