Thank you Michael,
I also think there is a lot to be said too about the results of
adjusting for potential confounders. Obviously I could elaborate but two
issues are important.
1. Adjustment for potential confounders should be as thorough as
possible. e.g ever/never smoked does not adjust for smoking history for
most diseases - some but by no means most. Therefore smoking remains a
potent potential confounder after adjustment. So look for inadequate
adjustments before believing causation.
2. When adjustment is as through as is feasible for all known
confounders are properly measured and there remains a residual
significant effect then interpretation may lean on by how much the
estimate changes before and after adjustment. If by only a small amount
then the association looks causal. Clearly this is not fool proof
because orthogonal unknown confounders could still explain the effect.
But such things are probably quite rare on current knowledge. Certainly
most things correlate a little - certainly known and unknown risk
factors might. But really if an unknown uncorrelated risk factor exists,
or is likely to exist, the burden then shifts somewhat to suggesting a
plausible hypothesis for what it might be. That can't be a bad thing
either.
All best and thanks again
Klim
>>> Michael Stuart <[log in to unmask]> 18/03/2009 21:35 >>>
Dear Colleagues:
We would like to receive feedback from list serve readers on criteria
we have devised for assessing causality in non-experimental studies.
Like many others, we assess efficacy of interventions by starting with
RCTs, first evaluating studies for validity; then, if a study is rated
as valid, we examine the reported results for clinically usefulness.
We have formulated a set of primary criteria that may be useful in
identifying situations in which, without experimental design, there is
likely to be a true cause and effect relationship between the exposure
or intervention and an observed outcome. We hypothesize that several
major criteria may be the most important considerations in determining
efficacy when valid RCTs are not available. We believe that these
criteria represent what appears to be a conservative approach, and we do
not believe they necessarily exclude other possibilities. We are
considering submitting a paper to this effect, but would like to get a
reaction from readers of this list serve (we are not including
references here) to see if the criteria seem to hold.
CriteriaTheorized as Possibly Reliable Indicators of Efficacy
(Uncertain Strength of Evidence)
Criteria 1.
1a) There is extremely low likelihood of improvement without some
intervention; and,
1b) Any effects are highly likely to be attributable to the
intervention (e.g., a single intervention or technology was utilized and
the likelihood of patients utilizing co-interventions is low or
co-interventions were equivalent in compared groups [note: equivalence
should take into account considerations such as administration,
duration, dosing, etc.]); and,
1c) Convincing sustained improvement is documented following use of the
intervention or technology.
OR
Criteria 2.
2a) Illness or condition has a predictable course or outcome (e.g.,
symptoms, diagnostic signs, progression or resolution of the condition
are predictable); and,
2b) Any effects in this population are likely to be attributable to the
intervention (e.g., plausible confounders have been considered, a single
intervention or technology was utilized and the likelihood of patients
utilizing co-interventions is low or co-interventions were equivalent in
groups being compared [note: equivalence should take into account
considerations such as administration, duration, dosing, etc.]); and,
2c) Convincing sustained improvement in this particular population is
documented following use of the intervention or technology with results
demonstrably and significantly earlier than what i
s predicted based on
reliable knowledge of natural history.
Secondary Considerations: Our Thinking
And then in postulating what might be reasonable secondary
considerations for determining efficacy in observational studies we came
up with the following:
Considerations Lending Support to the Evidence
Factoring in the following conditions may decrease uncertainty:
a. Immediacy of effectwhich may increase confidence in the
effectiveness of the intervention or technology not only because of
temporal proximity, but also because a delay in effect provides more
opportunity for application of a co-intervention which could be a
confounder. When considering immediacy, caution is urged because the
effect could be due to potential of confounding.
b. Dramatic effects(e.g., around 80% or greater relative risk
reduction). When considering dramatic effects, caution is urged because
the effects could be due to the strength of a confounder. For example,
The Cochrane Handbook states that observational evidence with large
effect sizes where a dose-response gradient exists and all plausible
confounding would most likely demonstrate a reduced effect size, may be
of moderate or even high quality (Higgins JPT, Green S (2008) Cochrane
Handbook for Systematic Reviews of Interventions Version 5.0.0 [updated
February 2008]. Available from www.cochrane-handbook.org (
http://www.cochrane-handbook.org/ ).accessed 08/07/2008. Section
12.2.3). However, we remain cautious about this because confounding
may result in large differences between groups.
c. Dose/response relationship. When considering dose/response
relationship, caution is urged because patient expectation bias may
result in increased perception of benefit or increased placebo effect if
patient is aware of the greater dosing through increase in side effects,
for example.
d. Consistent changes in multiple outcomes. When considering
consistent change, caution is urged because the change in multiple
outcomes may be due to confounding (e.g., a causal association between
an unknown confounder and each of the outcomes).
Again, we are most appreciative of peopleâs thoughts.
Michael E Stuart MD
President & Medical Director, Delfini Group
Clinical Asst Professor, UW School of Medicine
6831 31st Ave N.E.
Seattle, Washington 98115
206-854-3680 Mobile Phone
206-527-6146 Home Office
[log in to unmask]
www.delfini.org
Klim McPherson
Visiting Professor of Public Health Epidemiology
Fellow of New College
Nuffield Dept of Obstetrics & Gynaecology
Oxford University
Womens Centre, Level3
John Radcliffe Hospital
Headington
Oxford OX3 9DU
Tel: 01865 222937
Mobile: 07711335993
Home 01865 558743
[log in to unmask]
[log in to unmask]
