Thank you for your patience in attempting to understand the perspectives I
am trying to address. These are difficult yet important concepts which is
one reason why I believe discussing it one the mailbase is important as
opposed to private email correspondence. I am perhaps fortunate in being
able to access the considerable experience of many experts here at Cambridge
such as Professor's Nick Day (WHO Cancer trials) and KT Khaw (who was
schooled by the late Geoffrey Rose in the study of population based
cardiovascular disease) and the many distinguished statisticians from the
MRC Biostatistical Unit. I would hope that debating in this way will assist
other physiotherapists on this list to become more familiar with the
concepts supporting their treatment; an area that I am sure most will agree
is not their particular strength.
Is this important? Well hopefully it will achieve the realisation that such
concepts are integral to the clinical decision making process and cannot be
divorced from it. These concepts are not 'new' (being found in any
intermediate level medical statistical reference) however unfortunately are
not routinely embraced in physiotherapy.
Several important issues require clarification:
1] Statistical interpretation is central to the clinical decision making
process and so every clinician uses (or abuses?) statistics regardless or
not of whether they realise it. Therefore to be involved in statistics you
do not need to have completed a piece of research which you wish to analyse!
Perhaps the most common application of statistics in a clinical decision
making process is the use of theoretical knowledge and/or clinical
experience to diagnose 'y' when faced with a patient with 'x' symptoms. In
simple terms a frequentist statistician would say that the therapist would
use their knowledge and/or experience to reason that as a given number of
'y' patients have presented with 'x' symptoms previously this is the
probability that the diagnosis is correct. A bayesian statistician would
argue that the therapist would use subjective reasoning to estimate the
probability given 'x' symptoms that the patient has 'y' problem.
2] Statistical probability modelling is critical in clinical care as if the
wrong diagnosis is made, or a test result is incorrect, inappropriate
treatment will be given. This was discussed in my previous email and is
calculated by sensitivity, specificity and the positive and negative
predictive values. Whilst many clinicians do not realise it they go through
this process, at least in part, every time they make a decision. When it
goes wrong (i.e. sensitivity is low) in 'high risk' treatments (i.e.
surgery) or in cases where error is made (the numerous cases of breast
screening errors reported in the last few years) it makes headline news. The
fact is that 100% sensitivity and specificity are never achieved. I would
hope that in considering ethical principles most physiotherapists would aim
to give the correct diagnosis and/or treatment to their patients (i.e.
maximising sensitivity and specificity using clinical knowledge and
experience). Just because clinicians do not realise they are following this
process does not mean they are not!
3] The issue therefore is not so much does a particular phenomenon 'exist'
(or is a myth) but how close do we as clinicians come to the 'truth' when we
identify, assess, measure or diagnose it. To assume (or even believe) that
measurement and assessment is 100% accurate is reckless. Remember that when
you assess or even measure a 'thing' you are making an estimate. What you
and ultimately your patients are relying on is whether your estimate is
good. Confidence Intervals are invaluable in assessing this and although the
standard choice of quoting 95% limits (representing +/-1.96 standard
deviations of your estimate) is actually quite arbitrary I am sure you will
find on talking to most statisticians that they now quote Confidence
Intervals in favour of 'p' significance values. Note that this is now also
the policy of the BMJ and related journals.
[NB: This means that if you are assessing or attempting to measure
homeostasis you ARE calculating the likllihood of the events happening
(especially from a Bayesian perspective) and furthermore if you are
attempting to interpret homeostasis as a biological event you already
followed this process. I am not arguing whether something exists or not but
assessing the quality of the estimates (assessment, measurement etc.) that
we then go on and use in decision making or diagnosis]
4] I agree that precise definitions are important in all types of research.
I happened to use epidemiology as an example as this is the discipline that
uses the most rigorous process of definition (termed case ascertainment)
using a set of standards developed by Sir Austin Bradford Hill. Consequently
the criteria of temporality, strength, dose-response, reversibility,
consistency, biologic plausibility, specificity and analogy are used to
explore the concept of cause and effect associated with the particular
defined phenomenon. I concede that the Britannica definition of homeostasis
is useful for the lay person however strongly disagree that this definition
contains objective measures that may be investigated and tested against a
hypothesis. Thus whilst the Britannica definition is not incorrect, it would
have be rejected as a detailed objective quantifiable statement that could
be tested by a scientific investigation (i.e. it needs a lot more detail!)
5] Finally, with reference to the blood pressure example you concluded your
email by stating "however we unfortunately produce diagnoses using not only
mmHg numbers like 159 or 161." Ultimately it does not matter how many sets
of measurements we take as the conventional clinical process reduces down to
the dichotomous system I mentioned in my first email and where this debate
appeared to be going astray. That is although we as clinicians conceptualise
the biological phenomena we have assessed or measured down to two options
(patient has a assessed or measured problem so treat 'versus' patient
doesn't have a problem so don't treat) the fundamental problem with this
approach is that it depends on where we place the thresholds for 'disease'.
Thus appreciation that biological phenomena exist as a continuum is crucial
if we are to appreciate the needs of the patient whose systolic BP is
…..157,8 or 9 and 161, 2 and 3…… if the threshold for treatment is a BP of
160. Confidence Intervals are again useful in assessing the need for
intervention in borderline cases such as this in combination with
likelihood, relative and attributible risk which rely on estimation of the
normal population distribution referred to in my previous emails.
NB:- I have not included references as I fear this is already a lengthy
email however I will supply on request.
Alistair Grant
Institute of Public Health
University of Cambridge
>From: "Stanislav A. Korobov" <[log in to unmask]>
>Reply-To: PHYSIO - for physiotherapists in education and practice
> <[log in to unmask]>
>To: [log in to unmask]
>Subject: Re: The Myth of Homeostasis (for Alistair Grant)
>Date: Mon, 15 Jan 2001 22:35:21 +0000
>
>It seems to me (or: I hope) that I have been gradually commencing to
>understand your position. Probably you are currently looking at the
>phenomenon of discussion from mostly statistical point of view joined, as
>that often happens, with the theory of chances (hereof: "the classic bell
>shape distribution", "risk of an event happening", "tossing a coin", etc).
>If so, I was and am quite far away such a look at homeostasis. I take a
>little interest in calculating a likelihood of achieving the homeostasis in
>a given clinical situation or in the mankind population. I am rather
>interested to know whether such a biological phenomenon really exists (is
>not a "myth"). Of course, it would be interesting to know a "quantity of
>homeostasis" in a patient which is sitting near my table ("where is
>he/she -- in the middle of bell curve or somewhere around the tails?"). If
>this is (or: will be) possible, I'll find how to use such data in my
>practice or scientific meditations. However, since we have not yet similar
>quantitative descriptions as to a given organism, it is much for me to know
>that such phenomenon exists (by the way, I think this existance was proved
>by means of statistical methods among other ones).
>
>Regarding the word "relative". I used it exclusively in sense "not full",
>"not absolute". I.e. I did mean an extent of achieving the homeostasis and
>did not mean this word in its statistical sense (relativity of events).
>
>The importance of a precise definition of the phenomena under investigation
>is crucial to the quality of the research process not only in
>epidemiological research. It is related to any research. As to the
>Encyclopaedia Britannica's definition of homeostasis and other similar
>general definitions, they are very important as from at least two points of
>view: (1) they reflect a result of thousands of observations and
>conclusions
>(by the way -- using statistical approaches too); (2) they are milestones
>of
>scientific development, they attract investigators' attention to new
>subjects and issues, they are ski-jumps of future discoveries (sorry for
>this pathetics; I definitely dislike pathetics but the the item is too
>general and important simultaneously).
>
>And why you deem that the Britannica's definition is "not objective"?
>
>Your point about continuum as a form of existing biological processes is
>greatly essential, I think. This mathematical notion seems to be helping to
>understand and explain any phenomenon taking into account that the object
>has a non-discrete character with regards to its time and space variables.
>
>As to your blood pressure example, you are obviously right. However we
>fortunately produce diagnoses using not only mmHg numbers like 159 or 161.
>
>Stanislav A. Korobov, MD, PhD
>Physician-Physiotherapist
>P.O.Box 7, Odessa, 65089, Ukraine
>[log in to unmask]
_________________________________________________________________________
Get Your Private, Free E-mail from MSN Hotmail at http://www.hotmail.com.
