sham surgery and other "control" processes

[Alan Ruttenberg]

An issue came up in OBI today that I'd like to solicit opinion on.
The Role branch was discussing "The Control Role" in an experiment.

For example, in an experiment which is investigating the effect of a
drug treatment on a mouse, the mice that are untreated would be
considered the "control" (bearer of the role "Control") and would be
used to compare to the mice that do receive a treatment (bearer of
the role "treated", say).

Another element in the initial list was "Role of occurent", which
wouldn't be a role in the BFO sense. However the case is interesting.
An example that is given would be a case which was studying whether
some kind of surgical adjustment improves outcome for some condition.
In this case a mouse would be subjected to a surgical procedure that
was similar, but where the adjustment was not done (the sham
surgery). One can clearly say that the mouse that undergoes the sham
surgery bears the control role, but it was thought that the important
aspect of the "control" was the surgery - an occurrent.

So the question is, what do people think about this case - and are
there recommendations about what to say about the sham surgery?

-Alan

[Pierre Grenon]

Alan, see below (I'm not sure this will make it to the obi-role-brach,
feel free to forward)

On 4/14/07, Alan Ruttenberg <alanrut...@gmail.com> wrote:
>
> An issue came up in OBI today that I'd like to solicit opinion on.
> The Role branch was discussing "The Control Role" in an experiment.
>
> For example, in an experiment which is investigating the effect of a
> drug treatment on a mouse, the mice that are untreated would be
> considered the "control" (bearer of the role "Control") and would be
> used to compare to the mice that do receive a treatment (bearer of
> the role "treated", say).

seems to makes sense

> Another element in the initial list was "Role of occurent", which
> wouldn't be a role in the BFO sense. However the case is interesting.
> An example that is given would be a case which was studying whether
> some kind of surgical adjustment improves outcome for some condition.
> In this case a mouse would be subjected to a surgical procedure that
> was similar, but where the adjustment was not done (the sham
> surgery). One can clearly say that the mouse that undergoes the sham
> surgery bears the control role, but it was thought that the important
> aspect of the "control" was the surgery - an occurrent.

I'm not sure I see the difference with the first case, even if the
problem is to assess the effect of an intervention by not performing
it on a population.

Intuitively, I think it makes more sense to say:

Control = less intervention
Tested = more intervention

seems to me the mouse receiving treatment without the adjustment is
the control (these mouses have a role of that type), unless I
misunderstand the example. In the study all mice are receiving the
same treatment but for a group to which some surgical procedure is
applied in order to assess the effect of that procedure.

One idea could be to make Treated different from Tested.

In any event, it seems the control role is relation, i.e. wrt a
procedure. This is shown by the fact that attributing the role depends
on what is done and not done.

That doesn't mean that you want to attribute a role to the procedure
tested in the sense of BFO. I'm not sure why it is not enough to have
categorized the study or protocol as a test of a peculiar sort to have
de facto put the right emphasis on the relevant tested procedure.

> So the question is, what do people think about this case - and are
> there recommendations about what to say about the sham surgery?

no recommendation from me, but see if the above helps. I'm not too
sure I understand precisely the issue though

cheers
pierre

> -Alan
>
>
>
> >
>

[Matthew Pocock]

Is contol a role within the experimental procedure? Perhaps it is instead a
role within the analysis and interpretation of the experimental results?

Matthew

[William Bug]

Both Pierre's and Matthew's comments have been helpful already.

I would also point out that I instigated a thread on the topic on the obo-phenotype list (obo-ph...@lists.sourceforge.net) back in mid-Feburary (2/15 - 2/26) that Barry, Chris Mungall, Suzie Lewis, Nigam Shah and others all contributed to.  Our specification of such roles will require use of PATO.

http://sourceforge.net/mailarchive/forum.php?forum_name=obo-phenotype&max_rows=25&style=ultimate&viewmonth=200702

My sense is Pierre's recommendation for continuant role is pretty much the way we should proceed (for now - until we hit a stumbling block).  It will only get problematic if we over-specify the details as being inherent in the role per se as opposed to the collection of continuant relations required to define the role context.

The only fly-in-the-ointment regarding an "intervention" view of role is that many "experiment roles" don't involve ANY intervention per se beyond the experimenter performing a clustering of subjects by declared group qualities.  This is especially true in a clinical context.  As I detail more in that obo-phenotype thread, sometimes a control group will be defined based on a combination of observable qualities that had nothing to do with experimental intervention - "former smokers", "over weight males age 12 - 18", "age-matched controls" - the latter a particularly common control role categorization in clinical studies of neurodegenerative disease.

Regarding procedural roles, it's not completely clear yet to me, just as Pierre mentions, whether this is truly something that can only be specified as an inherent quality of a process.

Re: Matthew's comment, he brings up an important point.  If by analysis and interpretation you mean activities that post-date the actual experimental procedure performed, then no, there cannot EXCLUSIVELY be considered associated with analysis and interpretation.  They would be more properly thought of as associated with experiment design.  The design would require patient populations be chosen to meet some statistical criteria that will HOPEFULLY enable you to make a statistically significant evidentiary statement when you finish your analysis.  Of course, once you have completed the study population selection, you then "blind" everyone conducting the study and don't come back to your initial experimental categories until you are in the final phase of your analysis (typically).

I don't think approaching it in this way really works with the realist approach.  As Pierre points out, in the case of "sham operated" - or the other common control group for pharmacological experiements - "vehicle control" where only the saline solution vehicle is applied minus the drug under test - the role inheres in some way in the nature of the differential intervention performed.  I don't know how this would map to the other "experimental role clustering"

Cheers,

Bill

Bill Bug

Senior Research Analyst/Ontological Engineer

Laboratory for Bioimaging  & Anatomical Informatics

www.neuroterrain.org

Department of Neurobiology & Anatomy

Drexel University College of Medicine

2900 Queen Lane

Philadelphia, PA    19129

215 991 8430 (ph)

610 457 0443 (mobile)

215 843 9367 (fax)

Please Note: I now have a new email - Willi...@DrexelMed.edu

[William Bug]

For some reason, my first attempt to post this to the list failed.

Cheers,

Bill

[Matthew Pocock]

On Sunday 15 April 2007, William Bug wrote:
> I don't think approaching it in this way really works with the
> realist approach.

Oh? Could you elaborate on this? Which bit of this way of modelling
experiments clashes with which bit of a realist approach?

I'm half way through "epistomology" by Chrostopher Norris (ISBN 0-8264-7732-1)
which argues for a realist position for doing science, and unless I have
missunderstood him, Norris would definitely include experimental design and
interpretation of results in the things to be modelled as 'real'.

> As Pierre points out, in the case of "sham
> operated" - or the other common control group for pharmacological
> experiements - "vehicle control" where only the saline solution
> vehicle is applied minus the drug under test - the role inheres in
> some way in the nature of the differential intervention performed. I
> don't know how this would map to the other "experimental role
> clustering"

I'm at risk of telling us all what we already know for the rest of this
message. Sorry about that - just trying to make things explicit that are a
bit jumbled in my mind.

Are controls continuants or occurants? In the case of a drug trial, the
placebo (a continuant) is the control. In the case of surgery, the different
procedure performed (an occurant) is the control.

In biology, in well designed experiments, we normally 'control for' multiple
things. So, control is only defined relative to the possible source(s) of
experimental variation which it allows us to factor out of the experiment.
Either the control is a quality thing - if the control doesn't work, the
procedure is kaput, or it is a quantity thing - the control produces readings
just like the other parts of the experiment do, and the other results are
interpreted 'in the light of' the control to normalize them against the
source of error the control is for.

I guess a 'good' control allows a physical process to be carried out as near
to identical as possible between the 'untreated' and 'treated' cases. If
knowledge about which case are controls held by the agent performing the
procedure can potentially have a significant effect upon the outcome, then
this is 'controled for' by doing control/treatment randomisation (e.g.
double-blind medical trials).

Would it be worth hashing up a quick and dirty description of the processes
you go through during experimental design, doign the work, and then the
analysis and interpretation of the results, and then see if controls sit
naturally in a particular layer of this? I'd be happy to be involved in this.

Matthew

>
> Cheers,
> Bill
>

[William Bug]

Don't get me wrong, Matthew.  I wouldn't be writing these emails, if I didn't think there was a way to represent these experimental details using the realist approach.

What I meant by that comment was a reference to the paragraph or two that preceded it.  

I'm not certain how to apply the realist approach to specifying experiment population categories as related to "intervention", when the only intervention is the partitioning of a population performed based on arbitrarily chosen categorization created by the researcher.  In other words, in an age matched control for a clinical neuropathological study, the same experimental techniques are being applied - e.g., obtain fixed tissue from cadavers, treat it for a particular histological workup, do microscopic imaging, apply specific microscopic techniques to observe some detail of plaque deposition in the Substantia nigra.  The only differential intervention would be the way in which the researcher reached into the collection of cadavers and grouped them based on some criterion - e.g., those diagnosed with Parkinsonian Syndrome when alive matched to age-equivalent controls who received the same cognitive and/or motor assessments and were diagnosed PD negative.  I suppose the process of performing the diagnosis is a form of intervention, but I don't think it can be considered to have a direct effect on the biomaterial understudy (post-mortem fixed tissue).

This is what I meant by saying I don't think this would work when applying a realist approach to describing the qualities of these groups is there is REALLY nothing about those acts - diagnosing patients, choosing cadavers based on medical history - that is truly an intervention that would effect the quality of some continuant in the biomaterial under study - fixed, excised tissue from a specific region of the CNS.

This just implies to me that in the case of these forms of very common clinical experiments, describing the experimental populations based on the difference in intervention is probably not appropriate.

The following Wiki page I worked up a few months back gives a crude description of how one might use PATO to describe such experiment populations:

http://esw.w3.org/topic/HCLS/OntologyTaskForce/OboPhenotypeSyntaxExperiment

(scroll down to "The OBO Phenotype Syntax + PATO Quality way to represent experimental observations/research statements/claims")

Using "differential intervention" is very appropriate for the typical type of basic bio-science experiment done in the lab, where the experimenter does try to control the majority of factors - isolating one or two they intervene to manipulate - and then measure the differential effect when comparing observations made on subjects/specimens receiving those "highly" interventive treatments to analogous observations made on subjects/specimens receiving the "minimally" interventive treatments.

Cheers,

Bill

[Matthew Pocock]

On Monday 16 April 2007, William Bug wrote:
> Don't get me wrong, Matthew. I wouldn't be writing these emails, if
> I didn't think there was a way to represent these experimental
> details using the realist approach.

> This just implies to me that in the case of these forms of very

> common clinical experiments, describing the experimental populations
> based on the difference in intervention is probably not appropriate.

Fine. I think we're on the same page now. Seems I was confused about if you
thought it couldn't be done, or if the way it was presented was not
appropriate. Thanks.

> The following Wiki page I worked up a few months back gives a crude
> description of how one might use PATO to describe such experiment
> populations:
> http://esw.w3.org/topic/HCLS/OntologyTaskForce/
> OboPhenotypeSyntaxExperiment
> (scroll down to "The OBO Phenotype Syntax + PATO Quality way to
> represent experimental observations/research statements/claims")

This is useful. I'm reading through it now.

>
> Using "differential intervention" is very appropriate for the typical
> type of basic bio-science experiment done in the lab, where the
> experimenter does try to control the majority of factors - isolating
> one or two they intervene to manipulate - and then measure the
> differential effect when comparing observations made on subjects/
> specimens receiving those "highly" interventive treatments to
> analogous observations made on subjects/specimens receiving the
> "minimally" interventive treatments.

The phrases 'where the experimenter does try to control the majority of
factors' and '"minimally" interventive treatments' are interesting. Perhaps
we use control in multiple senses. We 'control for' variables we think may
introduce extraneous variability. We also have 'control conditions' that
provide base-lines (your "minimally" interveintive treatments). Also, some
specific processes or materials are called 'controls' by virtue of how we are
using them. There are probably other sorts of controls as well.

So, in the case of your dead people, an assumption has been made that the
age-matched cortexes are representative of base-line conditions, making the
conditions prevelant in the age-matched group a 'normal conditions control'
and the cortexes themselves 'control samples'. We tacitly assume that the
other cortexes can be described relative to this.

The (I assume quantative) PCR and Western blot analysis would have included,
among other things, a range of 'callibration controls' to ensure that
between-procedure comparrison can be done meaningfully. They probably also
had other 'internal controls' to check that the techinques where not
spectacularly failing.

Lots of different kinds of controls, at each level of the process from design
to what was done through to evaluation of the data. So, I'd sudgest that
control is only meaningfull with relation to what it controls for, and the
scope within which it controls for it.

Best,
Matthew

>
> Cheers,
> Bill

[William Bug]

Hi Matthew,

Thanks for your insightful follow-up.

Please see below for comments in-line.

Cheers,

Bill

On Apr 16, 2007, at 12:04 PM, Matthew Pocock wrote:

On Monday 16 April 2007, William Bug wrote:

Don't get me wrong, Matthew.  I wouldn't be writing these emails, if I didn't think there was a way to represent these experimental details using the realist approach.

This just implies to me that in the case of these forms of very common clinical experiments, describing the experimental populations based on the difference in intervention is probably not appropriate.

Fine. I think we're on the same page now. Seems I was confused about  if you thought it couldn't be done, or if the way it was presented was not appropriate. Thanks.

The following Wiki page I worked up a few months back gives a crude description of how one might use PATO to describe such experiment populations:

http://esw.w3.org/topic/HCLS/OntologyTaskForce/OboPhenotypeSyntaxExperiment

(scroll down to "The OBO Phenotype Syntax + PATO Quality way to represent experimental observations/research statements/claims")

This is useful. I'm reading through it now. 

Using "differential intervention" is very appropriate for the typical type of basic bio-science experiment done in the lab, where the experimenter does try to control the majority of factors - isolating one or two they intervene to manipulate - and then measure the differential effect when comparing observations made on subjects/specimens receiving those "highly" interventive treatments to analogous observations made on subjects/specimens receiving the "minimally" interventive treatments.

The phrases 'where the experimenter does try to control the majority of factors' and '"minimally" interventive treatments' are interesting. Perhaps we use control in multiple senses. We 'control for' variables we think may introduce extraneous variability. We also have 'control conditions' that provide base-lines (your "minimally" interveintive treatments). Also, some specific processes or materials are called 'controls' by virtue of how we are using them. There are probably other sorts of controls as well.

Absolutely - there are definitely multiple meanings to the word "control" when it comes to setting the base-line context for making empiricaly observations - sometimes based on interventive manipulation in the context of the experimental procedure conducted, sometimes based on our attempt to "tease out" a base-line from naturally occurring events not under our influence.  The latter, of course, is implicit in ALL experiments on biological material - possible all experiments that rely on the scientific method.  In other words, the investigator designing the domain of naturally occurring activity involving material continuants must play a sort of "mind game" by saying, we have controlled for X, Y, Z, thereby ignoring ALL the uncountable number of continuants, occurants, and associated qualities putatively not directly related to the phenomena understudy.  We have to assume given the enormous success of this fallacy over the last 400 years (choosing Bacon's updating of Aristotle - Novum Organum - as a somewhat arbitrary starting point - http://www.archive.org/details/novumorganum00bacouoft or http://www.sirbacon.org/novorgcom.html) that it is possible to gradually weave a web of understanding that is relevant to - and based on some fundamental facts about - these real world objects & activities.  My thinking on this topic is very much informed by Robert Rosen's treatise "Life Itself" - as well as the follow-up criticism.

So, in the case of your dead people, an assumption has been made that the age-matched cortexes are representative of base-line conditions, making the conditions prevelant in the age-matched group a 'normal conditions control' and the cortexes themselves 'control samples'. We tacitly assume that the other cortexes can be described relative to this.

yes - a loaded assumption, as I state above - but one that when executed carefully and consistently can be effective.

The (I assume quantative) PCR and Western blot analysis would have included, among other things, a range of 'callibration controls' to ensure that between-procedure comparrison can be done meaningfully. They probably also  had other 'internal controls' to check that the techinques where not  spectacularly failing.

Yes.  Which brings up the issue of normalization vs. control.  They do not play quite the same role in the context of an experiment, though their roles are certainly related.  Implied in a "control" (population, specimen, procedure, etc.) is the need for an "internally" proscribed baseline whose domain of use is limited to the particular experiment.  Normalization/Calibration procedures (and the materials used to execute them) imply a connection with the larger universe of experiments conducted using the same - or similar - techniques.  In fact, calibration can have a meaning outside the realm of any experiment - e.g., running periodic calibration procedures on a chromatagraphic devices, a pH meter, a microscope imaging system, etc.  The results of these calibrations performed outside the context of a given experiment - even some performed within the context of an experiment (calibrating a pH meter) - are typically not directly referred to when reducing/analyzing/interpreting results, except to sometimes state - "device X is periodically calibrated using procedure Y".  They can sometimes simply exist as an implicit fact known primarily to those who use such devices or techniques on a regular basis.

Lots of different kinds of controls, at each level of the process from design to what was done through to evaluation of the data. So, I'd sudgest that control is only meaningfull with relation to what it controls for, and the scope within which it controls for it.

I think this is one of THE CRITICAL issues when it comes to using BFO and/or PATO to describe control populations/specimens/procedures.  The PATO approach is designed to be terse.  It would be possible to represent all of this detail using BFO (though I'm not certain exactly where the continuant/occurent breakdown is for controls).  With PATO, the idea is to use ONLY that level of detail required to characterize the evidentiary observation and no more.  One must remember it is primarily - though not exclusively - designed for use in annotating the literature.  I was tempted to drill down to the detailed controls at the level of biochemical assays, BUT in the context of this particular article, the details were slim.  It wasn't clear there would be benefit to doing so.  I was guided in my inclination to do this by the primary use to which I'm putting BFO, PATO and other ontologies right now - which is as a means to formally semantically characterize details related to neuroimaging experiments.  In this context, there is definitely a need to represent both the experimental and control procedures/provenance and observations at a deep level of detail.  The over-arching goal is to make it possible to re-slice and re-dice this data for later meta-analysis as is typically done with bio-molecular data repositories (sequence and structure databases).  Even with microarray data, to a large extent, characterizing experimental detail down to the level of the original, segmented image spot arrays - and the subsequent threshold-based analysis - can be critical to properly characterizing the data for later re-analysis in a different context.  Hence the complexity of data models for storing gene expression data such as FuGE.

I don't have any absolute answers to this issue as of yet, except to say one needs to decide on the level of required detail based somewhat on how one intends for the ontology-based representation to be used later.

Best,

Matthew

Cheers,

Bill