Hi pafalafa-ga:
This question is in connection with my previous one about sources for
researching or screening possible adverse events of a product on the
market (post-marketing survelliance).
I would like to know your suggestions/ideas for building search
strategies in those databases.
Suppose the following situation:
I need to make a follow-up of a given product on a market, that is
screening activities in order to detect possible adverse events of a
drug produc as soon as possible. Which are the best search strategies
you envisage for this task? I will subdivide this question in four
parts:
1) General recomendations
2) I know that some medical works published in medical articles can be
mainly designed for other reason than for detecting adverse events,
but looking carefully to the results, you may be able to find adverse
events. How do suggest to overcome this issue?
3)I know that there are different kinds of adverse events, so a
different strategy can be conceived for each of the main adverse event
categories.
The answer I imagine to this question would give general search
strategies for every broad category of adverse event, that I could
apply to specific products I choose. In one of the reference links you
gave me in a previous answer, FDA recommends to take account of the
pharmacology of the product or drug under survelliance, in order to
predict the adverse event that may arise. How do you relate this with
potential search strategies in databases?
4)Is any particular thesaurus (e.g. MeSH, MedDRA) or codification
system (e.g. COSTART), preferred for the selection of terms to build
the search strategy?
Kind regards,
Tom
A final word: I think that the useful information that you (and also
tehuti) provided here will indirectly contribute to the early
detection of drug "adverse events", for the benefit of patients.
Kind regards,
Tom
...in case you don't have this link already:
http://www.ncbi.nlm.nih.gov/About/primer/pharm.html
My pleasure, Tom. It came in my link roundup (Pharmacogenomics) the
other day and I remembered your series of questions on adverse events.
These folks generate some useful linkage once a week if you care to
subscribe.
http://www.hedweb.com/siteoday/2003.htm
Also there's:
http://www.blackstump.com.au/anew2.htm
and:
http://lii.org/search/mntw
Enjoy!
V
Hello Tom,
Whew. I feel like I can write (or *am* writing!) a book. I’ll take
your questions one at a time, below. But if you post additional
questions in the future, there’s one I’d like you to answer as well:
–what’s more important to you – to be the very first one on your block
to know about an adverse reaction, or to be the one who knows the
most?
I’m sure your answer is going to be “both”, but give the question some
thought. It would be a help (to me, at least!) to know where you’re
coming from on this.
----------
1) General recommendations
[please forgive me if I’m stating the obvious here]
–use drug names *and* their synonyms for searching
Many search systems allow “OR” searching so that you can combine
several synonyms into a single search. For example, a search could be
for (tylenol OR acetaminophen), and the results would include
studies/articles that contain either term. The ChemIDplus system that
Tehuti-ga identified for you in an earlier question would be helpful
here:
http://answers.google.com/answers/main?cmd=threadview&id=172157
Be aware that drugs may be marketed overseas under different names
than the popular name used in the U.S.
–search first for the drug name (and synonyms) alone.
For many drugs, there simply may not be an extensive body of
literature – the number of published studies may be in the dozens.
For these drugs, it is feasible to simply amass every publication
available and scan them for relevant information.
–in other cases, a search on the drug name alone will reveal hundreds
or thousands of reports – too many to manually review. In this event,
search results can be narrowed down by:
1. Time – search only for articles in the past X weeks/months/years –
whatever time frame meets your research needs. Most search systems
provide some sort of date selection function.
2. Adverse effect – search for articles that include the drug’s name
as well as other key phrases of a general nature (“adverse effects”,
“side effects”, “case report” etc.) as well as more specific phrases
(such as “carcinogenicity”, “liver damage”, “reproductive toxin”,
etc.). The actual terms chosen depend, in large measure, on what you
are looking for, and what the thesaurus of the database looks like –
I’m sure Tehuti-ga can offer some more detailed advice on this.
3. Field searches – again, many search systems offer an option to
focus a search on particular fields – the title of an article,
keywords, journal source, etc. These can be useful in taking a list
of thousands of preliminary results, and paring it down to a
manageable number.
4. Proximity searches – this is less universal than other search
tools, but when it’s available, it can be awfully handy.
If X and Y are your two terms of interests, the proximity search
usually takes the form of [X NEAR Y] or [X w/# Y] or [X and Y in the
same line/sentence/paragraph/page]. The first will look for the terms
close together in a text (each search system has its own particular
protocol for interpreting “NEAR” – some searches also include ADJ for
“adjacent to”). The second will search for terms that are within a
specified number of words of one another: you get to select the value
of #. The third will look for the terms together in the same unit of
text, according to your selection. If you’re not familiar with these
types of searches, some good examples can be found at the search help
page at a Dept. of Energy site:
http://search.dis.anl.gov/plweb-cgi/iopcode_mhrex.pl
towards the bottom of the page is: Table 5 - Summary of Query
Operators.
----------
Of course, none of these suggestions help with the question of *where*
to search. As you are finding out (or perhaps, already knew) there
are many different sources of information regarding adverse effects of
medicines. The “where” question depends in very large measure on
what, particularly, you are trying to find out. Perhaps in a future
question you can provide some additional specifics, and ask a
researcher to craft a specific search strategy for you.
One major source that I want to mention here, though, is the Adverse
Events Reporting System – AERS – at FDA, which has already been
mentioned several times in other answers. I spoke to a staffer at FDA
who told me they handle about 270,000 reports in a typical year. The
entire AERS data set is available for purchase, if you’re interested
in managing a fairly sizable database. Information regarding
purchase, on either a one-time or ongoing subscription basis, can be
found at FDA’s site:
http://www.ntis.gov/search/product.asp?ABBR=SUB5460&starDB=GRAHIST
“The Adverse Event Reporting System (AERS) of the Center for Drug
Evaluation and Research is a computerized data base of drug adverse
reactions reported by health professionals and others. The system
contains only adverse reactions detected and reported after marketing
of the drug during the quarter. The information is not cumulative. The
primary purpose for the AERS data base is to serve as an early warning
or signaling system for adverse drug reactions not detected during
premarket testing. The files included on the CD-ROM are: (1)
Information on demographic and administrative information and the
initial report image ID number (if the image is available); (2) Drug
information on the case reports; (3) Reaction information on the
reports; (4) Patient outcome information on the reports; (5)
Information on the source of the reports. Historical Data may be
ordered through our Sales Desk at 1-800-553-6847. January-December
1999, Order Number PB2000-500054, price $360; January-December 2000,
Order Number PB2001-500048, price $360; January-December 2001, Order
Number PB2002-500060, price $360.”
----------------------------------------
2) I know that some medical works published in medical articles can be
mainly designed for other reason than for detecting adverse events,
but looking carefully to the results, you may be able to find adverse
events. How do suggest to overcome this issue?
Again, there’s a matter of numbers here. If you’ve conducted a search
and have uncovered three studies on a particular drug, then it’s
feasible to read each of the studies word for word and look for
evidence of adverse reactions that may not have been highlighted by
the authors.
But what to do if you have hundreds of potential studies to go
through? One answer is that the tools that I mentioned above under
“General recommendations” are the very same ones that could help you
here. An author may have dismissed a case of cancer, as a
happenstance, rather than an outcome related to drug exposure – in
other words, the study may treat the cancer as statistically not
significant, rather than as an “adverse reaction”. But nevertheless,
the presence of cancer in a test person or animal would not go
unmentioned, so that a search on the terms: [“drug name” and cancer]
should flag this report as one that is worthy of more detailed
attention.
I also want to mention – though I suspect you are already aware of
this – the field of meta-analysis. Meta-analysis combines information
from several related studies (in your case, studies of the same drug)
with the aim of increasing the overall statistical power of the
analysis. In theory, as well as in practice, a meta-analysis of
combined studies can reveal a statistically significant adverse effect
of a drug, that was not revealed by any of the individual studies.
An example of a drug-related meta-analysis can be found at the
Schaffer Library of Drug Policy site at:
http://www.druglibrary.org/schaffer/Misc/driving/s13p1.htm
“Medicines and Driver Fitness - Findings from a Metaanalysis of
Experimental Studies as Basic Information to Patients, Physicians and
Experts”.
I’m not aware of any means of automating meta-analysis – it is a form
of professional research by trained humans, rather than an outcome of
a clever machine. However, there is certainly the potential here to
clarify adverse effects that may otherwise have been missed, and for
this reason, may be of importance to your work.
3) In one of the reference links you gave me in a previous answer,
FDA recommends to take account of the pharmacology of the product or
drug under surveillance, in order to predict the adverse event that
may arise. How do you relate this with potential search strategies in
databases?
Once again, I want to emphasize that if you are researching a
relatively obscure drug that does not have a great many recent studies
in the literature, there may be no need for the pharmacological
approach you are describing, as it may be possible to review all
available information. Hence, there would be no need to narrow your
search by trying to predict relevant adverse effects.
I cannot go into a full discourse here on applying the tools of
pharmacology – pharmacokinetics, pharmacodynamics, and so on – to
predicting likely adverse effects. The basic approach here is to
combine an understanding of a drug’s chemistry, means of action,
metabolism in the body, and chemical energetics, to make reasonable
guesses about biologically-important drug properties, such as: how
does the drug partition in the body (proportion that goes to blood,
kidney, bones, etc); how long is the half-life of the drug in
different body compartments (is it water-soluble and quickly excreted,
or fat-soluble resided in the body for a long time); what are the
likely metabolic products of the drug (and what is the toxicity of
each of these chemicals).
The literature of a particular drug may well contain much of this type
of information, and could be isolated by using appropriate search
terms. But again, the actual search that you would want to conduct
depends very much on what drug you are researching, what search system
you are using, and what question(s) you want answered.
You can get a feel for what the pharmacological approach looks like by
reviewing the notes of a college-level course on pharmacokinetics:
http://www.boomer.org/c/p1/
“A First Course in Pharmacokinetics and Biopharmaceutics”.
I’m not sure what else to add on this topic – it is simply too broad
in scope to be able to readily narrow down to particular search
strategies for an unspecified chemical.
For this topic, as for all the information I’ve provided here, just
post a Request for Clarification if you want any additional
information.
And one last thing: somehow, in my earlier work, I overlooked an
important resource on adverse events reporting – the World Health
Organization. Information about their reporting system can be found
here:
http://www.who-umc.org/aboutumc.html
WHO adverse events database
It's worth a look.
So long...for now.
Dear pafalafa,
Answering your question:
"–what’s more important to you – to be the very first one on your
block
to know about an adverse reaction, or to be the one who knows the
most?
I’m sure your answer is going to be “both”, but give the question some
thought. It would be a help (to me, at least!) to know where you’re
coming from on this."
Yes, my answer is "both", but if I have to choose one answer, I would
choose, "being the very first to know". In other words, I prefer high
"recall", at the expenses of a bit less "precision". Of course, this
will imply the extra work of filtering the high recall output.
Your suggestions for search strategies are very much apprecited by me.
Thank you very much indeed!
Kind regards,
Tom
Tom,
You asked: Do you think it is possible today for a pharmaceutical
company
to get the info about an adverse event before it is made available in
public databases? I would post a question if you think you can provide
an answer to this issue.
My answer is a resounding "Maybe". Generally, what happens is this:
1. Someone submits an adverse event report to FDA
2. FDA staff summarize the report and remove all confidential
information (such as patient's name)
3. A quarterly electronic update of all new report summaries goes to
NTIS
4. NTIS makes the update publicly available.
Many weeks -- even months -- pass between steps 1 and 4. It is
*possible* that FDA would make info available on request at step 2,
putting you weeks ahead of the NTIS release. This could be done
informally, with a phone call, or formally through a Freedom of
Information request to FDA. The only way of knowing if this is a real
possibility would be having a conversation with FDA.
The woman who manages the whole adverse events reporting system is Min
Chen (not sure I have the right spelling) and her phone is:
301-827-3169
Give her a call and ask!
Alternatively, you could post a question here and have a researcher
ask BUT...I quite frankly think it would be better if you called
yourself, since you are in a position to explain to FDA the importance
of your project, while a Google Researcher would be at a real
disadvantage in this respect.
Goog luck.
Dear pafalafa,
Thank you very much for your feed-back.
I am interested in shortern the time for detecting an adverse event of
a marketed drug product. You recently shared some thoughts about this
issue. Do you think it is possible today for a pharmaceutical company
to get the info about an adverse event before it is made available in
public databases? I would post a question if you think you can provide
an answer to this issue.
Thanks again for your interest in these issues of pharamcovigilance!
Regards,
Tomh
Thanks Tehuti. Your comments are welcome, informative and appreciated.
paf
Tom,
Nice to hear back from you, and thanks so much for the kind words and
high rating. Even though I've been an adult for quite a while now, I
still get that warm, fuzzy kid's feeling when someone tells me "well
done".
Thanks, too, for commenting on my speed vs. depth question. The
reason for my asking is this: if speed is of the essence, the trick
is to try an intercept the information as soon as possible, which is
usually *well before* the time that it becomes available in a data
base.
I would guess that there is a lag time of several months between the
time an adverse reaction report is sent to FDA (or a drug company, or
an agency in another country) and the time it actually appears in a
publicly accessible data base.
It may well be possible to get access to key information from the
reports even before they make their way into public data systems.
This would involve a strategy for working the bureaucracy -- which
calls for a very different type of strategy than working on medical
data systems.
Just a thought to add to the pot you're brewing. Best of luck with
everything.
Dear voila-ga:
Thanks a lot for the excellent link!
Best regards,
Tom
I meant: "the suggestions for search strategies you made in your
answer were very much appreciated by me". Sorry for the missing words
in the sentence. Kind regards,
Tom
Dear pafalafa-ga:
Thank you very much for your work on this subject. I found the
information you posted very useful. Please, answer the first 3 parts
of the question, and I will consider the question as fully answered. I
will post a question to tehuiti-ga for the part 4, in order to get his
help on this particular issue, who gently added useful comments about
it. In case tehuti is reading this clarification, I would like to say
that you (Tehuti) also did a very good job answering my questions. I
asked both of you some similar question in order to get different (and
complementary) opionions about this interesting subject. Pafalafa,
regarding Lexis, I visited the website and read your instructions
posted in a previous answer. I like the system and I will try it
later. In case I have problems with it, I 'd probably post you a
question, in order to ask for your help.
Note: in the rating section of the question #173514, I meant "this is
exactly what I was looking for"
Thanks again, and please go ahead with your answers to part 1, 2 and 3
of my question!
Regards,
Tom
Tom,
Two things, if I may:
1. I've had a chance to read through your quesiton and think about it
a bit. I'm comfortable with the first three parts, but I'm not very
comfortable with number 4. I'm simply not well versed enough in the
details of the different data dictionaries to provide a meaningful
answer here.
Please let me know if this is essential to your needs. If so, you may
want to repost the question and open it up to all researchers (Tehuti
may have some insights here) so that you can get the best answer
possible.
2. I could more directly address the first three parts of your
question if I knew what shortcomings you have experienced (or
anticipate) in searching for information with the sources and search
strategies you are already aware of.
The best tools and strategies to use depend very much on what, in
particular, you are trying to accomplish. The more you can let me (or
another researcher) know, the better we can focus an answer.
Sure thing.
Dear pafalafa-ga:
I couldn't go through your answer yet. I will do it as soon as
possible, when I get the time to read it carefully. Although I
couldn't go through it yet and therefor cannot comment on it, I thank
you very much for the additional info you provided, which seems to be
very interesting. Hope I could post my feed-back today.
Kind regards,
Tom
Boy, you really keep 'em coming. I'll take a look at this soon, but
it may be a while (tomorrow?) before I can get back to you with a
reply.
Tom and pafalafa,
With respect to part 4, I hope I am not stepping out of line in making
a comment. I feel free to do so since pafalafa has indicated less
familiarity with this part of the question.
MeSH or any other thesaurus attains most relevance when you conduct a
search on the database using that thesaurus. Thus, MeSH terms will
enable the greatest precision when used with a Medline search.
For example, if you search for a drug name on the MeSH Browser, and
then select the detailed display for that entry, you will find a whole
set of subheadings you can use with that drug name eg:
http://www.ncbi.nlm.nih.gov/entrez/meshbrowser.cgi?term=aspirin&retrievestring=%22aspirin%22%5BMESH%5D&mbdetail=y
You can see that it is possible to select "adverse effects" (NB
effects not events in MeSH!), "toxicity" and "poisoning" (less
relevant!!!) as ways to narrow the search on Medline to records which
contain these qualifiers added on to the name of the drug that
interests you. This will ensure that you will only retrieve records
for which the indexer thought it necessary (a) to index that
particular drug, and (b) to assign one or more of these subheadings to
it. However, a minor mention of adverse effects in a paper might not
necessarily receive this indexing. I've heard Stuart Nelson, the head
of the MeSH department at NLM speak about the strictness with which
indexers must determine the "aboutness" of an article. Remember that
indexers have to follow certain rules, among which is one which
defines the maximum number of index terms that can be assigned to a
record.
However, looking at these qualifiers in MeSH will also give you ideas
for terms to use in a free text search on any database. A free text
search on "adverse effects" in Medline will pull up only records using
this MeSH qualifier in the assigned index terms (so the advantage of
limiting a search to MeSH terms lies primarily in its power to exclude
more irrelevant material), and it will also pull up records in which
the term "adverse effects" occurs in the abstract or title even if not
in the assigned MeSH terms (this should not really happen, since a
mention in the abstract or title should guarantee the index term being
assigned, however, indexers, being human, are sometimes inconsistent).
The same term will also be useful on other databases, in that it will
retrieve articles with the term in the title or abstract, even though
a different expression might be used in the thesaurus of that
database.
Of course, the best approach would be to optimise your search by using
terms specific to the database thesaurus. However, not every database
user will necessarily have access to a thesaurus.
If you are limited to free text, you have to bear in mind that you
will not have the advantage of having everything codified into one
agreed terminology. Thus, an abstract might refer to "adverse
effects", "adverse events", "undesirable effects", "side effects". If
your search needs to be as exhaustive as possible, you will need to
take all of these into account.
Until such time as all databases agree on a unified thesaurus, a
thesaurus will work with the database for which it is produced, and at
best only give you suggested search terms for other databases.
Each database+thesaurus combination has its own strengths and
weaknesses for different types of searches.
Tom,
I wanted to add one more piece of information to the answer for your
third question. The FDA (and others) make use of an emerging
technique called computational toxicology to predict adverse effects.
A description of it by an FDA official can be found here:
http://caat.jhsph.edu/programs/workshops/testsmart/pharm/contrera.htm
"Computational Toxicology And Knowledge Bases at FDA/CDER"
In brief, this approach looks at the structure-activity relationships
of complex molecules to predict possible adverse effects. That is, it
looks at the detailed molecular structure of a chemical or drug to
identify any sub-structures (molecular groups) that are identical or
similar to known groups with a well defined biochemical activity. By
methodically searching for these molecular similarities, scientists
have had some good successes in improving their ability to predict
toxic effects.
Oops, a superfluous "only" crept in, making nonsense of para 5, the
beginning of which should read:
However, looking at these qualifiers in MeSH will also give you ideas
for terms to use in a free text search on any database. A free text
search on "adverse effects" in Medline will pull up records using
this MeSH qualifier in the assigned index terms (so the advantage of
limiting a search to MeSH terms lies primarily in its power to exclude
more irrelevant material), and it will also pull up records in which
the term "adverse effects" occurs in the abstract or title even if not
in the assigned MeSH terms
Dear voila-ga
Thanks a lot! The sites you provide here are really enjoyable ones!
Thanks again for your kind attention to share these sites with me.
Kindest regards,
Tom
Dear pafalafa: I appreciate your answer. Thank you very much!
Tom
The researcher provided very useful information, and worked hard to
satisfy my information needs. The information provided is highly
relevant. A well deserved 5 star rating! Thank you for your help!
 I could use some help again...
 Lost Licences |
Post-marketing survelliance - search strategies **Note: only for pafalafa-ga **