Direct versus indirect method of standardization in estimation of risk

[Margaret]

Dear list members

I would be interested to receive your views on the advantages and
disadvantages of using the indirect as opposed to the direct method in
standardization of risk estimates. I appreciate that there is a lack
of consensus in this area. For example, in Section 3 of the BMJ
article 'Epidemiology for the Uninitiated' (see
http://www.bmj.com/epidem/epid.3.dtl#pgfId=1001728), the indirect
method is described as yielding more stable results in the majority of
cases. However, you may hold alternative views. Either way, I would be
happy to hear from you.

Best wishes

Margaret

[Martin Holt]

Dear Margaret,

Googling out of interest I came across this:

"There are two main methods of standardisation: direct and indirect. Direct
standardisation is often the preferred method especially in epidemiological
contexts. In this report it is used for example for the analysis of cancer
incidence and mortality. However in the context of case mix adjustment for
the outcome indicators presented in Sections C, D and E of this report -
readmission rates and post-admission and post-operative mortality - it has
one overwhelming drawback. Direct standardisation is inadvisable if the
number of cases in any of the cells of the cross-classification of the
variables used to standardise is small. Thus if one is standardising for
age, sex and deprivation and there is a possibility of very low numbers in
any combination of age, sex and deprivation categories, direct
standardisation should be avoided. If there is a possibility that there are
no cases in any of the cells of classification (zero cells) then direct
standardisation is entirely ruled out. Indirect standardisation is highly
robust in the context of small cell numbers.

It cannot be stressed strongly enough that despite the possible implication
in the very terms ('indirect' vs 'direct') that indirect standardisation is
somehow less powerful than direct standardisation, in the current context,
that of case mix adjustment of clinical outcome indicators for multiple
factors indirect standardisation is the more robust method."

....and there is more....

http://www.indicators.scot.nhs.uk/Trends_Jan_2009/Standard.htm

HTH,
Martin Holt

[BXC (Bendix Carstensen)]

Direct standardization is used when only estimated rates are available.
It is just an weighted average of the rates, and is not based on a model. It requres a set of weights (which incidentally are not often stated by users of the method).

Indirect standardization is really based on a model for the rates, and you need to have access to both cases (numerator) and person-years (denominator).

Indirect standardization therefore has the advantage that you base your conclusions on a clearly specified model where you are able explicitly state your assumptions (this may be considered a disadvantage by some).

Hence:
1) Use a model and specify it.
2) State you assumptions as embedded in the chosen model.
3) State your conclusions as derived from the model.
4) Avoid outdated terminology from the hand-held calculator days of epidemiology,
that is never use the word "standardization" in any context.

Best regards,
Bendix Carstensen

[Abhaya Indrayan]

To put it simply, my rule is as follows:

Prefer direct method of standardization unless the group-specific rates are unstable. The latter occurs in many clinical studies because of relatively small sample. Then indirect method is more appropriate. In most epidemiological studies, the sample is large enough to provide stable group-specific rates.

~Abhaya Indrayan

[MMD]

Dear Martin

Thank you very much for taking time out to join me in my googling
expedition. This is most kind and appreciated. However, I was really
rather keen to hear group members' personal views based on their
experience of using these methods.

Every best wish

Margaret

> > Margaret- Hide quoted text -
>
> - Show quoted text -

[MMD]

Dear Abhaya

Thanks for sharing your view. This is entirely in keeping with the one
shared in the reference I mentioned.

Best wishes

Margaret

[MMD]

I should perhaps also add, Abhaya, that I support the view in the
original BMJ article I cited, and that I rarely, if at all, have the
privilege of access to large enough groups to favour the direct
approach.

Thanks for sharing your view.

Best wishes

Margaret

On 5 Oct, 15:40, Abhaya Indrayan <a.indra...@gmail.com> wrote:

[Frank]

My opinion is that to some extent standardization represents an
avoidance of raw data and of proper conditioning. A multivariable
model using raw data, where covariate adjustment is used to condition
on or hold other factors constant, is much preferred to
standardization.

Frank Harrell

[Abhaya Indrayan]

Dear Margaret:

You are absolutely right. I was only trying to put well-known rule into simple words in the hope that this will minimise the confusion (if any). I work with medical professionals and try to keep statistics simple--sometimes perhaps overly simple. 

To provoke further discussion, may I add that I sometimes wonder how such simple issues get lost in the web of complex terminology and intricate methodological issues. My knowledge of such complexities is very limited.

~Abhaya Indrayan