Comments received about the QAMC project web page

Comments received about the QAMC project web page

If you have just sent feedback and cannot see it in this file, or you would like to guarantee viewing the most up-to-date version of this file, press the Reload button on your browser.

[ Read guide] [QAMC Publications] [Send feedback] [QAMC home page]


Feedback from: Rudy (rjal)
These comments were received on Tue Sep 2 03:17:27 1997:
- Title says likelihood of failure to progress in labour:
According to the models you use, i.e neural network, log.regr.
and lookup table you estimate post.probabilities (and in case
of the lookup table you can predict them exactly at least for the
data given), so I suggest:
Probability of failure to progress in labour OR
Estimated incidence of failure to progress in labour.

- It might be useful (?) to give a bit more info (not too much in
depth though) about the
different models in order to interpret them in a correct
manner, e.g. explain a lookup table and also why you smooth
results to give better generalisation, Log.Regression is
a linear model whilst the neural network allows to fit a non-linear
model, etc.

- Some of your ICD9-coded predictors are mutually exclusive,
e.g. 6523 and 6528, 6411 and 6418, 6565 and 6566, etc.
These will probably never occur together in the database
(unless there are inconsistencies) but it would still look
slightly more `professional' if they wouldn't appear together
on the next page with the results and even better that you
actually cannot select them together (remember that mainly health
professionals specialised in obs and gynae will read this page)

- To end: I'm very impressed with this page and the modelling
behind it: good work!

Feedback from: Marianne Mead (m.m.p.mead@herts.ac.uk) .
These comments were received on Wed Oct 1 08:28:59 1997:

I have just completed a comparison of the intrapartum care and outcome of women who have delivered in four local maternity units during 1994. Nearly 13000 cases were gradually reduced to include only women who had no previous medical history and no pregnancy pathology or previous history such as NND or CS.
For all intent and purposes, the remaining 4500 women could be identified as suitable for full midwifery care and although there were socio-economic differences as measured by marital status and smoking, women who delivered in the maternity unit that had the higher socio-economic indicator also had the higher caesarean section rate, following the highest rate of intervention in labour.
I wonder how policies, procedures and philosophies of care are going to be taken into consideration in the modelling of the risk of failure to progress. I feel that it may be dangerous to ignore individual practitioners' preferences for particular type of intrapartum care and their influence on failure to progress.
I would be grateful if you could comment on this.


Feedback from: Daniel Crespin (dcrespin@euler.ciens.ucv.ve) http://euler.ciens.ucv.ve/~dcrespin/Pub/.
These comments were received on Mon Oct 13 15:34:52 1997:

Very interesting!!
Iam interested to gain access to the database you used or at least to part of it. My purpose is to build a neural network and e.mail it to you.

Regards

Daniel Crespin


Feedback from: Alfredo de Almeida Cunha () .
These comments were received on Tue Feb 10 09:20:48 1998:

I enjoyed very much the idea of the page. I am trying to learn all it ccn teach.


Feedback from: Saji Jacob (saji@indigo.ie) .
These comments were received on Fri Nov 5 23:39:38 1999:

It is interesting note that even with 1.2 million saple size on your database the prior probability and accuracy of prediction is not sufficent enough for clinical use. I agree that the ANN models are giving the best accuracy compred to logistic regression. What type of ensemble of ANNs are you using. Do you use bootstrap ensembles.