Occupational Asthma Reference
|
Bright P, Pantin CFA, Newton D, Burge PS,
Interpretation of serial peak flow records for the diagnosis of occupational asthma – a method based on pattern recognition technology.,
Unpublished,
1998;1:1-20,
|
|
(Plain text:
Bright P, Pantin CFA, Newton D, Burge PS,
Interpretation of serial peak flow records for the diagnosis of occupational asthma - a method based on pattern recognition technology.,
Unpublished)
|
Keywords: Oasys-N, neural net, diagnosis, UK,
Known Authors
If you would like to become a known author and have your picture displayed along with your papers then please get in touch from the contact page. Known authors can choose to receive emails when their papers receive comments.
Abstract
Several attempts have been made to produce a statistical analysis of occupational serial peak flow charts but few have approached the sensitivity and specificity of human experts. The problem is essentially one of pattern recognition. We have designed and tested a neural network to interpret PEF records plotted as daily maximum, mean and minimum.
The records of 248 subjects investigated for the presence of occupational asthma were used. Subjects were asked to record PEF (best of 3) 2 hourly through the day, for a minimum of 21 days. Each record was divided into work-rest-work (WRW) or rest-work-rest (RWR) periods and scored blind by a human expert for the presence of work-related effect. The records were then divided into 3 sets (A-172, B-33, C-43), sets A and B were used to “teach” and “test” the neural network, set C to judge its success. The resulting network was then used to score records from a group of subjects with and without occupational asthma (gold standard set 1), diagnosed by other means to ascertain a cut-off score. Finally the sensitivity and specificity of the neural network was obtained by scoring a further set of gold standard records (set 2).
The weighted kappa for WRW periods from set C was 0.83, 0.74 for RWR periods. 2 cut-off points were chosen, at a whole record score of 2 and 3, with a “grey” area between. The sensitivities and specificities for sets 1 and 2 are shown at these points, and the midpoint of 2.5.
Gold standard set 1 Gold standard set 2
Cut point Sensitivity Specificity Sensitivity Specificity
2 100 78.6 100 97.8
2.5 93.8 91.1 89.3 100
3 81.3 100 80.4 100
These results show that the neural network is at least as good as the human expert in interpreting occupational serial peak flow records for the presence of work-related effect.
Full Text
Full text of this reference not available
Please Log In or Register to add the full text to this reference
Comments
|
This paper was never published, we are not sure why as it was an improvement on Oasys-2. Perhaps the requirement to manipulate the input data to 5 days at work, 2 days off and 5 further days at work was thought to be a bit removed from the source data. We went on to develop the ABC score which seemed even better.
|
Please sign in or register to add your thoughts.