Prediction of seven-year survival by artificial neural network and logistic regression: A comparison of results from medical and social data among 70-year-olds in Göteborg, Sweden.

• Problem: Are there, for practical uses, any benefits of Artificial Neural Network analyses (ANN) compared with logistic regression analyses? Data: A random sample of 2,294 70-year-old persons from Göteborg, Sweden, was investigated through interviews and medical examinations. Methods: Seven-year mortality was studied by neural network analysis using the SPSS module Clementine 9.0 and by standard logistic regression analysis. The guiding problem was as follows: How do examples of ANN analyses perform compared with logistic regression models in the analyses of one set of social variables and one set of biological and health variables from the same individuals? Result: The ANN generally did not perform better than logistic regression analyses, but in the data set with biological and health predictors, some ANN analyses produced much better results than logistic regression models when odds ratios were compared. Discussion and Conclusion: The ANN can be used as a heuristic method to evaluate if there are hidden structures in data that are not revealed by regression methods and thus call for further analyses, either by more adequate regression models or by other methods. The ANN models could be used as predictors of outcomes of multifactor genesis, which has not been well understood using other investigation methods.

Ämnesord

SAMHÄLLSVETENSKAP  -- Sociologi (hsv//swe)

SOCIAL SCIENCES  -- Sociology (hsv//eng)

Nyckelord

Survival

logistic regression

artificial neural networks

random sample

Publikations- och innehållstyp

vet (ämneskategori)

art (ämneskategori)