Survival prediction in the intensive care unit: A comparison of neural networks and binary logit regression

Ernest Preston Goss, Harish Ramchandani

Research output: Contribution to journalArticlepeer-review

6 Scopus citations


As a component of hospital cost containment policy, the National Institutes for Health recommends that hospitals limit intensive care unit (ICU) resources to patients who have a reasonable probability of recovery. In order to provide more ethical and objective measures of the likelihood of ICU recovery, hospitals have turned increasingly to decision support software such as APACHE (acute physiology and chronic health evaluation), which derives predictions from binary logit regression (BLR), a parametric estimation technique. Rapid advancements in computer software and hardware technology have encouraged researchers to use more computationally intensive non-parametric techniques such as neural networks (NNs) whose prediction capabilities are purported to be greater than those of parametric models. The present study applies this methodology to a sample of ICU patients and shows that neural networks appear to more correctly predict survival than does BLR.

Original languageEnglish (US)
Pages (from-to)189-198
Number of pages10
JournalSocio-Economic Planning Sciences
Issue number3
StatePublished - 1998

All Science Journal Classification (ASJC) codes

  • Geography, Planning and Development
  • Economics and Econometrics
  • Strategy and Management
  • Statistics, Probability and Uncertainty
  • Management Science and Operations Research


Dive into the research topics of 'Survival prediction in the intensive care unit: A comparison of neural networks and binary logit regression'. Together they form a unique fingerprint.

Cite this