Computation modes of multivariate positive predictive characteristics

Marek Malik, Timothy R Church

Research output: Contribution to journalArticlepeer-review

1 Scopus citations


Multivariate receiver operator characteristics (ROCs) and positive predictive characteristics (PPCs), based on combination of two or more clinical variables, are usually computed by varying dichotomy limits for each variable independently. This approach has o similar number of degrees of freedom (i.e., uses the same number of programmable parameters) as the approach which defines test positive cases based on a linear combination of all the clinical variables involved. Either approach can be implemented without any assumption about the underlying probability distributions by using an exhaustive computer search. Both approaches were compared in a demonstration study of predicting 2-year all-cause mortality after acute myocardial infarction, based on applying various time- and spectral-domain indices of signal-averaged ECGs from a research survey. The results showed that the optimum mode for the computation of ROCs and PPCs depends on the character of the particular data used. Therefore, in order to increase the precision of the retrospective multifactoral studies, both approaches to ROC and PPC computation should be used and compared in each individual investigation.

Original languageEnglish (US)
Pages (from-to)1708-1713
Number of pages6
JournalPACE - Pacing and Clinical Electrophysiology
Issue number6
StatePublished - Jul 18 1997


  • Positive predictive accuracy
  • Sensitivity
  • Specificity
  • optimization
  • risk stratification


Dive into the research topics of 'Computation modes of multivariate positive predictive characteristics'. Together they form a unique fingerprint.

Cite this