Statistical methods for comparing mortality among ESRD patients: Examples of regional/international variations

Edward F. Vonesh, Douglas E. Schaubel, Wenli Hao, Allan J. Collins

Research output: Contribution to journalArticlepeer-review

45 Scopus citations

Abstract

There have been a number of recent large-scale registry studies comparing mortality between patients receiving hemodialysis (HD) and peritoneal dialysis (PD). Results from these studies are mixed with some indicating a difference in mortality in favor of hemodialysis, others finding no difference in mortality and still others observing a difference in mortality in favor of peritoneal dialysis. Much of the apparent discrepancy between studies might be attributed to differences in study design and analytical methods. In this paper, we review and summarize three key methodologies used in the analysis of patient survival data. These are 1) the type of statistical model used to compare mortality (Cox proportional hazards regression versus Poisson regression); 2) the type of analysis used (intent-to-treat versus as-treated); and 3) the type of patient to be studied (prevalent versus incident). The impacts these methodologies have on the results of patient survival analysis are illustrated using national registry data from the U.S. and Canada. We demonstrate that when applied under the same conditions, survival analysis using Poisson regression versus Cox regression yields essentially equivalent results. However, results of survival analysis may or may not differ according to whether one uses an intent-to-treat versus an as-treated analysis, and they almost certainly will differ when carried out on prevalent versus incident patients. We also demonstrate that when analyzed using the same methodology, results obtained in different countries can be quite similar. One can use either Poisson or Cox regression to carry out patient survival analysis and still achieve similar results. In cases where one suspects that the relative risk of death between PD and HD varies with time, an interval Poisson or interval Cox regression that includes a modality by time interaction term is recommended. Other recommendations for standardizing patient survival analysis include performing both an intent-to-treat and as-treated analysis and confining such analyses to incident-only patients.

Original languageEnglish (US)
Pages (from-to)S19-S27
JournalKidney International, Supplement
Volume57
Issue number74
DOIs
StatePublished - 2000

Keywords

  • Cox proportional hazards
  • End-stage renal disease
  • Mortality statistics
  • Poisson regression
  • Relative risk

Fingerprint

Dive into the research topics of 'Statistical methods for comparing mortality among ESRD patients: Examples of regional/international variations'. Together they form a unique fingerprint.

Cite this