Open Access

Incorrect predictions, not incorrect statistics!

  • Reinier G Hoff1Email author,
  • Thomas Lumley2,
  • Gabriel JE Rinkel3,
  • Bon H Verweij4,
  • Cor J Kalkman1 and
  • Ale Algra3, 5
Critical Care200913:406

https://doi.org/10.1186/cc7773

Published: 29 April 2009

In a recent study on nurses' ability to predict volume status after subarachnoid haemorrhage, we found very low predictive values [1]. In a letter to the editor, Dane [2] argued that an error in statistics 'makes any reasonable conclusions impossible'. Because Dr Dane was the statistical reviewer of our manuscript, both the authors and the editors were aware of Dane's comments on this issue well before publication.

The substantive point made by Dane is that the standard errors (SEs) estimated in our paper may not be correct. We studied 350 combinations of volume predictions and measurements, obtained in 43 patients with about 170 participating nurses. Therefore, a certain amount of correlation exists, as we explicitly acknowledged in the report. The impact of multiple measurements per patient will be to overestimate, and the impact of multiple measurements per nurse will be to underestimate SEs somewhat. Even if 'within-nurse' correlations were perfect and each nurse made as many as four assessments, the SEs would not increase by more than a factor of two. In any realistic scenario the impact would be much smaller.

More importantly, all results presented in Table 2 of our report and all point estimates of predictive accuracy in Table 3 would remain completely unchanged if the identities of the nurses were recorded and used in analyses. These tables show that the relationship between volume and assessment is weak enough that the precise level of statistical uncertainty in the estimates is beside the point. Dane's statistical arguments are therefore irrelevant to the clinical situation. We maintain our conclusion that nurses' assessment does not adequately predict hypovolaemia or hypervolaemia after subarachnoid haemorrhage.

Authors' information

RGH is an anesthesiologist-intensivist at the UMC Utrecht, The Netherlands. TL is an Associate Professor in Biostatistics at the University of Washington, Seattle, WA, USA. GJER is a Professor in Neurology at the UMC Utrecht, The Netherlands. BHV is a neurosurgeon at the UMC Utrecht, The Netherlands. CJK is a Professor in Anesthesiology at the UMC Utrecht, The Netherlands. AA is a Professor in Clinical Epidemiology both at the UMC Utrecht and at the Leiden UMC, The Netherlands.

Declarations

Authors’ Affiliations

(1)
Department of Perioperative & Emergency Care, Rudolf Magnus Institute of Neuroscience, University Medical Center Utrecht
(2)
Department of Biostatistics, University of Washington
(3)
Department of Neurology, Rudolf Magnus Institute of Neuroscience, University Medical Center Utrecht
(4)
Department of Neurosurgery, Rudolf Magnus Institute of Neuroscience, University Medical Center Utrecht
(5)
Julius Center for Health Sciences and Primary Care, University Medical Center Utrecht

References

  1. Hoff RG, Rinkel GJE, Verweij BH, Algra A, Kalkman CJ: Nurses' prediction of volume status after aneurysmal subarachnoid hemorrhage: a prospective cohort study. Crit Care 2008, 12: R153. 10.1186/cc7142PubMed CentralView ArticlePubMedGoogle Scholar
  2. Dane FC: Type II errors in 'Nurses' prediction of volume status after aneurysmal subarachnoid hemorrhage: a prospective cohort study'. Crit Care 2009, 13: 402. 10.1186/cc7705PubMed CentralView ArticlePubMedGoogle Scholar

Copyright

© BioMed Central Ltd 2009

Advertisement