Skip to content

Advertisement

  • Poster presentation
  • Open Access

Three generations of mortality prediction models: accuracy for outcome prediction in the critically ill obstetric patient

  • 1 and
  • 2
Critical Care200610 (Suppl 1) :P400

https://doi.org/10.1186/cc4747

  • Published:

Keywords

  • Hospital Mortality
  • Discriminatory Power
  • Retrospective Chart Review
  • Pregnant Patient
  • Prospective Data

Introduction

Mortality prediction models (MPM) [13] are generalistic severity of illness scoring systems. No score is computed, and a logistic regression equation directly provides a probability of hospital mortality. Three generations of MPM scores are already available, and assess mortality at admission to the ICU (first hour).

Objective

To determine the accuracy of the different MPM systems in the critically ill obstetric patient.

Patients and methods

Prospective data collection of the parameters to calculate MPM1-H0 and MPM2-H0 [1, 2], and retrospective chart review of one of two additional parameters necessary to calculate MPM3-H0 [3] from the MPM2-H0 score [2]. Study period: January 1996–December 2003. Inclusion criteria: all obstetric patients hospitalized in a multidisciplinary ICU and with a length of stay >1 hour. Exclusion criteria: same as those published in the original references [1, 2].

Performances of the three systems were compared, using the area under the receiver–operator characteristic curve (AUROC) to assess the discriminatory power and the Hosmer–Lemeshow (HL) goodness-of-fit test for calibration. Data was computed on R version 2.1. P < 0.05 was considered significant.

Results

See Table 1.

Table 1

 

MPM1

MPM2

MPM3

AUROC

0.801

0.853

0.885

HL

0.036

0.156

0.37

Discussion and conclusion

There is no significant difference in the performances of MPM2 and MPM3. They are clearly better than the oldest model. The two new parameters included in the MPM3 did not significantly influence the performance of the system. Many reasons could explain these findings: when computing the MPM3 system, addition of a 'zero–factor' term [3] for patients with no risk factors other than age does not improve model performance in our population, because our database is composed of young woman issued from a homogeneous case mix. The fact that the 'Full Code' factor was assessed retrospectively and that we deal with young pregnant patients where very few Do Not Resuscitate orders were given explains that this parameter got little influence. We can conclude that MPM1 is outdated, and MPM3 tends to be better than the previous version without having a statistically significant difference. Adding known prognostic factors not included in MPM systems could have enhanced performances of MPM3 in our particular case mix.

Authors’ Affiliations

(1)
Groupe Hospitalier Pitié Salpêtrière, Paris, France
(2)
National Institute of Neurology, Tunis, France

References

  1. Crit Care Med. 1988, 16: 470-477. 10.1097/00003246-198805000-00002Google Scholar
  2. JAMA. 1993, 270: 2478-2486. 10.1001/jama.270.20.2478Google Scholar
  3. Crit Care. 2005,9(Suppl 1):S97. 10.1186/cc3160Google Scholar

Copyright

© BioMed Central Ltd 2006

Advertisement