Skip to content

Advertisement

  • Poster presentation
  • Open Access

Predicting the response to therapy from a mathematical model

  • 1,
  • 2,
  • 3,
  • 2,
  • 3 and
  • 2
Critical Care20048 (Suppl 1) :P206

https://doi.org/10.1186/cc2673

  • Published:

Keywords

  • Statistical Model
  • Thrombin
  • Tissue Factor
  • Host Response
  • Bacterial Load

Objectives

To determine the feasibility and usefulness of computer simulations in evaluating therapeutic strategies and patient selection in clinical trials of sepsis.

Methods

We simulated an interventional trial of a neutralizing body against tissue necrosis factor (anti-TNF) in sepsis based on a mechanistic mathematical model that includes a bacterial infection, the host response, and a therapeutic intervention. Simulated cases differed by bacterial load and virulence, as well as individual propensity to mount and modulate an inflammatory response. We submitted 1000 cases to three doses and durations of anti-TNF, and present the results of the simulation. To evaluate the usefulness of modeling to improve patient selection, we constructed a logistic model with a four-valued outcome: (1) helped by treatment, (2) survives irrespective of treatment, (3) dies irrespective of treatment, and (4) harmed by treatment. Independent predictors were 'measured' at the time of disease detection and 60 min later, and included serum TNF, IL-10, IL-6, activated protein C, thrombin, tissue factor, blood pressure and cell counts. All results from the statistical model are reported in a validation cohort.

Results

Control survival was 62.9% at 1 week. Depending on the dose and duration of treatment, survival ranged from 57.1% to 80.8%. Higher doses of anti-TNF, although effective, also resulted in considerable harm (Fig. 1). Discrimination of serum IL-6 to identify cases harmed by treatment was moderate with ROC = 0.72, compared with 0.98 for a statistical model. Following the statistical model's recommendation for treatment, only 26.5% of the cohort would have received treatment and 2.4% patients would have been harmed, whereas indiscriminate administration would have harmed 14.8% of patients.

Figure 1

Conclusions

Our models points out how to improve patient selection and how therapy could be individualized.

Authors’ Affiliations

(1)
UPMC, Pittsburgh, Pennsylvania, USA
(2)
University of Pittsburgh, Pennsylvania, USA
(3)
Immunetrics, Inc, Pittsburgh, Pennsylvania, USA

Copyright

© BioMed Central Ltd. 2004

Advertisement