Skip to main content

Table 3 Generalised linear model obtained in our study

From: Model for predicting short-term mortality of severe sepsis

Main effect

Beta estimate

Odds ratio

95% CI

P value

Intercept

-4.9419

-

< 10-4

Parameters on the day of severe sepsis

   

   LOD (per point)

0.1951

1.22 (1.16 to 1.27)

< 10-4

   Septic shock

0.3335

1.40 (1.08 to 1.81)

0.01

   First episode of severe sepsis

-

-

-

   Second episode of severe sepsis

0.2304

1.26 (0.96 to 1.66)

0.10

   Third or fourth episode of severe sepsis

0.9719

2.64 (1.71 to 4.08)

< 10-4

   Multiple sites of infection

0.3734

1.45 (1.04 to 2.03)

0.03

Variables at ICU admission

   

   SAPS (per point)

0.0244

1.02 (1.01 to 1.03)

< 10-4

   Fatal illness by McCabe Score(score 2 or 3)

0.6749

1.96 (1.43 to 2.70)

< 10-4

   No chronic illness

-

-

-

   Exactly one chronic illness

0.5592

1.75 (1.25 to 2.45)

0.001

   Two or more chronic illnesses

0.8084

2.24 (1.39 to 3.62)

0.001

  1. The area under the Receiver-Operating Characteristics curve was 0.822 and the Hosmer-Lemeshow chi-squared test was 8.6 (P > 0.05, 8 df), indicating good discrimination and good calibration of the final model in the training cohort. The following variables were tested in the generalized linear model: Logistic Organ Dysfunction (LOD), Sequential Organ Failure Assessment (SOFA), septic shock, high-dose vasoactive drugs (epinephrine and/or norepinephrine > 0.1 γ/kg/min), multiple sites of infection, Simplified Acute Physiology Score (SAPS) II, age, number of chronic organ failures (none, exactly one or two or more), arterial, central venous line or Swan-Ganz catheter, diagnosis at intensive care unit (ICU) admission, year of admission, centre, early effective antibiotic therapy, corticosteroid therapy, male gender, main symptom (multiple organ failure and cardiogenic shock), metastatic cancer, mechanical ventilation, urinary tract catheter, sedation, extrarenal replacement therapy, McCabe score, nature of the microorganism (E. coli, Candida species and methicillin-susceptible S. aureus), infection site and LOD increase from the day before to the day of severe sepsis diagnosis.
  2. To calculate the predicted risk of death for each patient:
  3. - compute the logit: logit = sum ('Beta estimate' multiplied by value of corresponding parameter)
  4. - compute the probability, using the logit: P = (exp (logit)) divided by (1+exp(logit))