[Prediction of outcome in critically ill patients using haemodynamic and oxyphoretic parameters]

To find a predictive model for short time mortality and survival of haemodinamically instable patients.

Prospective study on two consecutive series of critically ill patients admitted to ICU.

115 critically ill patients, subdivided in two series of 83 (47 survivors and 36 non survivors after 20 days from admission) and 32 patients (19 survivors and 13 non survivors), respectively. INTERVENTIONTS: All the patients have been monitored with a Swan-Ganz catheter.

Oxyphoretic parameters were measured at 7 different times (T): T0 (start), T1 (after 12 hours from T0, T2 (24 hrs), T3 (48 hrs), T4 (72 hrs), T5 (96 hrs) and T6 (120 hrs). A total number of 401 recorded values of the first patients' series were used to create the predictive model for outcome (0 non survivors, 1 survivors). Stepwise logistic regression was used to create the model. Model calibration and discrimination was assessed. The model was validated using data collected from the second series.

The probability of survival after 20 days was Pr (survive) = e(logit)/1+e(logit), with "logit" = -5,106+(-6,58E-02 SVI)+(-2,76E-03 SVR)+(0,1379 LVSWI)+(0,8933 Hb)+(-3,25E-02 VO2I)+(9,09E-02 O2ER)+(-7,89E-02 SHUNT)+PAT. For medical patients PAT = -0.568; for surgical patients PAT = -1.0525 and for politrauma PAT = 1.593. Goodness-of-fit test showed a good calibration: chi 2 = 4.267, p = 0.832. The area under the ROC curve was 0.831. The model used in the validation data set, with 199 recorded values, also showed a good calibration and discrimination (Goodness-of-fit test: chi 2 = 15.65, p = 0.111; area under the curve 0.798).

The mathematical model we found has been validated also in the second series and the discrimination capability increases with time. Using this model we can evaluate the probability of survival at every time. Its application at different times permits a better evaluation of haemodynamically instable patient trend.