Daly et al developed a model based on logistic regression to identify patients who may have been discharged from the intensive care unit (ICU) too early. The goal is to identify patients who may benefit from a longer period of intensive care. The authors are from St. Thomas's and St. George's Hospitals in London.

Patient selection (from the ICU at Guy's Hospital):

(1) Patients in the ICU for more than 3 days.

(2) Patients had been at risk for death within 48 hours before discharge.

(3) Excluded patients who died in the ICU or who were discharged to the ward as terminal.



(1) age

(2) chronic health points (from APACHE II)

(3) acute physiology points (from APACHE II)

(4) cardiac surgery

(5) length of stay in the ICU



• I could not find the time for recording the acute physiology points. My guess is that it is at the time of discharge from the ICU.





cardiac surgery performed prior to admission to the ICU







X =

= (0.0532 * (age in years)) + (0.2501 * (chronic health points)) + (0.1556 * (acute physiology points)) – (2.1084 * (points for cardiac surgery)) + (0.0447 * (number of days in ICU)) – 4.5821


probability value =

= 1 / (1 + EXP((-1) * X))



• A cutoff of >= 0.60 (60%) gave the best sensitivity (65.5%) and specificity (87.6%) for identifying a high risk group for post-discharge mortality.

• In the study group, mortality was reduced by 39% if patients were kept in the ICU two additional days. Keeping these patients for the additional time would have increased bed usage 16% (page 4, from eBMJ).



• The probability increases as the number of days increases in the equation. Since age, chronic health points and cardiac surgery are constant, the only other variable is acute physiology points. The supposition is that a person's acute physiology points will decrease with the additional ICU days.

To read more or access our algorithms and calculators, please log in or register.