The purpose of this study was to derive an early, highly sensitive and spec
ific prediction rule for the development of post-traumatic acute lung injur
y (ALI). In a prospective, non-interventional study a convenience sample of
92 adults admitted to the resuscitation room following blunt trauma was st
udied in order to derive this prediction rule. The study was conducted in t
he emergency department of a university hospital in the New Territories of
Hong Kong. One emergency physician assessed each patient for 25 variables,
which might predict post-traumatic ALI, and the primary outcome measure was
the presence or absence of ALI 48 h post-injury. Eleven variables associat
ed with ALI were entered into a classification and regression tree (CART) i
n order to derive models predictive of ALI. Two models were developed and u
sed to derive the decision rule. Acute lung injury was likely if either: (1
) the patient had an ISS > 27 and a haematocrit < 0.37, or (2) the patient
had a haematocrit < 0.36 and a total leucocyte count > 15. The first guidel
ine had a classification rate of 96.7% (95% confidence interval (CI), 90.8-
99.3%), a sensitivity of 100% (CI 65.2-100%) and a specificity of 96.5% (CI
90.0-99.3%). The second guideline had a classification rate of 96.7% (CI 9
0.8-99.3%), a sensitivity of 85.7% (CI 42.1-99.6%) and a specificity of 97.
7% (CI 91.8-99.7%). Practical highly sensitive and specific prediction guid
elines for post-traumatic acute lung injury have been derived and now requi
re prospective validation. (C) 1999 Elsevier Science Ireland Ltd. All right
s reserved.