Other studies have identified factors associated with duration of ventilation, then applied them as predictive criteria and analyzed their classic predictive performance characteristics. With this methodology, Troche et al. [7] reported that the Lung Injury Score may have high negative predictive value for surgical patients to need prolonged ventilation. In a study limited to a surgical ICU population, they followed 195 patients and analyzed multiple candidate variables including diagnosis leading to mechanical ventilation and severity of illness scores assessed at the time of admission and intubation. Only the need for emergent intubation and the Lung Injury Score were independently associated with duration of ventilation > 14 days. In their subsequent validation cohort of 128 patients requiring emergent intubation, a Lung Injury Score ≥1 predicted >14 days of ventilation with sensitivity of 0.88, specificity of 0.28, positive predictive value of 0.24, and negative predictive value of 0.91. In contrast, Clark et al. [8] found a high positive predictive value to require prolonged ventilation for medical patients when four of their predictive criteria were met. In their retrospective study of 99 medical ICU patients, excluding those dying before day 14, 27 common clinical and laboratory variables (diagnosis not included) were collected from the day of intubation. By multivariate analysis, intubation in the ICU, heart rate >100/min, blood urea nitrogen (BUN) > 25 mg/dl, creatinine> 2 mg/dl, pH < 7.25, and HCO₃ < 20 mEq/l were each associated with a duration of ventilation ≥14 days. A predictive model consisting of the number (0–4) of these criteria met, applied to the same derivation sample, resulted on sensitivity of 0.16, specificity of 1, positive predictive value of 1, and negative predictive value of 0.72 when four criteria were met. Lower numbers of criteria met resulted in progressively higher sensitivity and lower specificities. The area under the receiver operating characteristics (ROC) curve for this model was 0.75.

The largest study reported to date suggested that both diagnosis category and the degree of physiologic derangement could be important predictors [9]. This study not only aimed to identify predictive factors but also to use them to develop an equation to predict the precise duration of mechanical ventilation. In this retrospective analysis of the APACHE III database prospectively collected from 40 hospitals’ ICUs, 5,915 patients who were on mechanical ventilation on their first ICU day had many variables extracted from that day. The total duration on the ventilator was precisely measured for patients spending ≤ 7 days on the ventilator, while it was estimated for patients with longer durations. Of 11 variables that were found to be independently associated with mechanical ventilation duration, the primary reason for ICU admission (selected from the 78 APACHE III disease categories) and the Acute Physiology Score (a component of the APACHE III score) accounted for most of the relative contributions to this association. An equation to predict precise duration on mechanical ventilation was then developed by the authors. In internal cross-validation, this equation was shown to be accurate (R² 0.94) to predict average duration in patient groups classified by illness severity, but inaccurate (R² 0.18) in individual patients.