A prospective, multicenter, observational cohort study with the participation of 13 polyvalent ICUs was carried out. The Clinical Research Ethics Committee (CREC) of the coordinating center (Cuenca Health Area CREC #10/09) approved the study and exempted it from the need to obtain informed consent, in view of the anonymous and observational nature of the survey. Such approval in turn was extended to the rest of the participating institutions.

Between 1 June 2006 and 31 May 2008, we included all adults admitted to any of the participating ICUs and who required invasive mechanical ventilation for more than 24h. The exclusion criteria were: patients under 18years of age; limitation of therapeutic effort considerations prior to inclusion; patients participating in some other study involving some intervention capable of influencing the results obtained; and critical burn patients.

Prolonged mechanical ventilation was defined as the need for invasive mechanical ventilation for 7 or more days. A patient was considered to be free from ventilation when remaining under spontaneous ventilation for more than 48h after extubation. Those patients requiring reintubation during the first 48h were not regarded as being free of invasive ventilatory support. In turn, those requiring reintubation more than 48h after extubation were defined as “late reintubation” cases. All the patients were followed-up on until death or hospital discharge.

During the first 24h of invasive mechanical ventilation we documented the following variables: the Acute Physiology and Chronic Health Evaluation II (APACHE II) score, Sequential Organ Failure Assessment (SOFA) score, age, gender, localization prior to admission to the ICU (Emergency Department, clinical ward, surgical ward, operating room, intermediate care unit awakening/resuscitation room, other hospital), failure of noninvasive ventilation (NIV) prior to endotracheal intubation, comorbidity as established by the Charlson comorbidity index,13 functional capacity according to the Barthel index,14 duration of stay in hospital prior to admission to the ICU, the need for vasoactive drugs before admission to the ICU, and the reason for mechanical ventilation according to the International Study Group mechanical ventilation classification15,16: 1. – Acute respiratory failure: 1.a. – Postoperative: patients requiring continued mechanical ventilation after surgery due to risk derived from underlying disease, age, or high risk surgery; 1.b. – Pneumonia: defined by the development of new alveolar infiltration or worsening of previous alveolar infiltration, accompanied by fever/hypothermia and leukocytosis/leukopenia; 1.c. – Heart failure: patients with dyspnea, bilateral alveolar infiltration, hypoxemia and evidence of heart disease or patients with cardiogenic shock; 1.d. – Acute lung injury/acute respiratory distress syndrome (ALI/ARDS): according to the criteria of the American-European Consensus Conference.17 1.e. – Sepsis: according to the criteria of the consensus conference of the American College of Chest Physicians/Society of Critical Care Medicine18; 1.f. – Trauma: mechanical ventilation due to polytraumatism; 1.g. – Cardiac arrest: mechanical ventilation due to sudden and unexpected cardiopulmonary arrest. 2. – Coma: patients requiring mechanical ventilation due to worsened consciousness secondary to organic (cerebrovascular stroke, meningoencephalitis, head injuries) or toxic-metabolic causes. 3. – Exacerbated chronic respiratory failure: patients with basal chronic disease requiring mechanical ventilation (exacerbated chronic obstructive pulmonary disease (COPD), asthma, chronic respiratory disease other than COPD). 4. – Neuromuscular disease: patients with respiratory failure due to alterations of the peripheral nervous system, neuromuscular junction or myopathy. 5. – Other causes: mechanical ventilation due to other causes.

A univariate analysis was performed to compare the variables between the patients ventilated for <7days and those ventilated for ≥7days. The Student t-test was used to explore differences between means. In turn, the Mann–Whitney U-test was applied to those variables with a non-normal distribution, while the chi-squared test was used for evaluating differences between qualitative variables.

A multivariate risk model was developed in order to identify variables capable of differentiating between patients with a high probability of requiring prolonged mechanical ventilation (≥7days) and those with a high probability of early extubation (<7days) or of death before day 7 of mechanical ventilation. As response we considered an outcome variable with three categories corresponding to the three possible conditions to be taken into account:

• 1.

  Early mortality (fatal outcome before day 7 of mechanical ventilation).

• 2.

  Early extubation (extubation before day 7 of mechanical ventilation).

• 3.

  Prolonged mechanical ventilation (mechanical ventilation for 7 or more days).

As reference category we used prolonged mechanical ventilation, and estimated the coefficients for each of the two models obtained, as well as the predicted probabilities of each condition. The model excluded those variables which the Wald statistic did not consider to make a significant contribution to the feasibility of the multivariate model.

The global sample was divided into two sub-samples on a randomized basis: a training or design subsample and a validation subsample. The former was used to develop the model, and an evaluation was made of its discriminating capacity in the validation subsample.

The parameters of the multivariate model were estimated based on multinomial logistic regression analysis, within the context of the generalized linear models. The generalized logit function was used as the linking function.

The discriminating capacity of the model was analyzed following the methodology of Obuchowski et al.,19 for the case in which the response has multiples conditions, consisting of an extension of the Wilcoxon–Mann–Whitney estimator of the area under the receiver operating characteristic (ROC) curve. The estimators were obtained by pairs (death vs early extubation; death vs prolonged mechanical ventilation; extubation vs prolonged mechanical ventilation) among the three conditions to estimate the discriminating capacity between conditions–the global discriminating capacity being regarded as the weighted mean of the partial indices. Weighting was based on the relative prevalences of each condition in the sample, and we studied two functions referred to error penalization in distinguishing between two conditions: firstly, maximum penalization due to classification error between any of the conditions, and secondly no penalization of erroneous distinction for death versus early extubation and maximum penalization due to erroneous distinction between death versus prolonged mechanical ventilation and early extubation versus prolonged mechanical ventilation.

Calculation of the probability corresponding to each of the three conditions was based on the following equations:

• •

  Prob (early mortality/X1)=exp(α1+(Σβ1×X1))/1+[exp(α1+(Σβ1×X1))+exp(α2+(Σβ2×X2))].

• •

  Prob (early extubation/X1)=exp(α2+(Σβ2×X2))/1+[exp(α1+(Σβ1×X1))+exp(α2+(Σβ2×X2))]

• •

  Prob (prolonged mechanical ventilation/X1)=1/1+[exp(α1+(Σβ1×X1))+exp(α2+(Σβ2×X2))]

The statistical analysis was carried out using the SAS statistical package (SAS Institute, release 9.1), with the nonbinROC application of the language program R version 2.9.1 (2009-06-26) (Copyright (C) 2009. The R Foundation for Statistical Computing, ISBN 3-900051-07-0).

Of the 13 participating centers, four were university hospitals and 12 were qualified for postgraduate (resident) training in Intensive Care Medicine. The mean number of beds per participating hospital was 625 (range (r)=250–1450), while the mean number of beds per ICU was 15 (r=8–44). A total of 1661 patients were included. Of these, 58.3% (n=969) received prolonged mechanical ventilation during 20.5±16.8days. In turn, 31.1% (n=517) were extubated before day 7, with a mean duration of mechanical ventilation of 3.4±1.3days. A total of 10.5% (n=175) died before day 7 in the ICU, with a duration of mechanical ventilation of 3.5±1.4days. The rest of the demographic and clinical characteristics are shown in Table 1. The main indications of mechanical ventilation were acute respiratory failure (68.8%, n=1144), coma (17.6%, n=294), exacerbated chronic respiratory failure (7.7%, n=129), neuromuscular disease (1.1%, n=19), and other causes (4.8%, n=75) (Table 2).

The total APACHE II and SOFA scores were higher among the patients subjected to prolonged ventilation than in those on mechanical ventilation for less than 7days (p=0.04 and p=0.0001, respectively). Noninvasive ventilation failure was associated to prolonged mechanical ventilation (p=0.005) (Table 3).

A total of 1289 patients (77.6%) were assigned to the training sample and 372 (22.4%) to the validation sample. A multivariate model was generated with the following variables: APACHE II, SOFA, NIV failure, hospital localization before admission to the ICU (intermediate care unit, anesthetic awakening/resuscitation room), reason for mechanical ventilation: ALI/ARDS, heart failure, pneumonia, sepsis, polytraumatism, cardiac arrest, metabolic coma, intoxication coma, cerebrovascular stroke, head injury, neuromuscular disease, and other indications of mechanical ventilation. The model with its coefficients and significance is shown in Table 4.

The discriminating capacity of the model is expressed in Table 5, which shows the areas under the curve of the predictive probabilities of one condition versus another, in each of the two samples (training and validation), together with the global precision. Fig. 1 in turn presents the ROC curves, showing: probability 1 (early mortality vs prolonged mechanical ventilation) and probability 2 (early withdrawal vs prolonged mechanical ventilation) of the training and validation samples.

Figure 1.

ROC curves showing the area under the curve (AUC) of one condition versus another in training and validation samples. A. Probability 1 (early mortality vs prolonged mechanical ventilation). B. Probability 2 (early extubation vs prolonged mechanical ventilation).

(0.25MB).

The likelihood ratios (LRs) referred to the optimum cutoff points were as follows in the early extubation model: LR (+): 2.37 (95%CI 1.77–3.19), LR (−): 0.47 (95%CI: 0.41–0.55) for a cutoff point of 0.65 in the training sample; and: LR (+): 2.2 (95%CI 1.7–2.8), LR (−): 0.41 (95%CI: 0.31–0.55) for a cutoff point of 0.71 in the validation sample.

In turn, LRs referred to the optimum cutoff points were as follows in the early mortality model: LR (+): 2.64 (95%CI: 2.01–3.4), LR (−): 0.39 (95%CI: 0.30–0.51) for a cutoff point of 0.73 in the training sample; and: LR (+): 3.02 (95%CI: 2.15–4.22); LR (−): 0.48 (95%CI: 0.32–0.73) for a cutoff point of 0.61 in the validation sample.

In order to facilitate the calculation of each probability, a spreadsheet was developed for computation of the probabilities of each condition. This tool (.xls) is shown in Appendix A.