Clinical model for predicting prolonged mechanical ventilation.pdf

Clinical model for predicting prolonged mechanical ventilation, Paul A. Clark DO a, Christopher J. Lettieri MDb,c aDepartment of Medicine, Walter Reed National Military Medical Center, Bethesda, MD 20889, USA bPulmonary, Critical Care and Sleep Medicine, Walter Reed National Military Medical Center, Bethesda, MD 20889, USA cDepartment of Medicine, Uniformed Services University, Bethesda, MD, USA Keywords: Mechanical ventilatory support; Prolonged mechanical ventilation; Tracheostomy; Predictive model; Medical intensive care unit Abstract Introduction: Mechanical ventilation (MV) predisposes patients to numerous complications, which increases with longer durations of treatment. Identifying individuals more likely to require prolonged MV (PMV) may alter ventilation strategies or potentially minimize the duration of therapy and its associated complications. Our aim was to identify clinical variables at the time of intubation that could identify individuals who will require PMV. Methods: One hundred thirty consecutive adult patients requiring MV support in a medical intensive care unit (ICU)were retrospectively assessed. Prolonged MV was defined as MV support more than 14 days. Results: Mean age was 62.3 ±21.1 years, 64.6% were men, and mean duration of MV support was 11.4± 11.9 days. Prolonged MV was required in 31.3%. Requiring intubation after admission to the ICU, heart rate greater than 110, blood urea nitrogen more than 25 mg/dL, serum pH less than 7.25, serum creatinine more than 2.0 mg/dL, and a HCO3less than 20 mEq/L were the only variables independently associated with PMV. Specificity for predicting PMV was 100% with 4 or more of these variables. Conclusion: The novel predictive model, using Intubation in the ICU, Tachycardia, Renal dysfunction, Acidemia, elevated Creatinine, and a decreased HCO3, was highly specific in identifying patients who subsequently required PMV support and performed better than Acute Physiology Age Chronic Health Evaluation III. Published by Elsevier Inc. 1. Introduction The requirement for prolonged mechanical ventilation (PMV) support is common among intensive care unit (ICU) patients and is predicted to increase to 625298 patients per year by 2020 1. Prolonged mechanical ventilation increases the risk for complications and is often difficult to predict. Approximately 5% to 25% of all mixed ICU patients acutely requiring mechanical ventilation (MV) will remain intubated for more than 5 days, and 3% to 14% will require PMV, typically described as more than 14 or more than 21 days of MV support 2-9. The prevalence of PMV among a pure medical ICU (MICU) population has not been well defined but is likely more common. Financial support: None. Disclosure: None. Corresponding author. Tel.: +1 301 295 4630. E-mail address: Paul.albert.clarkus.army.mil (P.A. Clark). 0883-9441/$ see front matter. Published by Elsevier Inc. http:/dx.doi.org/10.1016/j.jcrc.2013.03.013 Journal of Critical Care (2013) 28, 880.e1880.e7 Although lifesaving, MV predisposes patients to numer- ous complications, such as ventilator-induced lung injury, ventilator-associated pneumonia, and other nosocomial in- fections, venothrombotic events, pressure ulcerations, gastri- tis, and increased length of stay 10-12. The probability of developing these complications increases with longer durations of MV support in a near linear relationship 13. To help mitigate these risks, several strategies have been developed and implemented to reduce the duration of MV. These often encompass minimizing sedation and performing daily spontaneous breathing trials to facilitate liberation from MV support as soon as the reason for intubation resolves 14,15. However, not all conditions requiring MV will resolve quickly. In these instances, aggressive measures to minimize the number of ventilator days may actually be counterproductive; lead to more pain, discomfort, or anxiety; andmayincreaseintubationtime16,17.Earlyidentification of those individuals who will ultimately require PMV would likely alter traditional ventilator and sedation management or identify those who may benefit from early tracheostomy. Unfortunately, there are limited models that can reliable predict PMV, especially on an individual patient basis. Although the Acute Physiology Score (APS) and Acute Physiology Age Chronic Health Evaluation III (APACHE III) score serve as a reliable index of disease severity and predictor of mortality, it is cumbersome to use in clinical practice. More importantly, Acute Physiology Age Chronic Health Evaluation and APS were not developed as predictors of PMV. Seneff et al 13 attempted to create an equation to predict the duration of MV support using clinical variables. In this study, the authors observed that 20% of patients receiving MV support required PMV, which they defined as more than 7 days. Similar to other studies exploring the duration of MV support, the authors included a mixed population of medical, surgical, and cardiac ICU patients. The prevalence of PMV and identifying factors that could predict longer durations of support among a pure MICU population remains unknown. A predictive model identifying those at risk for PMV could help clinicians better manage MV support. Our study had 2 objectives. First, we sought to determine the prevalence of PMV among a pure MICU population. Second, we attempted to develop an assessment tool using readily available clinical information at the time of intubation to predict patients who would subsequently require MV more than 14 days. 2. Materials and methods 2.1. Study design This retrospective review of all patients intubated in the MICU occurred at a single institution starting in January 2009 extending over a period of 1.5 years. Our institution (Walter Reed Army Medical Center, Washington, DC) is an academic, tertiary care center, with a closed, 16-bed MICU. During the 1.5 years, 231 consecutive patients were intubated in the MICU. Of the entire cohort, 101 MV patients (44%) were excluded for the following reasons: trauma patient, surgical patient boarding in MICU, and 31 patients (13%) who died before 14 days of MV, which resulted in 99 consecutive adult, nonsurgical, MICU patients (43%) requiring MV (Fig. 1). Those who died during the first 14 days on MV support were not included, as we could not determine if they would have required PMV. Otherwise, no records were excluded from the final analysis. The protocol was approved by our institution's Department of Clinical Investigation (Institutional Review Board IRB, Scientific Review Committee, and Human Use Committee, approval no. 364496). No external funding was used. 2.2. Measured variables All data were recorded using a closed electronic medical record system, and all measured variables were collected at the time of intubation. We recorded 27 commonly obtained clinical and laboratory variables in each patient. These included age, sex, weight, body mass index, core body temperature, heart rate (HR), mean arterial blood pressure, respiratory rate, saturation of peripheral oxygen, fraction of inspired oxygen (FIO2), and Glasgow Coma Scale score. Exclusion, N = 101 Trauma patients (N = 73) Surgical boarding patients (N = 28) MICU intubations N = 231 Death before MV day 14 N = 31 PMV N = 31 Non-PMV N = 68 Patients enrolled N = 99 Fig. 1Consort flow diagram. 880.e2P.A. Clark, C.J. Lettieri Laboratory values included serum levels of leukocytes, hematocrit, blood urea nitrogen (BUN), sodium, creatinine, albumin, glucose, total bilirubin, bicarbonate (HCO3), potassium, and lactate. Measurements of arterial pH, PaO2, PCO2, and the PaO2:FIO2ratios were also included. All included laboratory tests were collected within the 2 hours preceding intubation. Whether the patient was in the MICU at the time of intubation (divided into 2 categories: in MICU N1 day before intubation vs admitted for MV and intubated on admission) or admitted to the MICU because of the need for MV also served as an independent variable. Patients who required intubation were in the MICU for more than 1 day, before developing need for intubation. We also included other clinical data not related to the development of a predictive model of PMV. These included the duration of MV, length of MICU stay, length of hospitalization mortality, APS, and APACHE III score. When applicable, the timing of tracheotomy place- ment was recorded. 2.3. End points We defined PMV as patients requiring 14 days or more of MV support. Using a univariate analysis, we compared each measured variable between those receiving 14 or more and 14 days or less of MV to identify which were independently associated with PMV. Variables with a P b .10 during univariate analysis underwent multivariate assessment to develop a model that could predict the need for PMV. Once identified, this model was used to calculate the sensitivity and specificity for predicting PMV in each individual. The included variables were used to calculate an APS and APACHE III score. For internal validation, our predictive model was compared with APS and APACHE III scores. Given that early tracheostomy can reduce the duration of MV support, we applied our model to those who underwent tracheostomies for further validation. Early tracheostomy is considered controversial; when performed, it can increase ventilator-free days but not necessarily change long-term outcomes 18,19. 2.4. Statistical analysis All data are presented as the mean with SD. Comparisons between categorical variables were performed using 2or Fisher exact tests as appropriate. For continuous variables, differences between means were assessed using 2-sample t tests.Weperformedbothunivariateandmultivariateanalyses to identify potential predictors of PMV. All variables that achieved a P value .10 in univariate analysis were entered into a multivariate model. Multivariate modeling was performed using logistic regression. Receiver operating characteristic (ROC) curves were constructed for these variables. Receiver operating characteristic curves were compared via comparison of the area under the curve (AUC). The sensitivity, specificity, positive predictive value, and negative predictive value were determined for each number of positive variables. Data were analyzed using PASW 17.0 (SPSS, Inc, Chicago, Ill). 3. Results Amongthecohort(n=130),themeanagewas62.3±21.1 years, and 64.6% were men (Table 1). The most common indications for intubation were acute respiratory failure (30.0%), sepsis (20.0%), gastrointestinal bleeding (10.0%), cardiac arrest (6.9%), cerebral vascular accident (6.2%), and substance/medication overdose (3.8%) (Table 2). Thirty-one patients (23.8%) died during the first 14 days of MV support and were excluded from the final analysis. Of Table 1Clinical variables of patients receiving mechanical ventilatory support Variablesb14 d (n = 68) 14 d (n = 31) P No. of ventilator days5.1 ± 3.3625.3 ± 12.4b.001 Age (y)61.9 ± 22.363.0 ± 18.2.794 Men (%)60.274.2.093 Weight (kg)82.3 ± 23.087.4 ± 26.2.288 BMI (kg/m2)28.5 ± 7.926.6 ± 6.3.326 Temperature (°F)37.0 ± 1.137.1 ± 0.87.810 Respiratory rate23.3 ± 9.522.5 ± 8.8.588 HR (beats per min)94.4 ± 25.299.6 ± 29.2.069 MAP (mm Hg)92.9 ± 23.688.5 ± 21.4.294 Oxygen saturation (%)95.2 ± 6.596.9 ± 2.9.430 FIO2(%)0.639 ± 0.280.692 ± 0.28.354 PaO2/FIO2243.2 ± 196.5 245.5 ± 194.3.956 WBC (L)12.6 ± 6.914.2 ± 7.8.326 Hematocrit (%)33.0 ± 8.232.4 ± 7.7.708 Sodium (mEq/L)138.3 ± 4.8137.8 ± 7.4.982 Potassium (mEq/L)4.41 ± 0.834.52 ± 0.71.529 Glucose (mg/dL)133.4 ± 58.8122.8 ± 48.3.380 BUN (mg/dL)25.5 ± 19.836.5 ± 29.9.032 Serum creatinine (mg/dL)1.5 ± 1.52.3 ± 29.9.071 Albumin (g/dL)3.0 ± 0.823.1 ± 3.7.458 Total bilirubin (mg/dL)1.01 ± 2.551.56 ± 2.88.352 HCO3(mEq/L)25.4 ± 7.522.1 ± 5.8.010 pH7.36 ± 0.127.31 ± 0.12.088 pCO2(mm Hg)47.7 ± 24.446.1 ± 16.1.762 pO2(mm Hg)138.3 ± 103.4 150.5 ± 132.3.568 Lactate (mg/dL)2.1 ± 2.21.7 ± 1.1.408 Glasgow Coma Scale11.3 ± 3.4712.3 ± 3.62.189 Intubated in MICU (%)29.448.4b.001 Tracheostomy placed (%) 2.941.9b.001 APACHE III score46.3 ± 21.248.1 ± 19.5.648 APS33.3 ± 15.735.2 ± 18.6.544 BMI indicates body mass index; MAP, mean arterial blood pressure; WBC, white blood cell count. 880.e3Clinical model for predicting PMV the remaining cohort (n = 99), the mean duration of MV support was 10.2 ± 10.8 days. Prolonged MV was required in 31.3% of the cohort. Most of the measured variables were similar between those who did and did not require PMV (Table 2). On multivariate analysis, only 6 of the measured variables were found to be independent predictors of PMV (Table 3). These included an HR more than 110, BUN more than 25 mg/dL, serum pH less than 7.25, serum creatinine more than 2.0 mg/dL, and a HCO3less than 20 mEq/L. Admission to the MICU before developing the need for intubation was also independent variable associated with requiring MV more than 14 days. Using these variables as a predictive model was highly specific for identifying patients requiring PMV (Fig. 2). Specifically, we found that the presence of 3 or more variables was associated with a specificity of 89.7% and that of 4 or more variables was associated with excellent performance characteristics. In contrast, sensitivity progres- sively decreased with a greater number of positive variables and was only 32.3% in those with 3 or more findings. As demonstrated in Table 4, the specificity, positive predictive value, and negative predictive value for 4 or more variables were 100%, 100%, and 72.3%, respectively, for predicting PMV. For additional validation, screening tools were evaluated with ROC curves (Table 5). Intubation in the ICU, Tachycardia, Renal dysfunction, Acidemia, elevated Creatinine, and a decreased HCO3(I-TRACH) showed greatest AUC of all screening tools, 0.754 (Fig. 3). For validation, we compared the accuracy of our predictive model to APACHE III and APS scores. Both APACHE III and APS showed a near linear relationship to ventilator days. Specifically, an APACHE III score more than 68 was highly predictive of more than 14 days of MV support. Similarly, an APS score more than 48 correlated with and had similar need for PMV. Despite its clinical simplicity, I-TRACH had moderate correlation with similar accuracy to the APACHE III and APS scores for predicting MV support longer than 14 days (Fig. 4). Tracheostomy was performed in 15.2% of the cohort (mean age was 60.0 ± 19.3 years, and 73.3% were men). Of those receiving tracheostomies, the procedure was performed 11.5 ± 4.0 days following intubation, and the mean total duration of MV therapy was 22.7 ± 9.0 days. Applying this predictive model to those who underwent tracheostomy at the time they were intubated would have been predicted PMV. Specifically, among patients receiv- ing tracheostomies, 3 or more criteria at time of intubation was associated with a specificity of 89.7%, and 4 or more criteria at the time of intubation was associated with a spe