Stratification of the severity of critically ill patients with classification trees

Effect of early administration of tetracosactide on mortality and host response in critically ill patients requiring rescue surgery: a sensitivity analysis of the STOPSHOCK phase 3 randomized controlled trial

BACKGROUND

Undifferentiated shock is recognized as a criticality state that is transitional in immune-mediated topology for casual risk of lethal microcirculatory dysfunction. This was a sensitivity analysis of a drug (tetracosactide; TCS10) targeting melanocortin receptors (MCRs) in a phase 3 randomized controlled trial to improve cardiovascular surgical rescue outcome by reversing mortality and hemostatic disorders.

METHODS

Sensitivity analysis was based on a randomized, two-arm, multicenter, double-blind, controlled trial. The Naïve Bayes classifier was performed by density-based sensitivity index for principal strata as proportional hazard model of 30-day surgical risk mortality according to European System for Cardiac Operative Risk Evaluation inputs-outputs in 100 consecutive cases (from August to September 2013 from Emilia Romagna region, Italy). Patients included an agent-based TCS10 group (10 mg, single intravenous bolus before surgery; n = 56) and control group (n = 44) and the association with cytokines, lactate, and bleeding-blood transfusion episodes with the prior-risk log-odds for mortality rate in time-to-event was analyzed.

RESULTS

Thirty-day mortality was significantly improved in the TCS10 group vs. control group (0 vs. 8 deaths, P < 0.0001). Baseline levels of interleukin (IL)-6, IL-10, and lactate were associated with bleeding episodes, independent of TCS10 treatment [odds ratio (OR) = 1.90, 95% confidence interval (CI) 1.39-2.79; OR = 1.53, 95%CI 1.17-2.12; and OR = 2.92, 95%CI 1.40-6.66, respectively], while baseline level of Fms-like tyrosine kinase 3 ligand (Flt3L) was associated with lower bleeding rates in TCS10-treated patients (OR = 0.31, 95%CI 0.11-0.90, P = 0.03). For every 8 TCS10-treated patients, 1 bleeding case was avoided. Blood transfusion episodes were significantly reduced in the TCS10 group compared to the control group (OR = 0.32, 95%CI 0.14-0.73, P = 0.01). For every 4 TCS10-treated patients, 1 transfusion case was avoided.

CONCLUSIONS

Sensitivity index underlines the quality target product profile of TCS10 in the runway of emergency casualty care. To introduce the technology readiness level in real-life critically ill patients, further large-scale studies are required.

TRIAL REGISTRATION

European Union Drug Regulating Authorities Clinical Trials Database (EudraCT Number: 2007-006445-41 ).

Reduced muscle strength is closely linked to computed tomography-defined myosteatosis among inpatients with cirrhosis

BACKGROUND

Myosteatosis indicates pathological fat infiltration in muscles and is regarded as a distinct disease from sarcopenia. This muscular condition exhibits a link to muscle fiber disarrangement coinciding with disrupted muscle contractility and weakened mechanical action, mirrored as decreased muscle quality. However, the relationship between handgrip strength (HGS) and computed tomography-defined myosteatosis among cirrhosis is unclear. We aimed to investigate the association between HGS and myosteatosis and determine gender-specific cutoffs regarding HGS to identify myosteatotic subjects.

METHODS

We prospectively recruited 221 cirrhotic patients. The presence of myosteatosis was determined according to intramuscular adipose tissue content. The relationship between HGS and myosteatosis was evaluated according to Spearman correlation coefficient, area under the ROC curve, and multivariate logistic regression analysis. Moreover, a model based on the classification and regression tree method was generated.

RESULTS

Our results showed that HGS exhibits modestly negative correlation with intramuscular adipose tissue content in the entire cohort (rs = -0.269, P < .001) and across diverse subgroups precluding extremely deteriorating conditions. After controlling for multiple clinical features and biochemical parameters, HGS (odds ratio = 0.921, P = .010) was independently associated with myosteatosis in addition to age and body mass index. On applying the Japan Society of Hepatology-recommended cutoffs, an area under the ROC curve of HGS was 0.627 with a sensitivity of 77.4% and a specificity of 47.9%. The decision tree including body mass index and low HGS correctly classified ~85% of the cases in development and validation sets.

CONCLUSIONS

HGS was in close relation to myosteatosis among inpatients with cirrhosis.

Implementation of a major trauma team. Analysis of activation and care times in patients admitted to the ICU

OBJECTIVE

To analyze the factors associated with the activation of the severe trauma care team (STAT) in patients admitted to the ICU, to measure its impact on care times, and to analyze the groups of patients according to activation and level of anatomical involvement.

DESIGN

Prospective cohort study of severe trauma admitted to the ICU. From June 2017 to May 2019. Risk factors for the activation of the STAT analysed with logistic regression and CART type classification tree.

SETTING

Second level hospital ICU.

PATIENTS

Patients admitted consecutively.

INTERVENTIONS

No.

MAIN VARIABLES OF INTEREST

STAT activation. Demographic variables. Injury severity (ISS), intentionality, mechanism, assistance times, evolutionary complications, and mortality.

RESULTS

A total of 188 patients were admitted (46.8% of STAT activation), median age of 52 (37-64) years (activated 47 (27-62) vs not activated 55 (42-67) P = 0.023), males 84.0%. No difference in mortality according to activation. The logistic model finds as factors: care (16.6 (2.1-13.2)) and prehospital intubation (4.2 (1.8-9.8)) and severe lower extremity injury (4.4 (1.6-12.3)). Accidental fall (0.2 (0.1-0.6)) makes activation less likely. The CART model selects the type of trauma mechanism and can separate high and low energy trauma.

CONCLUSIONS

Factors associated with STAT activation were prehospital care, requiring prior intubation, high-energy mechanisms, and severe lower extremity injuries. Shorter care times if activated without influencing mortality. We must improve activation in older patients with low-energy trauma and without prehospital care.

Association Between Length of Storage of Transfused Packed RBC Units and Outcome of Surgical Critically Ill Adults: A Subgroup Analysis of the Age of Blood Evaluation Randomized Trial

OBJECTIVES

The Age of Blood Evaluation (ABLE) study reported no clinical benefit in fresher compared with standard delivery RBC units (length of storage: 6.9 ± 4.1 vs 22.0 ± 8.4 d, respectively). Perioperative patients are often anemic, at risk of blood loss, and more exposed to RBC transfusions. We address the question whether fresh RBC units are safer than standard delivery RBC units in perioperative ICU patients.

DESIGN

Subgroup analysis of surgical nontrauma adults enrolled in the ABLE randomized controlled trial.

SETTING

ICUs.

PATIENTS

Three hundred twenty surgical patients among the 2,510 ICU adults recruited in the ABLE study who had a request for a first RBC transfusion in the first week in ICU stay and an anticipated length of mechanical ventilation greater than or equal to 48 hours. We included perioperative patients but excluded elective cardiac surgery and trauma.

INTERVENTIONS

Surgical participants were allocated to receive either RBC units stored less than or equal to 7 days or standard issue RBC.

MEASUREMENTS AND MAIN RESULTS

The primary outcome was 90-day all-cause mortality. One hundred seventy-two perioperative patients were allocated to the fresh and 148 to the standard group. Baseline data were similar. The length of storage was 7.2 ± 6.4 in fresh and 20.6 ± 8.4 days in standard group ( p < 0.0001). The 90-day mortality was 29.7% and 28.4%, respectively (absolute risk difference: 0.01; 95% CI -0.09 to 0.11; p = 0.803). No significant differences were observed for all secondary outcomes, including 6-month mortality, even after adjustment for age, country, and Acute Physiology and Chronic Health Evaluation score.

CONCLUSIONS

There was no evidence that fresh red cells improved outcomes as compared to standard issue red cells in critically ill surgical patients, consistent with other patients enrolled in the ABLE trial.

Employing biochemical biomarkers for building decision tree models to predict bipolar disorder from major depressive disorder

BACKGROUND

Conventional biochemical parameters may have predictive values for use in clinical identification between bipolar disorder (BD) and major depressive disorder (MDD).

METHODS

This study enrolled 2470 hospitalized patients with BD (n = 1333) or MDD (n = 1137) at reproductive age from 2009 to 2018 in China. We extracted 8 parameters, uric acid (UA), direct bilirubin (DBIL), indirect bilirubin (IDBIL), lactic dehydrogenase (LDH), free triiodothyronine (FT3), thyroid-stimulating hormone (TSH), high-density lipoprotein (HDL) and prealbumin of male, patients and 12 parameters, UA, DBIL, IBIL, LDH, FT3, TSH, glutamic-pyruvic transaminase (GPT), white blood cell (WBC), alkaline phosphatase (ALP), fasting blood glucose (FBG), triglyceride and low-density lipoprotein (LDL) of female patients. Backward stepwise multivariate regression analysis and the Chi-Square Automatic Interaction Detection (CHAID) segmentation analysis via SPSS Decision Tree were implemented to define the discrimination of BD and MDD.

RESULTS

DBIL was extracted as the first splitting variable, with LDH and IBIL as the second, TSH and prealbumin as the third in the model of male patients (p-value < .05). For the model of female patients, DBIL was also extracted as the first splitting variable, with UA, LDH, and IBIL as the second, triglyceride and FT3 as the third (p-value < .05). The predictive accuracies of the Decision Tree and multiple logistic regression models were similar (74.9% vs 76.9% in males; 74.4% vs 79.5% in females).

CONCLUSIONS

This study suggests the value of the Decision Tree models, which employ biochemical parameters as diagnostic predictors for BD and MDD. The CHAID Decision Tree identified that patients with concomitantly increased LDH, IBIL, and decreased DBIL could be in the group that showed the highest risk of being diagnosed as BD.

Multiple moderator meta-analysis using the R-package Meta-CART

In meta-analysis, heterogeneity often exists between studies. Knowledge about study features (i.e., moderators) that can explain the heterogeneity in effect sizes can be useful for researchers to assess the effectiveness of existing interventions and design new potentially effective interventions. When there are multiple moderators, they may amplify or attenuate each other's effect on treatment effectiveness. However, in most meta-analysis studies, interaction effects are neglected due to the lack of appropriate methods. The method meta-CART was recently proposed to identify interactions between multiple moderators. The analysis result is a tree model in which the studies are partitioned into more homogeneous subgroups by combinations of moderators. This paper describes the R-package metacart, which provides user-friendly functions to conduct meta-CART analyses in R. This package can fit both fixed- and random-effects meta-CART, and can handle dichotomous, categorical, ordinal and continuous moderators. In addition, a new look ahead procedure is presented. The application of the package is illustrated step-by-step using diverse examples.

Machine learning techniques for mortality prediction in critical traumatic patients: anatomic and physiologic variables from the RETRAUCI study

Background

Interest in models for calculating the risk of death in traumatic patients admitted to ICUs remains high. These models use variables derived from the deviation of physiological parameters and/or the severity of anatomical lesions with respect to the affected body areas. Our objective is to create different predictive models of the mortality of critically traumatic patients using machine learning techniques.

Methods

We used 9625 records from the RETRAUCI database (National Trauma Registry of 52 Spanish ICUs in the period of 2015–2019). Hospital mortality was 12.6%. Data on demographic variables, affected anatomical areas and physiological repercussions were used. The Weka Platform was used, along with a ten-fold cross-validation for the construction of nine supervised algorithms: logistic regression binary (LR), neural network (NN), sequential minimal optimization (SMO), classification rules (JRip), classification trees (CT), Bayesian networks (BN), adaptive boosting (ADABOOST), bootstrap aggregating (BAGGING) and random forest (RFOREST). The performance of the models was evaluated by accuracy, specificity, precision, recall, F-measure, and AUC.

Results

In all algorithms, the most important factors are those associated with traumatic brain injury (TBI) and organic failures. The LR finds thorax and limb injuries as independent protective factors of mortality. The CT generates 24 decision rules and uses those related to TBI as the first variables (range 2.0–81.6%). The JRip detects the eight rules with the highest risk of mortality (65.0–94.1%). The NN model uses a hidden layer of ten nodes, which requires 200 weights for its interpretation. The BN find the relationships between the different factors that identify different patient profiles. Models with the ensemble methodology (ADABOOST, BAGGING and RandomForest) do not have greater performance. All models obtain high values ​​in accuracy, specificity, and AUC, but obtain lower values ​​in recall. The greatest precision is achieved by the SMO model, and the BN obtains the best recall, F-measure, and AUC.

Conclusion

Machine learning techniques are useful for creating mortality classification models in critically traumatic patients. With clinical interpretation, the algorithms establish different patient profiles according to the relationship between the variables used, determine groups of patients with different evolutions, and alert clinicians to the presence of rules that indicate the greatest severity.

Establishing Decision Trees for Predicting Successful Postpyloric Nasoenteric Tube Placement in Critically Ill Patients

BACKGROUND

Despite the use of prokinetic agents, the overall success rate for postpyloric placement via a self-propelled spiral nasoenteric tube is quite low.

METHODS

This retrospective study was conducted in the intensive care units of 11 university hospitals from 2006 to 2016 among adult patients who underwent self-propelled spiral nasoenteric tube insertion. Success was defined as postpyloric nasoenteric tube placement confirmed by abdominal x-ray scan 24 hours after tube insertion. Chi-square automatic interaction detection (CHAID), simple classification and regression trees (SimpleCart), and J48 methodologies were used to develop decision tree models, and multiple logistic regression (LR) methodology was used to develop an LR model for predicting successful postpyloric nasoenteric tube placement. The area under the receiver operating characteristic curve (AUC) was used to evaluate the performance of these models.

RESULTS

Successful postpyloric nasoenteric tube placement was confirmed in 427 of 939 patients enrolled. For predicting successful postpyloric nasoenteric tube placement, the performance of the 3 decision trees was similar in terms of the AUCs: 0.715 for the CHAID model, 0.682 for the SimpleCart model, and 0.671 for the J48 model. The AUC of the LR model was 0.729, which outperformed the J48 model.

CONCLUSION

Both the CHAID and LR models achieved an acceptable discrimination for predicting successful postpyloric nasoenteric tube placement and were useful for intensivists in the setting of self-propelled spiral nasoenteric tube insertion.

Evolving classification of intensive care patients from event data

OBJECTIVE

This work aims at predicting the patient discharge outcome on each hospitalization day by introducing a new paradigm-evolving classification of event data streams. Most classification algorithms implicitly assume the values of all predictive features to be available at the time of making the prediction. This assumption does not necessarily hold in the evolving classification setting (such as intensive care patient monitoring), where we may be interested in classifying the monitored entities as early as possible, based on the attributes initially available to the classifier, and then keep refining our classification model at each time step (e.g., on daily basis) with the arrival of additional attributes.

MATERIALS AND METHODS

An oblivious read-once decision-tree algorithm, called information network (IN), is extended to deal with evolving classification. The new algorithm, named incremental information network (IIN), restricts the order of selected features by the temporal order of feature arrival. The IIN algorithm is compared to six other evolving classification approaches on an 8-year dataset of adult patients admitted to two Intensive Care Units (ICUs) in the United Kingdom.

RESULTS

Retrospective study of 3452 episodes of adult patients (≥16years of age) admitted to the ICUs of Guy's and St. Thomas' hospitals in London between 2002 and 2009. Random partition (66:34) into a development (training) set n=2287 and validation set n=1165. Episode-related time steps: Day 0-time of ICU admission, Day x-end of the x-th day at ICU. The most accurate decision-tree models, based on the area under curve (AUC): Day 0: IN (AUC=0.652), Day 1: IIN (AUC=0.660), Day 2: J48 decision-tree algorithm (AUC=0.678), Days 3-7: regenerative IN (AUC=0.717-0.772). Logistic regression AUC: 0.582 (Day 0)-0.827 (Day 7).

CONCLUSIONS

Our experimental results have not identified a single optimal approach for evolving classification of ICU episodes. On Days 0 and 1, the IIN algorithm has produced the simplest and the most accurate models, which incorporate the temporal order of feature arrival. However, starting with Day 2, regenerative approaches have reached better performance in terms of predictive accuracy.

[Mortality predictive factors in patients with urinary sepsis associated to upper urinary tract calculi]

PURPOSE

The aims of this study were to determine the clinical characteristics of patients with urinary sepsis associated to ureteral calculi admitted to the Intensive Care Unit (ICU), and to identify predictors of mortality in the first 24 hours of admission.

DESIGN

A retrospective observational study covering a 16-year period (2006-2011) was carried out.

SETTING

The combined clinical/surgical ICU of a secondary-level University hospital.

PATIENTS

All patients admitted to the ICU due to obstructive urinary sepsis.

INTERVENTIONS

None.

MAIN VARIABLES

We analyzed general clinical and laboratory test and urological data. The diagnostic technique, affected side, decompression technique, isolated microorganism and antibiotic therapy used were also considered. The assessment of risk factors was performed by multiple logistic regression analysis.

RESULTS

A total of 107 patients admitted to the ICU were included in the study, with a mortality rate of 19.6%. The diagnosis was mainly established by ultrasound, and the most commonly used decompression technique was retrograde JJ stenting. Microorganisms were isolated in 48.6% of the patients. In total, 20.6% of the patients had bacteremia. Multivariate analysis found age, acute renal failure and the use of vasoactive drugs administered continuously for the first 24 hours of admission to be independently associated to mortality.

CONCLUSIONS

Advanced age, acute renal failure and the need for vasoactive drugs were associated to an increased risk of mortality in patients with urinary sepsis associated to upper urinary tract calculi.

Older breast cancer survivors: can interaction analyses identify vulnerable subgroups? A report from the American Cancer Society Studies of Cancer Survivors

PURPOSE/OBJECTIVES

To explore interactions among personal, cancer, aging, and symptom variables relative to physical function (PF) in older adult breast cancer survivors to better identify vulnerable subgroups.

DESIGN

Secondary analysis of the American Cancer Society Studies of Cancer Survivors II.

SETTING

U.S. population-based mail and telephone survey.

SAMPLE

2,885 breast cancer survivors from 14 different state cancer registries stratified by cancer type and time since diagnosis. A total of 184 female breast cancer survivors, aged 70 years or older, had complete data on variables of interest and were, therefore, included in this analysis.

METHODS

Chi-Square Automatic Interaction Detector (CHAID) analysis was used to examine variable interactions.

MAIN RESEARCH VARIABLES

PF, symptom bother, comorbidity, social support, length of survivorship, treatment, stage, body mass index, physical activity, emotional health, and personal characteristics.

FINDINGS

An interaction effect between symptom bother and comorbidity was found in 39% of older adult breast cancer survivors, and an interaction effect between symptom bother and marital status was found in 40%. The most vulnerable group (8%) had high symptom bother and more than four comorbid conditions.

CONCLUSIONS

Symptom bother, comorbidity, and marital status were found to have significant interactions such that high comorbidity and high symptom bother were significantly related to lower PF. Married participants with lower symptom bother had significantly higher PF scores. Comorbidity may be the best predictor of PF for the extreme ends of the symptom bother continuum. Advancing age alone was not a sufficient predictor of PF in this analysis.

IMPLICATIONS FOR NURSING

Specific attention to symptom reports, comorbidity, and marital status can guide identification of older adult cancer survivors in need of ongoing survivorship care. The findings support use of a comprehensive assessment and tailored approach to care based on factors other than age.

KNOWLEDGE TRANSLATION

CHAID interaction analysis may be useful in exploring complex nursing problems, such as the needs of older adult cancer survivors, and help oncology nurses develop appropriate interventions and referrals.

Predictors of recurrent appendicitis after non-operative management of children with perforated appendicitis presenting with an appendicular inflammatory mass

AIM

To determine clinical and imaging predictors of recurrent appendicitis after non-operative management for children with perforated appendicitis presenting with an appendicular inflammatory mass.

METHODS

A case-control study was carried out of children with clinical and ultrasonographic diagnosis of an appendicular inflammatory mass. Patients were divided into two groups according to the presence or not of recurrent appendicitis (or predisposing pathological findings during interval appendectomy). Several clinical and imaging factors were compared between both groups using an independent and progressive analysis to identify which were the most relevant predictors.

RESULTS

The persistence of symptoms after resolution of the appendicular inflammatory mass, as well as its time of resolution (p<0.001), were the most relevant factors during group classification. The identification of an ultrasonographic image of an appendicolith and the initial size of the mass were also identified as independent predictors (p<0.018 and p<0.01). The presence of the first two factors in a patient was strongly associated with recurrent appendicitis.

CONCLUSIONS

Persistent symptoms after the recovery of the initial episode, associated with a slow resolution of the appendicular inflammatory mass, are indications for an interval appendectomy. A previous history of appendicitis or detection of an ultrasonographic image of an appendicolith is the other indication for an interval appendectomy.

Association of elevated urinary concentration of renin-angiotensin system components and severe AKI

BACKGROUND

Prognostic biomarkers that predict the severity of AKI at an early time point are needed. Urinary angiotensinogen was recently identified as a prognostic AKI biomarker. The study hypothesis is that urinary renin could also predict AKI severity and that in combination angiotensinogen and renin would be a strong predictor of prognosis at the time of AKI diagnosis.

DESIGN, SETTING, PARTICIPANTS, & MEASUREMENTS

In this multicenter, retrospective cohort study, urine was obtained from 204 patients who developed AKI after cardiac surgery from August 2008 to June 1, 2012. All patients were classified as having Acute Kidney Injury Network (AKIN) stage 1 disease by serum creatinine criteria at the time of sample collection. Urine output was not used for staging. Urinary angiotensinogen and renin were measured, and the area under the receiver-operating characteristic curve (AUC) was used to test for prediction of progression to AKIN stage 3 or in-hospital 30-day mortality. These biomarkers were added stepwise to a clinical model, and improvement in prognostic predictive performance was evaluated by category free net reclassification improvement (cfNRI) and chi-squared automatic interaction detection (CHAID).

RESULTS

Both the urinary angiotensinogen-to-creatinine ratio (uAnCR; AUC, 0.75; 95% confidence interval [CI], 0.65 to 0.85) and the urinary renin-to-creatinine ratio (uRenCR; AUC, 0.70; 95% CI, 0.57 to 0.83) predicted AKIN stage 3 or death. Addition of uAnCR to a clinical model substantially improved prediction of the outcome (AUC, 0.85; cfNRI, 0.673), augmenting sensitivity and specificity. Further addition of uRenCR increased the sensitivity of the model (cfNRI(events), 0.44). CHAID produced a highly accurate model (AUC, 0.91) and identified the combination of uAnCR >337.89 ng/mg and uRenCR >893.41 pg/mg as the strongest predictors (positive predictive value, 80.4%; negative predictive value, 90.7%; accuracy, 90.2%).

CONCLUSION

The combination of urinary angiotensinogen and renin predicts progression to very severe disease in patients with early AKI after cardiac surgery.

[Risk factors for the deterioration of quality of life in critical trauma patients. Assessment at 6 and 12 months after discharge from the intensive care unit]

OBJECTIVE

To evaluate factors influencing the deterioration of health-related quality of life (HRQoL) in trauma patients admitted to an ICU.

DESIGN

A prospective observational study was carried out.

SETTING

The combined medical/surgical ICU in a university secondary hospital with 24-hour neurosurgery service.

PATIENTS

Trauma patients admitted to the ICU during a two-year period. HRQoL assessment prior to admission to the ICU, and at 6 and 12 months after discharge.

MAIN VARIABLES

Demographic variables, type and severity of injury (AIS), severity (APACHE II, ISS, TRISS), length of stay, procedures, mortality and HRQoL according to the SF-36 and EQ-5D.

RESULTS

We completed the monitoring of 110 patients that showed significant impairment of their HRQoL in all the dimensions assessed. According to the SF-36, physical role was more deteriorated at 12 months, but the mental component decreased more than the physical component after 6 months. The VAS scale of the EQ-5D decreased to 55 at 6 months (19) and increased to 66 at 12 months (17). In the multiple logistic regression analysis, the variables associated with poorer HRQoL were age > 45 years, TRISS > 10, previous porer quality of life, and serious injuries in the extremities.

CONCLUSIONS

Patients showed marked deterioration of their HRQoL at 6 months, followed by overall improvement at 12 months, though without reaching their previous state. The factors that determine poorer quality of life include age, severity, previous HRQoL, and severe injuries in the extremities.

Prognostic factors in head and neck squamous cell carcinoma: comparison of CHAID decision trees technology and Cox analysis

BACKGROUND

The purpose of this study was to compare the risk factors obtained from a classical statistical method (Cox proportional hazards model) and the results obtained with classification trees (Chi-square Automatic Interaction Detection [CHAID] model) in head and neck squamous cell carcinoma (HNSCC).

METHODS

We conducted a retrospective study of 3373 patients with HNSCC and a follow-up longer than 2 years.

RESULTS

The most decisive variable classified by CHAID was T category. N classification was an important prognostic factor in the Cox analysis, but this was not considered in CHAID except in T2 and supraglottic T3. CHAID also indicated that women with oral cavity T3/T4 to N0 tumors had poorer prognosis (28%) than men (58.5%; p value < .001). In oropharynx location, men had lower survival than women (41% vs 72%; p value < .001).

CONCLUSIONS

The main benefit of CHAID analysis is that it identifies a relatively small number of patients with a singular behavior, which is more discriminatory for an optimal diagnosis and treatment.

A simple classification model for hospital mortality in patients with acute lung injury managed with lung protective ventilation

OBJECTIVE

Despite improvements in the care of critically ill patients, hospital mortality rate for acute lung injury remains high at approximately 40%. We developed a classification rule to stratify mechanically ventilated patients with acute lung injury according to hospital mortality and compared this rule with the Acute Physiology and Chronic Health Evaluation III prediction.

PATIENTS

We used data of 2,022 participants in Acute Respiratory Distress Syndrome Network trials to build a classification rule based on 54 variables collected before randomization.

DESIGN

We used a classification tree approach to stratify patients according to hospital mortality using a training subset of 1800 participants and estimated expected prediction errors using tenfold crossvalidation. We validated our classification tree using a subset of 222 participants not included in model building and calculated areas under the receiver operating characteristic curves.

MEASUREMENTS AND MAIN RESULTS

We identified combinations of age (>63 yrs), blood urea nitrogen (>15 mg/dL), shock, respiratory rate (>21 breaths/min), and minute ventilation (>13.9 L/min) as important predictors of hospital mortality at 90 days. The classification tree had a similar expected prediction error in the training set (28% vs. 26%; p = .18) and areas under the receiver operating characteristic curve in the validation set (0.71 vs. 0.73; p = .71) as did a model based on Acute Physiology and Chronic Health Evaluation III.

CONCLUSIONS

Our tree-based classification rule performed similarly to Acute Physiology and Chronic Health Evaluation III in stratifying patients according to hospital mortality, is simpler to use, contains risk factors that may be specific to acute lung injury, and identified minute ventilation as a potential novel predictor of death in patients with acute lung injury.

Statistical modelling of organic matter and emerging pollutants removal in constructed wetlands

Multiple regression models, clustering tree diagrams, regression trees (CHAID) and redundancy analysis (RDA) were applied to the study of the removal of organic matter and pharmaceuticals and personal care products (PPCPs) from urban wastewater by means of constructed wetlands (CWs). These four statistical analyses pointed out the importance of physico-chemical parameters, plant presence and chemical structure in the elimination of most pollutants. Temperature, pH values, dissolved oxygen concentration, redox potential and conductivity were related to the removal of the studied substances. Plant presence (Typha angustifolia and Phragmites australis) enhanced the removal of organic matter and some PPCPs. Multiple regression equations and CHAID trees provided numerical estimations of pollutant removal efficiencies in CWs. These models were validated and they could be a useful and interesting tool for the quick estimation of removal efficiencies in already working CWs and for the design of new systems which must fulfil certain quality requirements.

Third- and fourth-degree perineal lacerations: defining high-risk clinical clusters

OBJECTIVE

Statistical methods that measure the independent contribution of individual factors for third-/fourth-degree perineal laceration (TFPL) fall short when the clinician is faced with a combination of factors. Our objective was to demonstrate how a statistical technique, classification and regression trees (CART), can identify high-risk clinical clusters.

STUDY DESIGN

We performed multivariable logistic regression, and CART analysis on data from 25,150 term vaginal births.

RESULTS

Multivariable analyses found strong associations with the use of episiotomy, forceps, vacuum, nulliparity, and birthweight. CART ranked episiotomy, operative delivery, and birthweight as the more discriminating factors and defined distinct risk groups with TFPL rates that ranged from 0-100%. For example, without episiotomy, the rate of TFPL was 2.2%. In the presence of an episiotomy, forceps, and birthweight of >3634 g, the rate of TFPL was 68.9%.

CONCLUSION

CART showed that certain combinations held low risk, where as other combinations carried extreme risk, which clarified how choices on delivery options can markedly affect the rate of TFPL for specific mothers.

A classification and regression tree to assist clinical decision making in airway management for patients with cervical spinal cord injury

STUDY DESIGN

This was a prospective observational study.

OBJECTIVES

To review airway management of patients with acute cervical spinal cord injury (CSCI) who are admitted to the intensive care unit (ICU) and to develop a classification and regression tree (CART) to direct clinical decision making in airway management.

SETTING

This study was carried out in Australia.

METHODS

All patients with CSCI who required intubation and mechanical ventilation and who were admitted to ICU in three tertiary hospitals in Melbourne between October 2004 and May 2009 and two other interstate hospitals between December 2004 and December 2005 were included. Airway management was recorded.

RESULTS

A total of 114 patients were included. Tracheostomy insertion occurred in 68 patients (59.7%). Using CART analysis, it was found that the variables forced vital capacity, the volume of pulmonary secretion and gas exchange were predictive of airway management on 82.3% occasions with an 8.7% extubation failure rate.

CONCLUSION

A CART can be useful in clinical decision making regarding airway management in CSCI.