A predictive model for the early identification of patients at risk for a prolonged intensive care unit length of stay

Comparison of regression methods for modeling intensive care length of stay

Intensive care units (ICUs) are increasingly interested in assessing and improving their performance. ICU Length of Stay (LoS) could be seen as an indicator for efficiency of care. However, little consensus exists on which prognostic method should be used to adjust ICU LoS for case-mix factors. This study compared the performance of different regression models when predicting ICU LoS. We included data from 32,667 unplanned ICU admissions to ICUs participating in the Dutch National Intensive Care Evaluation (NICE) in the year 2011. We predicted ICU LoS using eight regression models: ordinary least squares regression on untransformed ICU LoS,LoS truncated at 30 days and log-transformed LoS; a generalized linear model with a Gaussian distribution and a logarithmic link function; Poisson regression; negative binomial regression; Gamma regression with a logarithmic link function; and the original and recalibrated APACHE IV model, for all patients together and for survivors and non-survivors separately. We assessed the predictive performance of the models using bootstrapping and the squared Pearson correlation coefficient (R2), root mean squared prediction error (RMSPE), mean absolute prediction error (MAPE) and bias. The distribution of ICU LoS was skewed to the right with a median of 1.7 days (interquartile range 0.8 to 4.0) and a mean of 4.2 days (standard deviation 7.9). The predictive performance of the models was between 0.09 and 0.20 for R2, between 7.28 and 8.74 days for RMSPE, between 3.00 and 4.42 days for MAPE and between -2.99 and 1.64 days for bias. The predictive performance was slightly better for survivors than for non-survivors. We were disappointed in the predictive performance of the regression models and conclude that it is difficult to predict LoS of unplanned ICU admissions using patient characteristics at admission time only.

EM for regularized zero-inflated regression models with applications to postoperative morbidity after cardiac surgery in children

This paper proposes a new statistical approach for predicting postoperative morbidity such as intensive care unit length of stay and number of complications after cardiac surgery in children. In a recent multi-center study sponsored by the National Institutes of Health, 311 children undergoing cardiac surgery were enrolled. Morbidity data are count data in which the observations take only nonnegative integer values. Often, the number of zeros in the sample cannot be accommodated properly by a simple model, thus requiring a more complex model such as the zero-inflated Poisson regression model. We are interested in identifying important risk factors for postoperative morbidity among many candidate predictors. There is only limited methodological work on variable selection for the zero-inflated regression models. In this paper, we consider regularized zero-inflated Poisson models through penalized likelihood function and develop a new expectation-maximization algorithm for numerical optimization. Simulation studies show that the proposed method has better performance than some competing methods. Using the proposed methods, we analyzed the postoperative morbidity, which improved the model fitting and identified important clinical and biomarker risk factors.

Evaluation of the PaO2/FiO2 ratio after cardiac surgery as a predictor of outcome during hospital stay

BACKGROUND

The arterial partial pressure of O2 and the fraction of inspired oxygen (PaO2/FiO2) ratio is widely used in ICUs as an indicator of oxygenation status. Although cardiac surgery and ICU scores can predict mortality, during the first hours after cardiac surgery few instruments are available to assess outcome. The aim of this study was to evaluate the usefulness of PaO2/FIO2 ratio to predict mortality in patients immediately after cardiac surgery.

METHODS

We prospectively studied 2725 consecutive cardiac surgery patients between 2004 and 2009. PaO2/FiO2 ratio was measured on admission and at 3 h, 6 h, 12 h and 24 h after ICU admission, together with clinical data and outcomes.

RESULTS

All PaO2/FIO2 ratio measurements differed between survivors and non-survivors (p < 0.001). The PaO2/FIO2 at 3 h after ICU admission was the best predictor of mortality based on area under the curve (p < 0.001) and the optimum threshold estimation gave an optimal cut-off of 222 (95% Confidence interval (CI): 202-242), yielding three groups of patients: Group 1, with PaO2/FIO2 > 242; Group 2, with PaO2/FIO2 from 202 to 242; and Group 3, with PaO2/FIO2 < 202. Group 3 showed higher in-ICU mortality and ICU length of stay and Groups 2 and 3 also showed higher respiratory complication rates. The presence of a PaO2/FIO2 ratio < 202 at 3 h after admission was shown to be a predictor of in-ICU mortality (OR:1.364; 95% CI:1.212-1.625, p < 0.001) and of worse long-term survival (88.8% vs. 95.8%; Log rank p = 0.002. Adjusted Hazard ratio: 1.48; 95% CI:1.293-1.786; p = 0.004).

CONCLUSIONS

A simple determination of PaO2/FIO2 at 3 h after ICU admission may be useful to identify patients at risk immediately after cardiac surgery.

Management of nurse shortage and its impact on pathogen dissemination in the intensive care unit

INTRODUCTION

Studies provide evidence that reduced nurse staffing resources are associated to an increase in health care-associated infections in intensive care units, but tools to assess the contribution of the mechanisms driving these relations are still lacking. We present an agent-based model of pathogen spread that can be used to evaluate the impact on nosocomial risk of alternative management decisions adopted to deal with transitory nurse shortage.

MATERIALS AND METHODS

We constructed a model simulating contact-mediated dissemination of pathogens in an intensive-care unit with explicit staffing where nurse availability could be temporarily reduced while maintaining requisites of patient care. We used the model to explore the impact of alternative management decisions adopted to deal with transitory nurse shortage under different pathogen- and institution-specific scenarios. Three alternative strategies could be adopted: increasing the workload of working nurses, hiring substitute nurses, or transferring patients to other intensive-care units. The impact of these decisions on pathogen spread was examined while varying pathogen transmissibility and severity of nurse shortage.

RESULTS

The model-predicted changes in pathogen prevalence among patients were impacted by management decisions. Simulations showed that increasing nurse workload led to an increase in pathogen spread and that patient transfer could reduce prevalence of pathogens among patients in the intensive-care unit. The outcome of nurse substitution depended on the assumed skills of substitute nurses. Differences between predicted outcomes of each strategy became more evident with increasing transmissibility of the pathogen and with higher rates of nurse shortage.

CONCLUSIONS

Agent-based models with explicit staff management such as the model presented may prove useful to design staff management policies that mitigate the risk of healthcare-associated infections under episodes of increased nurse shortage.

APACHE IV is superior to MELD scoring system in predicting prognosis in patients after orthotopic liver transplantation

This study aims to compare the efficiency of APACHE IV with that of MELD scoring system for prediction of the risk of mortality risk after orthotopic liver transplantation (OLT). A retrospective cohort study was performed based on a total of 195 patients admitted to the ICU after orthotopic liver transplantation (OLT) between February 2006 and July 2009 in Guangzhou, China. APACHE IV and MELD scoring systems were used to predict the postoperative mortality after OLT. The area under the receiver operating characteristic curve (AUC) and the Hosmer-Lemeshow C statistic were used to assess the discrimination and calibration of APACHE IV and MELD, respectively. Twenty-seven patients died during hospitalization with a mortality rate of 13.8%. The mean scores of APACHE IV and MELD were 42.32 ± 21.95 and 18.09 ± 10.55, respectively, and APACHE IV showed better discrimination than MELD; the areas under the receiver operating characteristic curve for APACHE IV and MELD were 0.937 and 0.694 (P < 0.05 for both models), which indicated that the prognostic value of APACHE IV was relatively high. Both models were well-calibrated (The Hosmer-Lemeshow C statistics were 1.568 and 6.818 for APACHE IV and MELD, resp.; P > 0.05 for both). The respective Youden indexes of APACHE IV, MELD, and combination of APACHE IV with MELD were 0.763, 0.430, and 0.545. The prognostic value of APACHE IV is high but still underestimates the overall hospital mortality, while the prognostic value of MELD is poor. The function of the APACHE IV is, thus, better than that of the MELD.

Mortality prediction models for pediatric intensive care: comparison of overall and subgroup specific performance

AIM

To validate paediatric index of mortality (PIM) and pediatric risk of mortality (PRISM) models within the overall population as well as in specific subgroups in pediatric intensive care units (PICUs).

METHODS

Variants of PIM and PRISM prediction models were compared with respect to calibration (agreement between predicted risks and observed mortality) and discrimination (area under the receiver operating characteristic curve, AUC). We considered performance in the overall study population and in subgroups, defined by diagnoses, age and urgency at admission, and length of stay (LoS) at the PICU. We analyzed data from consecutive patients younger than 16 years admitted to the eight PICUs in the Netherlands between February 2006 and October 2009. Patients referred to another ICU or deceased within 2 h after admission were excluded.

RESULTS

A total of 12,040 admissions were included, with 412 deaths. Variants of PIM2 were best calibrated. All models discriminated well, also in patients <28 days of age (neonates), with overall higher AUC for PRISM variants (PIM = 0.83, PIM2 = 0.85, PIM2-ANZ06 = 0.86, PIM2-ANZ08 = 0.85, PRISM = 0.88, PRISM3-24 = 0.90). Best discrimination for PRISM3-24 was confirmed in 13 out of 14 subgroup categories. After recalibration PRISM3-24 predicted accurately in most (12 out of 14) categories. Discrimination was poorer for all models (AUC < 0.73) after LoS of >6 days at the PICU.

CONCLUSION

All models discriminated well, also in most subgroups including neonates, but had difficulties predicting mortality for patients >6 days at the PICU. In a western European setting both the PIM2(-ANZ06) or a recalibrated version of PRISM3-24 are suited for overall individualized risk prediction.

On the role of length of stay in healthcare-associated bloodstream infection

DESIGN

We conducted a retrospective cohort study to examine the role played by length of hospital stay in the risk of healthcare-associated bloodstream infection (BSI), independent of demographic and clinical risk factors for BSI.

PATIENTS

We employed data from 113,893 admissions from inpatients discharged between 2006 and 2008.

SETTING

Large tertiary healthcare center in New York City.

METHODS

We estimated the crude and adjusted hazard of BSI by conducting logistic regression using a person-day data structure. The covariates included in the fully adjusted model included age, sex, Charlson score of comorbidity, renal failure, and malignancy as static variables and central venous catheterization, mechanical ventilation, and intensive care unit stay as time-varying variables.

RESULTS

In the crude model, we observed a nonlinear increasing hazard of BSI with increasing hospital stay. This trend was reduced to a constant hazard when fully adjusted for demographic and clinical risk factors for BSI.

CONCLUSION

The association between longer length of hospital stay and increased risk of infection can largely be explained by the increased duration of stay among those who have underlying morbidity and require invasive procedures. We should take caution in attributing the association between length of stay and BSI to a direct negative impact of the healthcare environment.

Sedation in adults receiving mechanical ventilation: physiological and comfort outcomes

OBJECTIVE

To describe the relationships among sedation, stability in physiological status, and comfort during a 24-hour period in patients receiving mechanical ventilation.

METHODS

Data from 169 patients monitored continuously for 24 hours were recorded at least every 12 seconds, including sedation levels, physiological status (heart rate, respiratory rate, oxygen saturation by pulse oximetry), and comfort (movement of arms and legs as measured by actigraphy). Generalized linear mixed-effect models were used to estimate the distribution of time spent at various heart and respiratory rates and oxygen saturation and actigraphy intervals overall and as a function of level of sedation and to compare the percentage of time in these intervals between the sedation states.

RESULTS

Patients were from various intensive care units: medical respiratory (52%), surgical trauma (35%), and cardiac surgery (13%). They spent 42% of the time in deep sedation, 38% in mild/moderate sedation, and 20% awake/alert. Distributions of physiological measures did not differ during levels of sedation (deep, mild/moderate, or awake/alert: heart rate, P = .44; respirations, P = .32; oxygen saturation, P = .51). Actigraphy findings differed with level of sedation (arm, P < .001; leg, P = .01), with less movement associated with greater levels of sedation, even though patients spent the vast majority of time with no arm movement or leg movement.

CONCLUSIONS

Level of sedation most likely does not affect the stability of physiological status but does have an effect on comfort.

[Relationship between PAO2/FIO2 and SATO2/FIO2 with mortality and duration of admission in critically ill children]

OBJECTIVES

The aim of this study is to analyse the relationships and the association between PaO(2)/FiO(2) and SatO(2)/FiO(2with) the duration of admission in Paediatric Intensive Care Units (PICU) and mortality, and to study the relationships between both ratios.

MATERIAL AND METHODS

A retrospective study was conducted on PICU patients in whom a gas analysis was performed in the first twenty-four hours of admission. Demographic, clinical and ventilation variables were collected, and the relationship between PaO(2)/FiO(2) and SatO(2)/FiO(2) with days of admission and mortality was determined. Finally, the best cut-off points of SatO(2)/FiO(2) were determined for PaO(2)/FiO(2) values greater and less than 200.

RESULTS

Of 512 patients admitted during one year, a gas analysis was performed on 358, 65% of those in arterial blood. The median duration of hospitalization was two days and there were 11 patient deaths. There was a low negative correlation between the values of PaO(2)/FiO(2) and SatO(2/)FiO(2) on admission to PICU and with duration of admission, and an inverse association with mortality (P<.01). This association was stronger for the PaO(2)/FiO(2) ratio in patients with heart disease, those undergoing invasive mechanical ventilation, and for arterial blood samples. PaO(2)/FiO(2) and SatO(2)/FiO(2) ratios were significantly correlated with each other. A cut-off of 200 for SatO(2)/FiO(2) had a sensitivity of 97.5% for classifying patients with PaO(2)/FiO(2) values lower or higher than 200.

CONCLUSIONS

PaO(2)/FiO(2) and SatO(2)/FiO(2) index are markers of severity in critically ill patients. In patients who do not have an arterial line, SatO(2)/FiO(2) index can be used for assessment of oxygenation as an indicator of severity in children in critical condition.