Predictive models for pressure ulcers from intensive care unit electronic health records using Bayesian networks

Background

We develop predictive models enabling clinicians to better understand and explore patient clinical data along with risk factors for pressure ulcers in intensive care unit patients from electronic health record data. Identifying accurate risk factors of pressure ulcers is essential to determining appropriate prevention strategies; in this work we examine medication, diagnosis, and traditional Braden pressure ulcer assessment scale measurements as patient features. In order to predict pressure ulcer incidence and better understand the structure of related risk factors, we construct Bayesian networks from patient features. Bayesian network nodes (features) and edges (conditional dependencies) are simplified with statistical network techniques. Upon reviewing a network visualization of our model, our clinician collaborators were able to identify strong relationships between risk factors widely recognized as associated with pressure ulcers.

Methods

We present a three-stage framework for predictive analysis of patient clinical data: 1) Developing electronic health record feature extraction functions with assistance of clinicians, 2) simplifying features, and 3) building Bayesian network predictive models. We evaluate all combinations of Bayesian network models from different search algorithms, scoring functions, prior structure initializations, and sets of features.

Results

From the EHRs of 7,717 ICU patients, we construct Bayesian network predictive models from 86 medication, diagnosis, and Braden scale features. Our model not only identifies known and suspected high PU risk factors, but also substantially increases sensitivity of the prediction - nearly three times higher comparing to logistical regression models - without sacrificing the overall accuracy. We visualize a representative model with which our clinician collaborators identify strong relationships between risk factors widely recognized as associated with pressure ulcers.

Conclusions

Given the strong adverse effect of pressure ulcers on patients and the high cost for treating pressure ulcers, our Bayesian network based model provides a novel framework for significantly improving the sensitivity of the prediction model. Thus, when the model is deployed in a clinical setting, the caregivers can suitably respond to conditions likely associated with pressure ulcer incidence.

     

Diagnosis of breast cancer using Bayesian networks: a case study

We evaluate the effectiveness of seven Bayesian network classifiers as potential tools for the diagnosis of breast cancer using two real-world databases containing fine-needle aspiration of the breast lesion cases collected by a single observer and multiple observers, respectively. The results show a certain ingredient of subjectivity implicitly contained in these data: we get an average accuracy of 93.04% for the former and 83.31% for the latter. These findings suggest that observers see different things when looking at the samples in the microscope; a situation that significantly diminishes the performance of these classifiers in diagnosing such a disease.     

Prediction of prolonged mechanical ventilation in patients in the intensive care unit A cohort study

Introduction:

There are no established guidelines for selecting patients for early tracheostomy. The aim was to determine the factors that could predict the possibility of intubation longer than 7 days in critically ill adult patients.

Methods:

This is cohort study made at a general intensive care unit. Patients who required at least 48 hours of mechanical ventilation were included. Data on the clinical and physiologic features were collected for every intubated patient on the third day. Uni- and multivariate statistical analyses were conducted to determine the variables associated with extubation.

Results:

163 (62%) were male, and the median age was 59±17 years. Almost one-third (36%) of patients required mechanical ventilation longer than 7 days. The variables strongly associated with prolonged mechanical ventilation were: age (HR 0.97 (95% CI 0.96-0.99); diagnosis of surgical emergency in a patient with a medical condition (HR 3.68 (95% CI 1.62-8.35), diagnosis of surgical condition-non emergency (HR 8.17 (95% CI 2.12-31.3); diagnosis of non-surgical-medical condition (HR 5.26 (95% CI 1.85-14.9); APACHE II (HR 0.91 (95% CI 0.85-0.97) and SAPS II score (HR 1.04 (95% CI 1.00-1.09) The area under ROC curve used for prediction was 0.52. 16% of patients were extubated after day 8 of intubation.

Conclusions:

It was not possible to predict early extubation in critically ill adult patients with invasive mechanical ventilation with common clinical scales used at the ICU. However, the probability of successfully weaning patients from mechanical ventilation without a tracheostomy is low after the eighth day of intubation.     

Rating organ failure via adverse events using data mining in the intensive care unit

OBJECTIVE: The main intensive care unit (ICU) goal is to avoid or reverse the organ failure process by adopting a timely intervention. Within this context, early identification of organ impairment is a key issue. The sequential organ failure assessment (SOFA) is an expert-driven score that is widely used in European ICUs to quantify organ disorder. This work proposes a complementary data-driven approach based on adverse events, defined from commonly monitored biometrics. The aim is to study the impact of these events when predicting the risk of ICU organ failure. MATERIALS AND METHODS: A large database was considered, with a total of 25,215 daily records taken from 4425 patients and 42 European ICUs. The input variables include the case mix (i.e. age, diagnosis, admission type and admission from) and adverse events defined from four bedside physiologic variables (i.e. systolic blood pressure, heart rate, pulse oximeter oxygen saturation and urine output). The output target is the organ status (i.e. normal, dysfunction or failure) of six organ systems (respiratory, coagulation, hepatic, cardiovascular, neurological and renal), as measured by the SOFA score. Two data mining (DM) methods were compared: multinomial logistic regression (MLR) and artificial neural networks (ANNs). These methods were tested in the R statistical environment, using 20 runs of a 5-fold cross-validation scheme. The area under the receiver operator characteristic (ROC) curve and Brier score were used as the discrimination and calibration measures. RESULTS: The best performance was obtained by the ANNs, outperforming the MLR in both discrimination and calibration criteria. The ANNs obtained an average (over all organs) area under the ROC curve of 64, 69 and 74% and Brier scores of 0.18, 0.16 and 0.09 for the dysfunction, normal and failure organ conditions, respectively. In particular, very good results were achieved when predicting renal failure (ROC curve area of 76% and Brier score of 0.06). CONCLUSION: Adverse events, taken from bedside monitored data, are important intermediate outcomes, contributing to a timely recognition of organ dysfunction and failure during ICU length of stay. The obtained results show that it is possible to use DM methods to get knowledge from easy obtainable data, thus making room for the development of intelligent clinical alarm monitoring.     

Diagnosis of nosocomial pneumonia in intensive care unit patients

The optimal technique for diagnosing nosocomial bacterial pneumonia in critically ill patients cared for in the intensive care unit remains unclear, especially in the subgroup of patients requiring mechanical ventilation. An important advance has been the development of the protected specimen brush technique. We and others have demonstrated that secretions obtained using this technique and evaluated by quantitative cultures are useful in distinguishing patients with and without pneumonia. However, this procedure has important limitations in that results are not available immediately and in that a few false negative or false positive results may be observed. Recently, the use of bronchoalveolar lavage has been suggested to be of value in establishing the diagnosis of pneumonia, since the cells and liquid recovered can be examined microscopically immediately after the procedure and are also suitable for quantitative culture. We believe that microscopic identification of bacteria within cells recovered by lavage may provide a sensitive and specific means for early and rapid diagnosis of pneumonia in this setting and that the lavage technique can be conveniently incorporated into a protocol along with the quantitative culture of samples obtained using the protected specimen brush. This combination will probably improve the overall accuracy of diagnosis while allowing the administration of prompt empiric antimicrobial therapy in the majority of patients with pneumonia.     

Review of data of the patients receiving total parenteral nutrition at the intensive care unit of a university hospital

Parenteral nutrition is commonly administered to patients in intensive care units who cannot be fed gastrointestinally. Several problems might be encountered during parenteral nutrition. We designed this study to evaluate two years' data of the patients who received parenteral nutrition at the intensive care unit of the university hospital. Forty-five patients who were treated at this hospital between January 1, 2004 and December 31, 2005 were included in this study. Patient data were collected via questionnaires designed based on the information in the literature. Blood, urinary, oropharyngeal and catheter entry site cultures were obtained and analyzed on the third and seventh days of the treatment. We found the following results: 31.1% of the patients received parenteral nutrition due to renal insufficiency; ready-made amino acid/lipid solutions were used in 86.7% of the patients; 77.8% of the solutions were administered through a peripheral vein; 88.6% of total parenteral nutrition solutions given thorough the peripheral vein had higher osmolarities than 800 mOsmol/L; routine Fe and Fe binding capacity, prothrombin time, cholesterol and triglyceride level assessments were not performed before the initiation of treatment; and the culture tests most commonly revealed Staphylococcus epidermidis in the blood, Candido species in urine, Streptococcus in throat, and Staphylococcus aureus at catheter entry sites. Based on these results, we suggest that organizing a nutrition support team would be useful in order to improve the quality of the nursery and to provide close and rational management and follow-up of the patients receiving total parenteral nutrition.     

Multilevel Bayesian networks for the analysis of hierarchical health care data

ObjectiveLarge health care datasets normally have a hierarchical structure, in terms of levels, as the data have been obtained from different practices, hospitals, or regions. Multilevel regression is the technique commonly used to deal with such multilevel data. However, for the statistical analysis of interactions between entities from a domain, multilevel regression yields little to no insight. While Bayesian networks have proved to be useful for analysis of interactions, they do not have the capability to deal with hierarchical data. In this paper, we describe a new formalism, which we call multilevel Bayesian networks; its effectiveness for the analysis of hierarchically structured health care data is studied from the perspective of multimorbidity.MethodsMultilevel Bayesian networks are formally defined and applied to analyze clinical data from family practices in The Netherlands with the aim to predict interactions between heart failure and diabetes mellitus. We compare the results obtained with multilevel regression.ResultsThe results obtained by multilevel Bayesian networks closely resembled those obtained by multilevel regression. For both diseases, the area under the curve of the prediction model improved, and the net reclassification improvements were significantly positive. In addition, the models offered considerable more insight, through its internal structure, into the interactions between the diseases.ConclusionsMultilevel Bayesian networks offer a suitable alternative to multilevel regression when analyzing hierarchical health care data. They provide more insight into the interactions between multiple diseases. Moreover, a multilevel Bayesian network model can be used for the prediction of the occurrence of multiple diseases, even when some of the predictors are unknown, which is typically the case in medicine.     

A case of influenza A community-acquired pneumonia in an elderly subject hospitalised in intensive care unit

Influenza pneumonia and influenza-associated severe exacerbation of pre-existing heart and lung disease are responsible for major complications that may require intensive care unit admission. Here, we report the case of a diabetic 70 year-old man hospitalised in the intensive care unit (ICU) of the University Medical Center of Reims (France) for a severe bilateral and alveolar pneumonia requiring mechanical ventilation. This patient had received a classical antibiotic treatment by amoxycillin (3 g/24 hours per os); 48 hours later, he was admitted in ICU for a respiratory failure that evolved rapidly towards an acute respiratory distress syndrome. Because of the context of a winter influenza outbreak, a nasal swabbing sample was tested for the presence of Influenzavirus nucleocapsid-antigens (Immunochromatographic test; BinaxNow Flu A & B, Binax, Portland, USA). This rapid assay revealed the presence of an Influenzavirus A respiratory infection five days after the beginning of the respiratory syndrome. This rapid viral diagnosis will be further confirmed in post mortem by the positive Influenza strain isolation onto lung tissues by classical cell culture techniques (Influenzavirus A strain, H3N2). Influenza pneumonia is a significant cause of morbidity and mortality, especially during influenza epidemics. The use of commercially available rapid diagnostic tests for influenza associated pneumonia, allows the potential use of new specific anti-neuraminidase drugs, which can be efficient during the 30 hours after the beginning of the clinical influenza syndrome.     

Nosocomial infection of malnourished patients in an intensive care unit

Malnutrition is one of the most important factors for the development of nosocomial infection (NI). We performed a study of the correlation between abnormal nutritional factors and NI risk by investigating the patients who stayed longer than 3 days in the intensive care unit (ICU) of our university hospital. The patients were classified into three groups based on serum albumin levels and total lymphocyte counts (TLC). The criteria of Group I (well nourished group) were serum albumin level of 3.5 g/dl or higher and TLC of 1,400/mm3 or higher. The criteria of Group III (severely malnourished group) were serum albumin of less than 2.8 g/dl and TLC of less than 1,000/mm3. The other patients were classified as Group II (moderately malnourished group). The occurrences of NI were monitored during the study period and the APACHE III Score was calculated. The probability of first NI infection in Group III was 2.4 times higher than that in Groups I and II. The mortality rate of 20.5% was more significantly correlated with APACHE III Score than nutritional status. Nineteen (53%) of the total 36 NI patients were infected within 10 days after ICU admission and they all belonged to Group III. When we compared the gap period between infections, the time to first infection was significant.     

A comparison of learning algorithms for Bayesian networks: a case study based on data from an emergency medical service

Due to the uncertainty of many of the factors that influence the performance of an emergency medical service, we propose using Bayesian networks to model this kind of system. We use different algorithms for learning Bayesian networks in order to build several models, from the hospital manager’s point of view, and apply them to the specific case of the emergency service of a Spanish hospital. This first study of a real problem includes preliminary data processing, the experiments carried out, the comparison of the algorithms from different perspectives, and some potential uses of Bayesian networks for management problems in the health service.     

Autoimmune diseases in the intensive care unit. An update

Autoimmune diseases (ADs) are a challenge at the intensive care unit. The management of patients with these diseases in the critical care setting has improved over time since there are new and more aggressive alternatives to treat and diagnose them. We aimed to review the current causes of admission, clinical features, outcomes and variables associated with mortality of patients with ADs admitted to the intensive care unit (ICU). International classification criteria for ADs were used to include patients. Search was done through PubMed, SCOPUS, SciELO, and LILACS databases up to December of 2011.Twenty-nine case series and forty-one case reports were analyzed after quality assessment. Respiratory involvement was the leading cause of admission. Systemic lupus erythematosus (SLE) (33.5% of reported patients), rheumatoid arthritis (25%) and systemic vasculitis (15%) were the most frequent ADs in patients admitted to the ICU in the last decade. Mortality ranged from 17% to 55% in case series including all ADs, but in the ones that only included patients with a specific AD, such as SLE, it reached up to 79%. High APACHE score, multi-organ dysfunction, older age and cytopenia were the most reported variables associated with mortality. In conclusion, ADs should always be considered in patients with life threatening conditions that warrant critical care. Variables influencing mortality should be promptly identified in order to improve the patients' outcomes.     

Scoring systems in the intensive care unit: A compendium

Severity scales are important adjuncts of treatment in the intensive care unit (ICU) in order to predict patient outcome, comparing quality-of-care and stratification for clinical trials. Even though disease severity scores are not the key elements of treatment, they are however, an essential part of improvement in clinical decisions and in identifying patients with unexpected outcomes. Prediction models do face many challenges, but, proper application of these models helps in decision making at the right time and in decreasing hospital cost. In fact, they have become a necessary tool to describe ICU populations and to explain differences in mortality. However, it is also important to note that the choice of the severity score scale, index, or model should accurately match the event, setting or application; as mis-application, of such systems can lead to wastage of time, increased cost, unwarranted extrapolations and poor science. This article provides a brief overview of ICU severity scales (along with their predicted death/survival rate calculations) developed over the last 3 decades including several of them which has been revised accordingly.     

Biofilm formation of Candida spp. isolates from patients at a cardiothoracic intensive care unit

Invasive Candida infections in seriously ill patients are rare but associated with high mortality, while Candida colonization of the mucocuteneous surfaces is common in patients admitted to intensive care units. To investigate biofilm formation as a possible virulence factor, we analyzed the biofilm formation of 128 non-invasive isolates from long-term ICU patients and that of 106 invasive bloodstream isolates. Candida biofilms were quantified by the percent transmittance (%Tbloc) method. Crystal Violet (CV) staining was used as marker of biofilm thickness, and XTT assay was used as a marker of the metabolic activity of Candida cells. The ability of biofilm formation was 99.2% in the non-invasive isolates versus 96.2% in the bloodstream isolates (%Tbloc vs.%Tbloc not significant). However, high biofilm production (%Tbloc, ≥35) was more frequent among the non-invasive isolates compared to the bloodstream isolates (78.1 vs. 72.5, p<0.001).     

Epidemic of Serratia marcescens bacteremia in a cardiac intensive care unit.

From 16 July through 27 September 1988, seven cases of nosocomial Serratia marcescens bacteremia occurred in a cardiac care unit. In all seven case patients, S. marcescens was isolated from blood cultures. Two of the seven had other microorganisms identified in the blood culture in which S. marcescens was recovered; one had Enterobacter cloacae, and one had Klebsiella pneumoniae. A case-control study was conducted to identify risk factors for bloodstream infection. Case patients were more likely than controls to have been exposed to an intra-aortic balloon pump pressure transducer (7 of 7 versus 6 of 21; P = 0.001) and to a pulmonary arterial pressure transducer (7 of 7 versus 8 of 21; P = 0.005). Cultures of in-use and in-storage transducers revealed bacterial contamination of the pressure-sensitive membranes of the transducers. S. marcescens blood culture isolates obtained from five of the seven case patients, as well as six S. marcescens isolates from cultured transducers, belonged to serotypes Oundetermined:H1 and Oundetermined:H18. E. cloacae isolates from one case patient and from two stored and two in-use transducers had identical antimicrobial suceptibility patterns. Review of cardiac care unit disinfection practices revealed that the transducers were not processed with high-level disinfection or sterilization between patient uses. We concluded that the transducers had served as reservoirs for this outbreak of bloodstream infection. Because intra-aortic balloon pumps with pressure transducers are being used more frequently in the management of critically ill cardiac patients, their role as infectious reservoirs should be considered in the investigation of nosocomial bacteremia.     

Stenotrophomonas maltophilia in the respiratory tract of medical intensive care unit patients

The purpose of this study was to investigate characteristics of critically ill patients with Stenotrophomonas maltophilia (S. maltophilia) isolated from the respiratory tract, to identify risk factors for S. maltophilia-pneumonia and intensive care unit (ICU) mortality and to analyze antibiotic susceptibility of S. maltophilia. This was a retrospective analysis of 64 medical ICU patients with S. maltophilia in the respiratory tract. Thirty-six patients fulfilled the criteria for diagnosis of pneumonia. A significantly higher lung injury score (LIS) was observed in patients with pneumonia compared to patients with colonization (p=0.010). Independent risk factors for S. maltophilia-pneumonia were higher Sequential Organ Failure Assessment (SOFA) score (p=0.009) and immunosuppression (p=0.014). Patients with S. maltophilia-pneumonia had higher ICU mortality within a 28-day follow-up (p=0.040) and higher hospital mortality (p=0.018) than patients with colonization. The highest antibiotic susceptibility rates were observed to trimethoprim-sulfamethoxazole, tigecycline, and moxifloxacin. Higher SOFA score when S. maltophilia was isolated (p=0.001) and development of renal failure (p=0.021) were independent risk factors for ICU mortality. Higher SOFA score and immunosuppression are independent risk factors for S. maltophilia-pneumonia. Patients with S. maltophilia-pneumonia have a significantly higher ICU mortality within a 28-day follow-up, hospital mortality and LIS compared to patients with S. maltophilia-colonization.     

Epidemiologic trends in nosocomial bacteremia in a neonatal intensive care unit.

The primary goal of this study was to analyze the epidemiologic features of nosocomial bloodstream infection (NBSI) in a neonatal intensive care unit over a 7-year period. All neonatal patients with NBSI treated from January 1997 to December 2003 were retrospectively analyzed. 232 NBSI episodes were diagnosed in 208 patients. The average NBSI patient-day rates were 4.69 and 2.59 per 1000 patient-days in 1997-1999 and 2000-2003, respectively. The average NBSI rates were 5.00 and 1.50 per 1000 patient days in neonates <1500 g and > or =1500 g, respectively. The proportion of Gram-positive organisms increased from 24% in 1997-2001 to 41% in 2002-2003, whereas the proportion of Gram-negative isolates decreased from 65% in 1997-2001 to 47% in 2002-2003. The implementation of measures for the prevention of nosocomial infection was associated with the reduction of NBSI rates. Low birth weight was demonstrated to be a significant risk factor for NBSI. The fact that Gram-positive organisms were isolated in increasing frequency may impact on the appropriate selection of empiric antimicrobial therapy for NBSI in the neonatal intensive care unit.