Computer-assisted decision support for the diagnosis and treatment of infectious diseases in intensive care units

A study on the risk stratification for patients within 24 hours of admission for risk of hospital-acquired urinary tract infection using Bayesian network models

Early identification of patients at risk of hospital-acquired urinary tract infections (HA-UTI) enables the initiation of timely targeted preventive and therapeutic strategies. Machine learning (ML) models have shown great potential for this purpose. However, existing ML models in infection control have demonstrated poor ability to support explainability, which challenges the interpretation of the result in clinical practice, limiting the adaption of the ML models into a daily clinical routine. In this study, we developed Bayesian Network (BN) models to enable explainable assessment within 24 h of admission for risk of HA-UTI. Our dataset contained 138,250 unique hospital admissions. We included data on admission details, demographics, lifestyle factors, comorbidities, vital parameters, laboratory results, and urinary catheter. Models developed from a reduced set of five features were characterized by transparency compared to models developed from a full set of 50 features. The expert-based clinical BN model over the reduced feature space showed the highest performance (area under the curve = 0.746) compared to the naïve- and tree-augmented-naïve BN models. Moreover, models developed from expert-based knowledge were characterized by enhanced explainability.

The Ethics of Algorithms in Healthcare

The amount of data available to healthcare practitioners is growing, and the rapid increase in available patient data is becoming a problem for healthcare practitioners, as they are often unable to fully survey and process the data relevant for the treatment or care of a patient. Consequently, there are currently several efforts to develop systems that can aid healthcare practitioners with reading and processing patient data and, in this way, provide them with a better foundation for decision-making about the treatment and care of patients. There are also efforts to develop algorithms that provide suggestions for such decisions. However, the development of these systems and algorithms raises several concerns related to the privacy of patients, the patient-practitioner relationship, and the autonomy of healthcare practitioners. The aim of this article is to provide a foundation for understanding the ethical challenges related to the development of a specific form of data-processing systems, namely clinical algorithms.

Application of Surgical Decision Model for Patients With Childhood Cataract: A Study Based on Real World Data

Reliable validated methods are necessary to verify the performance of diagnosis and therapy-assisted models in clinical practice. However, some validated results have research bias and may not reflect the results of real-world application. In addition, the conduct of clinical trials has executive risks for the indeterminate effectiveness of models and it is challenging to finish validated clinical trials of rare diseases. Real world data (RWD) can probably solve this problem. In our study, we collected RWD from 251 patients with a rare disease, childhood cataract (CC) and conducted a retrospective study to validate the CC surgical decision model. The consistency of the real surgical type and recommended surgical type was 94.16%. In the cataract extraction (CE) group, the model recommended the same surgical type for 84.48% of eyes, but the model advised conducting cataract extraction and primary intraocular lens implantation (CE + IOL) surgery in 15.52% of eyes, which was different from the real-world choices. In the CE + IOL group, the model recommended the same surgical type for 100% of eyes. The real-recommended matched rates were 94.22% in the eyes of bilateral patients and 90.38% in the eyes of unilateral patients. Our study is the first to apply RWD to complete a retrospective study evaluating a clinical model, and the results indicate the availability and feasibility of applying RWD in model validation and serve guidance for intelligent model evaluation for rare diseases.

Hemogram data as a tool for decision-making in COVID-19 management: applications to resource scarcity scenarios

Background

COVID-19 pandemics has challenged emergency response systems worldwide, with widespread reports of essential services breakdown and collapse of health care structure. A critical element involves essential workforce management since current protocols recommend release from duty for symptomatic individuals, including essential personnel. Testing capacity is also problematic in several countries, where diagnosis demand outnumbers available local testing capacity.

Purpose

This work describes a machine learning model derived from hemogram exam data performed in symptomatic patients and how they can be used to predict qRT-PCR test results.

Methods

Hemogram exams data from 510 symptomatic patients (73 positives and 437 negatives) were used to model and predict qRT-PCR results through Naïve-Bayes algorithms. Different scarcity scenarios were simulated, including symptomatic essential workforce management and absence of diagnostic tests. Adjusts in assumed prior probabilities allow fine-tuning of the model, according to actual prediction context.

Results

Proposed models can predict COVID-19 qRT-PCR results in symptomatic individuals with high accuracy, sensitivity and specificity, yielding a 100% sensitivity and 22.6% specificity with a prior of 0.9999; 76.7% for both sensitivity and specificity with a prior of 0.2933; and 0% sensitivity and 100% specificity with a prior of 0.001. Regarding background scarcity context, resources allocation can be significantly improved when model-based patient selection is observed, compared to random choice.

Conclusions

Machine learning models can be derived from widely available, quick, and inexpensive exam data in order to predict qRT-PCR results used in COVID-19 diagnosis. These models can be used to assist strategic decision-making in resource scarcity scenarios, including personnel shortage, lack of medical resources, and testing insufficiency.

Best Feature Selection for Horizontally Distributed Private Biomedical Data Based on Genetic Algorithms

Due to the growing success of machine learning in the healthcare domain, medical institutions are striving to share their patients' data in the intention to build more accurate models which will be used to make better decisions. However, due to the privacy of the data, they are reluctant. To build the best models, they have to make the best feature selection for horizontally distributed private biomedical data. The previous proposed solutions are based on data perturbation techniques with the loss of performance. In this article, the researchers propose an original solution without perturbation. This is so the data utility is preserved and therefore the performance. The proposed solution uses a genetic algorithm, a distributed Naïve Bayes classifier, and a trusted third-party. The results obtained by the proposed approach surpass those obtained by other researchers, for the same problem.

PPCD: Privacy-preserving clinical decision with cloud support

With the prosperity of machine learning and cloud computing, meaningful information can be mined from mass electronic medical data which help physicians make proper disease diagnosis for patients. However, using medical data and disease information of patients frequently raise privacy concerns. In this paper, based on single-layer perceptron, we propose a scheme of privacy-preserving clinical decision with cloud support (PPCD), which securely conducts disease model training and prediction for the patient. Each party learns nothing about the other's private information. In PPCD, a lightweight secure multiplication is presented and introduced to improve the model training. Security analysis and experimental results on real data confirm the high accuracy of disease prediction achieved by the proposed PPCD without the risk of privacy disclosure.

PPSDT: A Novel Privacy-Preserving Single Decision Tree Algorithm for Clinical Decision-Support Systems Using IoT Devices

Medical service providers offer their patients high quality services in return for their trust and satisfaction. The Internet of Things (IoT) in healthcare provides different solutions to enhance the patient-physician experience. Clinical Decision-Support Systems are used to improve the quality of health services by increasing the diagnosis pace and accuracy. Based on data mining techniques and historical medical records, a classification model is built to classify patients' symptoms. In this paper, we propose a privacy-preserving clinical decision-support system based on our novel privacy-preserving single decision tree algorithm for diagnosing new symptoms without exposing patients' data to different network attacks. A homomorphic encryption cipher is used to protect users' data. In addition, the algorithm uses nonces to avoid one party from decrypting other parties' data since they all will be using the same key pair. Our simulation results have shown that our novel algorithm have outperformed the Naïve Bayes algorithm by 46.46%; in addition to the effects of the key value and size on the run time. Furthermore, our model is validated by proves, which meet the privacy requirements of the hospitals' datasets, frequency of attribute values, and diagnosed symptoms.

RDAD: A Machine Learning System to Support Phenotype-Based Rare Disease Diagnosis

DNA sequencing has allowed for the discovery of the genetic cause for a considerable number of diseases, paving the way for new disease diagnostics. However, due to the lack of clinical samples and records, the molecular cause for rare diseases is always hard to identify, significantly limiting the number of rare Mendelian diseases diagnosed through sequencing technologies. Clinical phenotype information therefore becomes a major resource to diagnose rare diseases. In this article, we adopted both a phenotypic similarity method and a machine learning method to build four diagnostic models to support rare disease diagnosis. All the diagnostic models were validated using the real medical records from RAMEDIS. Each model provides a list of the top 10 candidate diseases as the prediction outcome and the results showed that all models had a high diagnostic precision (≥98%) with the highest recall reaching up to 95% while the models with machine learning methods showed the best performance. To promote effective diagnosis for rare disease in clinical application, we developed the phenotype-based Rare Disease Auxiliary Diagnosis system (RDAD) to assist clinicians in diagnosing rare diseases with the above four diagnostic models. The system is freely accessible through http://www.unimd.org/RDAD/.

Privacy Preserving Feature Selection for Vertically Distributed Medical Data based on Genetic Algorithms and Naïve Bayes

Machine learning is a powerful tool to mine useful knowledge from vast databases. Many establishments in the medical area such as hospitals, laboratories want to join their efforts with the ambition to extract models that are more accurate. However, this approach faces problems. Due to the laws protecting patient privacy or other similar concerns, parties are reluctant to share their data. In vast amounts of data, which are useful and pertinent in constructing accurate data mining models? In this article, the researchers deal with these challenges for vertically distributed medical data. They propose an original secure wrapper solution to perform feature selection based on genetic algorithms and distributed Naïve Bayes. Contrary to the previous solutions, the original data is not perturbed. Therefore, the data utility and performance are preserved. They prove that the proposed solution selects relevant attributes to increase performance, preserving patient privacy.

Sooner is better: use of a real-time automated bedside dashboard improves sepsis care

BACKGROUND

Minimizing the interval between diagnosis of sepsis and administration of antibiotics improves patient outcomes. We hypothesized that a commercially available bedside clinical surveillance visualization system (BSV) would hasten antibiotic administration and decrease length of stay (LOS) in surgical intensive care unit (SICU) patients.

METHODS

A BSV, integrated with the electronic medical record and displayed at bedside, was implemented in our SICU in July 2016. A visual sepsis screen score (SSS) was added in July 2017. All patients admitted to SICU beds with bedside displays equipped with a BSV were analyzed to determine mean SSS, maximum SSS, time from positive SSS to antibiotic administration, SICU LOS, and mortality.

RESULTS

During the study period, 232 patients were admitted to beds equipped with the clinical surveillance visualization system. Thirty patients demonstrated positive SSS followed by confirmed sepsis (23 Pre-SSS versus 7 Post-SSS). Mean and maximum SSS were similar. Time from positive SSS to antibiotic administration was decreased in patients with a visual SSS (55.3 ± 15.5 h versus 16.2 ± 9.2 h; P < 0.05). ICU and hospital LOS was also decreased (P < 0.01).

CONCLUSIONS

Implementation of a visual SSS into a BSV led to a decreased time interval between the positive SSS and administration of antibiotics and was associated with shorter SICU and hospital LOS. Integration of a visual decision support system may help providers adhere to Surviving Sepsis Guidelines.

Development and Validation of Risk Matrices for Crohn's Disease Outcomes in Patients Who Underwent Early Therapeutic Interventions

INTRODUCTION

The establishment of prognostic models for Crohn's disease [CD] is highly desirable, as they have the potential to guide physicians in the decision-making process concerning therapeutic choices, thus improving patients' health and quality of life. Our aim was to derive models for disabling CD and reoperation based solely on clinical/demographic data.

METHODS

A multicentric and retrospectively enrolled cohort of CD patients, subject to early surgery or immunosuppression, was analysed in order to build Bayesian network models and risk matrices. The final results were validated internally and with a multicentric and prospectively enrolled cohort.

RESULTS

The derivation cohort included a total of 489 CD patients [64% with disabling disease and 18% who needed reoperation], while the validation cohort included 129 CD patients with similar outcome proportions. The Bayesian models achieved an area under the curve of 78% for disabling disease and 86% for reoperation. Age at diagnosis, perianal disease, disease aggressiveness and early therapeutic decisions were found to be significant factors, and were used to construct user-friendly matrices depicting the probability of each outcome in patients with various combinations of these factors. The matrices exhibit good performance for the most important criteria: disabling disease positive post-test odds = 8.00 [2.72-23.44] and reoperation negative post-test odds = 0.02 [0.00-0.11].

CONCLUSIONS

Clinical and demographical risk factors for disabling CD and reoperation were determined and their impact was quantified by means of risk matrices, which are applicable as bedside clinical tools that can help physicians during therapeutic decisions in early disease management.

Individualized Approaches Are Needed for Optimized Blood Cultures

Many strategies and technologies are available to improve blood culture (BC)-based diagnostics. The ideal approach to BCs varies between healthcare institutions. Institutions need to examine clinical needs and practices in order to optimize BC-based diagnostics for their site. Before laboratories consider offering rapid matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-ToF MS) or expensive rapid panel-based molecular BC diagnostics, they should optimize preanalytical, analytical, and postanalytical processes and procedures surrounding BC systems. Several factors need to be considered, including local resistance rates, antibiotic prescribing patterns, patient- and provider-types, laboratory staffing, and personnel available to liaise with clinicians to optimize antibiotic use. While there is much excitement surrounding new high-technology diagnostics, cost-neutral benefits can be realized by optimizing existing strategies and using available tools in creative ways. Rapid BC diagnostics should be implemented in a manner that optimizes impact. Strategies to optimize these BC diagnostics in individual laboratories are presented here.

A Critical Review of Mechanical Ventilation Virtual Simulators: Is It Time to Use Them?

BACKGROUND

Teaching mechanical ventilation at the bedside with real patients is difficult with many logistic limitations. Mechanical ventilators virtual simulators (MVVS) may have the potential to facilitate mechanical ventilation (MV) training by allowing Web-based virtual simulation.

OBJECTIVE

We aimed to identify and describe the current available MVVS, to compare the usability of their interfaces as a teaching tool and to review the literature on validation studies.

METHODS

We performed a comparative evaluation of the MVVS, based on a literature/Web review followed by usability tests according to heuristic principles evaluation of their interfaces as performed by professional experts on MV.

RESULTS

Eight MVVS were identified. They showed marked heterogeneity, mainly regarding virtual patient's anthropomorphic parameters, pulmonary gas exchange, respiratory mechanics and muscle effort configurations, ventilator terminology, basic ventilatory modes, settings alarms, monitoring parameters, and design. The Hamilton G5 and the Xlung covered a broader number of parameters, tools, and have easier Web-based access. Except for the Xlung, none of the simulators displayed monitoring of arterial blood gases and alternatives to load and save the simulation. The Xlung obtained the greater scores on heuristic principles assessments and the greater score of easiness of use, being the preferred MVVS for teaching purposes. No strong scientific evidence on the use and validation of the current MVVS was found.

CONCLUSIONS

There are only a few MVVS currently available. Among them, the Xlung showed a better usability interface. Validation tests and development of new or improvement of the current MVVS are needed.

Clinical and laboratory considerations for the rapid detection of carbapenem-resistant Enterobacteriaceae

Carbapenem resistance among the Enterobacteriaceae has become a significant clinical and public health dilemma. Rapid administration of effective antimicrobials and implementation of supplemental infection control practices is required to both improve patient outcomes and limit the spread of these highly resistant organisms. However, carbapenem-resistant Enterobacteriaceae (CRE)-infected patients are predominantly identified by routine culture methods, which take days to perform. Rapid genomic and phenotypic methods are currently available to accelerate the identification of carbapenemase-producing CRE. Effective use of these technologies is reliant on close collaboration between clinical microbiology, infection prevention, antimicrobial stewardship and infectious diseases specialists. This review discusses the performance characteristics of these technologies to date, and describes strategies for their optimal implementation.

Using what you get: dynamic physiologic signatures of critical illness

The development and resolution of cardiopulmonary instability take time to become clinically apparent, and the treatments provided take time to have an impact. The characterization of dynamic changes in hemodynamic and metabolic variables is implicit in physiologic signatures. When primary variables are collected with high enough frequency to derive new variables, this data hierarchy can be used to develop physiologic signatures. The creation of physiologic signatures requires no new information; additional knowledge is extracted from data that already exist. It is possible to create physiologic signatures for each stage in the process of clinical decompensation and recovery to improve outcomes.

Predicting bacteraemia in validated models--a systematic review

Bacteraemia is associated with high mortality. Although many models for predicting bacteraemia have been developed, not all have been validated, and even when they were, the validation processes varied. We identified validated models that have been developed; asked whether they were successful in defining groups with a very low or high prevalence of bacteraemia; and whether they were used in clinical practice. Electronic databases were searched to identify studies that underwent validation on prediction of bacteraemia in adults. We included only studies that were able to define groups with low or high probabilities for bacteraemia (arbitrarily defined as below 3% or above 30%). Fifteen publications fulfilled inclusion criteria, including 59 276 patients. Eleven were prospective and four retrospective. Study populations and the parameters included in the different models were heterogeneous. Ten studies underwent internal validation; the model performed well in all of them. Twelve performed external validation. Of the latter, seven models were validated in a different hospital, using a new independent database. In five of these, the model performed well. After contacting authors, we found that none of the models was implemented in clinical practice. We conclude that heterogeneous studies have been conducted in different defined groups of patients with limited external validation. Significant savings to the system and the individual patient can be gained by refraining from performing blood cultures in groups of patients in which the probability of true bacteraemia is very low, while the probability of contamination is constant. Clinical trials of existing or new models should be done to examine whether models are helpful and safe in clinical use, preferably multicentre in order to secure utility and safety in diverse clinical settings.

Long-term effect of computer-assisted decision support for antibiotic treatment in critically ill patients: a prospective 'before/after' cohort study

OBJECTIVES

Antibiotic resistance has risen dramatically over the past years. For individual patients, adequate initial antibiotic therapy is essential for clinical outcome. Computer-assisted decision support systems (CDSSs) are advocated to support implementation of rational anti-infective treatment strategies based on guidelines. The aim of this study was to evaluate long-term effects after implementation of a CDSS.

DESIGN

This prospective 'before/after' cohort study was conducted over four observation periods within 5 years. One preinterventional period (pre) was compared with three postinterventional periods: directly after intensive implementation efforts (post1), 2 years (post2) and 3 years (post3) after implementation.

SETTING

Five anaesthesiological-managed intensive care units (ICU) (one cardiosurgical, one neurosurgical, two interdisciplinary and one intermediate care) at a university hospital.

PARTICIPANTS

Adult patients with an ICU stay of >48 h were included in the analysis. 1316 patients were included in the analysis for a total of 12,965 ICU days.

INTERVENTION

Implementation of a CDSS.

OUTCOME MEASURES

The primary end point was percentage of days with guideline adherence during ICU treatment. Secondary end points were antibiotic-free days and all-cause mortality compared for patients with low versus high guideline adherence.

MAIN RESULTS

Adherence to guidelines increased from 61% prior to implementation to 92% in post1, decreased in post2 to 76% and remained significantly higher compared with baseline in post3, with 71% (p=0.178). Additionally, antibiotic-free days increased over study periods. At all time periods, mortality for patients with low guideline adherence was higher with 12.3% versus 8% (p=0.014) and an adjusted OR of 1.56 (95% CI 1.05 to 2.31).

CONCLUSIONS

Implementation of computerised regional adapted guidelines for antibiotic therapy is paralleled with improved adherence. Even without further measures, adherence stayed high for a longer period and was paralleled by reduced antibiotic exposure. Improved guideline adherence was associated with reduced ICU mortality.

TRIAL REGISTRATION NUMBER

ISRCTN54598675.

Clinical microbiology informatics

The clinical microbiology laboratory has responsibilities ranging from characterizing the causative agent in a patient's infection to helping detect global disease outbreaks. All of these processes are increasingly becoming partnered more intimately with informatics. Effective application of informatics tools can increase the accuracy, timeliness, and completeness of microbiology testing while decreasing the laboratory workload, which can lead to optimized laboratory workflow and decreased costs. Informatics is poised to be increasingly relevant in clinical microbiology, with the advent of total laboratory automation, complex instrument interfaces, electronic health records, clinical decision support tools, and the clinical implementation of microbial genome sequencing. This review discusses the diverse informatics aspects that are relevant to the clinical microbiology laboratory, including the following: the microbiology laboratory information system, decision support tools, expert systems, instrument interfaces, total laboratory automation, telemicrobiology, automated image analysis, nucleic acid sequence databases, electronic reporting of infectious agents to public health agencies, and disease outbreak surveillance. The breadth and utility of informatics tools used in clinical microbiology have made them indispensable to contemporary clinical and laboratory practice. Continued advances in technology and development of these informatics tools will further improve patient and public health care in the future.

Use of computerized decision support systems to improve antibiotic prescribing

This decade will see the emergence of the electronic medical record, electronic prescribing and computerized decision support in the hospital setting. Current opinion from key infectious diseases bodies supports the use of computerized decision support systems as potentially useful tools in antibiotic stewardship programs. However, although antibiotic decision support systems appear beneficial for improving the quality of prescribing and reducing the costs of antibiotic prescribing, their overall cost-effectiveness, impact on patient outcome and antimicrobial resistance is much less certain. This review describes computerized decision support systems used to assist with antibiotic prescribing, the evidence for their effectiveness and the current and future roles.

Application of probabilistic and fuzzy cognitive approaches in semantic web framework for medical decision support

This study aimed to focus on medical knowledge representation and reasoning using the probabilistic and fuzzy influence processes, implemented in the semantic web, for decision support tasks. Bayesian belief networks (BBNs) and fuzzy cognitive maps (FCMs), as dynamic influence graphs, were applied to handle the task of medical knowledge formalization for decision support. In order to perform reasoning on these knowledge models, a general purpose reasoning engine, EYE, with the necessary plug-ins was developed in the semantic web. The two formal approaches constitute the proposed decision support system (DSS) aiming to recognize the appropriate guidelines of a medical problem, and to propose easily understandable course of actions to guide the practitioners. The urinary tract infection (UTI) problem was selected as the proof-of-concept example to examine the proposed formalization techniques implemented in the semantic web. The medical guidelines for UTI treatment were formalized into BBN and FCM knowledge models. To assess the formal models' performance, 55 patient cases were extracted from a database and analyzed. The results showed that the suggested approaches formalized medical knowledge efficiently in the semantic web, and gave a front-end decision on antibiotics' suggestion for UTI.

Diagnosis support system based on clinical guidelines: comparison between case-based fuzzy cognitive maps and Bayesian networks

Several studies have described the prevalence and severity of diagnostic errors. Diagnostic errors can arise from cognitive, training, educational and other issues. Examples of cognitive issues include flawed reasoning, incomplete knowledge, faulty information gathering or interpretation, and inappropriate use of decision-making heuristics. We describe a new approach, case-based fuzzy cognitive maps, for medical diagnosis and evaluate it by comparison with Bayesian belief networks. We created a semantic web framework that supports the two reasoning methods. We used database of 174 anonymous patients from several European hospitals: 80 of the patients were female and 94 male with an average age 45±16 (average±stdev). Thirty of the 80 female patients were pregnant. For each patient, signs/symptoms/observables/age/sex were taken into account by the system. We used a statistical approach to compare the two methods.

A Bayesian approach to estimate the prevalence of Schistosomiasis japonica infection in the Hubei Province Lake Regions, China

A Bayesian inference model was introduced to estimate community prevalence of Schistosomiasis japonica infection based on the data of a large-scale survey of Schistosomiasis japonica in the lake region in Hubei Province. A multistage cluster random sampling approach was applied to the endemic villages in the lake regions of Hubei Province in 2011. IHA test and Kato-Katz test were applied for the detection of the S. japonica infection in the sampled population. Expert knowledge on sensitivities and specificities of IHA test and Kato-Katz test were collected based on a two-round interview. Prevalence of S. japonica infection was estimated by a Bayesian hierarchical model in two different situations. In Situation 1, Bayesian estimation used both IHA test data and Kato-Katz test data to estimate the prevalence of S. japonica. In Situation 2, only IHA test data was used for Bayesian estimation. Finally 14 cities and 46 villages from the lake regions of Hubei Province including 50,980 residents were sampled. Sensitivity and specificity for IHA test ranged from 80% to 90% and 70% to 80%, respectively. For the Kato-Katz test, sensitivity and specificity were from 20% to 70% and 90% to 100%, respectively. Similar estimated prevalence was obtained in the two situations. Estimated prevalence among sampled villages was almost below 13% in both situations and varied from 0.95% to 12.26% when only using data from the IHA test. The study indicated that it is feasible to apply IHA test only combining with Bayesian method to estimate the prevalence of S. japonica infection in large-scale surveys.

A new statistical approach to predict bacteremia using electronic medical records

BACKGROUND

Previous attempts to predict bacteremia have focused on selecting significant variables. However, these approaches have had limitations such as poor reproducibility in prediction accuracy and inconsistency in predictor selection. Here we propose a Bayesian approach to predict bacteremia based on the statistical distributions of clinical variables of previous patients, which has recently become possible through the adoption of electronic medical records.

METHODS

In a derivation cohort, Bayesian prediction models were derived and their discriminative performance was compared with previous models under varying combinations of predictors. Then the Bayesian models were prospectively tested in a validation cohort. According to Bayesian probabilities of bacteremia, patients in both cohorts were grouped into bacteremia risk groups.

RESULTS

Using the same prediction variables, the Bayesian predictions were more accurate than conventional rule-based predictions. Moreover, their better discriminative performance remained consistent despite variations in clinical variables. The receiver operating characteristic (ROC) area of the Bayesian model with 20 predictors was 0.70 ± 0.007 in the derivation cohort and 0.70 ± 0.018 in the validation cohort. The prevalence of bacteremia in groups I, II, and VI (grouped according to probability ratio) were 1.9%, 3.4%, and 20.0% in the derivation cohort, and 0.4%, 3.2%, and 18.4% in the validation cohort, respectively. The overall prevalence of bacteremia was 6.9% in both cohorts.

CONCLUSIONS

In the present study, the Bayesian prediction model showed stable performance in predicting bacteremia and identifying risk groups, as the previous models did. The clinical significance of the Bayesian approach is expected to be demonstrated through a multicenter trial.

Fuzzy cognitive map software tool for treatment management of uncomplicated urinary tract infection

Uncomplicated urinary tract infection (uUTI) is a bacterial infection that affects individuals with normal urinary tracts from both structural and functional perspective. The appropriate antibiotics and treatment suggestions to individuals suffer of uUTI is an important and complex task that demands a special attention. How to decrease the unsafely use of antibiotics and their consumption is an important issue in medical treatment. Aiming to model medical decision making for uUTI treatment, an innovative and flexible approach called fuzzy cognitive maps (FCMs) is proposed to handle with uncertainty and missing information. The FCM is a promising technique for modeling knowledge and/or medical guidelines/treatment suggestions and reasoning with it. A software tool, namely FCM-uUTI DSS, is investigated in this work to produce a decision support module for uUTI treatment management. The software tool was tested (evaluated) in a number of 38 patient cases, showing its functionality and demonstrating that the use of the FCMs as dynamic models is reliable and good. The results have shown that the suggested FCM-uUTI tool gives a front-end decision on antibiotics' suggestion for uUTI treatment and are considered as helpful references for physicians and patients. Due to its easy graphical representation and simulation process the proposed FCM formalization could be used to make the medical knowledge widely available through computer consultation systems.

ICU 2020: five interventions to revolutionize quality of care in the ICU

Intensive care units (ICUs) are an essential and unique component of modern medicine. The number of critically ill individuals, complexity of illness, and cost of care continue to increase with time. In order to meet future demands, maintain quality, and minimize medical errors, intensivists will need to look beyond traditional medical practice, seeking lessons on quality assurance from industry and aviation. Intensivists will be challenged to keep pace with rapidly advancing information technology and its diverse roles in ICU care delivery. Modern ICU quality improvement initiatives include ensuring evidence-based best practice, participation in multicenter ICU collaborations, employing state-of-the-art information technology, providing point-of-care diagnostic testing, and efficient organization of ICU care delivery. This article demonstrates that each of these initiatives has the potential to revolutionize the quality of future ICU care in the United States.

Impact of protocol watch on compliance with the surviving sepsis campaign

PURPOSE

Clinical decision support systems are intended to improve patients' care and outcomes, particularly when such systems are present at the point of care. Protocol Watch was developed as a bedside clinical decision support system to improve clinicians' adherence to the Surviving Sepsis Campaign guidelines. This pre/post-intervention pilot study was done to evaluate the effect of Protocol Watch on compliance with 5 guidelines from the Surviving Sepsis Campaign.

METHODS

Preintervention data on rates and time to complete the resuscitation and management bundles from the Surviving Sepsis Campaign and time to administer antibiotics were collected from intensive care units at 2 large teaching hospitals in the United States. Training on the Protocol Watch application was then provided to clinical staff in the units, and Protocol Watch was installed at all critical care beds in both hospitals. Data were collected on rates and time to completion for 5 Surviving Sepsis Campaign guidelines after installation of Protocol Watch, and univariate analyses were done to evaluate the effect of Protocol Watch on compliance with the guidelines.

RESULTS

Implementation of Protocol Watch was associated with significant improvements in compliance with the resuscitation bundle (P = .01) and decreased time to administer antibiotics (P = .006). No significant changes were achieved for compliance with the management bundle or time to complete the resuscitation or management bundles.

CONCLUSIONS

Clinical decision support systems such as Protocol Watch may improve adherence to the Surviving Sepsis Campaign guidelines, which potentially may contribute to reduced morbidity and mortality for critically ill patients with sepsis.

Computational classifiers for predicting the short-term course of Multiple sclerosis

BACKGROUND

The aim of this study was to assess the diagnostic accuracy (sensitivity and specificity) of clinical, imaging and motor evoked potentials (MEP) for predicting the short-term prognosis of multiple sclerosis (MS).

METHODS

We obtained clinical data, MRI and MEP from a prospective cohort of 51 patients and 20 matched controls followed for two years. Clinical end-points recorded were: 1) expanded disability status scale (EDSS), 2) disability progression, and 3) new relapses. We constructed computational classifiers (Bayesian, random decision-trees, simple logistic-linear regression-and neural networks) and calculated their accuracy by means of a 10-fold cross-validation method. We also validated our findings with a second cohort of 96 MS patients from a second center.

RESULTS

We found that disability at baseline, grey matter volume and MEP were the variables that better correlated with clinical end-points, although their diagnostic accuracy was low. However, classifiers combining the most informative variables, namely baseline disability (EDSS), MRI lesion load and central motor conduction time (CMCT), were much more accurate in predicting future disability. Using the most informative variables (especially EDSS and CMCT) we developed a neural network (NNet) that attained a good performance for predicting the EDSS change. The predictive ability of the neural network was validated in an independent cohort obtaining similar accuracy (80%) for predicting the change in the EDSS two years later.

CONCLUSIONS

The usefulness of clinical variables for predicting the course of MS on an individual basis is limited, despite being associated with the disease course. By training a NNet with the most informative variables we achieved a good accuracy for predicting short-term disability.

Computer-assisted decision support for changing practice in severe sepsis and septic shock

Computer-assisted decision support systems (CDSS) are designed to improve infection management. The aim of this prospective, clinical pre- and post-intervention study was to investigate the influence of CDSS on infection management of severe sepsis and septic shock in intensive care units (ICUs). Data were collected for a total of 180 days during two study periods in 2006 and 2007. Of the 186 patients with severe sepsis or septic shock, 62 were stratified into a low adherence to infection management standards group (LAG) and 124 were stratified into a high adherence group (HAG). ICU mortality was significantly increased in LAG versus HAG patients (Kaplan-Meier analysis). Following CDSS implementation, adherence to standards increased significantly by 35%, paralleled with improved diagnostics, more antibiotic-free days and a shortened time until antibiotics were administered. In conclusion, adherence to infection standards is beneficial for patients with severe sepsis or septic shock and CDSS is a useful tool to aid adherence.

[The new method for detecting the outbreak sign of MRSA]

Conventional outbreak detection laboratory-based made one unit from the beginning of the month to the end of the month, totaled and analyzed, cannot correctly detect outbreaks continued during two months. The real-time analysis (RTA) we devised adapts to methicillin-resistant Staphylococcus aureus (MRSA) and avoids the problems of conventional detection. RTA analyzes all data for the last 30 days when MRSA is newly isolated 48 hours or more after hospital admission. In the three years from April 2006 to March 2009, we compared the day and number of MRSA outbreaks newly isolated 48 hours or more after hospital admission in 572 subjects using the conventional method and RTA. We also calculated the RTA infection prevention effect. The number of outbreaks detected conventionally numbered 68 cases and those detected by RTA numbered 106 cases. The number of outbreaks newly detected by RTA numbered 38 cases in three years, averaging 4.3 days earlier than conventional detection using conventional method A an average of 15.7 days earlier than conventionally which totals for every end of the month using conventional method B. The effect of infection prevention in the change of RTA from conventional method A presumably decreases MRSA infection to 14-18 persons and it in the change of RTA from conventional method B decreases MRSA infection to 18-25 persons in one year. These results suggested that outbreak detection by RTA could help prevent MRSA outbreak and decrease MRSA infection frequency.

Simulation evaluation of an enhanced bedside monitor display for patients with sepsis

BACKGROUND

Most standard bedside monitors in critical care settings display multiple clinical parameters and do not provide directional signaling to alert clinicians to relevant changes in physiologic parameters. The complexity of information may delay identification of clinical changes and initiation of interventions. Clinical decision support system (CDSS) tools can present information at appropriate intervals in formats that may enhance clinical decision making.

OBJECTIVE

A 2-group, quasi-experimental design compared the effects of 2 different monitor displays on time required for nurses to recognize and initiate treatment of sepsis in response to a computer simulation.

METHODS

A convenience sample of 75 critical care nurses was randomly assigned to a standard or an enhanced bedside monitor (EBM) display during a computer-simulated sepsis scenario. Time to recognize symptoms and initiate treatment of sepsis was analyzed between the 2 displays.

RESULTS

Time to recognize symptoms and initiate sepsis treatment was significantly shorter for nurses exposed to the enhanced bedside monitor.

CONCLUSIONS

CDSS tools such as EBM may improve nurses' time to recognize and initiate treatment of sepsis.

A regret theory approach to decision curve analysis: a novel method for eliciting decision makers' preferences and decision-making

BACKGROUND

Decision curve analysis (DCA) has been proposed as an alternative method for evaluation of diagnostic tests, prediction models, and molecular markers. However, DCA is based on expected utility theory, which has been routinely violated by decision makers. Decision-making is governed by intuition (system 1), and analytical, deliberative process (system 2), thus, rational decision-making should reflect both formal principles of rationality and intuition about good decisions. We use the cognitive emotion of regret to serve as a link between systems 1 and 2 and to reformulate DCA.

METHODS

First, we analysed a classic decision tree describing three decision alternatives: treat, do not treat, and treat or no treat based on a predictive model. We then computed the expected regret for each of these alternatives as the difference between the utility of the action taken and the utility of the action that, in retrospect, should have been taken. For any pair of strategies, we measure the difference in net expected regret. Finally, we employ the concept of acceptable regret to identify the circumstances under which a potentially wrong strategy is tolerable to a decision-maker.

RESULTS

We developed a novel dual visual analog scale to describe the relationship between regret associated with "omissions" (e.g. failure to treat) vs. "commissions" (e.g. treating unnecessary) and decision maker's preferences as expressed in terms of threshold probability. We then proved that the Net Expected Regret Difference, first presented in this paper, is equivalent to net benefits as described in the original DCA. Based on the concept of acceptable regret we identified the circumstances under which a decision maker tolerates a potentially wrong decision and expressed it in terms of probability of disease.

CONCLUSIONS

We present a novel method for eliciting decision maker's preferences and an alternative derivation of DCA based on regret theory. Our approach may be intuitively more appealing to a decision-maker, particularly in those clinical situations when the best management option is the one associated with the least amount of regret (e.g. diagnosis and treatment of advanced cancer, etc).

Clinical diagnostics in human genetics with semantic similarity searches in ontologies

The differential diagnostic process attempts to identify candidate diseases that best explain a set of clinical features. This process can be complicated by the fact that the features can have varying degrees of specificity, as well as by the presence of features unrelated to the disease itself. Depending on the experience of the physician and the availability of laboratory tests, clinical abnormalities may be described in greater or lesser detail. We have adapted semantic similarity metrics to measure phenotypic similarity between queries and hereditary diseases annotated with the use of the Human Phenotype Ontology (HPO) and have developed a statistical model to assign p values to the resulting similarity scores, which can be used to rank the candidate diseases. We show that our approach outperforms simpler term-matching approaches that do not take the semantic interrelationships between terms into account. The advantage of our approach was greater for queries containing phenotypic noise or imprecise clinical descriptions. The semantic network defined by the HPO can be used to refine the differential diagnosis by suggesting clinical features that, if present, best differentiate among the candidate diagnoses. Thus, semantic similarity searches in ontologies represent a useful way of harnessing the semantic structure of human phenotypic abnormalities to help with the differential diagnosis. We have implemented our methods in a freely available web application for the field of human Mendelian disorders.

Computer says 2.5 litres--how best to incorporate intelligent software into clinical decision making in the intensive care unit?

What will be the role of the intensivist when computer-assisted decision support reaches maturity? Celi's group reports that Bayesian theory can predict a patient's fluid requirement on day 2 in 78% of cases, based on data collected on day 1 and the known associations between those data, based on observations in previous patients in their unit. There are both advantages and limitations to the Bayesian approach, and this test study identifies areas for improvement in future models. Although such models have the potential to improve diagnostic and therapeutic accuracy, they must be introduced judiciously and locally to maximize their effect on patient outcome. Efficacy is thus far undetermined, and these novel approaches to patient management raise new challenges, not least medicolegal ones.

A hybrid hierarchical decision support system for cardiac surgical intensive care patients. Part I: Physiological modelling and decision support system design

OBJECTIVE

To develop a clinical decision support system (CDSS) that models the different levels of the clinician's decision-making strategies when controlling post cardiac surgery patients weaned from cardio pulmonary bypass.

METHODS

A clinical trial was conducted to define and elucidate an expert anesthetists' decision pathway utilised in controlling this patient population. This data and derived knowledge were used to elicit a decision-making model. The structural framework of the decision-making model is hierarchical, clearly defined, and dynamic. The decision levels are linked to five important components of the cardiovascular physiology in turn, i.e. the systolic blood pressure (SBP), central venous pressure (CVP), systemic vascular resistance (SVR), cardiac output (CO), and heart rate (HR). Progress down the hierarchy is dependent upon the normalisation of each physiological parameter to a value pre-selected by the clinician via fluid, chronotropes or inotropes. Since interventions at each and every level cause changes and disturbances in the other components, the proposed decision support model continuously refers back decision outcomes back to the SBP which is considered to be the overriding supervisory safety component in this hierarchical decision structure. The decision model was then translated into a computerised decision support system prototype and comprehensively tested on a physiological model of the human cardiovascular system. This model was able to reproduce conditions experienced by post-operative cardiac surgery patients including hypertension, hypovolemia, vasodilation and the systemic inflammatory response syndrome (SIRS).

RESULTS

In all the simulated patients scenarios considered the CDSS was able to initiate similar therapeutic interventions to that of the expert, and as a result, was also able to control the hemodynamic parameters to the prescribed target values.

An artificial intelligence tool to predict fluid requirement in the intensive care unit: a proof-of-concept study

Introduction

The goal of personalised medicine in the intensive care unit (ICU) is to predict which diagnostic tests, monitoring interventions and treatments translate to improved outcomes given the variation between patients. Unfortunately, processes such as gene transcription and drug metabolism are dynamic in the critically ill; that is, information obtained during static non-diseased conditions may have limited applicability. We propose an alternative way of personalising medicine in the ICU on a real-time basis using information derived from the application of artificial intelligence on a high-resolution database. Calculation of maintenance fluid requirement at the height of systemic inflammatory response was selected to investigate the feasibility of this approach.

Methods

The Multi-parameter Intelligent Monitoring for Intensive Care II (MIMIC II) is a database of patients admitted to the Beth Israel Deaconess Medical Center ICU in Boston. Patients who were on vasopressors for more than six hours during the first 24 hours of admission were identified from the database. Demographic and physiological variables that might affect fluid requirement or reflect the intravascular volume during the first 24 hours in the ICU were extracted from the database. The outcome to be predicted is the total amount of fluid given during the second 24 hours in the ICU, including all the fluid boluses administered.

Results

We represented the variables by learning a Bayesian network from the underlying data. Using 10-fold cross-validation repeated 100 times, the accuracy of the model in predicting the outcome is 77.8%. The network generated has a threshold Bayes factor of seven representing the posterior probability of the model given the observed data. This Bayes factor translates into p < 0.05 assuming a Gaussian distribution of the variables.

Conclusions

Based on the model, the probability that a patient would require a certain range of fluid on day two can be predicted. In the presence of a larger database, analysis may be limited to patients with identical clinical presentation, demographic factors, co-morbidities, current physiological data and those who did not develop complications as a result of fluid administration. By better predicting maintenance fluid requirements based on the previous day's physiological variables, one might be able to prevent hypotensive episodes requiring fluid boluses during the course of the following day.

Clinical decision support software for chronic heart failure

Critical care pathways, protocols, and guidelines have become an everyday feature of clinical practice and represent a distillation of the best available evidence. Chronic heart failure guidelines can be complex, and it is acknowledged that a combination of knowledge and expert advice, in addition to guidelines, is required to optimally treat these patients. This current article describes the potential value of clinical decision support software (CDSS) in the treatment of patients with chronic heart failure and practical aspects of using such a tool. Barriers to implementation of our tool included relatively low computer skills among family physicians and a lack of complexity within CDSS in addressing the wider nonmedical needs of patients. Improving computer skills, integrating CDSS into referral pathways, and requests for investigation may be ways of enhancing the use of this technology.

From inverse problems in mathematical physiology to quantitative differential diagnoses

The improved capacity to acquire quantitative data in a clinical setting has generally failed to improve outcomes in acutely ill patients, suggesting a need for advances in computer-supported data interpretation and decision making. In particular, the application of mathematical models of experimentally elucidated physiological mechanisms could augment the interpretation of quantitative, patient-specific information and help to better target therapy. Yet, such models are typically complex and nonlinear, a reality that often precludes the identification of unique parameters and states of the model that best represent available data. Hypothesizing that this non-uniqueness can convey useful information, we implemented a simplified simulation of a common differential diagnostic process (hypotension in an acute care setting), using a combination of a mathematical model of the cardiovascular system, a stochastic measurement model, and Bayesian inference techniques to quantify parameter and state uncertainty. The output of this procedure is a probability density function on the space of model parameters and initial conditions for a particular patient, based on prior population information together with patient-specific clinical observations. We show that multimodal posterior probability density functions arise naturally, even when unimodal and uninformative priors are used. The peaks of these densities correspond to clinically relevant differential diagnoses and can, in the simplified simulation setting, be constrained to a single diagnosis by assimilating additional observations from dynamical interventions (e.g., fluid challenge). We conclude that the ill-posedness of the inverse problem in quantitative physiology is not merely a technical obstacle, but rather reflects clinical reality and, when addressed adequately in the solution process, provides a novel link between mathematically described physiological knowledge and the clinical concept of differential diagnoses. We outline possible steps toward translating this computational approach to the bedside, to supplement today's evidence-based medicine with a quantitatively founded model-based medicine that integrates mechanistic knowledge with patient-specific information.

Although quantitative physiology has developed numerous mathematical descriptions of components of the human body, their application in clinical medicine has been limited to a few mostly primitive and physiologically inaccurate models. One reason for this is that the inverse problem of identifying unknown model parameters and states from prior knowledge and clinical observations does not usually have a unique solution. Hypothesizing that this non-uniqueness might actually convey clinically useful information, we used a simplified mathematical model of the cardiovascular system and its control, in combination with Bayesian inference techniques, to simulate the diagnosis of low blood pressure in an acute care setting. The inference procedure yielded a distribution of physiologically interpretable model parameters and states that exhibited multiple peaks. The key observation was that these peaks corresponded directly to clinically relevant differential diagnoses, enabling a quantitative, probabilistic assessment of the simulated patient's condition. We conclude that the proposed probabilistic approach to the inverse problem in quantitative physiology may not only be useful for quantitative interpretation of clinical data, and eventually allow model-based prediction and therapy optimization, but also provides a novel link between mathematically described physiological mechanisms and the clinical concept of differential diagnoses based on patient-specific information.

A Bayesian decision-support system for diagnosing ventilator-associated pneumonia

OBJECTIVE

To determine the diagnostic performance of a Bayesian Decision-Support System (BDSS) for ventilator-associated pneumonia (VAP).

DESIGN

A previously developed BDSS, automatically obtaining patient data from patient information systems, provides likelihood predictions of VAP. In a prospectively studied cohort of 872 ICU patients, VAP was diagnosed by two infectious-disease specialists using a decision tree (reference diagnosis). After internal validation daily BDSS predictions were compared with the reference diagnosis. For data analysis two approaches were pursued: using BDSS predictions (a) for all 9422 patient days, and (b) only for the 238 days with presumed respiratory tract infections (RTI) according to the responsible physicians.

MEASUREMENTS AND RESULTS

157 (66%) of 238 days with presumed RTI fulfilled criteria for VAP. In approach (a), median daily BDSS likelihood predictions for days with and without VAP were 77% [Interquartile range (IQR) = 56-91%] and 14% [IQR 5-42%, p < 0.001, Mann-Whitney U-test (MWU)], respectively. In receiver operating characteristics (ROC) analysis, optimal BDSS cut-off point for VAP was 46%, and with this cut-off point positive predictive value (PPV) and negative predictive value (NPV) were 6.1 and 99.6%, respectively [AUC = 0.857 (95% CI 0.827-0.888)]. In approach (b), optimal cut-off for VAP was 78%, and with this cut-off point PPV and NPV were 86 and 66%, respectively [AUC = 0.846 (95% CI 0.794-0.899)].

CONCLUSIONS

As compared with the reference diagnosis, the BDSS had good test characteristics for diagnosing VAP, and might become a useful tool for assisting ICU physicians, both for routinely daily assessment and in patients clinically suspected of having VAP. Empirical validation of its performance is now warranted.

Clinical decision support systems and antibiotic use

AIM

To review and appraise randomised controlled trials (RCT) and 'before and after' studies published on clinical decision support systems (CDSS) used to support the use of antibiotics.

METHODS

A literature search was carried out in October 2006 using MEDLINE including Medical Subject Heading (MeSH) terms (1966-2006), EMBASE (Excerpta Medica, 1980-2006) and International Pharmaceutical Abstracts (IPA, 1970-2006) using the combinations of the following terms: (Decision support systems) or (CDSS) AND (antibiotics) or (anti-infectives) or (antibacterials) or (antimicrobials). Only English language papers were selected. Editorials, letters and case reports/series were excluded. The reference sections of all retrieved articles were also searched for any further relevant articles.

RESULTS

Forty articles were identified. Five RCT and six 'before and after' studies were retrieved. In the RCTs, three studies used computer-based CDSS, one paper-based CDSS and one a combination of both. Two studies were conducted in primary care and three within secondary care. The primary outcomes for each study were different and only three studies were significant in the favour of the use of CDSS. 'Before and after' studies were used where RCT were not feasible. One 'before and after' study was excluded because it did not include any control group. The remaining five included historical control groups and evaluated the use of computer-based CDSS within secondary care. Their primary outcomes also varied but all concluded significant benefits of CDSS. Only three of ten studies were conducted outside the USA; one in Switzerland and two in Australia.

CONCLUSION

CDSS could be a powerful tool to improve clinical care and patient outcomes. It presents a promising future for optimising antibiotic use. However, it is difficult to generalise as most studies were conducted in the United States. Although RCT are the 'gold standard' in research, they may not be feasible to conduct. Realising that different study designs answer different questions would allow researchers to choose the most appropriate study design to evaluate CDSS in a specified setting.

Doctors – A species on the verge of extinction?

Medicine is undergoing profound change, but the basic format of the medical encounter has remained unchanged. Nevertheless, medicine in the 22nd century may be fully computerized, and a possible model is shortly depicted in this paper. Computer applications are constantly increasing their share in medical diagnosis, and may ultimately replace physicians. Treatment decisions have been submitted to standardized treatment guidelines, which may be applied more efficiently by computer applications. Although hundreds of studies have evaluated computerized tools in diagnosis and treatment, the possibility that computer applications may replace human physicians in the future is rarely raised. The effects of this process on doctors and medicine may be tremendous and will probably be felt even in early stages, and therefore, this process should be a subject of open discussion.

Nosocomial pneumonia

PURPOSE OF REVIEW

This review describes the most recent advances in the management and prevention of nosocomial pneumonia. The new ATS guidelines in particular are most likely to affect clinical practice outside the USA.

RECENT FINDINGS

The problem of multidrug-resistant bacteria causing nosocomial pneumonia seems to be increasing. This is particularly true for methicillin-resistant Staphylococcus aureus. While the diagnosis of ventilator associated pneumonia remains a conflictive issue, serial tracheobronchial aspirates may improve the selection of adequate antimicrobial treatment. Combined beta-lactam and aminoglycoside therapy is inferior to beta-lactam monotherapy, both in terms of clinical outcome and in the prevention of resistance during treatment; in addition, it carries an increased risk of nephrotoxicity.

SUMMARY

The updated ATS guidelines will considerably impact clinical approaches to nosocomial and healthcare-related pneumonia. Serial tracheobronchial aspirates can be used to guide selection of antimicrobial treatment in ventilator associated pneumonia. The combination of beta-lactams and aminoglycosides is likely to be abandoned in the future. New potent treatment options for pneumonia due to nonfermenters are urgently needed.