Evaluation of a clinical tool for early etiology identification in status epilepticus

OBJECTIVES

Because early etiologic identification is critical to select appropriate specific status epilepticus (SE) management, we aim to validate a clinical tool we developed that uses history and readily available investigations to guide prompt etiologic assessment.

METHODS

This prospective multicenter study included all adult patients treated for SE of all but anoxic causes from four academic centers. The proposed tool is designed as a checklist covering frequent precipitating factors for SE. The study team completed the checklist at the time the patient was identified by electroencephalography (EEG) request. Only information available in the emergency department or at the time of in-hospital SE identification was used. Concordance between the etiology indicated by the tool and the determined etiology at hospital discharge was analyzed, together with interrater agreement.

RESULTS

Two hundred twelve patients were included. Concordance between the etiology hypothesis generated using the tool and the finally determined etiology was 88.7% (95% confidence interval (CI) 86.4-89.8) (κ = 0.88). Interrater agreement was 83.3% (95% CI 80.4-96) (κ = 0.81).

SIGNIFICANCE

This tool is valid and reliable for identification early the etiology of an SE. Physicians managing patients in SE may benefit from using it to identify promptly the underlying etiology, thus facilitating selection of the appropriate treatment.

The standardization debate: A conflation trap in critical care electroencephalography

PURPOSE

Persistent uncertainty over the clinical significance of various pathological continuous electroencephalography (cEEG) findings in the intensive care unit (ICU) has prompted efforts to standardize ICU cEEG terminology and an ensuing debate. We set out to understand the reasons for, and a satisfactory resolution to, this debate.

METHOD

We review the positions for and against standardization, and examine their deeper philosophical basis.

RESULTS

We find that the positions for and against standardization are not fundamentally irreconcilable. Rather, both positions stem from conflating the three cardinal steps in the classic approach to EEG, which we term "description", "interpretation", and "prescription". Using real-world examples we show how this conflation yields muddled clinical reasoning and unproductive debate among electroencephalographers that is translated into confusion among treating clinicians. We propose a middle way that judiciously uses both standardized terminology and clinical reasoning to disentangle these critical steps and apply them in proper sequence.

CONCLUSION

The systematic approach to ICU cEEG findings presented herein not only resolves the standardization debate but also clarifies clinical reasoning by helping electroencephalographers assign appropriate weights to cEEG findings in the face of uncertainty.

Sparse extreme learning machine for classification

Extreme learning machine (ELM) was initially proposed for single-hidden-layer feedforward neural networks (SLFNs). In the hidden layer (feature mapping), nodes are randomly generated independently of training data. Furthermore, a unified ELM was proposed, providing a single framework to simplify and unify different learning methods, such as SLFNs, least square support vector machines, proximal support vector machines, and so on. However, the solution of unified ELM is dense, and thus, usually plenty of storage space and testing time are required for large-scale applications. In this paper, a sparse ELM is proposed as an alternative solution for classification, reducing storage space and testing time. In addition, unified ELM obtains the solution by matrix inversion, whose computational complexity is between quadratic and cubic with respect to the training size. It still requires plenty of training time for large-scale problems, even though it is much faster than many other traditional methods. In this paper, an efficient training algorithm is specifically developed for sparse ELM. The quadratic programming problem involved in sparse ELM is divided into a series of smallest possible sub-problems, each of which are solved analytically. Compared with SVM, sparse ELM obtains better generalization performance with much faster training speed. Compared with unified ELM, sparse ELM achieves similar generalization performance for binary classification applications, and when dealing with large-scale binary classification problems, sparse ELM realizes even faster training speed than unified ELM.

Weighing the value of memory loss in the surgical evaluation of left temporal lobe epilepsy: a decision analysis

OBJECTIVES

Anterior temporal lobectomy is curative for many patients with disabling medically refractory temporal lobe epilepsy, but carries an inherent risk of disabling verbal memory loss. Although accurate prediction of iatrogenic memory loss is becoming increasingly possible, it remains unclear how much weight such predictions should have in surgical decision making. Here we aim to create a framework that facilitates a systematic and integrated assessment of the relative risks and benefits of surgery versus medical management for patients with left temporal lobe epilepsy.

METHODS

We constructed a Markov decision model to evaluate the probabilistic outcomes and associated health utilities associated with choosing to undergo a left anterior temporal lobectomy versus continuing with medical management for patients with medically refractory left temporal lobe epilepsy. Three base-cases were considered, representing a spectrum of surgical candidates encountered in practice, with varying degrees of epilepsy-related disability and potential for decreased quality of life in response to post-surgical verbal memory deficits.

RESULTS

For patients with moderately severe seizures and moderate risk of verbal memory loss, medical management was the preferred decision, with increased quality-adjusted life expectancy. However, the preferred choice was sensitive to clinically meaningful changes in several parameters, including quality of life impact of verbal memory decline, quality of life with seizures, mortality rate with medical management, probability of remission following surgery, and probability of remission with medical management.

SIGNIFICANCE

Our decision model suggests that for patients with left temporal lobe epilepsy, quantitative assessment of risk and benefit should guide recommendation of therapy. In particular, risk for and potential impact of verbal memory decline should be carefully weighed against the degree of disability conferred by continued seizures on a patient-by-patient basis.

Aspirin for secondary prevention after stroke of unknown etiology in resource-limited settings

OBJECTIVE

To analyze the potential impact of aspirin therapy for long-term secondary prevention after stroke of undetermined etiology in resource-limited settings without access to neuroimaging to distinguish ischemic stroke from intracerebral hemorrhage (ICH).

METHODS

We conducted a decision analysis using a Markov state transition model. Sensitivity analyses were performed across the worldwide reported range of the proportion of strokes due to ICH and the 95% confidence intervals (CIs) of aspirin-associated relative risks in patients with ICH.

RESULTS

For patients with stroke of undetermined etiology, long-term aspirin was the preferred treatment strategy across the worldwide reported range of the proportion of strokes due to ICH. At 34% of strokes due to ICH (the highest proportion reported in a large epidemiologic study), the benefit of aspirin remained beyond the upper bounds of the 95% CIs of aspirin-associated post-ICH relative risks most concerning to clinicians (ICH recurrence risk and mortality risk if ICH recurs on aspirin). Based on the estimated 11,590,204 strokes in low- and middle-income countries in 2010, our model predicts that aspirin therapy for secondary stroke prevention in all patients with stroke in these countries could lead to an estimated yearly decrease of 84,492 recurrent strokes and 4,056 stroke-related mortalities.

CONCLUSIONS

The concern that the risks of aspirin in patients with stroke of unknown etiology could outweigh the benefits is not supported by our model, which predicts that aspirin for secondary prevention in patients with stroke of undetermined etiology in resource-limited settings could lead to decreased stroke-related mortality and stroke recurrence.

Aspirin for acute stroke of unknown etiology in resource-limited settings: a decision analysis

OBJECTIVE

To analyze the potential impact of aspirin on outcome at hospital discharge after acute stroke in resource-limited settings without access to neuroimaging to distinguish ischemic stroke from intracerebral hemorrhage (ICH).

METHODS

A decision analysis was conducted to evaluate aspirin use in all patients with acute stroke of unknown type for the duration of initial hospitalization. Data were obtained from the International Stroke Trial and Chinese Acute Stroke Trial. Predicted in-hospital mortality and stroke recurrence risk were determined across the worldwide reported range of the proportion of strokes caused by ICH. Sensitivity analyses were performed on aspirin-associated relative risks in patients with ICH.

RESULTS

At the highest reported proportion of strokes due to ICH from a large epidemiologic study (34% in sub-Saharan Africa), aspirin initiation after acute stroke of undetermined etiology is predicted to reduce in-hospital mortality (from 85/1,000 without treatment to 81/1,000 with treatment), in-hospital stroke recurrence (58/1,000 to 50/1,000), and combined risk of in-hospital mortality or stroke recurrence (127/1,000 to 114/1,000). Benefits of aspirin therapy remained in sensitivity analyses across a range of plausible parameter estimates for relative risks associated with aspirin initiation after ICH.

CONCLUSION

Aspirin treatment for the period of initial hospitalization after acute stroke of undetermined etiology is predicted to decrease acute stroke-related mortality and in-hospital stroke recurrence even at the highest reported proportion of acute strokes due to ICH. In the absence of clinical trials to test this approach empirically, clinical decisions require patient-specific evaluation of risks and benefits of aspirin in this context.

The probability of seizures during EEG monitoring in critically ill adults

OBJECTIVE

To characterize the risk for seizures over time in relation to EEG findings in hospitalized adults undergoing continuous EEG monitoring (cEEG).

METHODS

Retrospective analysis of cEEG data and medical records from 625 consecutive adult inpatients monitored at a tertiary medical center. Using survival analysis methods, we estimated the time-dependent probability that a seizure will occur within the next 72-h, if no seizure has occurred yet, as a function of EEG abnormalities detected so far.

RESULTS

Seizures occurred in 27% (168/625). The first seizure occurred early (<30min of monitoring) in 58% (98/168). In 527 patients without early seizures, 159 (30%) had early epileptiform abnormalities, versus 368 (70%) without. Seizures were eventually detected in 25% of patients with early epileptiform discharges, versus 8% without early discharges. The 72-h risk of seizures declined below 5% if no epileptiform abnormalities were present in the first two hours, whereas 16h of monitoring were required when epileptiform discharges were present. 20% (74/388) of patients without early epileptiform abnormalities later developed them; 23% (17/74) of these ultimately had seizures. Only 4% (12/294) experienced a seizure without preceding epileptiform abnormalities.

CONCLUSIONS

Seizure risk in acute neurological illness decays rapidly, at a rate dependent on abnormalities detected early during monitoring. This study demonstrates that substantial risk stratification is possible based on early EEG abnormalities.

SIGNIFICANCE

These findings have implications for patient-specific determination of the required duration of cEEG monitoring in hospitalized patients.

Interrater agreement for Critical Care EEG Terminology

OBJECTIVE

The interpretation of critical care electroencephalography (EEG) studies is challenging because of the presence of many periodic and rhythmic patterns of uncertain clinical significance. Defining the clinical significance of these patterns requires standardized terminology with high interrater agreement (IRA). We sought to evaluate IRA for the final, published American Clinical Neurophysiology Society (ACNS)-approved version of the critical care EEG terminology (2012 version). Our evaluation included terms not assessed previously and incorporated raters with a broad range of EEG reading experience.

METHODS

After reviewing a set of training slides, 49 readers independently completed a Web-based test consisting of 11 identical questions for each of 37 EEG samples (407 questions). Questions assessed whether a pattern was an electrographic seizure; pattern location (main term 1), pattern type (main term 2); and presence and classification of eight other key features ("plus" modifiers, sharpness, absolute and relative amplitude, frequency, number of phases, fluctuation/evolution, and the presence of "triphasic" morphology).

RESULTS

IRA statistics (κ values) were almost perfect (90-100%) for seizures, main terms 1 and 2, the +S modifier (superimposed spikes/sharp waves or sharply contoured rhythmic delta activity), sharpness, absolute amplitude, frequency, and number of phases. Agreement was substantial for the +F (superimposed fast activity) and +R (superimposed rhythmic delta activity) modifiers (66% and 67%, respectively), moderate for triphasic morphology (58%), and fair for evolution (21%).

SIGNIFICANCE

IRA for most terms in the ACNS critical care EEG terminology is high. These terms are suitable for multicenter research on the clinical significance of critical care EEG patterns. A PowerPoint slide summarizing this article is available for download in the Supporting Information section http://dx.doi.org/10.1111/epi.12653/supinfo.

The information theoretic perspective on medical diagnostic inference

The goal of this work is to present information theory, specifically Claude Shannon's mathematical theory of communication, in a clinical context and elucidate its potential contributions to understanding the process of diagnostic inference. We use probability theory, information theory, and clinical examples to develop information theory as a means to examine uncertainty in diagnostic testing situations. We begin our discussion with a brief review of probability theory as it relates to diagnostic testing. An outline of Shannon's theory of communication theory and how it directly translates to the medical diagnostic process serves as the essential justification for this article. Finally, we introduce the mathematical tools of information theory that allow for an understanding of diagnostic uncertainty and test effectiveness in a variety of contexts. We show that information theory provides a quantitative framework for understanding uncertainty that readily extends to medical diagnostic contexts.

Spectrogram screening of adult EEGs is sensitive and efficient

OBJECTIVE

Quantitatively evaluate whether screening with compressed spectral arrays (CSAs) is a practical and time-effective protocol for assisting expert review of continuous EEG (cEEG) studies in hospitalized adults.

METHODS

Three neurophysiologists reviewed the reported findings of the first 30 minutes of 118 cEEGs, then used CSA to guide subsequent review ("CSA-guided review" protocol). Reviewers viewed 120 seconds of raw EEG data surrounding suspicious CSA segments. The same neurophysiologists performed independent page-by-page visual interpretation ("conventional review") of all cEEGs. Independent conventional review by 2 additional, more experienced neurophysiologists served as a gold standard. We compared review times and detection rates for seizures and other pathologic patterns relative to conventional review.

RESULTS

A total of 2,092 hours of cEEG data were reviewed. Average times to review 24 hours of cEEG data were 8 (±4) minutes for CSA-guided review vs 38 (±17) minutes for conventional review (p < 0.005). Studies containing seizures required longer review: 10 (±4) minutes for CSA-guided review vs 44 (±20) minutes for conventional review (p < 0.005). CSA-guided review was sensitive for seizures (87.3%), periodic epileptiform discharges (100%), rhythmic delta activity (97.1%), focal slowing (98.7%), generalized slowing (100%), and epileptiform discharges (88.5%).

CONCLUSIONS

CSA-guided review reduces cEEG review time by 78% with minimal loss of sensitivity compared with conventional review.

CLASSIFICATION OF EVIDENCE

This study provides Class IV evidence that screening of cEEG with CSAs efficiently and accurately identifies seizures and other EEG abnormalities as compared with standard cEEG visual interpretation.

Burst suppression in sleep in a routine outpatient EEG

Burst suppression (BS) is an electroencephalogram (EEG) pattern that is characterized by brief bursts of spikes, sharp waves, or slow waves of relatively high amplitude alternating with periods of relatively flat EEG or isoelectric periods. The pattern is usually associated with coma, severe encephalopathy of various etiologies, or general anesthesia. We describe an unusual case of anoxic brain injury in which a BS pattern was seen during behaviorally defined sleep during a routine outpatient EEG study.

Automated sleep apnea quantification based on respiratory movement

Obstructive sleep apnea (OSA) is a prevalent and treatable disorder of neurological and medical importance that is traditionally diagnosed through multi-channel laboratory polysomnography(PSG). However, OSA testing is increasingly performed with portable home devices using limited physiological channels. We tested the hypothesis that single channel respiratory effort alone could support automated quantification of apnea and hypopnea events. We developed a respiratory event detection algorithm applied to thoracic strain-belt data from patients with variable degrees of sleep apnea. We optimized parameters on a training set (n=57) and then tested performance on a validation set (n=59). The optimized algorithm correlated significantly with manual scoring in the validation set (R2=0.73 for training set, R2=0.55 for validation set; p<0.05). For dichotomous classification, the AUC was >0.92 and >0.85 using apnea-hypopnea index cutoff values of 5 and 15, respectively. Our findings demonstrate that manually scored AHI values can be approximated from thoracic movements alone. This finding has potential applications for automating laboratory PSG analysis as well as improving the performance of limited channel home monitors.

SpikeGUI: software for rapid interictal discharge annotation via template matching and online machine learning

Detection of interictal discharges is a key element of interpreting EEGs during the diagnosis and management of epilepsy. Because interpretation of clinical EEG data is time-intensive and reliant on experts who are in short supply, there is a great need for automated spike detectors. However, attempts to develop general-purpose spike detectors have so far been severely limited by a lack of expert-annotated data. Huge databases of interictal discharges are therefore in great demand for the development of general-purpose detectors. Detailed manual annotation of interictal discharges is time consuming, which severely limits the willingness of experts to participate. To address such problems, a graphical user interface "SpikeGUI" was developed in our work for the purposes of EEG viewing and rapid interictal discharge annotation. "SpikeGUI" substantially speeds up the task of annotating interictal discharges using a custom-built algorithm based on a combination of template matching and online machine learning techniques. While the algorithm is currently tailored to annotation of interictal epileptiform discharges, it can easily be generalized to other waveforms and signal types.

Real-time closed-loop control in a rodent model of medically induced coma using burst suppression

BACKGROUND

A medically induced coma is an anesthetic state of profound brain inactivation created to treat status epilepticus and to provide cerebral protection after traumatic brain injuries. The authors hypothesized that a closed-loop anesthetic delivery system could automatically and precisely control the electroencephalogram state of burst suppression and efficiently maintain a medically induced coma.

METHODS

In six rats, the authors implemented a closed-loop anesthetic delivery system for propofol consisting of: a computer-controlled pump infusion, a two-compartment pharmacokinetics model defining propofol's electroencephalogram effects, the burst-suppression probability algorithm to compute in real time from the electroencephalogram the brain's burst-suppression state, an online parameter-estimation procedure and a proportional-integral controller. In the control experiment each rat was randomly assigned to one of the six burst-suppression probability target trajectories constructed by permuting the burst-suppression probability levels of 0.4, 0.65, and 0.9 with linear transitions between levels.

RESULTS

In each animal the controller maintained approximately 60 min of tight, real-time control of burst suppression by tracking each burst-suppression probability target level for 15 min and two between-level transitions for 5-10 min. The posterior probability that the closed-loop anesthetic delivery system was reliable across all levels was 0.94 (95% CI, 0.77-1.00; n = 18) and that the system was accurate across all levels was 1.00 (95% CI, 0.84-1.00; n = 18).

CONCLUSION

The findings of this study establish the feasibility of using a closed-loop anesthetic delivery systems to achieve in real time reliable and accurate control of burst suppression in rodents and suggest a paradigm to precisely control medically induced coma in patients.

Inferring seizure frequency from brief EEG recordings

Routine EEGs remain a cornerstone test in caring for people with epilepsy. Although rare, a self-limited seizure (clinical or electrographic only) may be observed during such brief EEGs. The implications of observing a seizure in this situation, especially with respect to inferring the underlying seizure frequency, are unclear. The issue is complicated by the inaccuracy of patient-reported estimations of seizure frequency. The treating clinician is often left to wonder whether the single seizure indicates very frequent seizures, or if it is of lesser significance. We applied standard concepts of probabilistic inference to a simple model of seizure incidence to provide some guidance for clinicians facing this situation. Our analysis establishes upper and lower bounds on the seizure rate implied by observing a single seizure during routine EEG. Not surprisingly, with additional information regarding the expected seizure rate, these bounds can be further constrained. This framework should aid the clinician in applying a more principled approach toward decision making in the setting of a single seizure on a routine EEG.

Real-time segmentation of burst suppression patterns in critical care EEG monitoring

OBJECTIVE

Develop a real-time algorithm to automatically discriminate suppressions from non-suppressions (bursts) in electroencephalograms of critically ill adult patients.

METHODS

A real-time method for segmenting adult ICU EEG data into bursts and suppressions is presented based on thresholding local voltage variance. Results are validated against manual segmentations by two experienced human electroencephalographers. We compare inter-rater agreement between manual EEG segmentations by experts with inter-rater agreement between human vs automatic segmentations, and investigate the robustness of segmentation quality to variations in algorithm parameter settings. We further compare the results of using these segmentations as input for calculating the burst suppression probability (BSP), a continuous measure of depth-of-suppression.

RESULTS

Automated segmentation was comparable to manual segmentation, i.e. algorithm-vs-human agreement was comparable to human-vs-human agreement, as judged by comparing raw EEG segmentations or the derived BSP signals. Results were robust to modest variations in algorithm parameter settings.

CONCLUSIONS

Our automated method satisfactorily segments burst suppression data across a wide range adult ICU EEG patterns. Performance is comparable to or exceeds that of manual segmentation by human electroencephalographers.

SIGNIFICANCE

Automated segmentation of burst suppression EEG patterns is an essential component of quantitative brain activity monitoring in critically ill and anesthetized adults. The segmentations produced by our algorithm provide a basis for accurate tracking of suppression depth.

Calculating the risk benefit equation for aggressive treatment of non-convulsive status epilepticus

OBJECTIVE

To address the question: does non-convulsive status epilepticus warrant the same aggressive treatment as convulsive status epilepticus?

METHODS

We used a decision model to evaluate the risks and benefits of treating non-convulsive status epilepticus with intravenous anesthetics and ICU-level aggressive care. We investigated how the decision to use aggressive versus non-aggressive management for non-convulsive status epilepticus impacts expected patient outcome for four etiologies: absence epilepsy, discontinued antiepileptic drugs, intraparenchymal hemorrhage, and hypoxic ischemic encephalopathy. Each etiology was defined by distinct values for five key parameters: baseline mortality rate of the inciting etiology; efficacy of non-aggressive treatment in gaining control of seizures; the relative contribution of seizures to overall mortality; the degree of excess disability expected in the case of delayed seizure control; and the mortality risk of aggressive treatment.

RESULTS

Non-aggressive treatment was favored for etiologies with low morbidity and mortality such as absence epilepsy and discontinued antiepileptic drugs. The risk of aggressive treatment was only warranted in etiologies where there was significant risk of seizure-induced neurologic damage. In the case of post-anoxic status epilepticus, expected outcomes were poor regardless of the treatment chosen. The favored strategy in each case was determined by strong interactions of all five model parameters.

CONCLUSIONS

Determination of the optimal management approach to non-convulsive status epilepticus is complex and is ultimately determined by the inciting etiology.

Estimating cerebral microinfarct burden from autopsy samples

OBJECTIVE

To estimate whole-brain microinfarct burden from microinfarct counts in routine postmortem examination.

METHODS

We developed a simple mathematical method to estimate the total number of cerebral microinfarcts from counts obtained in the small amount of tissue routinely examined in brain autopsies. We derived estimates of total microinfarct burden from autopsy brain specimens from 648 older participants in 2 community-based clinical-pathologic cohort studies of aging and dementia.

RESULTS

Our results indicate that observing 1 or 2 microinfarcts in 9 routine neuropathologic specimens implies a maximum-likelihood estimate of 552 or 1,104 microinfarcts throughout the brain. Similar estimates were obtained when validating in larger sampled brain volumes.

CONCLUSIONS

The substantial whole-brain burden of cerebral microinfarcts suggested by even a few microinfarcts on routine pathologic sampling suggests a potential mechanism by which these lesions could cause neurologic dysfunction in individuals with small-vessel disease. The estimation framework developed here may generalize to clinicopathologic correlations of other imaging-negative micropathologies.

Modulation of EEG functional connectivity networks in subjects undergoing repetitive transcranial magnetic stimulation

Transcranial magnetic stimulation (TMS) is a noninvasive brain stimulation technique that utilizes magnetic fluxes to alter cortical activity. Continuous theta-burst repetitive TMS (cTBS) results in long-lasting decreases in indices of cortical excitability, and alterations in performance of behavioral tasks. We investigated the effects of cTBS on cortical function via functional connectivity and graph theoretical analysis of EEG data. Thirty-one channel resting-state EEG recordings were obtained before and after 40 s of cTBS stimulation to the left primary motor cortex. Functional connectivity between nodes was assessed in multiple frequency bands using lagged max-covariance, and subsequently thresholded to construct undirected graphs. After cTBS, we find widespread decreases in functional connectivity in the alpha band. There are also simultaneous increases in functional connectivity in the high-beta bands, especially amongst anterior and interhemispheric connections. The analysis of the undirected graphs reveals that interhemispheric and interregional connections are more likely to be modulated after cTBS than local connections. There is also a shift in the topology of network connectivity, with an increase in the clustering coefficient after cTBS in the beta bands, and a decrease in clustering and increase in path length in the alpha band, with the alpha-band connectivity primarily decreased near the site of stimulation. cTBS produces widespread alterations in cortical functional connectivity, with resulting shifts in cortical network topology.

Real-time segmentation and tracking of brain metabolic state in ICU EEG recordings of burst suppression

We provide a method for estimating brain metabolic state based on a reduced-order model of EEG burst suppression. The model, derived from previously suggested biophysical mechanisms of burst suppression, describes important electrophysiological features and provides a direct link to cerebral metabolic rate. We design and fit the estimation method from EEG recordings of burst suppression from a neurological intensive care unit and test it on real and synthetic data.

Information theoretic quantification of diagnostic uncertainty

Diagnostic test interpretation remains a challenge in clinical practice. Most physicians receive training in the use of Bayes' rule, which specifies how the sensitivity and specificity of a test for a given disease combine with the pre-test probability to quantify the change in disease probability incurred by a new test result. However, multiple studies demonstrate physicians' deficiencies in probabilistic reasoning, especially with unexpected test results. Information theory, a branch of probability theory dealing explicitly with the quantification of uncertainty, has been proposed as an alternative framework for diagnostic test interpretation, but is even less familiar to physicians. We have previously addressed one key challenge in the practical application of Bayes theorem: the handling of uncertainty in the critical first step of estimating the pre-test probability of disease. This essay aims to present the essential concepts of information theory to physicians in an accessible manner, and to extend previous work regarding uncertainty in pre-test probability estimation by placing this type of uncertainty within a principled information theoretic framework. We address several obstacles hindering physicians' application of information theoretic concepts to diagnostic test interpretation. These include issues of terminology (mathematical meanings of certain information theoretic terms differ from clinical or common parlance) as well as the underlying mathematical assumptions. Finally, we illustrate how, in information theoretic terms, one can understand the effect on diagnostic uncertainty of considering ranges instead of simple point estimates of pre-test probability.

Propagation of uncertainty in Bayesian diagnostic test interpretation

OBJECTIVES

Bayesian interpretation of diagnostic test results usually involves point estimates of the pretest probability and the likelihood ratio corresponding to the test result; however, it may be more appropriate in clinical situations to consider instead a range of possible values to express uncertainty in the estimates of these parameters. We thus sought to demonstrate how uncertainty in sensitivity, specificity, and disease pretest probability can be accommodated in Bayesian interpretation of diagnostic testing.

METHODS

We investigated three questions: How does uncertainty in the likelihood ratio propagate to the posttest probability range, assuming a point estimate of pretest probability? How does uncertainty in the sensitivity and specificity of a test affect uncertainty in the likelihood ratio? How does uncertainty propagate when present in both the pretest probability and the likelihood ratio?

RESULTS

Propagation of likelihood ratio uncertainty depends on the pretest probability and is more prominent for unexpected test results. Uncertainty in sensitivity and specificity propagates into the calculation of likelihood ratio prominently as these parameters approach 100%; even modest errors of ± 10% caused dramatic propagation. Combining errors of ± 20% in the pretest probability and in the likelihood ratio exhibited modest propagation to posttest probability, suggesting a realistic target range for clinical estimations.

CONCLUSIONS

The results provide a framework for incorporating ranges of uncertainty into Bayesian reasoning. Although point estimates simplify the implementation of Bayesian reasoning, it is important to recognize the implications of error propagation when ranges are considered in this multistep process.

Absence of early epileptiform abnormalities predicts lack of seizures on continuous EEG

OBJECTIVE

To determine whether the absence of early epileptiform abnormalities predicts absence of later seizures on continuous EEG monitoring of hospitalized patients.

METHODS

We retrospectively reviewed 242 consecutive patients without a prior generalized convulsive seizure or active epilepsy who underwent continuous EEG monitoring lasting at least 18 hours for detection of nonconvulsive seizures or evaluation of unexplained altered mental status. The findings on the initial 30-minute screening EEG, subsequent continuous EEG recordings, and baseline clinical data were analyzed. We identified early EEG findings associated with absence of seizures on subsequent continuous EEG.

RESULTS

Seizures were detected in 70 (29%) patients. A total of 52 patients had their first seizure in the initial 30 minutes of continuous EEG monitoring. Of the remaining 190 patients, 63 had epileptiform discharges on their initial EEG, 24 had triphasic waves, while 103 had no epileptiform abnormalities. Seizures were later detected in 22% (n = 14) of studies with epileptiform discharges on their initial EEG, vs 3% (n = 3) of the studies without epileptiform abnormalities on initial EEG (p < 0.001). In the 3 patients without epileptiform abnormalities on initial EEG but with subsequent seizures, the first epileptiform discharge or electrographic seizure occurred within the first 4 hours of recording.

CONCLUSIONS

In patients without epileptiform abnormalities during the first 4 hours of recording, no seizures were subsequently detected. Therefore, EEG features early in the recording may indicate a low risk for seizures, and help determine whether extended monitoring is necessary.

Should risky treatments be reserved for secondary prevention? Theoretical considerations regarding risk-benefit tradeoffs

OBJECTIVE

Clinical intuition suggests that risk-reducing treatments are more beneficial for patients with greater risk of disease. This intuition contributes to our rationale for tolerating greater adverse event risk in the setting of secondary prevention of certain diseases such as myocardial infarction or stroke. However, under certain conditions treatment benefits may be greater in primary prevention, even when the treatment carries harmful adverse effect potential.

STUDY DESIGN AND SETTING

We present simple decision-theoretic models that illustrate conditions of risk and benefit under which a treatment is predicted to be more beneficial in primary than in secondary prevention.

RESULTS

The models cover a spectrum of possible clinical circumstances, and demonstrate that net benefit in primary prevention can occur despite no benefit (or even net harm) in secondary prevention.

CONCLUSION

This framework provides a rationale for extending the familiar concept of balancing risks and benefits to account for disease-specific considerations of primary vs. secondary prevention.

Exploration and modulation of brain network interactions with noninvasive brain stimulation in combination with neuroimaging

Much recent work in systems neuroscience has focused on how dynamic interactions between different cortical regions underlie complex brain functions such as motor coordination, language and emotional regulation. Various studies using neuroimaging and neurophysiologic techniques have suggested that in many neuropsychiatric disorders, these dynamic brain networks are dysregulated. Here we review the utility of combined noninvasive brain stimulation and neuroimaging approaches towards greater understanding of dynamic brain networks in health and disease. Brain stimulation techniques, such as transcranial magnetic stimulation and transcranial direct current stimulation, use electromagnetic principles to alter brain activity noninvasively, and induce focal but also network effects beyond the stimulation site. When combined with brain imaging techniques such as functional magnetic resonance imaging, positron emission tomography and electroencephalography, these brain stimulation techniques enable a causal assessment of the interaction between different network components, and their respective functional roles. The same techniques can also be applied to explore hypotheses regarding the changes in functional connectivity that occur during task performance and in various disease states such as stroke, depression and schizophrenia. Finally, in diseases characterized by pathologic alterations in either the excitability within a single region or in the activity of distributed networks, such techniques provide a potential mechanism to alter cortical network function and architectures in a beneficial manner.

Should a sentinel node biopsy be performed in patients with high-risk breast cancer?

A negative sentinel lymph node (SLN) biopsy spares many breast cancer patients the complications associated with lymph node irradiation or additional surgery. However, patients at high risk for nodal involvement based on clinical characteristics may remain at unacceptably high risk of axillary disease even after a negative SLN biopsy result. A Bayesian nomogram was designed to combine the probability of axillary disease prior to nodal biopsy with customized test characteristics for an SLN biopsy and provides the probability of axillary disease despite a negative SLN biopsy. Users may individualize the sensitivity of an SLN biopsy based on factors known to modify the sensitivity of the procedure. This tool may be useful in identifying patients who should have expanded upfront exploration of the axilla or comprehensive axillary irradiation.

Revising the "Rule of Three" for inferring seizure freedom

PURPOSE

How long after starting a new medication must a patient go without seizures before they can be regarded as seizure-free? A recent International League Against Epilepsy (ILAE) task force proposed using a "Rule of Three" as an operational definition of seizure freedom, according to which a patient should be considered seizure-free following an intervention after a period without seizures has elapsed equal to three times the longest preintervention interseizure interval over the previous year. This rule was motivated in large part by statistical considerations advanced in a classic 1983 paper by Hanley and Lippman-Hand. However, strict adherence to the statistical logic of this rule generally requires waiting much longer than recommended by the ILAE task force. Therefore, we set out to determine whether an alternative approach to the Rule of Three might be possible, and under what conditions the rule may be expected to hold or would need to be extended.

METHODS

Probabilistic modeling and application of Bayes' rule.

KEY FINDINGS

We find that an alternative approach to the problem of inferring seizure freedom supports using the Rule of Three in the way proposed by the ILAE in many cases, particularly in evaluating responses to a first trial of antiseizure medication, and to favorably-selected epilepsy surgical candidates. In cases where the a priori odds of success are less favorable, our analysis requires longer seizure-free observation periods before declaring seizure freedom, up to six times the average preintervention interseizure interval. The key to our approach is to take into account not only the time elapsed without seizures but also empirical data regarding the a priori probability of achieving seizure freedom conferred by a particular intervention.

SIGNIFICANCE

In many cases it may be reasonable to consider a patient seizure-free after they have gone without seizures for a period equal to three times the preintervention interseizure interval, as proposed on pragmatic grounds in a recent ILAE position paper, although in other commonly encountered cases a waiting time up to six times this interval is required. In this work we have provided a coherent theoretical basis for modified criterion for seizure freedom, which we call the "Rule of Three-To-Six."

Cerebral infarction due to smoker's polycythemia

A 65-year-old man presented with fluctuating focal neurological deficits and neuroimaging findings of multiple small cerebral infarctions. His medical investigation revealed a >100 pack/year smoking history, and a haematocrit >60. Subsequent investigations led to a diagnosis of cerebral infarction due to smoker's polycythemia, the third such case reported in the medical literature. The patient's neurological deficits resolved completely with subsequent haematocrit reduction. This brief report reviews the differential diagnosis of polycythemia, current knowledge of the mechanisms by which smoker's polycythemia may lead to ischemic stroke, and recommendations for management.

Reversible vasoconstriction syndrome with bilateral basal ganglia hemorrhages

Reversible cerebral vasoconstriction syndrome (RCVS) is an increasingly recognized acute cerebrovascular condition that may produce myriad transient and sustained neurologic deficits as well as a host of radiologic features. We report the case of a woman with RCVS and a severe clinical syndrome with bilateral basal ganglia hemorrhages, cerebral infarctions, and marked vascular abnormalities. The patient made a near complete clinical recovery, representing an extreme and illustrative form of RCVS.

Classification algorithms for predicting sleepiness and sleep apnea severity

Identifying predictors of subjective sleepiness and severity of sleep apnea are important yet challenging goals in sleep medicine. Classification algorithms may provide insights, especially when large data sets are available. We analyzed polysomnography and clinical features available from the Sleep Heart Health Study. The Epworth Sleepiness Scale and the apnea-hypopnea index were the targets of three classifiers: k-nearest neighbor, naive Bayes and support vector machine algorithms. Classification was based on up to 26 features including demographics, polysomnogram, and electrocardiogram (spectrogram). Naive Bayes was best for predicting abnormal Epworth class (0-10 versus 11-24), although prediction was weak: polysomnogram features had 16.7% sensitivity and 88.8% specificity; spectrogram features had 5.3% sensitivity and 96.5% specificity. The support vector machine performed similarly to naive Bayes for predicting sleep apnea class (0-5 versus >5): 59.0% sensitivity and 74.5% specificity using clinical features and 43.4% sensitivity and 83.5% specificity using spectrographic features compared with the naive Bayes classifier, which had 57.5% sensitivity and 73.7% specificity (clinical), and 39.0% sensitivity and 82.7% specificity (spectrogram). Mutual information analysis confirmed the minimal dependency of the Epworth score on any feature, while the apnea-hypopnea index showed modest dependency on body mass index, arousal index, oxygenation and spectrogram features. Apnea classification was modestly accurate, using either clinical or spectrogram features, and showed lower sensitivity and higher specificity than common sleep apnea screening tools. Thus, clinical prediction of sleep apnea may be feasible with easily obtained demographic and electrocardiographic analysis, but the utility of the Epworth is questioned by its minimal relation to clinical, electrocardiographic, or polysomnographic features.

Significance testing as perverse probabilistic reasoning

Truth claims in the medical literature rely heavily on statistical significance testing. Unfortunately, most physicians misunderstand the underlying probabilistic logic of significance tests and consequently often misinterpret their results. This near-universal misunderstanding is highlighted by means of a simple quiz which we administered to 246 physicians at two major academic hospitals, on which the proportion of incorrect responses exceeded 90%. A solid understanding of the fundamental concepts of probability theory is becoming essential to the rational interpretation of medical information. This essay provides a technically sound review of these concepts that is accessible to a medical audience. We also briefly review the debate in the cognitive sciences regarding physicians' aptitude for probabilistic inference.

Statin use following intracerebral hemorrhage: a decision analysis

CONTEXT

Statins are widely prescribed for primary and secondary prevention of ischemic cardiac and cerebrovascular disease. Although serious adverse effects are uncommon, results from a recent clinical trial suggested increased risk of intracerebral hemorrhage (ICH) associated with statin use. For patients with baseline elevated risk of ICH, it is not known whether this potential adverse effect offsets the cardiovascular and cerebrovascular benefits.

OBJECTIVE

To address the following clinical question: Given a history of prior ICH, should statin therapy be avoided?

DESIGN

A Markov decision model was used to evaluate the risks and benefits of statin therapy in patients with prior ICH.

MAIN OUTCOME MEASURE

Life expectancy, measured as quality-adjusted life-years. We investigated how statin use affects this outcome measure while varying a range of clinical parameters, including hemorrhage location (deep vs lobar), ischemic cardiac and cerebrovascular risks, and magnitude of ICH risk associated with statins.

RESULTS

Avoiding statins was favored over a wide range of values for many clinical parameters, particularly in survivors of lobar ICH who are at highest risk of ICH recurrence. In survivors of lobar ICH without prior cardiovascular events, avoiding statins yielded a life expectancy gain of 2.2 quality-adjusted life-years compared with statin use. This net benefit persisted even at the lower 95% confidence interval of the relative risk of statin-associated ICH. In patients with lobar ICH who had prior cardiovascular events, the annual recurrence risk of myocardial infarction would have to exceed 90% to favor statin therapy. Avoiding statin therapy was also favored, although by a smaller margin, in both primary and secondary prevention settings for survivors of deep ICH.

CONCLUSIONS

Avoiding statins should be considered for patients with a history of ICH, particularly those cases with a lobar location.

Automated surveillance for central line-associated bloodstream infection in intensive care units

OBJECTIVE

To develop and evaluate computer algorithms with high negative predictive values that augment traditional surveillance for central line-associated bloodstream infection (CLABSI).

SETTING

Barnes-Jewish Hospital, a 1,250-bed tertiary care academic hospital in Saint Louis, Missouri.

METHODS

We evaluated all adult patients in intensive care units who had blood samples collected during the period from July 1, 2005, to June 30, 2006, that were positive for a recognized pathogen on culture. Each isolate recovered from culture was evaluated using the definitions for nosocomial CLABSI provided by the National Healthcare Safety Network of the Centers for Disease Control and Prevention. Using manual surveillance by infection prevention specialists as the gold standard, we assessed the ability of various combinations of dichotomous rules to determine whether an isolate was associated with a CLABSI. Sensitivity, specificity, and predictive values were calculated.

RESULTS

Infection prevention specialists identified 67 cases of CLABSI associated with 771 isolates recovered from blood samples. The algorithms excluded approximately 40%-62% of the isolates from consideration as possible causes of CLABSI. The simplest algorithm, with 2 dichotomous rules (ie, the collection of blood samples more than 48 hours after admission and the presence of a central venous catheter within 48 hours before collection of blood samples), had the highest negative predictive value (99.4%) and the lowest specificity (44.2%) for CLABSI. Augmentation of this algorithm with rules for common skin contaminants confirmed by another positive blood culture result yielded in a negative predictive value of 99.2% and a specificity of 68.0%.

CONCLUSIONS

An automated approach to surveillance for CLABSI that is characterized by a high negative predictive value can accurately identify and exclude positive culture results not representing CLABSI from further manual surveillance.

Asymptotic Geometry of Multiple Hypothesis Testing

We present a simple geometrical interpretation for the solution to the multiple hypothesis testing problem in the asymptotic limit. Under this interpretation, the optimal decision rule is a nearest neighbor classifier on the probability simplex.

Achievable Rates for Pattern Recognition

Biological and machine pattern recognition systems face a common challenge: Given sensory data about an unknown pattern, classify the pattern by searching for the best match within a library of representations stored in memory. In many cases, the number of patterns to be discriminated and the richness of the raw data force recognition systems to internally represent memory and sensory information in a compressed format. However, these representations must preserve enough information to accommodate the variability and complexity of the environment, otherwise recognition will be unreliable. Thus, there is an intrinsic tradeoff between the amount of resources devoted to data representation and the complexity of the environment in which a recognition system may reliably operate. In this paper, we describe a mathematical model for pattern recognition systems subject to resource constraints, and show how the aforementioned resource-complexity tradeoff can be characterized in terms of three rates related to the number of bits available for representing memory and sensory data, and the number of patterns populating a given statistical environment. We prove single-letter information-theoretic bounds governing the achievable rates, and investigate in detail two illustrative cases where the pattern data is either binary or Gaussian.

The neural multiple access channel

In many neural systems, independently encoded information must at some point be transmitted over the spike train of one neuron. We introduce a method for quantitatively studying the effects of the signal encoding and transmission processes on the rates of transmission of multiple sources of information over one spike train, using the multiple access channel model from network information theory. To illustrate this method we study the effects of a small set of synaptic input patterns and input signal power spectra on the information capacity region of a simple three-neuron system.

Layer 4C in monkey V1 may linearize the output of the LGN

In primates, most LGN fibers terminate in cortical layer 4C, an anatomically prominent structure of unexplained function. We hypothesize that the enormous number of cells in layer 4C of monkey primate visual cortex functions as a neural network "hidden layer" that inverts distortions introduced by transmitting visual signals through the LGN. This hypothesis helps explain how simple cells respond (quasi-) linearly to visual inputs in spite of nonlinearities present in LGN responses. Linearization averts prematurely discarding visual information, in keeping with the role of primary visual cortex as the source of raw visual information to the rest of the brain.