Validation of the Asthma Predictive Index and comparison with simpler clinical prediction rules

Background The loose and stringent Asthma Predictive Indices (API), developed in Tucson, are popular rules to predict asthma in preschool children. To be clinically useful, they require validation in different settings. Objective To assess the predictive performance of the API in an independent population and compare it with simpler rules based only on preschool wheeze. Methods We studied 1954 children of the population-based Leicester Respiratory Cohort, followed up from age 1 to 10 years. The API and frequency of wheeze were assessed at age 3 years, and we determined their association with asthma at ages 7 and 10 years by using logistic regression. We computed test characteristics and measures of predictive performance to validate the API and compare it with simpler rules. Results The ability of the API to predict asthma in Leicester was comparable to Tucson: for the loose API, odds ratios for asthma at age 7 years were 5.2 in Leicester (5.5 in Tucson), and positive predictive values were 26% (26%). For the stringent API, these values were 8.2 (9.8) and 40% (48%). For the simpler rule early wheeze, corresponding values were 5.4 and 21%; for early frequent wheeze, 6.7 and 36%. The discriminative ability of all prediction rules was moderate (c statistic ≤ 0.7) and overall predictive performance low (scaled Brier score < 20%). Conclusion Predictive performance of the API in Leicester, although comparable to the original study, was modest and similar to prediction based only on preschool wheeze. This highlights the need for better prediction rules.

Weighted Majoritarian Rules for the Location of Multiple Public Facilities

We consider collective decision problems given by a profile of single-peaked preferences defined over the real line and a set of pure public facilities to be located on the line. In this context, Bochet and Gordon (2012) provide a large class of priority rules based on efficiency, object-population monotonicity and sovereignty. Each such rule is described by a fixed priority ordering among interest groups. We show that any priority rule which treats agents symmetrically — anonymity — respects some form of coherence across collective decision problems — reinforcement — and only depends on peak information — peakonly — is a weighted majoritarian rule. Each such rule defines priorities based on the relative size of the interest groups and specific weights attached to locations. We give an explicit account of the richness of this class of rules.

An Avron rule for fragments of R-mingle

Axiomatic bases of admissible rules are obtained for fragments of the substructural logic R-mingle. In particular, it is shown that a ‘modus-ponens-like’ rule introduced by Arnon Avron forms a basis for the admissible rules of its implication and implication–fusion fragments, while a basis for the admissible rules of the full multiplicative fragment requires an additional countably infinite set of rules. Indeed, this latter case provides an example of a three-valued logic with a finitely axiomatizable consequence relation that has no finite basis for its admissible rules.

Application and comparison of classification algorithms for recognition of Alzheimer's disease in electrical brain activity (EEG)

The early detection of subjects with probable Alzheimer's disease (AD) is crucial for effective appliance of treatment strategies. Here we explored the ability of a multitude of linear and non-linear classification algorithms to discriminate between the electroencephalograms (EEGs) of patients with varying degree of AD and their age-matched control subjects. Absolute and relative spectral power, distribution of spectral power, and measures of spatial synchronization were calculated from recordings of resting eyes-closed continuous EEGs of 45 healthy controls, 116 patients with mild AD and 81 patients with moderate AD, recruited in two different centers (Stockholm, New York). The applied classification algorithms were: principal component linear discriminant analysis (PC LDA), partial least squares LDA (PLS LDA), principal component logistic regression (PC LR), partial least squares logistic regression (PLS LR), bagging, random forest, support vector machines (SVM) and feed-forward neural network. Based on 10-fold cross-validation runs it could be demonstrated that even tough modern computer-intensive classification algorithms such as random forests, SVM and neural networks show a slight superiority, more classical classification algorithms performed nearly equally well. Using random forests classification a considerable sensitivity of up to 85% and a specificity of 78%, respectively for the test of even only mild AD patients has been reached, whereas for the comparison of moderate AD vs. controls, using SVM and neural networks, values of 89% and 88% for sensitivity and specificity were achieved. Such a remarkable performance proves the value of these classification algorithms for clinical diagnostics.

High specificity of line-immunoassay based algorithms for recent HIV-1 infection independent of viral subtype and stage of disease

Background Serologic testing algorithms for recent HIV seroconversion (STARHS) provide important information for HIV surveillance. We have shown that a patient's antibody reaction in a confirmatory line immunoassay (INNO-LIATM HIV I/II Score, Innogenetics) provides information on the duration of infection. Here, we sought to further investigate the diagnostic specificity of various Inno-Lia algorithms and to identify factors affecting it. Methods Plasma samples of 714 selected patients of the Swiss HIV Cohort Study infected for longer than 12 months and representing all viral clades and stages of chronic HIV-1 infection were tested blindly by Inno-Lia and classified as either incident (up to 12 m) or older infection by 24 different algorithms. Of the total, 524 patients received HAART, 308 had HIV-1 RNA below 50 copies/mL, and 620 were infected by a HIV-1 non-B clade. Using logistic regression analysis we evaluated factors that might affect the specificity of these algorithms. Results HIV-1 RNA <50 copies/mL was associated with significantly lower reactivity to all five HIV-1 antigens of the Inno-Lia and impaired specificity of most algorithms. Among 412 patients either untreated or with HIV-1 RNA ≥50 copies/mL despite HAART, the median specificity of the algorithms was 96.5% (range 92.0-100%). The only factor that significantly promoted false-incident results in this group was age, with false-incident results increasing by a few percent per additional year. HIV-1 clade, HIV-1 RNA, CD4 percentage, sex, disease stage, and testing modalities exhibited no significance. Results were similar among 190 untreated patients. Conclusions The specificity of most Inno-Lia algorithms was high and not affected by HIV-1 variability, advanced disease and other factors promoting false-recent results in other STARHS. Specificity should be good in any group of untreated HIV-1 patients.

Development and validation of decision rules to guide frequency of monitoring CD4 cell count in HIV-1 infection before starting antiretroviral therapy

Background Although CD4 cell count monitoring is used to decide when to start antiretroviral therapy in patients with HIV-1 infection, there are no evidence-based recommendations regarding its optimal frequency. It is common practice to monitor every 3 to 6 months, often coupled with viral load monitoring. We developed rules to guide frequency of CD4 cell count monitoring in HIV infection before starting antiretroviral therapy, which we validated retrospectively in patients from the Swiss HIV Cohort Study. Methodology/Principal Findings We built up two prediction rules (“Snap-shot rule” for a single sample and “Track-shot rule” for multiple determinations) based on a systematic review of published longitudinal analyses of CD4 cell count trajectories. We applied the rules in 2608 untreated patients to classify their 18 061 CD4 counts as either justifiable or superfluous, according to their prior ≥5% or <5% chance of meeting predetermined thresholds for starting treatment. The percentage of measurements that both rules falsely deemed superfluous never exceeded 5%. Superfluous CD4 determinations represented 4%, 11%, and 39% of all actual determinations for treatment thresholds of 500, 350, and 200×106/L, respectively. The Track-shot rule was only marginally superior to the Snap-shot rule. Both rules lose usefulness for CD4 counts coming near to treatment threshold. Conclusions/Significance Frequent CD4 count monitoring of patients with CD4 counts well above the threshold for initiating therapy is unlikely to identify patients who require therapy. It appears sufficient to measure CD4 cell count 1 year after a count >650 for a threshold of 200, >900 for 350, or >1150 for 500×106/L, respectively. When CD4 counts fall below these limits, increased monitoring frequency becomes advisable. These rules offer guidance for efficient CD4 monitoring, particularly in resource-limited settings.

Diagnostic performance of line-immunoassay based algorithms for incident HIV-1 infection

Background Serologic testing algorithms for recent HIV seroconversion (STARHS) provide important information for HIV surveillance. We have previously demonstrated that a patient's antibody reaction pattern in a confirmatory line immunoassay (INNO-LIA™ HIV I/II Score) provides information on the duration of infection, which is unaffected by clinical, immunological and viral variables. In this report we have set out to determine the diagnostic performance of Inno-Lia algorithms for identifying incident infections in patients with known duration of infection and evaluated the algorithms in annual cohorts of HIV notifications. Methods Diagnostic sensitivity was determined in 527 treatment-naive patients infected for up to 12 months. Specificity was determined in 740 patients infected for longer than 12 months. Plasma was tested by Inno-Lia and classified as either incident (< = 12 m) or older infection by 26 different algorithms. Incident infection rates (IIR) were calculated based on diagnostic sensitivity and specificity of each algorithm and the rule that the total of incident results is the sum of true-incident and false-incident results, which can be calculated by means of the pre-determined sensitivity and specificity. Results The 10 best algorithms had a mean raw sensitivity of 59.4% and a mean specificity of 95.1%. Adjustment for overrepresentation of patients in the first quarter year of infection further reduced the sensitivity. In the preferred model, the mean adjusted sensitivity was 37.4%. Application of the 10 best algorithms to four annual cohorts of HIV-1 notifications totalling 2'595 patients yielded a mean IIR of 0.35 in 2005/6 (baseline) and of 0.45, 0.42 and 0.35 in 2008, 2009 and 2010, respectively. The increase between baseline and 2008 and the ensuing decreases were highly significant. Other adjustment models yielded different absolute IIR, although the relative changes between the cohorts were identical for all models. Conclusions The method can be used for comparing IIR in annual cohorts of HIV notifications. The use of several different algorithms in combination, each with its own sensitivity and specificity to detect incident infection, is advisable as this reduces the impact of individual imperfections stemming primarily from relatively low sensitivities and sampling bias.

Pelvic radiography in ATLS algorithms: A diminishing role?

ABSTRACT: BACKGROUND: Pelvic x-ray is a routine part of the primary survey of polytraumatized patients according to Advanced Trauma Life Support (ATLS) guidelines. However, pelvic CT is the gold standard imaging technique in the diagnosis of pelvic fractures. This study was conducted to confirm the safety of a modified ATLS algorithm omitting pelvic x-ray in hemodynamically stable polytraumatized patients with clinically stable pelvis in favour of later pelvic examination by CT scan. METHODS: We conducted a retrospective analysis of all polytraumatized patients in our emergency room between 01.07.2004 and 31.01.2006. Inclusion criteria were blunt abdominal trauma, initial hemodynamic stability and a stable pelvis on clinical examination. We excluded patients requiring immediate intervention because of hemodynamic instability. RESULTS: We reviewed the records of n = 452 polytraumatized patients, of which n = 91 fulfilled inclusion criteria (56% male, mean age = 45 years). The mechanism of trauma included 43% road traffic accidents, 47% falls. In 68/91 (75%) patients, both a pelvic x-ray and a CT examination were performed; the remainder had only pelvic CT. In 6/68 (9%) patients, pelvic fracture was diagnosed by pelvic x-ray. None of these 6 patients was found having a false positive pelvic x-ray, i.e. there was no fracture on pelvic CT scan. In 3/68 (4%) cases a fracture was missed in the pelvic x-ray, but confirmed on CT (false negative on x-ray). None of the diagnosed fractures needed an immediate therapeutic intervention. 5 (56%) were classified type A fractures, and another 4 (44%) B 2.1 in computed tomography (AO classification). One A 2.1 fracture was found in a clinically stable patient who only received CT scan (1/23). CONCLUSION: While pelvic x-ray is an integral part of ATLS assessment, this retrospective study suggests that in hemodynamically stable patients with clinically stable pevis, its sensitivity is only 67% and it may safely be omitted in favor of a pelvic CT examination if such is planned in adjunct assessment and available. The results support the safety and utility of our modified ATLS algorithm. A randomized controlled trial using the algorithm can safely be conducted to confirm the results.

Individual patient data meta-analysis of diagnostic and prognostic studies in obstetrics, gynaecology and reproductive medicine

BACKGROUND: In clinical practice a diagnosis is based on a combination of clinical history, physical examination and additional diagnostic tests. At present, studies on diagnostic research often report the accuracy of tests without taking into account the information already known from history and examination. Due to this lack of information, together with variations in design and quality of studies, conventional meta-analyses based on these studies will not show the accuracy of the tests in real practice. By using individual patient data (IPD) to perform meta-analyses, the accuracy of tests can be assessed in relation to other patient characteristics and allows the development or evaluation of diagnostic algorithms for individual patients. In this study we will examine these potential benefits in four clinical diagnostic problems in the field of gynaecology, obstetrics and reproductive medicine. METHODS/DESIGN: Based on earlier systematic reviews for each of the four clinical problems, studies are considered for inclusion. The first authors of the included studies will be invited to participate and share their original data. After assessment of validity and completeness the acquired datasets are merged. Based on these data, a series of analyses will be performed, including a systematic comparison of the results of the IPD meta-analysis with those of a conventional meta-analysis, development of multivariable models for clinical history alone and for the combination of history, physical examination and relevant diagnostic tests and development of clinical prediction rules for the individual patients. These will be made accessible for clinicians. DISCUSSION: The use of IPD meta-analysis will allow evaluating accuracy of diagnostic tests in relation to other relevant information. Ultimately, this could increase the efficiency of the diagnostic work-up, e.g. by reducing the need for invasive tests and/or improving the accuracy of the diagnostic workup. This study will assess whether these benefits of IPD meta-analysis over conventional meta-analysis can be exploited and will provide a framework for future IPD meta-analyses in diagnostic and prognostic research.

Evaluating retraining rules for semi-supervised learning in neural network based cursive word recognition

Training a system to recognize handwritten words is a task that requires a large amount of data with their correct transcription. However, the creation of such a training set, including the generation of the ground truth, is tedious and costly. One way of reducing the high cost of labeled training data acquisition is to exploit unlabeled data, which can be gathered easily. Making use of both labeled and unlabeled data is known as semi-supervised learning. One of the most general versions of semi-supervised learning is self-training, where a recognizer iteratively retrains itself on its own output on new, unlabeled data. In this paper we propose to apply semi-supervised learning, and in particular self-training, to the problem of cursive, handwritten word recognition. The special focus of the paper is on retraining rules that define what data are actually being used in the retraining phase. In a series of experiments it is shown that the performance of a neural network based recognizer can be significantly improved through the use of unlabeled data and self-training if appropriate retraining rules are applied.

Dynamic Optimization of SLA-Based Services Scaling Rules

Current advanced cloud infrastructure management solutions allow scheduling actions for dynamically changing the number of running virtual machines (VMs). This approach, however, does not guarantee that the scheduled number of VMs will properly handle the actual user generated workload, especially if the user utilization patterns will change. We propose using a dynamically generated scaling model for the VMs containing the services of the distributed applications, which is able to react to the variations in the number of application users. We answer the following question: How to dynamically decide how many services of each type are needed in order to handle a larger workload within the same time constraints? We describe a mechanism for dynamically composing the SLAs for controlling the scaling of distributed services by combining data analysis mechanisms with application benchmarking using multiple VM configurations. Based on processing of multiple application benchmarks generated data sets we discover a set of service monitoring metrics able to predict critical Service Level Agreement (SLA) parameters. By combining this set of predictor metrics with a heuristic for selecting the appropriate scaling-out paths for the services of distributed applications, we show how SLA scaling rules can be inferred and then used for controlling the runtime scale-in and scale-out of distributed services. We validate our architecture and models by performing scaling experiments with a distributed application representative for the enterprise class of information systems. We show how dynamically generated SLAs can be successfully used for controlling the management of distributed services scaling.

Admissibility in Finitely Generated Quasivarieties

Checking the admissibility of quasiequations in a finitely generated (i.e., generated by a finite set of finite algebras) quasivariety Q amounts to checking validity in a suitable finite free algebra of the quasivariety, and is therefore decidable. However, since free algebras may be large even for small sets of small algebras and very few generators, this naive method for checking admissibility in Q is not computationally feasible. In this paper, algorithms are introduced that generate a minimal (with respect to a multiset well-ordering on their cardinalities) finite set of algebras such that the validity of a quasiequation in this set corresponds to admissibility of the quasiequation in Q. In particular, structural completeness (validity and admissibility coincide) and almost structural completeness (validity and admissibility coincide for quasiequations with unifiable premises) can be checked. The algorithms are illustrated with a selection of well-known finitely generated quasivarieties, and adapted to handle also admissibility of rules in finite-valued logics.

Code-specific learning rules improve action selection by populations of spiking neurons

Population coding is widely regarded as a key mechanism for achieving reliable behavioral decisions. We previously introduced reinforcement learning for population-based decision making by spiking neurons. Here we generalize population reinforcement learning to spike-based plasticity rules that take account of the postsynaptic neural code. We consider spike/no-spike, spike count and spike latency codes. The multi-valued and continuous-valued features in the postsynaptic code allow for a generalization of binary decision making to multi-valued decision making and continuous-valued action selection. We show that code-specific learning rules speed up learning both for the discrete classification and the continuous regression tasks. The suggested learning rules also speed up with increasing population size as opposed to standard reinforcement learning rules. Continuous action selection is further shown to explain realistic learning speeds in the Morris water maze. Finally, we introduce the concept of action perturbation as opposed to the classical weight- or node-perturbation as an exploration mechanism underlying reinforcement learning. Exploration in the action space greatly increases the speed of learning as compared to exploration in the neuron or weight space.