Adaptive Algorithms for Personalized Diabetes Treatment

Dynamic systems, especially in real-life applications, are often determined by inter-/intra-variability, uncertainties and time-varying components. Physiological systems are probably the most representative example in which population variability, vital signal measurement noise and uncertain dynamics render their explicit representation and optimization a rather difficult task. Systems characterized by such challenges often require the use of adaptive algorithmic solutions able to perform an iterative structural and/or parametrical update process towards optimized behavior. Adaptive optimization presents the advantages of (i) individualization through learning of basic system characteristics, (ii) ability to follow time-varying dynamics and (iii) low computational cost. In this chapter, the use of online adaptive algorithms is investigated in two basic research areas related to diabetes management: (i) real-time glucose regulation and (ii) real-time prediction of hypo-/hyperglycemia. The applicability of these methods is illustrated through the design and development of an adaptive glucose control algorithm based on reinforcement learning and optimal control and an adaptive, personalized early-warning system for the recognition and alarm generation against hypo- and hyperglycemic events.

Real-time adaptive models for the personalized prediction of glycemic profile in type 1 diabetes patients

Prediction of glycemic profile is an important task for both early recognition of hypoglycemia and enhancement of the control algorithms for optimization of insulin infusion rate. Adaptive models for glucose prediction and recognition of hypoglycemia based on statistical and artificial intelligence techniques are presented.

Prediction of Psychosis by Mismatch Negativity

BACKGROUND: To develop risk-adapted prevention of psychosis, an accurate estimation of the individual risk of psychosis at a given time is needed. Inclusion of biological parameters into multilevel prediction models is thought to improve predictive accuracy of models on the basis of clinical variables. To this aim, mismatch negativity (MMN) was investigated in a sample clinically at high risk, comparing individuals with and without subsequent conversion to psychosis. METHODS: At baseline, an auditory oddball paradigm was used in 62 subjects meeting criteria of a late risk at-state who remained antipsychotic-naive throughout the study. Median follow-up period was 32 months (minimum of 24 months in nonconverters, n = 37). Repeated-measures analysis of covariance was employed to analyze the MMN recorded at frontocentral electrodes; additional comparisons with healthy controls (HC, n = 67) and first-episode schizophrenia patients (FES, n = 33) were performed. Predictive value was evaluated by a Cox regression model. RESULTS: Compared with nonconverters, duration MMN in converters (n = 25) showed significantly reduced amplitudes across the six frontocentral electrodes; the same applied in comparison with HC, but not FES, whereas the duration MMN in in nonconverters was comparable to HC and larger than in FES. A prognostic score was calculated based on a Cox regression model and stratified into two risk classes, which showed significantly different survival curves. CONCLUSIONS: Our findings demonstrate the duration MMN is significantly reduced in at-risk subjects converting to first-episode psychosis compared with nonconverters and may contribute not only to the prediction of conversion but also to a more individualized risk estimation and thus risk-adapted prevention.

Time-lapse microscopy and classification of 2D human mesenchymal stem cells based on cell shape picks up myogenic from osteogenic and adipogenic differentiation

Current methods to characterize mesenchymal stem cells (MSCs) are limited to CD marker expression, plastic adherence and their ability to differentiate into adipogenic, osteogenic and chondrogenic precursors. It seems evident that stem cells undergoing differentiation should differ in many aspects, such as morphology and possibly also behaviour; however, such a correlation has not yet been exploited for fate prediction of MSCs. Primary human MSCs from bone marrow were expanded and pelleted to form high-density cultures and were then randomly divided into four groups to differentiate into adipogenic, osteogenic chondrogenic and myogenic progenitor cells. The cells were expanded as heterogeneous and tracked with time-lapse microscopy to record cell shape, using phase-contrast microscopy. The cells were segmented using a custom-made image-processing pipeline. Seven morphological features were extracted for each of the segmented cells. Statistical analysis was performed on the seven-dimensional feature vectors, using a tree-like classification method. Differentiation of cells was monitored with key marker genes and histology. Cells in differentiation media were expressing the key genes for each of the three pathways after 21 days, i.e. adipogenic, osteogenic and chondrogenic, which was also confirmed by histological staining. Time-lapse microscopy data were obtained and contained new evidence that two cell shape features, eccentricity and filopodia (= 'fingers') are highly informative to classify myogenic differentiation from all others. However, no robust classifiers could be identified for the other cell differentiation paths. The results suggest that non-invasive automated time-lapse microscopy could potentially be used to predict the stem cell fate of hMSCs for clinical application, based on morphology for earlier time-points. The classification is challenged by cell density, proliferation and possible unknown donor-specific factors, which affect the performance of morphology-based approaches. Copyright © 2012 John Wiley & Sons, Ltd.

Genome-wide prediction of childhood asthma and related phenotypes in a longitudinal birth cohort

Childhood wheezing and asthma vary greatly in clinical presentation and time course. The extent to which phenotypic variation reflects heterogeneity in disease pathways is unclear.

Prediction of whole-genome DNA-DNA similarity, determination of G+C content and phylogenetic analysis within the family Pasteurellaceae by multilocus sequence analysis (MLSA)

Genome predictions based on selected genes would be a very welcome approach for taxonomic studies, including DNA-DNA similarity, G+C content and representative phylogeny of bacteria. At present, DNA-DNA hybridizations are still considered the gold standard in species descriptions. However, this method is time-consuming and troublesome, and datasets can vary significantly between experiments as well as between laboratories. For the same reasons, full matrix hybridizations are rarely performed, weakening the significance of the results obtained. The authors established a universal sequencing approach for the three genes recN, rpoA and thdF for the Pasteurellaceae, and determined if the sequences could be used for predicting DNA-DNA relatedness within the family. The sequence-based similarity values calculated using a previously published formula proved most useful for species and genus separation, indicating that this method provides better resolution and no experimental variation compared to hybridization. By this method, cross-comparisons within the family over species and genus borders easily become possible. The three genes also serve as an indicator of the genome G+C content of a species. A mean divergence of around 1 % was observed from the classical method, which in itself has poor reproducibility. Finally, the three genes can be used alone or in combination with already-established 16S rRNA, rpoB and infB gene-sequencing strategies in a multisequence-based phylogeny for the family Pasteurellaceae. It is proposed to use the three sequences as a taxonomic tool, replacing DNA-DNA hybridization.

Dosimetric impact of geometric errors due to respiratory motion prediction on dynamic multileaf collimator-based four-dimensional radiation delivery

The synchronization of dynamic multileaf collimator (DMLC) response with respiratory motion is critical to ensure the accuracy of DMLC-based four dimensional (4D) radiation delivery. In practice, however, a finite time delay (response time) between the acquisition of tumor position and multileaf collimator response necessitates predictive models of respiratory tumor motion to synchronize radiation delivery. Predicting a complex process such as respiratory motion introduces geometric errors, which have been reported in several publications. However, the dosimetric effect of such errors on 4D radiation delivery has not yet been investigated. Thus, our aim in this work was to quantify the dosimetric effects of geometric error due to prediction under several different conditions. Conformal and intensity modulated radiation therapy (IMRT) plans for a lung patient were generated for anterior-posterior/posterior-anterior (AP/PA) beam arrangements at 6 and 18 MV energies to provide planned dose distributions. Respiratory motion data was obtained from 60 diaphragm-motion fluoroscopy recordings from five patients. A linear adaptive filter was employed to predict the tumor position. The geometric error of prediction was defined as the absolute difference between predicted and actual positions at each diaphragm position. Distributions of geometric error of prediction were obtained for all of the respiratory motion data. Planned dose distributions were then convolved with distributions for the geometric error of prediction to obtain convolved dose distributions. The dosimetric effect of such geometric errors was determined as a function of several variables: response time (0-0.6 s), beam energy (6/18 MV), treatment delivery (3D/4D), treatment type (conformal/IMRT), beam direction (AP/PA), and breathing training type (free breathing/audio instruction/visual feedback). Dose difference and distance-to-agreement analysis was employed to quantify results. Based on our data, the dosimetric impact of prediction (a) increased with response time, (b) was larger for 3D radiation therapy as compared with 4D radiation therapy, (c) was relatively insensitive to change in beam energy and beam direction, (d) was greater for IMRT distributions as compared with conformal distributions, (e) was smaller than the dosimetric impact of latency, and (f) was greatest for respiration motion with audio instructions, followed by visual feedback and free breathing. Geometric errors of prediction that occur during 4D radiation delivery introduce dosimetric errors that are dependent on several factors, such as response time, treatment-delivery type, and beam energy. Even for relatively small response times of 0.6 s into the future, dosimetric errors due to prediction could approach delivery errors when respiratory motion is not accounted for at all. To reduce the dosimetric impact, better predictive models and/or shorter response times are required.

Risk prediction of fever in neutropenia in children with cancer: a step towards individually tailored supportive therapy?

BACKGROUND: Fever in severe chemotherapy-induced neutropenia (FN) is the most frequent manifestation of a potentially lethal complication of current intensive chemotherapy regimens. This study aimed at establishing models predicting the risk of FN, and of FN with bacteremia, in pediatric cancer patients. METHODS: In a single-centre cohort study, characteristics potentially associated with FN and episodes of FN were retrospectively extracted from charts. Poisson regression accounting for chemotherapy exposure time was used for analysis. Prediction models were constructed based on a derivation set of two thirds of observations, and validated based on the remaining third of observations. RESULTS: In 360 pediatric cancer patients diagnosed and treated for a cumulative chemotherapy exposure time of 424 years, 629 FN were recorded (1.48 FN per patient per year, 95% confidence interval (CI), 1.37-1.61), 145 of them with bacteremia (23% of FN; 0.34; 0.29-0.40). More intensive chemotherapy, shorter time since diagnosis, bone marrow involvement, central venous access device (CVAD), and prior FN were significantly and independently associated with a higher risk to develop both FN and FN with bacteremia. The prediction models explained more than 30% of the respective risks. CONCLUSIONS: The two models predicting FN and FN with bacteremia were based on five easily accessible clinical variables. Before clinical application, they need to be validated by prospective studies.

A prediction model for personal radio frequency electromagnetic field exposure

Radio frequency electromagnetic fields (RF-EMF) in our daily life are caused by numerous sources such as fixed site transmitters (e.g. mobile phone base stations) or indoor devices (e.g. cordless phones). The objective of this study was to develop a prediction model which can be used to predict mean RF-EMF exposure from different sources for a large study population in epidemiological research. We collected personal RF-EMF exposure measurements of 166 volunteers from Basel, Switzerland, by means of portable exposure meters, which were carried during one week. For a validation study we repeated exposure measurements of 31 study participants 21 weeks after the measurements of the first week on average. These second measurements were not used for the model development. We used two data sources as exposure predictors: 1) a questionnaire on potentially exposure relevant characteristics and behaviors and 2) modeled RF-EMF from fixed site transmitters (mobile phone base stations, broadcast transmitters) at the participants' place of residence using a geospatial propagation model. Relevant exposure predictors, which were identified by means of multiple regression analysis, were the modeled RF-EMF at the participants' home from the propagation model, housing characteristics, ownership of communication devices (wireless LAN, mobile and cordless phones) and behavioral aspects such as amount of time spent in public transports. The proportion of variance explained (R2) by the final model was 0.52. The analysis of the agreement between calculated and measured RF-EMF showed a sensitivity of 0.56 and a specificity of 0.95 (cut-off: 90th percentile). In the validation study, the sensitivity and specificity of the model were 0.67 and 0.96, respectively. We could demonstrate that it is feasible to model personal RF-EMF exposure. Most importantly, our validation study suggests that the model can be used to assess average exposure over several months.

Kinetics of maternally acquired anti-hepatitis A antibodies: prediction of waning based on maternal or cord blood antibody levels

BACKGROUND Timing is critical for efficient hepatitis A vaccination in high endemic areas as high levels of maternal IgG antibodies against the hepatitis A virus (HAV) present in the first year of life may impede the vaccine response. OBJECTIVES To describe the kinetics of the decline of anti-HAV maternal antibodies, and to estimate the time of complete loss of maternal antibodies in infants in León, Nicaragua, a region in which almost all mothers are anti-HAV seropositive. METHODS We collected cord blood samples from 99 healthy newborns together with 49 corresponding maternal blood samples, as well as further blood samples at 2 and 7 months of age. Anti-HAV IgG antibody levels were measured by enzyme immunoassay (EIA). We predicted the time when antibodies would fall below 10 mIU/ml, the presumed lowest level of seroprotection. RESULTS Seroprevalence was 100% at birth (GMC 8392 mIU/ml); maternal and cord blood antibody concentrations were similar. The maternal antibody levels of the infants decreased exponentially with age and the half-life of the maternal antibody was estimated to be 40 days. The relationship between the antibody concentration at birth and time until full waning was described as: critical age (months)=3.355+1.969 × log(10)(Ab-level at birth). The survival model estimated that loss of passive immunity will have occurred in 95% of infants by the age of 13.2 months. CONCLUSIONS Complete waning of maternal anti-HAV antibodies may take until early in the second year of life. The here-derived formula relating maternal or cord blood antibody concentrations to the age at which passive immunity is lost may be used to determine the optimal age of childhood HAV vaccination.

A Nonstationary Space-Time Gaussian Process Model for Partially Converged Simulations

In the context of expensive numerical experiments, a promising solution for alleviating the computational costs consists of using partially converged simulations instead of exact solutions. The gain in computational time is at the price of precision in the response. This work addresses the issue of fitting a Gaussian process model to partially converged simulation data for further use in prediction. The main challenge consists of the adequate approximation of the error due to partial convergence, which is correlated in both design variables and time directions. Here, we propose fitting a Gaussian process in the joint space of design parameters and computational time. The model is constructed by building a nonstationary covariance kernel that reflects accurately the actual structure of the error. Practical solutions are proposed for solving parameter estimation issues associated with the proposed model. The method is applied to a computational fluid dynamics test case and shows significant improvement in prediction compared to a classical kriging model.

Improving the clinical prediction of psychosis by combining ultra-high risk criteria and cognitive basic symptoms

OBJECTIVE Cognitive impairments are regarded as a core component of schizophrenia. However, the cognitive dimension of psychosis is hardly considered by ultra-high risk (UHR) criteria. Therefore, we studied whether the combination of symptomatic UHR criteria and the basic symptom criterion "cognitive disturbances" (COGDIS) is superior in predicting first-episode psychosis. METHOD In a naturalistic 48-month follow-up study, the conversion rate to first-episode psychosis was studied in 246 outpatients of an early detection of psychosis service (FETZ); thereby, the association between conversion, and the combined and singular use of UHR criteria and COGDIS was compared. RESULTS Patients that met UHR criteria and COGDIS (n=127) at baseline had a significantly higher risk of conversion (hr=0.66 at month 48) and a shorter time to conversion than patients that met only UHR criteria (n=37; hr=0.28) or only COGDIS (n=30; hr=0.23). Furthermore, the risk of conversion was higher for the combined criteria than for UHR criteria (n=164; hr=0.56 at month 48) and COGDIS (n=158; hr=0.56 at month 48) when considered irrespective of each other. CONCLUSIONS Our findings support the merits of considering both COGDIS and UHR criteria in the early detection of persons who are at high risk of developing a first psychotic episode within 48months. Applying both sets of criteria improves sensitivity and individual risk estimation, and may thereby support the development of stage-targeted interventions. Moreover, since the combined approach enables the identification of considerably more homogeneous at-risk samples, it should support both preventive and basic research.

An Early-Warning System for Hypo-/Hyperglycemic Events Based on Fusion of Adaptive Prediction Models

Introduction: Early warning of future hypoglycemic and hyperglycemic events can improve the safety of type 1 diabetes mellitus (T1DM) patients. The aim of this study is to design and evaluate a hypoglycemia / hyperglycemia early warning system (EWS) for T1DM patients under sensor-augmented pump (SAP) therapy. Methods: The EWS is based on the combination of data-driven online adaptive prediction models and a warning algorithm. Three modeling approaches have been investigated: (i) autoregressive (ARX) models, (ii) auto-regressive with an output correction module (cARX) models, and (iii) recurrent neural network (RNN) models. The warning algorithm performs postprocessing of the models′ outputs and issues alerts if upcoming hypoglycemic/hyperglycemic events are detected. Fusion of the cARX and RNN models, due to their complementary prediction performances, resulted in the hybrid autoregressive with an output correction module/recurrent neural network (cARN)-based EWS. Results: The EWS was evaluated on 23 T1DM patients under SAP therapy. The ARX-based system achieved hypoglycemic (hyperglycemic) event prediction with median values of accuracy of 100.0% (100.0%), detection time of 10.0 (8.0) min, and daily false alarms of 0.7 (0.5). The respective values for the cARX-based system were 100.0% (100.0%), 17.5 (14.8) min, and 1.5 (1.3) and, for the RNN-based system, were 100.0% (92.0%), 8.4 (7.0) min, and 0.1 (0.2). The hybrid cARN-based EWS presented outperforming results with 100.0% (100.0%) prediction accuracy, detection 16.7 (14.7) min in advance, and 0.8 (0.8) daily false alarms. Conclusion: Combined use of cARX and RNN models for the development of an EWS outperformed the single use of each model, achieving accurate and prompt event prediction with few false alarms, thus providing increased safety and comfort.

Detecting determinism with improved sensitivity in time series: Rank-based nonlinear predictability score

The rank-based nonlinear predictability score was recently introduced as a test for determinism in point processes. We here adapt this measure to time series sampled from time-continuous flows. We use noisy Lorenz signals to compare this approach against a classical amplitude-based nonlinear prediction error. Both measures show an almost identical robustness against Gaussian white noise. In contrast, when the amplitude distribution of the noise has a narrower central peak and heavier tails than the normal distribution, the rank-based nonlinear predictability score outperforms the amplitude-based nonlinear prediction error. For this type of noise, the nonlinear predictability score has a higher sensitivity for deterministic structure in noisy signals. It also yields a higher statistical power in a surrogate test of the null hypothesis of linear stochastic correlated signals. We show the high relevance of this improved performance in an application to electroencephalographic (EEG) recordings from epilepsy patients. Here the nonlinear predictability score again appears of higher sensitivity to nonrandomness. Importantly, it yields an improved contrast between signals recorded from brain areas where the first ictal EEG signal changes were detected (focal EEG signals) versus signals recorded from brain areas that were not involved at seizure onset (nonfocal EEG signals).