Prediction Error Analysis Research Articles

Estimation of Future Glucose Concentrations with Subject-Specific Recursive Linear Models

Estimation of future glucose concentrations is a crucial task for diabetes management. Predicted glucose values can be used for early hypoglycemic/hyperglycemic alarms or for adjustment of insulin injections or insulin infusion rates of manual or automated pumps. Continuous glucose monitoring (CGM) technologies provide glucose readings at a high frequency and consequently detailed insight into the subject's glucose variations. The objective of this research is to develop reliable subject-specific glucose prediction models using CGM data. Two separate patient databases collected under hospitalized (disturbance-free) and normal daily life conditions are used for validation of the proposed glucose prediction algorithm. Both databases consist of glucose concentration data collected at 5-min intervals using a CGM device. Using time-series analysis, low-order linear models are developed from patients' own CGM data. The time-series models are integrated with recursive identification and change detection methods, which enables dynamic adaptation of the model to inter-/intra-subject variability and glycemic disturbances. Prediction performance is evaluated in terms of glucose prediction error and Clarke Error Grid analysis (CG-EGA). Prediction errors are significantly reduced with recursive identification of the models, and predictions are further improved with inclusion of a parameter change detection method. CG-EGA analysis results in accurate readings of 90% or more. Subject-specific glucose prediction strategy has been developed. Including a change detection method to the recursive algorithm improves the prediction accuracy. The proposed modeling algorithm with small number of parameters is a good candidate for installation in portable devices for early hypoglycemic/hyperglycemic alarms and for closing the glucose regulation loop with an insulin pump.

Comparative analysis of pharmaceuticals versus industrial chemicals acute aquatic toxicity classification according to the United Nations classification system for chemicals. Assessment of the (Q)SAR predictability of pharmaceuticals acute aquatic toxicity and their predominant acute toxic mode-of-action

Pharmaceuticals have been reported to be ubiquitously present in surface waters prompting concerns of effects of these bioactive substances. Meanwhile, there is a general scarcity of publicly available ecotoxicological data concerning pharmaceuticals. The aim of this paper was to compile a comprehensive database based on OECD's standardized measured ecotoxicological data and to evaluate if there is generally cause of greater concern with regards to pharmaceutical aquatic toxicological profiles relative to industrial chemicals. Comparisons were based upon aquatic ecotoxicity classification under the United Nations Global Harmonized System for classification and labeling of chemicals (GHS). Moreover, we statistically explored whether the predominant mode-of-action (MOA) for pharmaceuticals is narcosis. We found 275 pharmaceuticals with 569 acute aquatic effect data; 23 pharmaceuticals had chronic data. Pharmaceuticals were found to be more frequent than industrial chemicals in GHS category III. Acute toxicity was predictable (>92%) using a generic (Q)SAR ((Quantitative) Structure Activity Relationship) suggesting a narcotic MOA. Analysis of model prediction error suggests that 68% of the pharmaceuticals have a non-specific MOA. Additionally, the acute-to-chronic ratio (ACR) for 70% of the analyzed pharmaceuticals was below 25 further suggesting a non-specific MOA. Sub-lethal receptor-mediated effects may however have a more specific MOA.

Characteristic analysis of inter-frame prediction error for high frame-rate video signal based on information criterion

Characteristic analysis of inter-frame prediction error for high frame-rate video signal based on information criterion

A multiple model approach to respiratory motion prediction for real-time IGRT

Respiration induces significant movement of tumours in the vicinity of thoracic and abdominal structures. Real-time image-guided radiotherapy (IGRT) aims to adapt radiation delivery to tumour motion during irradiation. One of the main problems for achieving this objective is the presence of time lag between the acquisition of tumour position and the radiation delivery. Such time lag causes significant beam positioning errors and affects the dose coverage. A method to solve this problem is to employ an algorithm that is able to predict future tumour positions from available tumour position measurements. This paper presents a multiple model approach to respiratory-induced tumour motion prediction using the interacting multiple model (IMM) filter. A combination of two models, constant velocity (CV) and constant acceleration (CA), is used to capture respiratory-induced tumour motion. A Kalman filter is designed for each of the local models and the IMM filter is applied to combine the predictions of these Kalman filters for obtaining the predicted tumour position. The IMM filter, likewise the Kalman filter, is a recursive algorithm that is suitable for real-time applications. In addition, this paper proposes a confidence interval (CI) criterion to evaluate the performance of tumour motion prediction algorithms for IGRT. The proposed CI criterion provides a relevant measure for the prediction performance in terms of clinical applications and can be used to specify the margin to accommodate prediction errors. The prediction performance of the IMM filter has been evaluated using 110 traces of 4-minute free-breathing motion collected from 24 lung-cancer patients. The simulation study was carried out for prediction time 0.1–0.6 s with sampling rates 3, 5 and 10 Hz. It was found that the prediction of the IMM filter was consistently better than the prediction of the Kalman filter with the CV or CA model. There was no significant difference of prediction errors for the sampling rates 5 and 10 Hz. For these sampling rates, the errors of the IMM filter for 0.4 s prediction time were less than 2.1 mm in terms of the 95% CI criterion or 1.1 mm in terms of the standard deviation (SD) or root mean squared errors (RMSE) criterion. For the prediction time of 0.6 s the errors were less than 3.6 mm in terms of the 95% CI criterion or 1.8 mm in terms of the SD/RMSE criterion. The prediction error analysis showed that the average percentage of the target lies outside the 95% CI margin was 5.2% and outside the SD/RMSE margin was 24.3%. This indicates the effectiveness of the 95% CI criterion as a margining strategy to accommodate prediction errors.

Phosphorus Utilization and Environmental and Economic Implications of Reducing Phosphorus Pollution from Ontario Dairy Cows

A major source of environmental pollution has been overfeeding P to dairy cows, caused by the “safety margins” added to diets in order not to compromise the health and production of animals. An extant whole-animal model was evaluated using an experiment conducted in Ontario to assess its applicability for predicting P excretion. The objective of the study was to use the model to estimate P excretion levels and the economic and environmental implications of implementing mitigating options by following recommendations from studies that have reported sufficient levels of P inclusion in the diet. Mean square prediction error and concordance coefficient analysis showed that the overall predictions were close to the mean and that there was only a slight underprediction of fecal P output by the model. The majority of the error was random, with only 8.9% coming from error caused by deviation from the regression line, and the model did not show a systematic trend of over-or underprediction. The model was then used to predict P excretion in Ontario by using diets commonly fed to dairy cows on Ontario farms. It is estimated that Ontario dairy farms produce 7kt of P annually at current levels of P inclusion in the diet. Reducing P levels from the current 0.41% P of dry matter to 0.35% is estimated to save producers CAN $20/cow per year and the environment 1.3 kt/yr without impairing cow health or productivity. Additionally, the reductions might be from inorganic P sources added to the feed, which are more polluting than organic sources because of their water-soluble nature and liability to leaching and runoff.

Sub-kilometer length scales in coastal waters

Patchiness or spatial variability is ubiquitous in marine systems. With increasing anthropogenic impacts to coastal resources and coastal systems being disproportionately large contributors to ocean productivity, identifying the spatial scales of this patchiness, particularly in coastal waters, is of critical importance to understand coastal ecosystem dynamics. The current work focuses on fine scale structure in three coastal regions. More specifically, we utilize variogram analyses to identify sub-kilometer scales of variability in biological and physical parameters measured by an autonomous underwater vehicle (AUV) in the Mid-Atlantic Bight, Monterey Bay, and in San Luis Obispo Bay between 2001 and 2004. Critical scales of variability in density, turbidity, fluorescence, and bioluminescence are examined as a function of depth and distance offshore. Furthermore, the effects of undersampling are assessed using predictive error analysis. Results indicate the presence of scales of variability ranging from 10s to 100s of meters and provide valuable insight for sampling design and resource allocation for future studies.

Linear Predictive Detection for Power Line Communications Impaired by Colored Noise

Robust detection algorithms capable of mitigating the effects of colored noise are of primary interest in communication systems operating on power line channels. In this paper, we present a sequence detection scheme based on linear prediction to be applied in single-carrier power line communications impaired by colored noise. The presence of colored noise and the need for statistical sufficiency requires the design of an optimal front-end stage, whereas the need for a low-complexity solution suggests a more practical suboptimal front-end. The performance of receivers employing both optimal and suboptimal front-ends has been assessed by means of minimum mean square prediction error (MMSPE) analysis and bit-error rate (BER) simulations. We show that the proposed optimal solution improves the BER performance with respect to conventional systems and makes the receiver more robust against colored noise. As case studies, we investigate the performance of the proposed receivers in a low-voltage (LV) power line channel limited by colored background noise and in a high-voltage (HV) power line channel limited by corona noise.

Predictive error analysis for a water resource management model

In calibrating a model, a set of parameters is assigned to the model which will be employed for the making of all future predictions. If these parameters are estimated through solution of an inverse problem, formulated to be property posed through either pre-calibration or mathematical regularisation, then solution of this inverse problem will, of necessity, lead to a simplified parameter set that omits the details of reality, while still fitting historical data acceptably well Furthermore, estimates of parameters so obtained will be contaminated by measurement noise. Both of these phenomena will lead to errors in predictions made by the model, with the potential for error increasing with the hydraulic property detail on which the prediction depends. Integrity of model usage demands that model predictions be accompanied by some estimate of the possible errors associated with them. The present paper applies theory developed in a previous work to the analysis of predictive error associated with a real world, water resource management model. The analysis offers many challenges, including the fact that the model is a complex one that was partly calibrated by hand. Nevertheless, it is typical of models which are commonly employed as the basis for the making of important decisions, and for which such an analysis must be made. The potential errors associated with point-based and averaged water level and creek inflow predictions are examined, together with the dependence of these errors on the amount of averaging involved. Error associated with predictions made by the existing model are compared with optimized error variances that could have been obtained had calibration been undertaken in such a way as to minimize predictive error variance. The contributions by different parameter types to the overall error variance of selected predictions are also examined. (c) 2006 Elsevier B.V. All rights reserved.

Pharmacokinetic modelling of valproate from clinical data in Serbian epileptic patients

The aim of this study was to define a population pharmacokinetic model that could estimate the clearance of valproate (VPA) in a Serbian epileptic population. For the analysis, 97 steady-state concentrations of VPA were used, collected from 93 patients with epilepsy during routine clinical care in our hospital. To build a final model, we selected the ADVAN1 program subroutine from the NONMEM library for estimating the drug clearance (CL) and determining the influence of different covariates. This is a one-compartment model with first-order elimination and without absorption. Estimation of the predictive performances of the final model was performed on a validation set consisting of 20 epileptic patients. Typical mean value of CL of VPA estimated by the base model in our population was 0.368 l h(-1). The results of the final model show that the CL of VPA increased linearly with total body weight (TBW) and patients' age, while carbamazepine (CBZ) comedication did not affect it significantly. Interindividual variability (coefficient of variation) for CL was 27.2%. Residual error, including intraindividual variability, was 29.68%. The results of the validation process and the analysis of prediction errors suggest a good predictive performance of the final population model. The defined model describes CL of VPA in terms of specific Serbian patient characteristics, using serum concentration data obtained from routine therapeutic drug monitoring.

Comparison of Postoperative Refractive Outcome in Phacotrabeculectomy and Phacoemulsification With Posterior Chamber Intraocular Lens Implantation

To compare the postoperative refractive outcomes following phacotrabeculectomy and phacoemulsification with posterior chamber intraocular lens implantation. A retrospective comparative study of 90 consecutive patients (95 eyes) with cataract with or without co-existing glaucoma who had undergone uncomplicated phacotrabeculectomy or phacoemulsification by a single surgeon. The biometry prediction error was determined for each case by the difference between the actual postoperative refraction and the preoperative predicted refractive outcome, in spherical equivalent. This was followed by a comparative analysis of the mean biometry prediction error, and the number of cases with error >0.50 D or 1.00 D. All biometry measurements were performed by one of the two qualified optometrists. The mean and absolute mean biometry prediction error of the 25 eyes that received phacotrabeculectomy (+0.20 D, absolute error 0.96 D) was comparable to the 70 eyes that received phacoemulsification (-0.14 D, absolute error 0.68 D), P = 0.18 (absolute error, P = 0.12). The proportion of cases with prediction error >0.50 D or 1.00 D was 60% or 40% for phacotrabeculectomy, and 44% or 17% for phacoemulsification, respectively. Phacotrabeculectomy was more likely to have a prediction error >1.00 D (P = 0.02) and a myopic shift of >0.50 D or 1.00 D (P = 0.03 or 0.02, respectively). No significant differences were found with regard to the frequency of hyperopic shift or prediction error >0.50 D. There was no significant difference in either the mean error or the frequency of the different types of error for the different biometry operators, types of glaucoma, or postoperative intraocular pressure control success status. Myopically shifted prediction error was significantly more frequent following posterior chamber intraocular lens implantation with phacotrabeculectomy compared with phacoemulsification, even when surgery was uncomplicated and performed by the same surgeon.

TRACEr and TRACEr-lite: bridging the gap between incident investigation and performance prediction

Incident investigation and performance prediction clearly have the same goal, to reduce the possibility of future adverse events. The communities involved have a common interest in the contextual and cognitive influences on ‘human (un)reliability’. Despite this, the synergy between the two communities has been far less than would be expected; the techniques and terminology employed have been at odds. While no technique will fully fit the needs of both retrospective and predictive human error analysis, the opportunities for cooperation may have a greater impact on safety than the search for the perfect technique. TRACEr-lite (and its father technique TRACEr—Technique for the Retrospective and Predictive Analysis of Human Error) is an attempt to bridge the gap between performance prediction and retrospective human error analysis. This paper outlines the TRACEr-lite approach and illustrates a number of retrospective and predictive applications in the field of air traffic management and aviation. Future directions to enhance co-operation are suggested. It is proposed that in terms of the contribution of the safety community, the whole is greater than the sum of its parts. The techniques have already helped to contribute to important design and policy decisions, and are now being adapted to other industries.

Analysis of Nitrogen Fractions in Silage by Near-Infrared Spectroscopy

Total nitrogen, soluble nitrogen (SN), nonprotein nitrogen (NPN), and acid-detergent insoluble nitrogen (ADIN) were analyzed in grass silage by near-infrared (NIR) spectroscopy. A set of 144 samples was used to calibrate the instrument by modified partial least-squares regression, and the following statistical results were achieved: standard error of calibration (SEC) = 0.449 and square correlation coefficient (R (2)) = 0.98 for total nitrogen x 6.25, SEC = 0.425 and R (2) = 0.95 for SN x 6.25, SEC = 0.414 and R (2) = 0.94 for NPN x 6.25, and SEC = 0.139 and R (2) = 0.84 for ADIN x 6.25. To validate the calibration performed, a set of 48 silage samples was used. Standard errors of prediction were 0.76, 0.64, 0.63, and 0.25 for total nitrogen, SN, NPN, and ADIN (all of them multiplied by 6.25), respectively, and R (2) for the regression of measurements by reference method versus NIR analysis were 0.94, 0.92, 0.90, and 0.48 for total nitrogen, SN, NPN, and ADIN, respectively. To compare the results obtained by NIR spectroscopy with those obtained by the reference methods for total nitrogen, SN, and NPN of the validation set, linear regression and paired t tests were applied, and the results were not significantly different (p = 0.05). When mean square prediction error analysis was applied, it could be concluded that for total nitrogen, SN, and NPN, a robust calibration model was obtained and that the main error was unexplained error. Statistical data for ADIN were worse than those of the other parameters; as a result NIR spectroscopy is not an effective method for quantitative analyses of ADIN in silage; nevertheless, it may be an acceptable method for semiquantitative evaluation.

Setting the scene: CFD and symposium overview

The present situation of CWE(Computational Wind Engineering) and the papers presented to the CWE 2000 Symposium are reviewed from the following viewpoints; 1) topics treated, 2) utilization of commercial code (software), 3) incompleteness of CWE, 4) remaining research subjects, 5) prediction accuracy, 6) new fields of CWE application, etc. Firstly, new tendencies within CWE applications are indicated. Next, the over-attention being given to the application field and the lack of attention to fundamental problems, including prediction error analysis, are pointed out. Lastly, the future trends of CFD (Computational Fluid Dynamics) applications to wind engineering design are discussed.

Short-term prediction of the aggregated power output of wind farms—a statistical analysis of the reduction of the prediction error by spatial smoothing effects

We discuss the accuracy of the prediction of the aggregated power output of wind farms distributed over given regions. Our forecasting procedure provides the expected power output for a time horizon up to 48 h ahead. It is based on the large-scale wind field prediction which is generated operationally by the German weather service. Our investigation focuses on the statistical analysis of the power prediction error of an ensemble of wind farms compared to single sites. Due to spatial smoothing effects the relative prediction error decreases considerably. Using measurements of the power output of 30 wind farms in Germany we find that this reduction depends on the size of the region. To generalize these findings an analytical model based on the spatial correlation function of the prediction error is derived to describe the statistical characteristics of arbitrary configurations of wind farms. This analysis shows that the magnitude of the error reduction depends only weakly on the number of sites and is mainly determined by the size of the region, e.g for the size of a typical large utility (∼370 km in diameter) <50 sites are sufficient to have an error reduction of 63%. Towards a correction of systematic prediction errors an analysis of the temporal structure of the forecast error is performed. For this purpose the correlation of the errors for consecutive forecasts is analysed for single sites and ensembles. This knowledge on previous errors can be beneficially used to correct the actual ensemble forecast.

An analysis of prediction error for new prime-time television programmes: a comparative study between the USA and Korea

This study investigates possible factors in prediction error for audience share for new prime-time television programmes in the USA and Korea. While previous studies have focused on the factors affecting the size of a programme's actual audience, little attention has been paid to the accuracy of audience share prediction. A total of 193 new prime-time TV programmes from 1994 to 1997 for the USA and from 1998 to 2001 for Korea were examined. Advertising industry forecasters' predicted audience share, and actual share data were accumulated from Broadcasting & Cable and Weekly Nielsen Reports. Findings show that the ‘Programme Type’, ‘Returning Lead-Ins’ and ‘Returning Lead-Outs’ worked as significant factors for estimating error in audience share in the USA, and ‘Programme Type’ and ‘Programme Length’ played major roles in forecasting error in Korea. The degrees of relationship with forecasting error are consistent with previous studies' findings. Similarities and dissimilarities in prediction error in both countries were found and are discussed in this study.

An integrated framework to the predictive error analysis in emergency situation

Abstract This paper introduces an analysis framework and procedure to predictively analyze human errors in performing emergency tasks, which are mostly composed of cognitive activities, in nuclear power plants. The framework focuses on the cognitive errors and provides a new perspective in the utilization of context factors into cognitive error prediction. The basic viewpoint on the occurrence of cognitive error taken in this paper is that the cognitive function failures occur from the mismatch between operator's cognitive capability and the requirements of a given task and situational condition. In accordance with this viewpoint, performance influencing factors that influence the occurrence of human errors are classified into three groups, i.e. Performance Assisting Factors (PAF), Task Characteristic Factors (TCF), and Situational Factors (SF). This classification helps analysts view the overall task context in an integrative way by considering the level of PAF with the requirements of TCF and SF to predict the possibility of cognitive function failures. Further, it enables analysts to draw specific error reduction strategies. The framework suggested was applied to the analysis of cognitive error potential for the bleed and feed operation of emergency tasks in nuclear power plants.

Field evaluation of the equilibrium concentration technique (JTI method) for measuring ammonia emission from land spread manure or fertiliser

Three experiments were conducted in which intercomparisons were made between the equilibrium concentration technique, developed at JTI, Sweden, and the integrated horizontal flux technique for measuring ammonia emissions following applications of urea fertiliser, cattle slurry and solid pig manure to land. Mean square prediction error analysis was used to compare the emission rates measured by the two techniques. There were no significant differences between the measurement techniques, although there was some evidence that emission rates were overestimated by the equilibrium concentration method relative to the integrated horizontal flux technique at higher emission rates (>400 g.N ha −1 h −1). The equilibrium concentration method provides a practical and relatively inexpensive technique for measuring emissions under ambient conditions from small plots but good sampler preparation, adequate replication of emission measurements and appropriate choice of duration of sampling periods are necessities for obtaining reliable results.

A Comparison of Calculated and Measured Resting Energy Expenditure in Obese Women

Resting energy expenditure (REE) measured using indirect calorimetry was compared with REE calculated with the Harris-Benedict equation (HBE) using actual (ABW), ideal (IBW), and adjusted (AdjBW) body weight and with the Ireton-Jones Equation (IJE) in 19 obese females. Root mean square prediction error (RMSPE) analysis indicated that HBE using IBW and AdjBW were the most accurate predictors of measured REE. T test revealed no significant difference between HBE using IBW and measured REE. The IJE and HBE, using ABW and AdjBW, were significantly higher than measured REE. In conclusion, HBE using IBW or AdjBW is better than HBE using actual weight or IJE for estimating REE in obese women.

Moisture, solids-non-fat and fat analysis in butter by near infrared spectroscopy

Abstract Moisture, solids-non-fat and fat were analysed in butter by near infrared (NIR) spectroscopy without any previous sample treatment. A set of 102 samples was used to calibrate the instrument by modified partial least-squares regression. The following statistical results were achieved: standard error of calibration (SEC)=0.192 and square correlation coefficient (R2)=0.90 for moisture, SEC=0.086 and R2=0.93 for solids-non-fat, and SEC=0.168 and R2=0.94 for fat. To validate the calibration performed, a set of 24 butter samples was used. Standard errors of validation were 0.255, 0.071 and 0.377 for moisture, solids-non-fat and fat, respectively, and R2 for the regression of measurements by reference method versus NIR analysis were 0.83, 0.94 and 0.72 for moisture, solids-non-fat and fat, respectively. To compare, for all three components of the validation set, the results obtained by NIR spectroscopy with those obtained by the reference methods, linear regression and paired “t”-test were applied and the methods did not give significantly different results for p=0.05. When mean square prediction error analysis was applied, it could be concluded that a strong calibration model was obtained and that for all three components the main error was unexplained.

Current and future status of CWE

The current and future situation of CWE (Computational Wind Engineering) is reviewed from the following viewpoints ; 1) topics treated, 2) utilization of commercial code (software), 3) incompleteness of CWE, 4) remaining research subjects, 5) prediction accuracy, 6) new fields of CWE application, etc. Firstly, new tendencies within CWE application are indicated. Next, the over-attention being given to the application field and the lack of attention to the fundamental problems, including prediction error analysis are pointed out. Lastly, the future trend of CFD (Computational Fluid Dynamics) applications to wind engineering design are discussed.