Mean Absolute Prediction Error Research Articles

Kriging-based prediction of the Earth’s pole coordinates

Abstract Coordinates of the Earth’s pole represent two out of five Earth orientation parameters describing Earth’s rotation. They are necessary in transformation between celestial reference frame and terrestrial reference frame and what goes further in precise positioning and navigation, applications in astronomy, communication with outer space objects. Complexity of measuring techniques and data processing involved in the pole coordinates determination make it impossible to obtain them in real-time mode, hence a prediction problem of the polar motion emerges. In this study, geostatistical prediction methods, i. e., simple and ordinary kriging are applied. Millions of predictions have been performed to draw reasonable conclusions on prediction capabilities of applied kriging variants. The study is intended in ultra-short-term prediction (up to 15 days into the future) using the IERS EOP 14 C04 (IAU2000A) and IERS EOP 05 C04 (IAU2000A) series as a reference. Mean absolute prediction errors (for days 1–15) with respect to IERS 14 C04 are ranging 0.66–5.25 mas for PMx and 0.47–3.59 mas for PMy. On the other hand, for IERS 05 C04 the values are 0.60–4.95 mas and 0.44–3.29 mas for PMx and PMy; respectively. The results indicate competitiveness of the introduced methods with existing ones.

Predicting symptom complexity: Using autoregressive integrated moving average (ARIMA) models to create responsive clinic scheduling.

e13529 Background: Increasing cancer incidence, coupled with a trend in treating patients for longer periods of time, presents challenges in addressing all patients’ symptoms/concerns within the allotted time for ambulatory clinic appointments. Consequently, the ability to forecast and monitor the percentage of cancer patients with different symptom complexity levels is extremely valuable. Symptom complexity is a summary score that weighs the severity of all patient reported symptom scores at one time point. If a clinic could predict how many patients may need more time due to complex symptom management needs, clinic-scheduling templates could be adjusted to include a set number of longer appointments. Methods: Auto Regressive Integrated Moving Average (ARIMA) models were utilized to forecast the percentage of patients with a high symptom complexity level within one cancer clinic in Alberta, Canada. Goodness-of-fit measures such as Bayesian information criterion (BIC) and Ljung-Box test were used to determine optimal form for the ARIMA model. Following model selection, the autocorrelation function (ACF) was performed. These tests together verified that chosen AR, MA and differencing (I) were appropriate. Model performance on the historical data for model fit was summarized by Mean Absolute Error (MAE) and Root Squared Mean Error (RSME). Forecasting accuracy was assessed using mean absolute prediction error by comparing the forecasts with actual clinic data. Results: Of the multiple model structures tested, ARIMA (0, 0, 1) was selected, with the lowest BIC and non-significant Ljung-Box test. We obtained forecasts of the percentage of patients with high symptom complexity levels, with an MAE at 4.0%. To assess forecast accuracy, we calculated the absolute prediction error by comparing the forecasted percentages of patients with high symptom complexity levels to actual clinic visit data and the mean absolute prediction error was 5.9%. Conclusions: This forecasting model has important implications, allowing clinics to adjust scheduling templates to provide a select number of longer timeslots and therefore, be better prepared to meet the symptom management needs of cancer patients who are considered highly complex. This model could be applied to other clinical populations to allow for a tailored scheduling approach based on each clinic’s symptom complexity forecasting.

Pipeline Construction Cost Forecasting Using Multivariate Time Series Methods

Pipe material and labor costs constitute about 70% of pipeline construction costs. Pipe and labor costs are subject to considerable fluctuations over time. These fluctuations are problematic for cost estimation and bid preparation in pipeline projects, which are mostly large and long-term projects. The accurate prediction of pipe and labor costs is invaluable for cost estimators to prepare accurate bids and manage the cost contingencies. However, the existing literature does not take advantage of the leading indicators of pipeline construction cost time series to accurately forecast cost fluctuations in pipeline projects. The objective of this research is to identify the leading indicators of pipeline construction costs and develop multivariate time series models for forecasting cost fluctuations in pipeline projects. Nineteen potential leading indicators of pipe and labor costs were initially selected based on a comprehensive review of construction cost forecasting literature. The leading indicators were identified from this pool of potential leading indicators based on unit root tests and Granger causality tests. Multivariate time series models were developed based on the results of cointegration tests. Vector error correction (VEC) models were developed for the cointegrated variables, while vector autoregressive (VAR) models were developed for the non-cointegrated variables. Since multivariate time series models include information from the identified leading indicators, multivariate time series models are often expected to deliver more accurate forecasts than univariate time series models. The forecasting accuracies of multivariate time series models were compared with those of univariate time series models based on three common error measures: mean absolute prediction error (MAPE), root-mean-squared error (RMSE), and mean average error (MAE). The results show that multivariate time series models outperform univariate models for forecasting cost fluctuations in pipeline projects. The findings of this research contribute to the state of knowledge by identifying leading indicators of pipe and labor costs and developing multivariate time series models to forecast them. The multivariate time series models with leading indicators are more accurate than univariate models for forecasting cost fluctuations in pipeline projects. It is expected that the proposed multivariate time series forecasting models contribute to the enhancement of the theory and practice of pipeline construction cost forecasting and help cost engineers and investment planners to prepare more accurate bids, cost estimates, and budgets for pipeline projects.

Refractive prediction of four different intraocular lens calculation formulas compared between new swept source optical coherence tomography and partial coherence interferometry.

PurposeTo compare the biometry and prediction of postoperative refractive outcomes of four different formulae (Haigis, SRK/T, Holladay1, Barrett Universal II) obtained by swept-source optical coherence tomography (SS-OCT) biometers and partial coherence interferometry (PCI; IOLMaster ver 5.4).MethodsWe compared the biometric values of SS-OCT (ANTERION, Heidelberg Engineering Inc., Heidelberg, Germany) and PCI (IOLMaster, Carl Zeiss Meditec, Jena, Germany). Predictive errors calculated using four different formulae (Haigis, SRKT, Holladay1, Barrett Universal II) were compared at 1 month after cataract surgery.ResultsThe mean preoperative axial length (AL) showed no statistically significant difference between SS-OCT and PCI (SS-OCT: 23.78 ± 0.12 mm and PCI: 23.77 ± 0.12 mm). The mean anterior chamber depth (ACD) was 3.30 ± 0.04 mm for SS-OCT and 3.23 ± 0.04 mm for PCI, which was significantly different between the two techniques. The mean corneal curvature also differed significantly between the two techniques. The difference in mean arithmetic prediction error was significant in the Haigis, SRKT, and Holladay1 formulae. The difference in mean absolute prediction error was significant in all four formulae.ConclusionsSS-OCT and PCI demonstrated good agreement on biometric measurements; however, there were significant differences in some biometric values. These differences in some ocular biometrics can cause a difference in refractive error after cataract surgery. New type SS-OCT was not superior to the IOL power prediction calculated by PCI.

Assessment of patient specific information in the wild on fundus photography and optical coherence tomography

In this paper we analyse the performance of machine learning methods in predicting patient information such as age or sex solely from retinal imaging modalities in a heterogeneous clinical population. Our dataset consists of N = 135,667 fundus images and N = 85,536 volumetric OCT scans. Deep learning models were trained to predict the patient’s age and sex from fundus images, OCT cross sections and OCT volumes. For sex prediction, a ROC AUC of 0.80 was achieved for fundus images, 0.84 for OCT cross sections and 0.90 for OCT volumes. Age prediction mean absolute errors of 6.328 years for fundus, 5.625 years for OCT cross sections and 4.541 for OCT volumes were observed. We assess the performance of OCT scans containing different biomarkers and note a peak performance of AUC = 0.88 for OCT cross sections and 0.95 for volumes when there is no pathology on scans. Performance drops in case of drusen, fibrovascular pigment epitheliuum detachment and geographic atrophy present. We conclude that deep learning based methods are capable of classifying the patient’s sex and age from color fundus photography and OCT for a broad spectrum of patients irrespective of underlying disease or image quality. Non-random sex prediction using fundus images seems only possible if the eye fovea and optic disc are visible.

Model-Based Estimation of Iohexol Plasma Clearance for Pragmatic Renal Function Determination in the Renal Transplantation Setting

BackgroundIohexol plasma clearance-based glomerular filtration rate (GFR) determination provides an accurate method for renal function evaluation. This technique is increasingly advocated for clinical situations that dictate highly accurate renal function assessment, as an alternative to conventional serum creatinine-based methods with limited accuracy or poor feasibility. In the renal transplantation setting, this particularly applies to living renal transplant donor eligibility screening, renal transplant function monitoring and research purposes. The dependency of current iohexol GFR estimation techniques on extensive sampling, however, has limited its clinical application. We developed a population pharmacokinetic model and limited sampling schedules, implemented in the online InsightRX precision dosing platform, to facilitate pragmatic iohexol GFR assessment.MethodsIohexol concentrations (n = 587) drawn 5 min to 4 h after administration were available from 67 renal transplant recipients and 41 living renal transplant donor candidates with measured iohexol GFRs of 27–117 mL/min/1.73 m2. These were split into a model development (n = 72) cohort and an internal validation (n = 36) cohort. External validation was performed with 1040 iohexol concentrations from 268 renal transplant recipients drawn between 5 min and 4 h after administration, and extended iohexol curves up to 24 h from 11 random patients with impaired renal function. Limited sampling schedules based on one to four blood draws within 4 h after iohexol administration were evaluated in terms of bias and imprecision, using the mean relative prediction error and mean absolute relative prediction error. The total deviation index and percentage of limited sampling schedule-based GFR predictions within ± 10% of those of the full model (P10) were assessed to aid interpretation.ResultsIohexol pharmacokinetics was best described with a two-compartmental first-order elimination model, allometrically scaled to fat-free mass, with patient type as a covariate on clearance and the central distribution volume. Model validity was confirmed during the internal and external validation. Various limited sampling schedules based on three to four blood draws within 4 h showed excellent predictive performance (mean relative prediction error < ± 0.5%, mean absolute relative prediction error < 3.5%, total deviation index < 5.5%, P10 > 97%). The best limited sampling schedules based on three to four blood draws within 3 h showed reduced predictive performance (mean relative prediction error < ± 0.75%, mean absolute relative prediction error < 5.5%, total deviation index < 9.5%, P10 ≥ 85%), but may be considered for their enhanced clinical feasibility when deemed justified.ConclusionsOur online pharmacometric tool provides an accurate, pragmatic, and ready-to-use technique for measured GFR-based renal function evaluation for clinical situations where conventional methods lack accuracy or show limited feasibility. Additional adaptation and validation of our model and limited sampling schedules for renal transplant recipients with GFRs below 30 mL/min is warranted before considering this technique in these patients.Supplementary InformationThe online version contains supplementary material available at 10.1007/s40262-021-00998-z.

Limited Sampling Strategy for Estimation of Mycophenolic Acid Exposure in Adult Chinese Heart Transplant Recipients.

Background: With the increasing use of mycophenolic acid (MPA) formulations in organ transplantation, the need for personalized immunosuppressive therapy has become well recognized based on therapeutic drug monitoring (TDM) for avoidance of drug-related toxicity while maintaining efficacy. Few studies have assessed area under the 12 h concentration-time curve of MPA (MPA-AUC0–12h) in heart transplant recipients who received mycophenolate mofetil (MMF) dispersible tablets (MMFdt). The aim of the study was to investigate the pharmacokinetics (PK) of MMFdt combined with tacrolimus and further to develop a practical method for estimation of MPA-AUC0–12h using a limited sampling strategy (LSS). Methods: A prospective study in a single center was performed in patients who continuously administrated with MMFdt or MMF capsule (MMFc) for at least 7 days after cardiac transplantation from 2018 to 2020. A total of 48 Chinese adult heart transplant recipients were enrolled. Blood samples were collected before and 0.5, 1, 1.5, 2, 4, 6, 8, 10 and 12 h after MMF administration. The validated high-performance liquid chromatography combined with tandem mass spectrometry method was used to measure MPA concentrations. Non-compartmental pharmacokinetic (PK) analysis was applied to calculate the data obtained from individual recipients by WinNonlin. LSS models were developed for MPA-AUC0–12h prediction with multivariate stepwise regression analysis. Results: A large inter-individual variability was observed in AUC0–12h, Tmax, Cmax, MRT0–12h, t1/2 and CL/F after multiple dosing of MMFdt. However, no significant differences were observed between main PK parameters of MMFdt and MMFc. The best estimation of MPA-AUC0–12h was achieved with four points: MPA-AUC0–12h = 8.424 + 0.781 × C0.5 + 1.263 × C2 + 1.660 × C4 + 3.022 × C6 (R 2 = 0.844). The mean prediction error (MPE) and mean absolute prediction error (MAPE) of MPA-AUC0–12h were 2.09 ± 14.05% and 11.17 ± 8.52%, respectively. Both internal and external validations showed good applicability for four-point LSS equation. Conclusion: The results provide strong evidence for the use of LSS model other than a single time-point concentration of MPA when performing TDM. A four-point LSS equation using the concentrations at 0.5, 2, 4, 6 h is recommended to estimate MPA-AUC0–12h during early period after transplantation in Chinese adult heart transplant recipients receiving MMFdt or MMFc. However, proper internal and external validations with more patients should be conducted in the future.

Adaptive Susceptible-Infectious-Removed Model for Continuous Estimation of the COVID-19 Infection Rate and Reproduction Number in the United States: Modeling Study.

BackgroundThe dynamics of the COVID-19 pandemic vary owing to local population density and policy measures. During decision-making, policymakers consider an estimate of the effective reproduction number Rt, which is the expected number of secondary infections spread by a single infected individual.ObjectiveWe propose a simple method for estimating the time-varying infection rate and the Rt.MethodsWe used a sliding window approach with a Susceptible-Infectious-Removed (SIR) model. We estimated the infection rate from the reported cases over a 7-day window to obtain a continuous estimation of Rt. A proposed adaptive SIR (aSIR) model was applied to analyze the data at the state and county levels.ResultsThe aSIR model showed an excellent fit for the number of reported COVID-19 cases, and the 1-day forecast mean absolute prediction error was <2.6% across all states. However, the 7-day forecast mean absolute prediction error approached 16.2% and strongly overestimated the number of cases when the Rt was rapidly decreasing. The maximal Rt displayed a wide range of 2.0 to 4.5 across all states, with the highest values for New York (4.4) and Michigan (4.5). We found that the aSIR model can rapidly adapt to an increase in the number of tests and an associated increase in the reported cases of infection. Our results also suggest that intensive testing may be an effective method of reducing Rt.ConclusionsThe aSIR model provides a simple and accurate computational tool for continuous Rt estimation and evaluation of the efficacy of mitigation measures.

Ray-tracing Calculation Using Scheimpflug Tomography of Diffractive Extended Depth of Focus IOLs Following Myopic LASIK.

To evaluate a ray-tracing formula for intraocular lens (IOL) calculation of diffractive extended depth of focus IOLs after myopic laser in situ keratomileusis (LASIK) compared to formulas from an established online calculator. This retrospective, consecutive case series included patients after cataract surgery with implantation of an extended depth of focus (EDOF) IOL (AT LARA, Carl Zeiss Meditec; Symfony, Johnson & Johnson) and a history of myopic LASIK. Preoperative assessments included biometry (IOLMaster; Carl Zeiss Meditec) and corneal tomography, including true net power (TNP) (Pentacam; Oculus Optikgeräte GmbH). To evaluate the measurements, the simulated keratometry values (SimK) were compared to the TNP. Regarding IOL calculation, the mean prediction error, mean and median absolute prediction error (MAE and MedAE), and number of eyes within ±0.50, ±1.00, and ±2.00 diopters (D) from the Haigis-L, Shammas, and Barrett True K No History formulas to the Potvin-Hill and Haigis with TNP (Pentacam) formulas were compared. Thirty-six eyes matched the inclusion criteria with a mean spherical equivalent of -6.26 ± 3.25 diopters (D) preoperatively and -0.79 ± 0.75 D postoperatively. The mean difference from SimK and TNP was significantly different from zero (P < .001; -1.24 ± 0.81 D). The best performing formulas by MedAE were the Potvin-Hill and Barrett True K No History (0.39 ± 0.78 and 0.64 ± 1.00 D). The formula with the most eyes within ±0.50 D was the Potvin-Hill (64%), followed by the Barrett True K No History (44%). For MAE and percentage of eyes within ±0.50 D, the Potvin-Hill formula was significantly better than the Haigis-L, Shammas, and Haigis-TNP formulas (P < .05). Calculation of IOLs in patients who had LASIK remains less predicable than calculations for virgin eyes. Using ray-tracing to calculate diffractive EDOF IOLs after myopic LASIK, the Potvin-Hill formula outperformed established formulas in terms of the percentage within target refraction and the MAE. [J Refract Surg. 2021;37(4):231-239.].

Does Every Calculation Formula Fit for All Types of Intraocular Lenses? Optimization of Constants for Tecnis ZA9003 and ZCB00 Is Necessary.

Background and Objectives: To evaluate the performance of intraocular lenses (IOLs) using power calculation formulas on different types of IOL. Materials and Methods: 120 eyes and four IOL types (BioLine Yellow Accurate Aspheric IOL (i-Medical), TECNIS ZCB00, TECNIS ZA9003 (Johnson & Johnson) (3-piece IOL) and Softec HD (Lenstec)) were analyzed. The performance of Haigis, Barret Universal II and SKR-II formulas were compared between IOL types. The mean prediction error (ME) and mean absolute prediction error (MAE) were analyzed. Results: The overall percentage of eyes predicted within ±0.25 diopters (D) was 40.8% for Barret; 39.2% Haigis and 31.7% for SRK-II. Barret and Haigis had a significantly lower MAE than SRK-II (p < 0.05). The results differed among IOL types. The largest portion of eyes predicted within ±0.25 D was with the Barret formula in ZCB00 (33.3%) and ZA9003 (43.3%). Haigis was the most accurate in Softec HD (50%) and SRK-II in Biolline Yellow IOL (50%). ZCB00 showed a clinically significant hypermetropic ME compared to other IOLs. Conclusions: In general, Barret formulas had the best performance as a universal formula. However, the formula should be chosen according to the type of IOL in order to obtain the best results. Constant optimizations are necessary for the Tecnis IOL ZCB00 and ZA9003, as all of the analyzed formulas achieved a clinically significant poor performance in this type of IOL. ZCB00 also showed a hypermetropic shift in ME in all the formulas.

Machine learning models based on remote and proximal sensing as potential methods for in-season biomass yields prediction in commercial sorghum fields.

Crop yield monitoring demonstrated the potential to improve agricultural productivity through improved crop breeding, farm management and commodity planning. Remote and proximal sensing offer the possibility to cut crop monitoring costs traditionally associated with surveys and censuses. Fraction of absorbed photosynthetically active radiation (fAPAR), chlorophyll concentration (CI) and normalized difference vegetation (NDVI) indices were used in crop monitoring, but their comparative performances in sorghum monitoring is lacking. This work aimed therefore at closing this gap by evaluating the performance of machine learning modelling of in-season sorghum biomass yields based on Sentinel-2-derived fAPAR and simpler high-throughput optical handheld meters-derived NDVI and CI calculated from sorghum plants reflectance. Bayesian ridge regression showed good cross-validated performance, and high reliability (R2 = 35%) and low bias (mean absolute prediction error, MAPE = 0.4%) during the validation step. Hand-held optical meter-derived CI and Sentinel-2-derived fAPAR showed comparable effects on machine learning performance, but CI outperformed NDVI and was therefore considered as a good alternative to Sentinel-2's fAPAR. The best times to sample the vegetation indices were the months of June (second half) and July. The results obtained in this work will serve several purposes including improvements in plant breeding, farming management and sorghum biomass yield forecasting at extension services and policy making levels.

Research on the predictive effect of a combined model of ARIMA and neural networks on human brucellosis in Shanxi Province, China: a time series predictive analysis

BackgroundBrucellosis is a major public health problem that seriously affects developing countries and could cause significant economic losses to the livestock industry and great harm to human health. Reasonable prediction of the incidence is of great significance in controlling brucellosis and taking preventive measures.MethodsOur human brucellosis incidence data were extracted from Shanxi Provincial Center for Disease Control and Prevention. We used seasonal-trend decomposition using Loess (STL) and monthplot to analyse the seasonal characteristics of human brucellosis in Shanxi Province from 2007 to 2017. The autoregressive integrated moving average (ARIMA) model, a combined model of ARIMA and the back propagation neural network (ARIMA-BPNN), and a combined model of ARIMA and the Elman recurrent neural network (ARIMA-ERNN) were established separately to make predictions and identify the best model. Additionally, the mean squared error (MAE), mean absolute error (MSE) and mean absolute percentage error (MAPE) were used to evaluate the performance of the model.ResultsWe observed that the time series of human brucellosis in Shanxi Province increased from 2007 to 2014 but decreased from 2015 to 2017. It had obvious seasonal characteristics, with the peak lasting from March to July every year. The best fitting and prediction effect was the ARIMA-ERNN model. Compared with those of the ARIMA model, the MAE, MSE and MAPE of the ARIMA-ERNN model decreased by 18.65, 31.48 and 64.35%, respectively, in fitting performance; in terms of prediction performance, the MAE, MSE and MAPE decreased by 60.19, 75.30 and 64.35%, respectively. Second, compared with those of ARIMA-BPNN, the MAE, MSE and MAPE of ARIMA-ERNN decreased by 9.60, 15.73 and 11.58%, respectively, in fitting performance; in terms of prediction performance, the MAE, MSE and MAPE decreased by 31.63, 45.79 and 29.59%, respectively.ConclusionsThe time series of human brucellosis in Shanxi Province from 2007 to 2017 showed obvious seasonal characteristics. The fitting and prediction performances of the ARIMA-ERNN model were better than those of the ARIMA-BPNN and ARIMA models. This will provide some theoretical support for the prediction of infectious diseases and will be beneficial to public health decision making.

Predicting the EQ-5D utilities from the Kansas City Cardiomyopathy Questionnaire in patients with heart failure

Evaluation of health status benefits, cost-effectiveness, and value of new heart failure therapies is critical for supporting their use. The Kansas City Cardiomyopathy Questionnaire (KCCQ) measures patients' heart failure-specific health status but does not provide utilities needed for cost-effectiveness analyses. We mapped the KCCQ scores to EQ-5D scores so that estimates of societal-based utilities can be generated to support economic analyses. Using data from two US cohort studies, we developed models for predicting EQ-5D utilities (3L and 5L versions) from the KCCQ (23- and 12-item versions). In addition to predicting scores directly, we considered predicting the five EQ-5D health state items and deriving utilities from the predicted responses, allowing different countries' health state valuations to be used. Model validation was performed internally via bootstrap and externally using data from two clinical trials. Model performance was assessed using R2, mean prediction error, mean absolute prediction error, and calibration of observed vs. predicted values. The EQ-5D-3L models were developed from 1000 health status assessments in 547 patients with heart failure and reduced ejection fraction (HFrEF), while the EQ-5D-5L model was developed from 3925 patients with HFrEF. For both versions, models predicting individual EQ-5D items performed as well as those predicting utilities directly. The selected models for the 3L had internally validated R2 of 48.4-50.5% and 33.7-45.6% on external validation. The 5L version had validated R2 of 57.7%. Mappings from the KCCQ to the EQ-5D can yield the estimates of societal-based utilities to support cost-effectiveness analyses when EQ-5D data are not available.

MetabCombiner: Paired Untargeted LC-HRMS Metabolomics Feature Matching and Concatenation of Disparately Acquired Data Sets.

LC-HRMS experiments detect thousands of compounds, with only a small fraction of them identified in most studies. Traditional data processing pipelines contain an alignment step to assemble the measurements of overlapping features across samples into a unified table. However, data sets acquired under nonidentical conditions are not amenable to this process, mostly due to significant alterations in chromatographic retention times. Alignment of features between disparately acquired LC-MS metabolomics data could aid collaborative compound identification efforts and enable meta-analyses of expanded data sets. Here, we describe metabCombiner, a new computational pipeline for matching known and unknown features in a pair of untargeted LC-MS data sets and concatenating their abundances into a combined table of intersecting feature measurements. metabCombiner groups features by mass-to-charge (m/z) values to generate a search space of possible feature pair alignments, fits a spline through a set of selected retention time ordered pairs, and ranks alignments by m/z, mapped retention time, and relative abundance similarity. We evaluated this workflow on a pair of plasma metabolomics data sets acquired with different gradient elution methods, achieving a mean absolute retention time prediction error of roughly 0.06 min and a weighted per-compound matching accuracy of approximately 90%. We further demonstrate the utility of this method by comprehensively mapping features in urine and muscle metabolomics data sets acquired from different laboratories. metabCombiner has the potential to bridge the gap between otherwise incompatible metabolomics data sets and is available as an R package at https://github.com/hhabra/metabCombiner and Bioconductor.

Effect of the capsular tension ring on refractive outcome after phacoemulsification.

Purpose: To evaluate how capsular tension rings (CTR) affect refractive outcomes following cataract extraction by phacoemulsification complicated by zonular instability.Study design: Retrospective.Methods: In a retrospective case-control study, the 29 eyes of 29 patients with CTRs were compared with the 29 eyes of patients without zonular instability following routine phacoemulsification, with mean arithmetic refractive prediction error (ArRPE) and mean absolute refractive prediction error (AbRPE) as primary outcome measures.Results: A comparison of eyes with and without CTRs according to the SRK/ T formula revealed no statistically significant difference in ArRPE (0.52 vs. 0.45 D, p = 0.570) or AbRPE (0.52 vs. 0.55 D, p = 0.799). Postoperative hypermetropic shift occurred in most cases in both groups, although the mean difference between eyes with CTR (22/ 29) and without CTR (25/ 29) was not statistically significant (p = 0.315).Conclusion: Implanting CTRs did not consistently affect refractive outcomes versus routine phacoemulsification. Results suggest that intraocular lens power can be calculated as usual when CTRs are used.Abbreviations: CTR = capsular tension ring, ArRPE = arithmetic refractive prediction errors, AbRPE = absolute refractive prediction errors, IOL = intraocular lens, WTW = white-to-white, ACD = anterior chamber depths, AL = axial lengths, RPE = refractive prediction error

A predictive internet-based model for COVID-19 hospitalization census

The COVID-19 pandemic has strained hospital resources and necessitated the need for predictive models to forecast patient care demands in order to allow for adequate staffing and resource allocation. Recently, other studies have looked at associations between Google Trends data and the number of COVID-19 cases. Expanding on this approach, we propose a vector error correction model (VECM) for the number of COVID-19 patients in a healthcare system (Census) that incorporates Google search term activity and healthcare chatbot scores. The VECM provided a good fit to Census and very good forecasting performance as assessed by hypothesis tests and mean absolute percentage prediction error. Although our study and model have limitations, we have conducted a broad and insightful search for candidate Internet variables and employed rigorous statistical methods. We have demonstrated the VECM can potentially be a valuable component to a COVID-19 surveillance program in a healthcare system.

Food waste reduction and economic savings in times of crisis: The potential of machine learning methods to plan guest attendance in Swedish public catering during the Covid-19 pandemic

Food waste is a significant problem within public catering establishments in any normal situation. During spring 2020 the Covid-19 pandemic placed the public catering system under greater pressure, revealing weaknesses within the system and generation of food waste due to rapidly changing consumption patterns. In times of crisis, it is especially important to conserve resources and allocate existing resources to areas where they can be of most use, but this poses significant challenges. This study evaluated the potential of a forecasting model to predict guest attendance during the start and throughout the pandemic. This was done by collecting data on guest attendance in Swedish school and preschool catering establishments before and during the pandemic, and using a machine learning approach to predict future guest attendance based on historical data. Comparison of various learning methods revealed that random forest produced more accurate forecasts than a simple artificial neural network, with conditional mean absolute prediction error of <0.15 for the trained dataset. Economic savings were obtained by forecasting compared with a no-plan scenario, supporting selection of the random forest approach for effective forecasting of meal planning. Overall, the results obtained using forecasting models for meal planning in times of crisis confirmed their usefulness. Continuous use can improve estimates for the test period, due to the agile and flexible nature of these models. This is particularly important when guest attendance is unpredictable, so that production planning can be optimized to reduce food waste and contribute to a more sustainable and resilient food system.

Spatial Models of Adjusted Precipitation for Canada at Varying Time Scales

AbstractThis study presents spatial models (i.e., thin-plate spatially continuous spline surfaces) of adjusted precipitation for Canada at daily, pentad (5 day), and monthly time scales from 1900 to 2015. The input data include manual observations from 3346 stations that were adjusted previously to correct for snow water equivalent (SWE) conversion and various gauge-related issues. In addition to the 42 331 models for daily total precipitation and 1392 monthly total precipitation models, 8395 pentad models were developed for the first time, depicting mean precipitation for 73 pentads annually. For much of Canada, mapped precipitation values from this study were higher than those from the corresponding unadjusted models (i.e., models fitted to the unadjusted data), reflecting predominantly the effects of the adjustments to the input data. Error estimates compared favorably to the corresponding unadjusted models. For example, root generalized cross-validation (GCV) estimate (a measure of predictive error) at the daily time scale was 3.6 mm on average for the 1960–2003 period as compared with 3.7 mm for the unadjusted models over the same period. There was a dry bias in the predictions relative to recorded values of between 1% and 6.7% of the average precipitations amounts for all time scales. Mean absolute predictive errors of the daily, pentad, and monthly models were 2.5 mm (52.7%), 0.9 mm (37.4%), and 11.2 mm (19.3%), respectively. In general, the model skill was closely tied to the density of the station network. The current adjusted models are available in grid form at ~2–10-km resolutions.

Accuracy of Intraocular Lens Power Calculation in Children With Vitrectomized Eyes Undergoing Cataract Surgery

Purpose: To assess the predictability of desired postoperative refractive outcomes using the SRK-II formula for intraocular lens (IOL) power calculation in children undergoing cataract surgery in eyes with a previous pars plana vitrectomy (PPV). Methods: In this retrospective study, 68 eyes of 66 children who underwent cataract surgery and IOL implantation in eyes that had previous vitrectomy between January 2008 and December 2017 were included. Data were collected on preoperative and postoperative characteristics. The Mann-Whitney test, Kruskal-Wallis test, and multinominal logistic regression were used for comparing the results. Results: Absolute prediction error (APE) in the cohort was 1.29 ± 1.13. Desirable refractive outcome with insignificant prediction error of less than 0.50 diopters (D) was found in approximately one-third of the children. Patients were further subdivided by magnitude of APE as ≥ 0.50 to ≤ 1.00 D and &gt; 1.00 D. Age at the time of surgery, axial length, mean keratometry, silicone oil removal, IOL position, scleral buckle, and corneal suture did not affect APE and there was no significant difference between the groups ( P &gt; .05). Conclusions: The mean APE of the SRK-II formula for IOL power calculation in pediatric eyes that had vitrectomy is comparable to that reported in the literature for routine pediatric cataract surgeries. Factors such as age, axial length, corneal power, IOL position (bag/sulcus), scleral buckle, corneal suture, and silicone oil removal done prior to surgery or along with cataract surgery did not affect the APE. [ J Pediatr Ophthalmol Strabismus . 2021;58(2):126–131.]

Web Traffic Time Series Forecasting Using LSTM Neural Networks with Distributed Asynchronous Training

Evaluating web traffic on a web server is highly critical for web service providers since, without a proper demand forecast, customers could have lengthy waiting times and abandon that website. However, this is a challenging task since it requires making reliable predictions based on the arbitrary nature of human behavior. We introduce an architecture that collects source data and in a supervised way performs the forecasting of the time series of the page views. Based on the Wikipedia page views dataset proposed in a competition by Kaggle in 2017, we created an updated version of it for the years 2018–2020. This dataset is processed and the features and hidden patterns in data are obtained for later designing an advanced version of a recurrent neural network called Long Short-Term Memory. This AI model is distributed training, according to the paradigm called data parallelism and using the Downpour training strategy. Predictions made for the seven dominant languages in the dataset are accurate with loss function and measurement error in reasonable ranges. Despite the fact that the analyzed time series have fairly bad patterns of seasonality and trend, the predictions have been quite good, evidencing that an analysis of the hidden patterns and the features extraction before the design of the AI model enhances the model accuracy. In addition, the improvement of the accuracy of the model with the distributed training is remarkable. Since the task of predicting web traffic in as precise quantities as possible requires large datasets, we designed a forecasting system to be accurate despite having limited data in the dataset. We tested the proposed model on the new Wikipedia page views dataset we created and obtained a highly accurate prediction; actually, the mean absolute error of predictions regarding the original one on average is below 30. This represents a significant step forward in the field of time series prediction for web traffic forecasting.