Error Confidence Intervals Research Articles

Forecasting progression of mild cognitive impairment (MCI) and Alzheimer’s disease (AD) with digital twins

AbstractBackgroundMachine learning models can leverage historical data to forecast disease progression. These predictions can be integrated in clinical trial design to reduce sample size or increase power, speeding up the evaluation of new drugs. This is especially critical in AD where trials face challenges with enrollment and where no new therapies have emerged in 18 years. Recently, there has been a shift towards evaluating drugs in earlier stages of the disease using Clinical Dementia Rating Sum‐of‐Boxes (CDR‐SB) as the primary endpoint. We present a model that includes CDR‐SB and spans a broad range of baseline disease severity from MCI to mild‐to‐moderate AD.MethodWe used nearly 7,000 clinical records from placebo arms of AD clinical trials (in the C‐Path Online Data Repository for AD) and from observational studies (in the AD Neuroimaging Initiative) to train Conditional Restricted Boltzmann Machines (CRBMs). A CRBM is a generative machine learning model that learns a multivariate distribution over the relevant variables, and is particularly suited to model clinical data. A CRBM generates Digital Twins ‐ synthetic clinical records with baseline characteristics matched to those of actual trial subjects describing their likely progression under standard‐of‐care (SOC) with/without placebo. These forecasts include all components of CDR‐SB, Alzheimer’s Disease Assessment Scale ‐ Cognitive Subscale (ADAS‐Cog11) and Mini Mental State Examination (MMSE), along with other variables including labs, vitals, and biomarker status. We evaluated Digital Twins on a held‐out portion of the dataset (not included in the training dataset), stratified by baseline ADAS‐Cog11. For each variable, the mean change from baseline +/‐ 95% confidence interval error at 3‐month intervals was estimated by drawing 100 Digital Twins for each subject and averaging over the population.ResultCRBM‐generated predictions of ADAS‐Cog11, MMSE, and CDR‐SB progression up to 18 months were consistent within 95% confidence intervals with actual data across all scores, timepoints, and cohorts.ConclusionDigital Twins accurately model MCI and AD subjects’ progression on SOC with/without placebo across a broad range of baseline disease severity, and can be integrated into clinical trials as prognostic scores to increase power or reduce sample size required to achieve desired power.

Analysis of maturation features in fetal brain ultrasound via artificial intelligence for the estimation of gestational age

Optimal prenatal care relies on accurate gestational age dating. After the first trimester, the accuracy of current gestational age estimation methods diminishes with increasing gestational age. Considering that, in many countries, access to first trimester crown rump length is still difficult owing to late booking, infrequent access to prenatal care, and unavailability of early ultrasound examination, the development of accurate methods for gestational age estimation in the second and third trimester of pregnancy remains an unsolved challenge in fetal medicine. This study aimed to evaluate the performance of an artificial intelligence method based on automated analysis of fetal brain morphology on standard cranial ultrasound sections to estimate the gestational age in second and third trimester fetuses compared with the current formulas using standard fetal biometry. Standard transthalamic axial plane images from a total of 1394 patients undergoing routine fetal ultrasound were used to develop an artificial intelligence method to automatically estimate gestational age from the analysis of fetal brain information. We compared its performance-as stand alone or in combination with fetal biometric parameters-against 4 currently used fetal biometry formulas on a series of 3065 scans from 1992 patients undergoing second (n=1761) or third trimester (n=1298) routine ultrasound, with known gestational age estimated from crown rump length in the first trimester. Overall, 95% confidence interval of the error in gestational age estimation was 14.2 days for the artificial intelligence method alone and 11.0 when used in combination with fetal biometric parameters, compared with 12.9 days of the best method using standard biometrics alone. In the third trimester, the lower 95% confidence interval errors were 14.3 days for artificial intelligence in combination with biometric parameters and 17 days for fetal biometrics, whereas in the second trimester, the 95% confidence interval error was 6.7 and 7, respectively. The performance differences were even larger in the small-for-gestational-age fetuses group (14.8 and 18.5, respectively). An automated artificial intelligence method using standard sonographic fetal planes yielded similar or lower error in gestational age estimation compared with fetal biometric parameters, especially in the third trimester. These results support further research to improve the performance of these methods in larger studies.

155: Quantification of mitral regurgitation by calculation of regurgitant volume: 3D left ventricular echocardiography versus PISA

Quantification of mitral regurgitation (MR) by echocardiography is well established using several echo or Doppler parameters in which effective regurgitation office area (ERO) and regurgitant volume (RV) are the most commonly used. However, hemispheric assumptions allowing application of proximal isovelocity area (PISA) may be erroneous due to complex mitral valve morphology (i.e. mitral valve prolapse). We hypothesized that 3D left ventricular echocardiography associated with pulse Doppler could obtain RV with high reliability for comparison of 2 methods (3D vs PISA) in presence of mitral regurgitation. Methods: First, in 50 patients without MR, we compared LV ejection volumes (LVEj) from a full volume 3D echocardiographic acquisition and 2D Simpson method to the aortic stroke volume (ASV) obtained by Pulsed Doppler for validation of the 3D approach. Second, we analyzed 50 patients with different degree MR for comparison of the two approaches and verification of PISA RV values. Inter and intra observer variabilities were assessed for all techniques. Results: Correlations and Bland&Altman methods gave high adequacy between 3D LVEj and ASV compared to 2D and ASV (respectively 3D, r=0.96, y=0.91x+4.8, mean error (ME) and 95% confidence interval of error (CIE) [−0.84±6 ml] and 2D, r=0.81, y=0.78x+5.8, [−5.41±16 ml]. Variabilities average for 3D LVEj was 6±3% and 15+/-5% for 2D. In patients with mitral regurgitation, RV from PISA and 3D LVEj were 23.1±12 ml and 24.5±11 ml (p=0.37) and the fractional regurgitation 32±13% and 33±12% (p=0.63). However, mean error measurement was −1.37 ml and 95% confidence interval of error was 17.90 ml showing high discrepancy between the two methods. By selecting a cut-off of 5, 10, 15 ml, numbers of patients with uncorrected measurements was 48.5, 17 and 8.5%. When using ASE 4 grades classification, 25,7% of misclassified patients was obtained with PISA. Conclusion: 3D LVEj method is robust and reliable for calculation of mitral regurgitant volumes with significant differences compared than those obtained with PISA which seems to be less adequate for all types of MR.

Generalized Cornish-Fisher expansions

Cornish-Fisher expansions about the normal distribution provide accurate approximations for distributions of estimates and also for the level in the nominal error of confidence intervals. However, there is an advantage is expanding about a skew distribution like the chi-square, since the first order approximations become second order if the skewness is matched. Higher order approximations are also simplified. We demonstrate the method by approximating the distribution of standardized and Studentized linear combinations of means.

Estimating energy and protein requirements of thermally injured patients: art or science?

pected period of sustained hypermetabolism from thermal injury. The predictive method was considered precise if the 95% confidence interval (CI) for the root mean-square prediction error was within 15% of the measured energy expenditure. Unfortunately, none of the formulas met the criteria for being precise. None of the formulas’ CIs for error was within 20% of measured energy expenditure. Thirty-three percent of all of the methods were biased toward overpredicting measured energy expenditure, whereas about 20% of the formulas consistently underpredicted measured energy expenditure. The most precise, unbiased methods for estimating resting energy expenditure in our population 10 were the formulas of Xie et al. 11 (18 15% error, CI 12–24%) and Milner et al. 12 (16 15% error, CI 10 –22%). Both methods included body surface area (BSA) and percentage of total body surface area burn (BSAB). The method of Milner et al. also included the number of days postinjury and the basal metabolic rate according to Fleish’s standards for normal subjects.13 Because of its practicality in ease of use without a substantial difference in error, we chose the formula of Xie et al. 11 for estimation of goal caloric

Bootstrap Confidence Intervals In Nonlinear Regression Models When The Number of Observations is Fixed and The Variance Tends To 0. Application To Biadditive Models

We consider a parametric nonlinear regression model with independent and Gaussian errors. We assume that the number of observations is fixed and that the variance of errors tends to zero; we then derive the properties of confidence intervals for the parameters. These confidence intervals are calculated using both the quantiles of the estimator's asymptotic law and the quantiles of the estimator's bootstrap distribution. We show that if the pseudo-errors are simulated using the Gaussian distribution, then bootstrap can be applied successfully. The usual reduction in coverage error of confidence intervals is not, however, verified. A simulation study for a biadditive model shows the superiority of bootstrap when calculating confidence intervals for the interaction parameters.

An analysis of quantization error in digital filters based on interval algebras

Interval algebras are introduced as media in which to simulate digital filters for the purpose of analyzing quantization effects due to finite wordlength. The theory presented forms the basis of a computer-aided analysis scheme which generates confidence interval error bounds on the time domain response of a digital filter. The techniques are applicable to filters of arbitrary order and configuration implemented in either fixed or floating-point arithmetic. Experimental data is presented for a fourth-order low-pass Butterworth filter realized with fixed-point arithmetic in the direct, canonical direct, parallel, cascade, and ladder configurations. An example of Parker and Hess is used to illustrate the bounding limit cycles with interval techniques.