Lowest Mean Absolute Difference Research Articles

Balance diagnostics in propensity score analysis following multiple imputation: A new method.

The combination of propensity score analysis and multiple imputation has been prominent in epidemiological research in recent years. However, studies on the evaluation of balance in this combination are limited. In this paper, we propose a new method for assessing balance in propensity score analysis following multiple imputation. A simulation study was conducted to evaluate the performance of balance assessment methods (Leyrat's, Leite's, and new method). Simulated scenarios varied regarding the presence of missing data in the control or treatment and control group, and the imputation model with/without outcome. Leyrat's method was more biased in all the studied scenarios. Leite's method and the combine method yielded balanced results with lower mean absolute difference, regardless of whether the outcome was included in the imputation model or not. Leyrat's method had a higher false positive ratio and Leite's and combine method had higher specificity and accuracy, especially when the outcome was not included in the imputation model. According to simulation results, most of time, Leyrat's method and Leite's method contradict with each other on appraising the balance. This discrepancy can be solved using new combine method.

Improvement of Bone Age Assessment Using a Deep Learning Model in Young Children: Significance of Carpal Bone Analysis

Background: Deep learning methods used for bone age assessment (BAA) mostly employ the whole hand or regional convolutional neural networks without carpal bones; therefore, their application is insufficient in young children. Objectives: This study aimed to improve the accuracy of BAA in young children by integrating a carpal bone analysis and to achieve a similar BAA accuracy for all age groups. Patients and Methods: A hybrid Greulich-Pyle (GP) and modified Tanner-Whitehouse deep learning model for BAA was trained by integrating an additional carpal bone analysis of an open dataset. A total of 453 hand radiographs from a single institution were selected for external validation. To create the reference standard, three human experts conducted a BAA, based on the GP Atlas, and then, interobserver agreement was evaluated. The model performance was estimated by comparing the mean absolute difference (MAD) and the root mean square error (RMSE) between the two BAA models, including one with a carpal bone analysis (M1) and one without a carpal bone analysis (M2), and the reference standard. The MAD of each model was compared between sex and age groups with respect to four major developmental stages, that is, pre-puberty, early and mid-puberty, late puberty, and post-puberty. Results: The M1 model showed a higher accuracy with a lower MAD (0.366; 95% confidence interval [CI]: 0.337 - 0.395) compared to the M2 model (0.388; 95% CI: 0.358 - 0.418) for all age groups, with a significant difference (P < 0.001). The RMSE values versus the reference standard were 0.483 and 0.505 years for the M1 and M2 models, respectively. According to sex and developmental stage distributions, the M1 model had a greater predictive ability compared to the M2 model for pre-pubertal patients, regardless of sex (P = 0.008 for males and P = 0.022 for females). Conclusion: Based on the present findings, the integration of a carpal bone analysis into the BAA model improved its accuracy, especially in young children.

Restricted Inverse Kinematic Alignment Better Restores the Native Joint Line Orientation While Achieving Similar Balance, Laxity, and Arithmetic Hip-Knee-Ankle Angle to Gap Balancing Total Knee Arthroplasty.

Both restricted inverse kinematic alignment (iKA) and gap balancing aim for a balanced total knee arthroplasty by adjusting femoral component position based on ligamentous gaps. However, iKA targets a native tibial joint line vs resecting perpendicular to the mechanical axis. This study compares how these 2 techniques impact the balance and laxity throughout flexion and joint line obliquity (JLO), arithmetic hip-knee-ankle angle (aHKA), and the coronal plane alignment of the knee (CPAK). Two surgeons performed 75 robot-assisted iKA total knee arthroplasties. A digital joint tensioner collected laxity data throughout flexion before femoral resection. The femoral component position was determined using predictive gap-planning to optimize the balance throughout flexion. Planned gap balancing (pGB) simulations were performed for each case using neutral tibial resections. Mediolateral balance, laxity, and CPAK were compared among pGB, planned iKA (piKA), and final iKA. Both piKA and pGB had similar mediolateral balance and laxity, with mean differences <0.4 mm. piKA had a lower mean absolute difference from native JLO than pGB (3 ± 2° vs 7 ± 4°, P < .001). aHKA was similar (P > .05) between pGB and piKA. piKA recreated a more native CPAK distribution, with types I-V being the most common ones, while most pGB knees were of type V, VII, and III. Final iKA and piKA had similar mediolateral balance and laxity, with a root-mean-square error <1.4 mm. Although balance, laxity, and aHKA were similar between piKA and pGB, piKA better restored native JLO and CPAK phenotypes. The neutral tibial resection moved most pGB knees into types V, VII, and III. Surgeons should appreciate how the alignment strategy affects knee phenotypes.

Accuracy and Clinical Impact of Estimating Low-Density Lipoprotein-Cholesterol at High and Low Levels by Different Equations.

New more effective lipid-lowering therapies have made it important to accurately determine Low-density lipoprotein-cholesterol (LDL-C) at both high and low levels. LDL-C was measured by the β-quantification reference method (BQ) (N = 40,346) and compared to Friedewald (F-LDL-C), Martin (M-LDL-C), extended Martin (eM-LDL-C) and Sampson (S-LDL-C) equations by regression analysis, error-grid analysis, and concordance with the BQ method for classification into different LDL-C treatment intervals. For triglycerides (TG) &lt; 175 mg/dL, the four LDL-C equations yielded similarly accurate results, but for TG between 175 and 800 mg/dL, the S-LDL-C equation when compared to the BQ method had a lower mean absolute difference (mg/dL) (MAD = 10.66) than F-LDL-C (MAD = 13.09), M-LDL-C (MAD = 13.16) or eM-LDL-C (MAD = 12.70) equations. By error-grid analysis, the S-LDL-C equation for TG &gt; 400 mg/dL not only had the least analytical errors but also the lowest frequency of clinically relevant errors at the low (&lt;70 mg/dL) and high (&gt;190 mg/dL) LDL-C cut-points (S-LDL-C: 13.5%, F-LDL-C: 23.0%, M-LDL-C: 20.5%) and eM-LDL-C: 20.0%) equations. The S-LDL-C equation also had the best overall concordance to the BQ reference method for classifying patients into different LDL-C treatment intervals. The S-LDL-C equation is both more analytically accurate than alternative equations and results in less clinically relevant errors at high and low LDL-C levels.

783-P: Can Mood and Energy Levels Be Predicted by Preceding Glucose Values? Combining Ecological Momentary Assessment (EMA) and Continuous Glucose Monitoring (CGM)

For association with glycemic control, psychosocial research almost exclusively used retrospective questionnaire data to assess psychosocial outcomes. Little is known about the immediate, real-time effects of glucose values on psychosocial outcomes. In an observational study, 152 people with type 1 diabetes were equipped with CGM and completed a Smartphone-based EMA-survey 4-times a day over 18 days. Participants rated their mood and energy level separately on a single-item scale from 0 (very poor) to 10 (very good). Glucose data 90 minutes prior to each rating was extracted and mean glucose, time in range, time in hypoglycemic ranges (&amp;lt;70, &amp;lt;55 mg/dl), time in hyperglycemic ranges (&amp;gt;180, &amp;gt;250 mg/dl), and the mean absolute difference (MAD) of consecutive glucose values were calculated. Multilevel regression analyses were conducted with mood and energy levels as dependent variables and each parameter of glycemic control (90min prior) as independent variable, controlled for the overall level of glycemic control. Better mood was significantly predicted by lower mean glucose levels (p=.003), higher time in range (p=.003), and less hyperglycemic values (&amp;gt;180 mg/dl: p=.021; &amp;gt;250 mg/dl: p=.002). No significant associations were found for hypoglycemic values and MAD. Higher energy levels were significantly predicted by lower mean glucose levels (p=.012), higher time in range (p=.001), less hypoglycemic values (&amp;lt;70 mg/dl: p&amp;lt;.001; &amp;lt;54 mg/dl: p=.023), less values in the severe hyperglycemic range (&amp;gt;250 mg/dl: p&amp;lt;.001). Lower MAD predicted higher energy levels (p=.047). Ratings of mood and energy were significantly affected by immediately preceding glucose levels. While hyperglycemic values were more important for mood, hypoglycemic values were more important for energy. These preliminary findings have important implications for our understanding of real-time effects of hypo- and hyperglycemic glucose levels. Disclosure D. Ehrmann: Speaker’s Bureau; Self; Abbott, Berlin-Chemie AG, Glooko, Inc. Other Relationship; Self; Dexcom, Inc., Roche Diabetes Care. A.J. Schmitt: None. P. Rubertus: None. B. Kulzer: Research Support; Self; Abbott, AstraZeneca, Berlin-Chemie AG, Roche Diabetes Care. Speaker’s Bureau; Self; Abbott, Ascensia Diabetes Care, Berlin-Chemie AG, Eli Lilly and Company, Medtronic, Novo Nordisk Inc., Roche Diabetes Care. N. Hermanns: Advisory Panel; Self; Roche Diabetes Care. Research Support; Self; Abbott, AstraZeneca, Berlin-Chemie AG. Funding German Center for Diabetes Research (FKZ82DZD1102A)

Retrofitting Real-Life Dexcom G5 Data

We proposed in 2014 a retrofitting algorithm to retrospectively increase the accuracy of continuous glucose monitoring (CGM) data by using some blood glucose (BG) measurements. The method proved effective on Dexcom SEVEN Plus when about 10 highly accurate YSI measurements/session were available. In this study, we test the method on Dexcom G5 sensor in a more realistic setup, where only five capillary BG measurements (self-monitoring blood glucose [SMBG]) per 12 h-session are available. Furthermore, we investigate how accuracy is affected by the number of BG measurements. The algorithm was tested in 51 adults and 46 adolescents studied for 7 days with Dexcom G5. Each patient also underwent an ∼12-h hospital admission where frequent SMBG and YSI measurements were collected. First, five SMBGs per 12-h session were used to retrofit the CGM. Then, we varied the number of SMBGs provided to the method from 2 to 10 per 12-h session. Retrofitted CGM traces with five SMBGs per 12-h session have lower mean absolute difference than original CGM, reduced from 16.2 to 10.7 mg/dL (P < 0.001) in adults and from 17.6 to 11.5 mg/dL (P < 0.001) in adolescents, and mean absolute relative difference is reduced from 9.0% to 6.4% (P < 0.001) in adults and from 10.3% to 6.8% (P < 0.001) in adolescents. Reducing the number of BG measurements reduces improvement in the accuracy from >30% with 10 SMBGs per 12-h session to <16% with 2 SMBGs/day. The retrofitting method retrospectively improves the accuracy of CGM data, even if applied to one of the most accurate CGM sensors currently available on the market.

An Optimal Sample Data Usage Strategy to Minimize Overfitting and Underfitting Effects in Regression Tree Models Based on Remotely-Sensed Data

Regression tree models have been widely used for remote sensing-based ecosystem mapping. Improper use of the sample data (model training and testing data) may cause overfitting and underfitting effects in the model. The goal of this study is to develop an optimal sampling data usage strategy for any dataset and identify an appropriate number of rules in the regression tree model that will improve its accuracy and robustness. Landsat 8 data and Moderate-Resolution Imaging Spectroradiometer-scaled Normalized Difference Vegetation Index (NDVI) were used to develop regression tree models. A Python procedure was designed to generate random replications of model parameter options across a range of model development data sizes and rule number constraints. The mean absolute difference (MAD) between the predicted and actual NDVI (scaled NDVI, value from 0–200) and its variability across the different randomized replications were calculated to assess the accuracy and stability of the models. In our case study, a six-rule regression tree model developed from 80% of the sample data had the lowest MAD (MADtraining = 2.5 and MADtesting = 2.4), which was suggested as the optimal model. This study demonstrates how the training data and rule number selections impact model accuracy and provides important guidance for future remote-sensing-based ecosystem modeling.

Automatic versus manual lymphocyte fixation: impact on dose estimation using the cytokinesis-block micronucleus assay.

The lymphocyte cytokinesis-block micronucleus (CBMN) assay is a biodosemeter for the exposure to ionizing radiation. We examined the feasibility to implement a fully automated cell harvesting system for binucleate lymphocyte (BN) fixation. We compared fully automated versus manual BN fixation and evaluated its relevance on the accuracy of dose estimates using the CBMN. First, dose-response curves based on X-ray irradiated blood samples of ten healthy donors (0-4Gy, dose rate 1.0Gy/min) were established. BN was either prepared manually or fully automatically using the Hanabi cell harvester system PII. Slides were finally scored following an automatic or semi-automatic approach using the Metafer4 platform. The variance was calculated per dose and separately for each of the four fixation and scoring combinations. Thereafter, a serial of 16 blood samples of unknown exposure doses (0-3.9Gy X-ray) was analyzed. Employing the four fixation and scoring combinations, we compared the number of dose estimates lying outside the ±0.5Gy interval and the mean absolute difference (MAD) and examined sensitivity, specificity and accuracy of doses merged into binary dose categories of clinical significance. Irrespective of the fixation procedure, we observed at doses ≤1.0Gy about 2-4 times higher median variances for the automated scoring procedure over the semi-automated approach (p≤0.03). The lowest median variance was observed for automatic fixation+semi-automated scoring (135) which was even 2 times lower relative to manual fixation+semi-automated scoring (276, p=0.04). These differences became negligible after doses >1.0Gy. For the automatic fixation procedure, we also observed a tendency toward borderline significant higher numbers of dose estimates falling into the ±0.5Gy interval (25%, p=0.08) and lower MAD values (50%, p=0.09), which was predominantly caused by the accuracy of dose assessment >1.0Gy. Regarding the discrimination of binary dose categories of clinical significance, we observed a good agreement of both fixation procedures. The implementation of the automatic cell harvesting system considerably reduces the workload and results in dose estimates with a tendency of being slightly more accurate as they are after a manual fixation.

Comparative Performance Evaluation of the CONTOUR® NEXT Blood Glucose Monitoring System

Compare the performance of the CONTOUR® NEXT blood glucose monitoring system (BGMS) with other BGMSs. Study staff tested fingerstick samples from 146 subjects aged ≥18 years with diabetes using the CONTOUR® NEXT BGMS and 5 other BGMSs (Accu-Chek® Aviva Nano, Freestyle Lite®, OneTouch® Ultra® 2, OneTouch® Verio™ Pro and Truetrack®). All were compared to a blood sample from the same source assayed on a YSI. Subjects’ glucose levels were safely lowered or raised to provide a wide range of values; extreme glucose levels were achieved by blood sample glucose modification (i.e. glycolysis or addition of glucose solution). Analysis of variance (ANOVA) was conducted to compare the mean absolute difference (MAD) from the laboratory reference value across BGMSs. Comparisons of system accuracy were evaluated across a number of BG ranges (<3.9 mmol/L [70 mg/dL]; primary endpoint). In the low glucose range (<3.9 mmol/L [70 mg/dL]), CONTOUR® NEXT had a significantly lower MAD than all systems except Freestyle Lite®. Across the entire tested range (1.2 to 27.5 mmol/L [21 to 496 mg/dL]), CONTOUR® NEXT had a significantly lower MAD than all other BGMSs. By Parkes-Consensus Error Grid analysis, 100% of CONTOUR® NEXT, Accu-Chek® Aviva Nano and OneTouch® Verio™ Pro results were within Zone A. As assessed by MAD, the CONTOUR® NEXT was significantly more accurate than 4 of 5 BGMSs in the low glucose range. Across the entire glucose range tested (as assessed by MAD), CONTOUR® NEXT was significantly more accurate compared with the 5 other BGMSs evaluated.

Laboratory Intercomparison of the Cytokinesis-Block Micronucleus Assay

The focus of the study is an intercomparison of laboratories' dose-assessment performances using the cytokinesis-block micronucleus (CBMN) assay as a diagnostic triage tool for individual radiation dose assessment. Homogenously X-irradiated (240 kVp, 1 Gy/min) blood samples for establishing calibration data (0.25-5 Gy) as well as blind samples (0.1-6.4 Gy) were sent to the participants. The CBMN assay was performed according to protocols individually established and varying among participating laboratories. The time taken to report dose estimates was documented for each laboratory. Additional information concerning laboratory organization/characteristics as well as assay performance was collected. The mean absolute difference (MAD) was calculated and radiation doses were merged into four triage categories reflecting clinical aspects to calculate accuracy, sensitivity and specificity. The earliest report time was 4 days after sample arrival. The CBMN dose estimates were reported with high accuracy (MAD values of 0.20-0.50 Gy at doses below 6.4 Gy for both manual and automated scoring procedures), but showed a limitation of the assay at the dose point of 6.4 Gy, which resulted in a clear dose underestimation in all cases. The MAD values (without 6.4 Gy) differed significantly (P = 0.03) between manual (0.25 Gy, SEM = 0.06, n = 4) or automated scoring procedures (0.37 Gy, SEM = 0.08, n = 5), but lowest MAD were equal (0.2 Gy) for both scoring procedures. Likewise, both scoring procedures led to the same allocation of dose estimates to triage categories of clinical significance (about 83% accuracy and up to 100% specificity).

Estimation of daily and total lactation milk yield of Chios ewes from single morning or evening records

Linear models were developed and evaluated for the estimation of daily (24 h) and total lactation milk yield of dairy sheep from single morning (am) or evening (pm) milking records. A total of 22,908 individual test-day milk records of 5110 ewes of the Chios breed, raised in 46 flocks, were used. Depending on the model, different daily milk yield estimates were derived for each ewe, accounting for lactation number, stage of lactation, season of previous lambing and interval between successive milkings. Daily milk yield was also estimated from doubling the single am or pm record. Actual and estimated yields were compared using root mean square errors (RMSEs), mean absolute differences, an accuracy parameter defined as the ratio of the actual yield variance over the sum of the variances of actual yield and absolute difference, and the product moment correlation between estimated and actual yield. Results were validated on independent datasets. Linear models resulted in less biased and more accurate estimates of the daily milk yield than simply doubling the am or pm record. Root mean square errors decreased by 7–37% and the mean absolute difference was reduced by 1–4% of the average daily yield. Higher accuracy and correlation were obtained from linear model application than doubling. Total lactation milk yield was predicted based on actual daily yield and compared to predictions based on estimated daily yield from am or pm records, as well as directly on single milking records. Root mean square errors obtained when daily yield had been estimated with linear models were 26–35% lower compared to doubling the am or pm yield and 0–13% lower compared to estimating the total lactation yield directly from single milking records. Linear model application also resulted in lower mean absolute difference and higher accuracy and correlation than doubling the am or pm record. Recording the yield of a single milking (am or pm) instead of both can benefit milk recording by reducing its cost and increasing farmer participation. In this context, linear models developed in the present study can be used for the accurate estimation of daily (24 h) and total lactation milk yield from single milking records.

Analysis: dawn of real-time continuous glucose sensing.

381 NEAR-CONTINUOUS GLUCOSE MONITORING is useful in evaluating glucose trends during the day, and frequently identifies postprandial hyperglycemia and nocturnal hypoglycemia that would have been missed by the routine capillary blood glucose monitoring. The first commercially available near-continuous glucose monitoring system (CGMS) was the Medtronic Minimed CGMS introduced in 1999. This system is not marketed to patients, but is sold to health care professionals to be used similar to a Holter monitor. The CGMS provides glucose readings every 5 min, and it is critical to have software that can analyze and present these data in a meaningful, usable manner. Because the sensor can develop “biofouling” resulting in unreliable glucose readings, a retrospective analysis is necessary to assure the sensor has been functioning properly and the resulting information is sufficiently accurate to interpret glucose trends. Several recent articles have expressed concerns about the reliability of data generated by the CGMS, with one article questioning its precision (the ability of two sensors worn at the same time to give consistent results),1 and a second article questioned the ability of the CGMS to accurately detect nocturnal hypoglycemia.2 Medtronic Minimed has been actively working to resolve these concerns. In 2001 they revised their software with version 3.0. The previous software used a linear regression model utilizing at least three calibration values entered by the subject each calendar day from midnight to midnight. Because a new regression equation was created every 24 h, there was a discontinuity of glucose readings at midnight. The new software (version 3.0) uses a rolling average of glucose values, eliminating the discontinuity of readings at midnight, and uses more sophisticated regression equations. In this issue of the journal, Weinzimer et al.3 compared glucose values generated using Medronic Minimed software version 3.0 to values generated by the previous version (2.0). Using the new software there were 5–7% fewer skipped readings, and accuracy [as determined by a lower mean absolute difference (MAD) between sensor readings and calibration readings] was improved for both nondiabetic and diabetic subjects. Additional features of version 3.0 software include the ability to analyze results by time brackets (such as daytime or overnight), and by preand postprandial time brackets for each meal (as long as the patient enters a “event marker” when he or she ate). The software calculates average, minimum, and maximum values and graphs results around each meal/or time frame, which greatly aids the rapid review and analysis of the many data points. The software still has several shortcomings. To determine an “optimal” day (meeting Minimed criteria for sensor accuracy) the program continues to use a midnight to midnight day, instead of a 24-h time frame beginning when the sensor is initialized. The glucose readings are still manually entered by the subject, and wrong values can be entered, val-

Evolution of analytical performance in portable glucose meters in the last decade.

To assess and compare the technical accuracy of portable glucose meters during the last decade. One-thousand preprandial (pre) and postprandial (post) capillary whole-blood glucose values measured with meters owned mainly by diabetic patients were compared with a single laboratory method yearly from 1989 to 1999. A total of 21,950 capillary measurements and their corresponding laboratory reference values were analyzed at our clinic. The lowest mean absolute difference was found in 1989 (pre: 2 +/- 22 mg/dl, post: 9 +/- 31 mg/dl) (mean +/- SD). The highest mean absolute difference was observed in 1993 (pre: 31 +/- 33 mg/dl) and 1996 (post: 50 +/- 35 mg/dl). The highest mean relative deviation was observed in 1990 (pre: 16.4%) and 1996 (post: 20.6%). The highest percentage of readings that were within a 5% deviation limit were observed in 1998 (pre: 44.5%) and in 1997 (post: 36.7%). Based on blood glucose levels within +/-5 and +/-10% of laboratory values, the technical accuracy of meters was similar for 1989 and 1999 (P = 0.27 and 0.52, respectively). The percentage of pre values in zone A of Clarke's error grid analysis was >90% in 1989, 1997, 1998, and 1999. The analytical performance of glucose meters decreased between 1990 and 1996 but was restored between 1997 and 1999. Nevertheless, our data suggest that the technical accuracy of glucose meters has not significantly improved during the last decade. Complementary studies taking into account the preanalytical improvements of the recent meters, as well as their calibration method, appear necessary.

Mapping clay content for subsurface drainage in the Nile Delta

Drain envelope requirements for large-scale subsurface drainage projects are often estimated based on the surface area with clay percentages less than a given threshold value. Contouring clay percentage data can be done by computer using spatial interpolation methods. Cross-validation was used to evaluate the precision of estimation for clay percentage using five interpolation techniques (closest neighbor, moving average, weighted moving average, kriging, and cokriging) for a study area of 33,500 ha in the Nile Delta. The clay content was measured at 485 sites. Cokriging used the correlation between clay percentage and saturated hydraulic conductivity, which had been measured at 3488 sites. The lowest mean absolute difference (MAD) between observed and interpolated clay percentage values was obtained for cokriging (12.01). MAD values were very close to that of cokriging in the case of moving average (12.13), weighted moving average (12.23), and kriging (12.21), whereas for the closest neighbor method a MAD value of 16.44 was obtained. Surface areas under 30 and 40 percent clay were over-estimated for the closest neighbor method, and under-estimated for the moving average method. Surface areas and contour maps based on weighted moving average, kriging, and cokriging were very similar. Use of moving average resulted in a loss of precision due to the smoothing effect.

Spatial interpolation of soil salinity and sodicity for a saline soil in Southern Alberta

Soil salinity and sodicity data are usually represented as contour maps based on spatially interpolated data. This study was conducted to assess two commonly used spatial interpolation methods (moving average and weighted moving average) and three forms of two-dimensional (2D) and three-dimensional (3D) kriging. Soil was sampled from two depths on a 25-m grid at 40 sites in a saline and sodic soil of Southern Alberta. Samples were analyzed for electrical conductivity (ECe) and sodium absorption ratio (SARe) of saturation paste extracts. Methods were compared by cross-validation using the mean absolute difference (MAD) between estimated and observed values. Values of MAD resulting from 3D kriging were inferior to those obtained from moving averages. Moving average methods were very sensitive to the number of neighbors used in the estimation. For kriging methods, however, precision remained stable for more than about 10–12 neighbors. Three-dimensional ordinary kriging was the most precise and robust interpolator with a MAD value of 9.74 dS m−1 for ECe. For SARe the lowest MAD value of 10.55 was obtained with 3D ordinary co-kriging. Data representation produced by 3D kriging allowed easy identification of ECe spatial structure. Key words: Spatial interpolation, kriging, moving average, soil, salinity, sodicity