Search

PurposeWhole-exome (WES) and whole-genome sequencing (WGS) increase the diagnostic yield in autism spectrum disorder (ASD) compared to chromosomal microarray (CMA), but there have been no comprehensive cost analyses. The objective was to perform such an assessment of CMA, WES, and WGS and compare the incremental cost per additional positive finding in hypothetical testing scenarios.MethodsFive-year patient and program costs were estimated from an institutional perspective. WES and WGS estimates were based on HiSeq 2500 with an additional WGS estimate for HiSeq X platforms. Parameter uncertainty was assessed with probabilistic and deterministic sensitivity analysis.ResultsThe cost per ASD sample was CAD$1,655 (95% CI: 1,611; 1,699) for WES, CAD$2,851 (95% CI: 2,750; 2,956) for WGS on HiSeq X, and CAD$5,519 (95% CI: 5,244; 5,785) on HiSeq 2500, compared to CAD$744 (95% CI 714, 773) for CMA. The incremental cost was over CAD$25,000 per additional positive finding if CMA was replaced by newer technology.ConclusionWhile costs for WES and WGS remain high, future reductions in material and equipment costs, and increased understanding of newly discovered variants and variants of unknown significance will lead to improved value.

Aim: Thiopurine S-methyltransferase (TPMT) testing is used in patients receiving thiopurines to identify enzyme deficiencies and risk for adverse drug reactions. It is uncertain whether genotyping is superior to phenotyping. The objectives were to conduct a systematic review of TPMT-test performance studies. Materials & methods: Electronic and grey literature sources were searched for studies reporting test performance compared with a reference standard. Sixty-six eligible studies were appraised for quality. Results: Thirty phenotype–genotype and six phenotype–phenotype comparisons were of high quality. The calculated sensitivity and specificity for genotyping to identify a homozygous mutation ranged from 0.0–100.0% and from 97.8–100.0%, respectively. Conclusion: Clinical decision-makers require high-quality evidence of clinical validity and clinical utility of TPMT genotyping to ensure appropriate use in patients.

Background and Aims Screening and brief intervention (SBI) is a public health intervention that has been shown to be effective in reducing heavy alcohol consumption. The aim of this study is to estimate the cost-effectiveness of implementing universal alcohol SBI in primary care in Canada. Design We developed a microsimulation model of alcohol consumption and its effects on 18 alcohol-related causes of death. Setting The model simulates a Canadian population. Participants The model simulates individuals and their alcohol consumption on a continuous scale starting from age 17 years to death. Interventions The reference case assumes no SBI in Canada. The base case assumes screening was conducted using the Alcohol Use Disorders Identification Test (AUDIT) at a threshold score of 8. Additional analyses included evaluating SBI using the AUDIT at threshold scores between 4 and 8 or the Derived Alcohol Use Disorders Identification Test (AUDIT-C) at threshold scores between 3 and 7. Measurements The model estimates the direct health-care costs, life years gained and quality-adjusted life years (QALY) gained, which are then used to estimate the incremental cost-effectiveness ratio (ICER) of SBI versus no SBI. Findings SBI with AUDIT (at a threshold score of 8) had an ICER of $8729/QALY. Our results suggest that using AUDIT thresholds between 8 and 4, inclusive, would be cost-effective for the whole population, as well as for men and women individually. Our results suggest that the AUDIT-C would be cost-effective at thresholds of 7 to 3, inclusive, for men, women and the whole population. Conclusions In Canada, screening and brief intervention via Alcohol Use Disorders Identification Test (AUDIT) and Derived Alcohol Use Disorders Identification Test (AUDIT-C) to reduce heavy alcohol consumption appears to be cost-effective for men and women at Alcohol Use Disorders Identification Test (AUDIT) thresholds of 8 and lower and at Derived Alcohol Use Disorders Identification Test (AUDIT-C) thresholds of 7 and lower.

Rationale and Objectives To evaluate stratified random sampling (SRS) of screening mammograms by (1) Breast Imaging Reporting and Data System (BI-RADS) assessment categories, and (2) the presence of breast cancer in mammograms, for estimation of screening-mammography receiver operating characteristic (ROC) curves in retrospective observer studies. Materials and Methods We compared observer study case sets constructed by (1) random sampling (RS); (2) SRS with proportional allocation (SRS-P) with BI-RADS 1 and 2 noncancer cases accounting for 90.6% of all noncancer cases; (3) SRS with disproportional allocation (SRS-D) with BI-RADS 1 and 2 noncancer cases accounting for 10%–80%; and (4) SRS-D and multiple imputation (SRS-D + MI) with missing BI-RADS 1 and 2 noncancer cases imputed to recover the 90.6% proportion. Monte Carlo simulated case sets were drawn from a large case population modeled after published Digital Mammography Imaging Screening Trial data. We compared the bias, root-mean-square error, and coverage of 95% confidence intervals of area under the ROC curve (AUC) estimates from the sampling methods (200–2000 cases, of which 25% were cancer cases) versus from the large case population. Results AUC estimates were unbiased from RS, SRS-P, and SRS-D + MI, but biased from SRS-D. AUC estimates from SRS-P and SRS-D + MI had 10% smaller root-mean-square error than RS. Conclusions Both SRS-P and SRS-D + MI can be used to obtain unbiased and 10% more efficient estimate of screening-mammography ROC curves.

The purpose of this study was to investigate the effect of a noise injection method on the “overfitting” problem of artificial neural networks (ANNs) in two-class classification tasks. The authors compared ANNs trained with noise injection to ANNs trained with two other methods for avoiding overfitting: weight decay and early stopping. They also evaluated an automatic algorithm for selecting the magnitude of the noise injection. They performed simulation studies of an exclusive-or classification task with training datasets of 50, 100, and 200 cases (half normal and half abnormal) and an independent testing dataset of 2000 cases. They also compared the methods using a breast ultrasound dataset of 1126 cases. For simulated training datasets of 50 cases, the area under the receiver operating characteristic curve (AUC) was greater (by 0.03) when training with noise injection than when training without any regularization, and the improvement was greater than those from weight decay and early stopping (both of 0.02). For training datasets of 100 cases, noise injection and weight decay yielded similar increases in the AUC (0.02), whereas early stopping produced a smaller increase (0.01). For training datasets of 200 cases, the increases in the AUC were negligibly small for all methods (0.005). For the ultrasound dataset, noise injection had a greater average AUC than ANNs trained without regularization and a slightly greater average AUC than ANNs trained with weight decay. These results indicate that training ANNs with noise injection can reduce overfitting to a greater degree than early stopping and to a similar degree as weight decay.

We compare two methods of training artificial neural networks (ANNs) that potentially reduce the risk of the neural network overfitting the training data set. We refer to these methods as training with jitter. In one method of training with jitter, a new random noise vector is added to each training-data vector between successive iterations. In this work, we propose a different method of training with jitter, in which instead of adding different random noise vectors between iterations, a number of random vectors are used to expand the training data set prior to training. This artificially expanded data set is then used to train the artificial neural network in the conventional manner. These two methods are compared to the conventional method of training artificial neural networks. We find that although training with a single expanded training data set does increase the performance of the neural networks, overfitting can still occur after a large number of training iterations.

Health care systems require high quality economic evaluations to support wise allocation decisions to enhance child health within the population served. LMICs face the added jeopardy of having very limited resources (human and fiscal) to produce complex economic evaluations within highly constrained healthcare budgets that demand judicious expenditure. Furthermore, conducting economic evaluations in children is challenging methodologically and this is exacerbated in LMICs where healthcare funding and health priorities are influenced by external forces. World Health Organization initiatives such as generalized cost-effectiveness analysis have addressed many of the pertinent methodological issues. However, of 49 economic evaluations of preventive interventions for children published in 2012, only 17 were relevant to LMIC. Of 25 economic evaluations of active pediatric treatments, only four targeted an LMIC population. Greater collaboration between health researchers from LMICs and high-income countries promises, in future, to improve the quality of evidence available for decision-making.

To evaluate stratified random sampling (SRS) of screening mammograms by (1) Breast Imaging Reporting and Data System (BI-RADS) assessment categories, and (2) the presence of breast cancer in mammograms, for estimation of screening-mammography receiver operating characteristic (ROC) curves in retrospective observer studies.We compared observer study case sets constructed by (1) random sampling (RS); (2) SRS with proportional allocation (SRS-P) with BI-RADS 1 and 2 noncancer cases accounting for 90.6% of all noncancer cases; (3) SRS with disproportional allocation (SRS-D) with BI-RADS 1 and 2 noncancer cases accounting for 10%-80%; and (4) SRS-D and multiple imputation (SRS-D + MI) with missing BI-RADS 1 and 2 noncancer cases imputed to recover the 90.6% proportion. Monte Carlo simulated case sets were drawn from a large case population modeled after published Digital Mammography Imaging Screening Trial data. We compared the bias, root-mean-square error, and coverage of 95% confidence intervals of area under the ROC curve (AUC) estimates from the sampling methods (200-2000 cases, of which 25% were cancer cases) versus from the large case population.AUC estimates were unbiased from RS, SRS-P, and SRS-D + MI, but biased from SRS-D. AUC estimates from SRS-P and SRS-D + MI had 10% smaller root-mean-square error than RS.Both SRS-P and SRS-D + MI can be used to obtain unbiased and 10% more efficient estimate of screening-mammography ROC curves.

Rationale and Objectives We showed previously that maximum-likelihood (ML) and Bayesian (with a flat prior on a common parameterization of the model) estimates of “proper” binormal receiver operating characteristic (ROC) curves produce similar results. We propose a new prior that is flat over the area under the ROC curve (AUC) and investigate its effect on the Bayesian estimates. Materials and Methods In two simulation studies, we compared Bayesian estimation of the AUC with the two prior probability distributions against ML estimation in terms of root mean squared error (RMSE) and the coverage of 95% confidence (or credible) intervals (both abbreviated CIs). In the first study, we simulated categorical data that tend to be “well-behaved” and produce ROC curve estimates that most would consider reasonable. In the second study, we simulated coarsely discretized categorical data that often included so-called degenerate datasets that cause the ML estimate to be the perfect ROC curve. Results For the well-behaved datasets, all three AUC estimates were similar in terms of RMSE and 95% CI coverage. For the coarsely discretized datasets, the RMSE of ML was consistently greater than that of Bayesian estimation and the 95% CI coverage of ML estimation was consistently below nominal, whereas the 95% CI coverage of Bayesian estimation was consistently equal to, or greater than, nominal. Conclusion Bayesian estimation with a flat prior on the AUC can provide reasonable inference from datasets with coarsely categorized data that are prone to be degenerate and produce results similar to other estimation methods on well-behaved datasets.

We studied the effect of noise injection in overcoming the problem of overtraining in the training of artificial neural networks (ANNs) in comparison with other common approaches for overcoming this problem such as early stopping of the ANN training process and weight decay (which is similar to Bayesian artificial neural networks). We found from simulation studies and studies of a computer-aided diagnosis application that noise injection is effective in overcoming overtraining and is as effective as, or even more effective than, early stopping and weight decay.

Maximum likelihood estimation of receiver operating characteristic (ROC) curves using the "proper" binormal model can be interpreted in terms of Bayesian estimation as assuming a flat joint prior distribution on the c and d a parameters. However, this is equivalent to assuming a non-flat prior distribution for the area under the curve (AUC) that peaks at AUC = 1.0. We hypothesize that this implicit prior on AUC biases the maximum likelihood estimate (MLE) of AUC. We propose a Bayesian implementation of the "proper" binormal ROC curve-fitting model with a prior distribution that is marginally flat on AUC and conditionally flat over c . This specifies a non-flat joint prior for c and d a . We developed a Monte Carlo Markov chain (MCMC) algorithm to estimate the posterior distribution and the maximum a posteriori (MAP) estimate of AUC. We performed a simulation study using 500 draws of a small dataset (25 normal and 25 abnormal cases) with an underlying AUC value of 0.85. When the prior distribution was a flat joint prior on c and d a , the MLE and MAP estimates agreed, suggesting that the MCMC algorithm worked correctly. When the prior distribution was marginally flat on AUC, the MAP estimate of AUC appeared to be biased low. However, the MAP estimate of AUC for perfectly separable degenerate datasets did not appear to be biased. Further work is needed to validate the algorithm and refine the prior assumptions.