Bayesian bivariate meta-analysis of diagnostic test studies with interpretable priors

Diagnostic accuracy of salivary active matrix metalloproteinase (aMMP)-8 point-of-care test for detecting periodontitis in adults: A systematic review and meta-analysis

OBJECTIVE

To determine the accuracy of salivary active matrix metalloproteinase (aMMP)-8 point-of-care test (POCT) for detecting periodontitis in adults, through meta-analysis.

MATERIALS AND METHODS

Diagnostic studies evaluating the accuracy of salivary/oral rinse aMMP-8 POCT for detecting periodontitis in adults, when compared with clinical examination, were considered eligible. A comprehensive search was performed up to 31 August 2023 through five databases. Quality Assessment of Diagnostic Accuracy Studies 2 was utilized to evaluate the methodological quality of the included articles. Meta-analysis was performed using Bayesian bivariate hierarchical model and subgroup analysis.

RESULTS

From 368 screened studies, 6 studies (4 cross-sectional and 2 longitudinal studies) were included in the meta-analysis. Overall, the pooled sensitivity and specificity of salivary aMMP-8-POCT for detecting periodontitis were 0.63 (95% CI: 0.41-0.82) and 0.84 (95% CI: 0.65-0.95), respectively. Subgroup analyses revealed that the 95% CI for oral fluid types, predefined diagnostic thresholds and the POCT systems largely overlapped, indicating that the differences between them may not be significant.

CONCLUSION

Salivary aMMP-8 POCT shows fair accuracy for detecting periodontitis. The diagnostic accuracy cannot be significantly influenced by the types of oral fluids, predefined diagnostic thresholds or the specific POCT systems used. More research is needed to confirm the clinical utility and implementation of aMMP-8 POCT in the diagnosis of periodontitis.

A comprehensive meta-analysis and systematic review of breath analysis in detection of COVID-19 through Volatile organic compounds

BACKGROUND

The COVID-19 pandemic had profound global impacts on daily lives, economic stability, and healthcare systems. Diagnosis of COVID-19 infection via RT-PCR was crucial in reducing spread of disease and informing treatment management. While RT-PCR is a key diagnostic test, there is room for improvement in the development of diagnostic criteria. Identification of volatile organic compounds (VOCs) in exhaled breath provides a fast, reliable, and economically favorable alternative for disease detection.

METHODS

This meta-analysis analyzed the diagnostic performance of VOC-based breath analysis in detection of COVID-19 infection. A systematic review of twenty-nine papers using the grading criteria from Newcastle-Ottawa Scale (NOS) and PRISMA guidelines was conducted.

RESULTS

The cumulative results showed a sensitivity of 0.92 (95 % CI, 90 %-95 %) and a specificity of 0.90 (95 % CI 87 %-93 %). Subgroup analysis by variant demonstrated strong sensitivity to the original strain compared to the Omicron and Delta variant in detection of SARS-CoV-2 infection. An additional subgroup analysis of detection methods showed eNose technology had the highest sensitivity when compared to GC-MS, GC-IMS, and high sensitivity-MS.

CONCLUSION

Overall, these results support the use of breath analysis as a new detection method of COVID-19 infection.

Predicting EGFR Mutation Status in Non-Small Cell Lung Cancer Using Artificial Intelligence: A Systematic Review and Meta-Analysis

RATIONALE AND OBJECTIVES

Recent advancements in artificial intelligence (AI) render a substantial promise for epidermal growth factor receptor (EGFR) mutation status prediction in non-small cell lung cancer (NSCLC). We aimed to evaluate the performance and quality of AI algorithms that use radiomics features in predicting EGFR mutation status in patient with NSCLC.

MATERIALS AND METHODS

We searched PubMed (Medline), EMBASE, Web of Science, and IEEExplore for studies published up to February 28, 2022. Studies utilizing an AI algorithm (either conventional machine learning [cML] and deep learning [DL]) for predicting EGFR mutations in patients with NSLCL were included. We extracted binary diagnostic accuracy data and constructed a bivariate random-effects model to obtain pooled sensitivity, specificity, and 95% confidence interval. This study is registered with PROSPERO, CRD42021278738.

RESULTS

Our search identified 460 studies, of which 42 were included. Thirty-five studies were included in the meta-analysis. The AI algorithms exhibited an overall area under the curve (AUC) value of 0.789 and pooled sensitivity and specificity levels of 72.2% and 73.3%, respectively. The DL algorithms outperformed cML in terms of AUC (0.822 vs. 0.775) and sensitivity (80.1% vs. 71.1%), but had lower specificity (70.0% vs. 73.8%, p-value < 0.001) compared to cML. Subgroup analysis revealed that the use of positron-emission tomography/computed tomography, additional clinical information, deep feature extraction, and manual segmentation can improve diagnostic performance.

CONCLUSION

DL algorithms can serve as a novel method for increasing predictive accuracy and thus have considerable potential for use in predicting EGFR mutation status in patient with NSCLC. We also suggest that guidelines on using AI algorithms in medical image analysis should be developed with a focus on oncologic radiomics.

Muscle ultrasound to diagnose sarcopenia in chronic kidney disease: a systematic review and bayesian bivariate meta-analysis

OBJECTIVE

The aim of this systematic review was to assess the diagnostic test accuracy of muscle ultrasound for sarcopenia among chronic kidney disease (CKD) populations.

BACKGROUND

Sarcopenia has become a worldwide health issue, especially for CKD patients. Conventional techniques of muscle mass assessment often prove limited, thus prompts increasing interest in ultrasound suitability.

METHODS

We searched the Cochrane Library, PubMed and Embase for literature published up to June 2023. Ultrasound diagnosis of sarcopenia in CKD patients was included. Two independent investigators used the Quality Assessment Tool for Diagnosis Accuracy Studies (QUADAS-2) to assess the quality. We extracted valuable information from eligible studies. Using a Bayesian bivariate model, we pooled sensitivity and specificity values and summary receiver operating characteristic (SROC) curves.

RESULTS

Five articles, involving 428 participants at various stages of CKD were included. Three studies diagnosed by the cross-sectional area (CSA) of the rectus femoris, while two others by muscle thickness (MT) and shear wave elastography (SWE) from the same muscle, separately. Overall, CSA or SWE had a pooled sensitivity of 0.95 (95% CrI, 0.80, 1.00), and the specificity was 0.73 (95% CrI, 0.55, 0.88) for diagnosing sarcopenia in CKD patients.

CONCLUSIONS

Ultrasound measurements of CSA and SWE were more sensitive for diagnosing sarcopenia in the CKD population than in the general population. Ultrasound assessment from a single peripheral skeletal muscle site may serve as a rapid screening tool for identifying sarcopenic individuals within the CKD population, if a specific cut-off value could be determined.

Meta-analysis of microbial source tracking for the identification of fecal contamination in aquatic environments based on data-mining

Microbial source tracking (MST) technology represents an innovative approach employed to trace fecal contamination in environmental water systems. The performance of primers may be affected by amplification techniques, target primer categories, and regional differences. To investigate the influence of these factors on primer recognition performance, a meta-analysis was conducted on the application of MST in water environments using three databases: Web of Science, Scopus, and PubMed (n = 2291). After data screening, 46 studies were included in the final analysis. The investigation encompassed Polymerase Chain Reaction (PCR)/quantitative PCR (qPCR) methodologies, dye-based (SYBR)/probe-based (TaqMan) techniques, and geographical differences of a human host-specific (HF183) primer and other 21 additional primers. The results indicated that the primers analyzed were capable of differentiating host specificity to a certain degree. Nonetheless, by comparing sensitivity and specificity outcomes, it was observed that virus-based primers exhibited superior specificity and recognition capacity, as well as a stronger correlation with human pathogenicity in water environments compared to bacteria-based primers. This finding highlights an important direction for future advancements. Moreover, within the same category, qPCR did not demonstrate significant benefits over conventional PCR amplification methods. In comparing dye-based and probe-based techniques, it was revealed that the probe-based method's advantage lay primarily in specificity, which may be associated with the increased propensity of dye-based methods to produce false positives. Furthermore, the heterogeneity of the HF183 primer was not detected in China, Canada, and Singapore respectively, indicating a low likelihood of regional differences. The variation among the 21 other primers may be attributable to regional differences, sample sources, detection techniques, or alternative factors. Finally, we identified that economic factors, climatic conditions, and geographical distribution significantly influence primer performance.

Diagnostic performance of ultrasound risk stratification systems on thyroid nodules cytologically classified as indeterminate: a systematic review and meta-analysis

PURPOSE

Ultrasound (US) risk stratification systems (RSSs) are increasingly being utilized for the optimal management of thyroid nodules, including those with indeterminate cytology. The goal of this study was to evaluate the category-based diagnostic performance of US RSSs in identifying malignancy in indeterminate nodules.

METHODS

This systematic review and meta-analysis was registered on PROSPERO (CRD42021266195). PubMed, EMBASE, and Web of Science were searched through December 1, 2022. Original articles reporting data on the performance of US RSSs for indeterminate nodules were included. The numbers of nodules classified as true negative, true positive, false negative, and false positive were extracted.

RESULTS

Thirty-three studies evaluating 7,225 indeterminate thyroid nodules were included. The diagnostic accuracy was quantitatively synthesized using a Bayesian bivariate model based on the integrated nested Laplace approximation in R. For the intermediate- to high-risk category, the sensitivity levels of the American College of Radiology, the American Thyroid Association, the European Thyroid Association, the Korean Thyroid Association/Korean Society of Thyroid Radiology, and Kwak et al. were found to be 0.80, 0.72, 0.76, 0.96, and 0.97, respectively. The corresponding specificity measurements were 0.36, 0.50, 0.49, 0.28, and 0.17. Furthermore, for the high-risk category, the sensitivity values were 0.40, 0.46, 0.55, 0.47, and 0.10, while the specificity levels were 0.91, 0.90, 0.71, 0.91, and 0.99, respectively.

CONCLUSION

The overall diagnostic performance of the US RSSs was moderate in the differentiation of indeterminate nodules.

Diagnostic performance of shear wave elastography in thyroid nodules with indeterminate cytology: A systematic review and meta-analysis

Purpose

Thyroid nodules classified as indeterminate in previous fine-needle aspiration cytology often necessitate additional evaluation to determine their histology, while shear wave elastography (SWE) offers an alternative option in this regard. The objective of this study was to assess the diagnostic effectiveness of SWE in evaluating indeterminate nodules.

Methods

The PubMed, EMBASE, and Web of Science databases were searched from 1st January 1970 to 1st March 2023. The studies were reviewed and the data was extracted by two separate reviewers. A Bayesian bivariate model was utilized to quantitatively synthesize the diagnostic accuracy and yield of the studies in R.

Results

A total of seven studies, involving indeterminate thyroid nodules undergoing SWE were included, and the overall malignancy rate was 34.1% (307/900). The summarized estimates of sensitivity and specificity were 0.792 (95% credible interval [CI], 0.727-0.850) and 0.845 (95% CI, 0.797-0.887), respectively. The summarized estimate for the diagnostic odds ratio (DOR) was 17.8 (95% CI, 14.0-22.6). Summarized receiver operating characteristic (SROC) plots indicated a trade-off between sensitivity and specificity, and the estimate of AUC was 0.866 (95% CI, 0.834-0.895). The summary estimates for positive and negative likelihood ratios were 4.67 (95% CI, 3.98-5.85) and 0.26 (95% CI, 0.23-0.28), respectively.

Conclusions

The overall accuracy of SWE remains satisfactory in indeterminate thyroid nodules. However, it should be noted that the available data are still extremely limited, and more studies or guidelines are required to provide further insights.

Copula modelling with penalized complexity priors: the bivariate case

We explore the use of penalized complexity (PC) priors for assessing the dependence structure in a multivariate distribution F, with a particular emphasis on the bivariate case. We use the copula representation of F and derive the PC prior for the parameter governing the copula. We show that any $$\alpha $$ α -divergence between a multivariate distribution and its counterpart with independent components does not depend on the marginal distribution of the components. This implies that the PC prior for the parameters of the copula can be elicited independently of the specific form of the marginal distributions. This represents a useful simplification in the model building step and may offer a new perspective in the field of objective Bayesian methodology. We also consider strategies for minimizing the role of subjective inputs in the prior elicitation step. Finally, we explore the use of PC priors in Bayesian hypothesis testing. Our prior is compared with competing default priors both for estimation purposes and testing.

Flat Inferior Vena Cava on Computed Tomography for Predicting Shock and Mortality in Trauma: A Meta-Analysis

Hypovolemia may be underestimated due to compensatory mechanisms. In this systematic review and meta-analysis, we investigated the diagnostic accuracy of a flat inferior vena cava (IVC) on computed tomography (CT) for predicting the development of shock and mortality in trauma patients. Relevant studies were obtained by searching PubMed, EMBASE, and Cochrane databases (articles up to 16 September 2022). The number of 2-by-2 contingency tables for the index test were collected. We adopted the Bayesian bivariate random-effects meta-analysis model. Twelve studies comprising a total of 1706 patients were included. The flat IVC on CT showed 0.46 pooled sensitivity (95% credible interval [CrI] 0.32-0.63), 0.87 pooled specificity (95% CrI 0.78-0.94), and 0.78 pooled AUC (95% CrI 0.58-0.93) for the development of shock. The flat IVC for mortality showed 0.48 pooled sensitivity (95% CrI 0.21-0.94), 0.70 pooled specificity (95% CrI 0.47-0.88), and 0.60 pooled AUC (95% CrI 0.26-0.89). Regarding the development of shock, flat IVC provided acceptable accuracy with high specificity. Regarding in-hospital mortality, the flat IVC showed poor accuracy. However, these results should be interpreted with caution due to the high risk of bias and substantial heterogeneity in some included studies.

Differentiation between glioma recurrence and treatment effects using amide proton transfer imaging: A mini-Bayesian bivariate meta-analysis

Background

Amide proton transfer (APT) imaging as an emerging MRI approach has been used for distinguishing tumor recurrence (TR) and treatment effects (TEs) in glioma patients, but the initial results from recent studies are different.

Aim

The aim of this study is to systematically review and quantify the diagnostic performance of APT in assessing treatment response in patients with post-treatment gliomas.

Methods

A systematic search in PubMed, EMBASE, and the Web of Science was performed to retrieve related original studies. For the single and added value of APT imaging in distinguishing TR from TEs, we calculated pooled sensitivity and specificity by using Bayesian bivariate meta-analyses.

Results

Six studies were included, five of which reported on single APT imaging parameters and four of which reported on multiparametric MRI combined with APT imaging parameters. For single APT imaging parameters, the pooled sensitivity and specificity were 0.85 (95% CI: 0.75–0.92) and 0.88 (95% CI: 0.74–0.97). For multiparametric MRI including APT, the pooled sensitivity and specificity were 0.92 (95% CI: 0.85–0.97) and 0.83 (95% CI: 0.55–0.97), respectively. In addition, in the three studies reported on both single and added value of APT imaging parameters, the combined imaging parameters further improved diagnostic performance, yielding pooled sensitivity and specificity of 0.91 (95% CI: 0.80–0.97) and 0.92 (95% CI: 0.79–0.98), respectively, but the pooled sensitivity was 0.81 (95% CI: 0.65-0.93) and specificity was 0.82 (95% CI: 0.61–0.94) for single APT imaging parameters.

Conclusion

APT imaging showed high diagnostic performance in assessing treatment response in patients with post-treatment gliomas, and the addition of APT imaging to other advanced MRI techniques can improve the diagnostic accuracy for distinguishing TR from TE.

Diagnostic performance of C-TIRADS in malignancy risk stratification of thyroid nodules: A systematic review and meta-analysis

BACKGROUND

Chinese thyroid imaging reports and data systems (C-TIRADS) is a novel malignancy risk stratification used for thyroid nodule diagnosis and guiding thyroid fine needle aspiration (FNA). In this review, we aim to assess the performance of C-TIRADS in malignancy risk stratification of thyroid nodules.

METHODS

PubMed, Medline, Web of Science, Embase, CNKI, and Wanfang databases were searched until 1 April 2022. Original articles reporting data about C-TIRADS and setting FNA or histology as reference standards were included. C-TIRADS 4A, 4B, and 4C were set as thresholds, respectively, to obtain pooled sensitivity, specificity, positive likelihood ratio (LR+), negative likelihood ratio (LR-), diagnostic odds ratio (DOR), and the area under the curve (AUC). Integrated nested Laplace approximation was used for Bayesian bivariate meta-analysis of diagnostic tests.

RESULTS

Sixteen studies were included, evaluating 11,506 thyroid nodules. The rate of malignancy in each risk classification is comparable with that in C-TIRADS. C-TIRADS 4B appeared to have better diagnostic performance than C-TIRADS 4A and 4C. The pooled sensitivity, specificity, LR+, LR-, and DOR of C-TI-RADS 4B were 0.94 (95% CI: 0.89-0.97), 0.70 (95% CI: 0.60-0.79), 3.20 (95% CI: 2.28-4.39), 0.09 (95% CI: 0.05-0.15), and 33.71 (95% CI: 25.51-42.40), respectively. The area under the summary ROC curve was 0.94 (95% CI: 0.90-0.96).

CONCLUSION

C-TIRADS performed well in malignancy risk stratification of thyroid nodules. C-TIRADS 4B showed strong evidence of detecting malignancy.

Diagnostic value of supersonic shear impulse elastography for malignant cervical lymph nodes: a Bayesian analysis

Purpose

This study aimed to assess the diagnostic performance of supersonic impulse (SSI) elastography in differentiating malignant and benign cervical lymph nodes.

Methods

The Medline, Embase, and Cochrane Central databases were searched until December 1, 2020. Two different reviewers checked the studies and extracted the data. The diagnostic yields were quantitatively synthesized using a Bayesian bivariate model with an integrated nested Laplace approximation in R.

Results

In total, 590 patients with 892 cervical lymph nodes who underwent SSI elastography were included. The total prevalence of malignancy was 33.7% (301/892), and the four elastic modulus values (mean, maximum, minimum, and standard deviation) were significantly different between malignant and benign lymph nodes. For the mean elastic modulus, the summary estimates for sensitivity and specificity were 0.720 (95% credible interval [CrI], 0.592 to 0.824) and 0.877 (95% CrI, 0.727 to 0.969), respectively. The estimated area under the curve (AUC) was 0.845 (95% CrI, 0.672 to 0.914). For the maximum elastic modulus, the sensitivity and specificity were estimated to be 0.809 (95% CrI, 0.698 to 0.899) and 0.816 (95% CrI, 0.643 to 0.924), respectively. The estimated AUC was 0.834 (95% CrI, 0.579 to 0.938). The minimum and standard deviation of the elastic modulus and the outcomes of the positive and negative likelihood ratio, diagnostic odds ratio, and risk difference were also calculated.

Conclusion

SSI elastography is an acceptable imaging technique for diagnosing malignant cervical lymph nodes, and it can play a complementary role today. Both maximum and mean elastic modulus values should be taken into consideration to make a clinical judgment.

A Bayesian approach for diagnostic accuracy of malignant peripheral nerve sheath tumors: a systematic review and meta-analysis

Background

Malignant peripheral nerve sheath tumors (MPNST) carry a dismal prognosis and require early detection and complete resection. However, MPNSTs are prone to sampling errors and biopsies or resections are cumbersome and possibly damaging in benign peripheral nerve sheath tumor (BPNST). This study aimed to systematically review and quantify the diagnostic accuracy of noninvasive tests for distinguishing MPNST from BPNST.

Methods

Studies on accuracy of MRI, FDG-PET (fluorodeoxyglucose positron emission tomography), and liquid biopsies were identified in PubMed and Embase from 2000 to 2019. Pooled accuracies were calculated using Bayesian bivariate meta-analyses. Individual level-patient data were analyzed for ideal maximum standardized uptake value (SUV_max) threshold on FDG-PET.

Results

Forty-three studies were selected for qualitative synthesis including data on 1875 patients and 2939 lesions. Thirty-five studies were included for meta-analyses. For MRI, the absence of target sign showed highest sensitivity (0.99, 95% CI: 0.94-1.00); ill-defined margins (0.94, 95% CI: 0.88-0.98); and perilesional edema (0.95, 95% CI: 0.83-1.00) showed highest specificity. For FDG-PET, SUV_max and tumor-to-liver ratio show similar accuracy; sensitivity 0.94, 95% CI: 0.91-0.97 and 0.93, 95% CI: 0.87-0.97, respectively, specificity 0.81, 95% CI: 0.76-0.87 and 0.79, 95% CI: 0.70-0.86, respectively. SUV_max ≥3.5 yielded the best accuracy with a sensitivity of 0.99 (95% CI: 0.93-1.00) and specificity of 0.75 (95% CI: 0.56-0.90).

Conclusions

Biopsies may be omitted in the presence of a target sign and the absence of ill-defined margins or perilesional edema. Because of diverse radiological characteristics of MPNST, biopsies may still commonly be required. In neurofibromatosis type 1, FDG-PET scans may further reduce biopsies. Ideal SUV_max threshold is ≥3.5.

Robust modeling of additive and nonadditive variation with intuitive inclusion of expert knowledge

We propose a novel Bayesian approach that robustifies genomic modeling by leveraging expert knowledge (EK) through prior distributions. The central component is the hierarchical decomposition of phenotypic variation into additive and nonadditive genetic variation, which leads to an intuitive model parameterization that can be visualized as a tree. The edges of the tree represent ratios of variances, for example broad-sense heritability, which are quantities for which EK is natural to exist. Penalized complexity priors are defined for all edges of the tree in a bottom-up procedure that respects the model structure and incorporates EK through all levels. We investigate models with different sources of variation and compare the performance of different priors implementing varying amounts of EK in the context of plant breeding. A simulation study shows that the proposed priors implementing EK improve the robustness of genomic modeling and the selection of the genetically best individuals in a breeding program. We observe this improvement in both variety selection on genetic values and parent selection on additive values; the variety selection benefited the most. In a real case study, EK increases phenotype prediction accuracy for cases in which the standard maximum likelihood approach did not find optimal estimates for the variance components. Finally, we discuss the importance of EK priors for genomic modeling and breeding, and point to future research areas of easy-to-use and parsimonious priors in genomic modeling.

Robust modeling of additive and nonadditive variation with intuitive inclusion of expert knowledge

We propose a novel Bayesian approach that robustifies genomic modeling by leveraging expert knowledge (EK) through prior distributions. The central component is the hierarchical decomposition of phenotypic variation into additive and nonadditive genetic variation, which leads to an intuitive model parameterization that can be visualized as a tree. The edges of the tree represent ratios of variances, for example broad-sense heritability, which are quantities for which EK is natural to exist. Penalized complexity priors are defined for all edges of the tree in a bottom-up procedure that respects the model structure and incorporates EK through all levels. We investigate models with different sources of variation and compare the performance of different priors implementing varying amounts of EK in the context of plant breeding. A simulation study shows that the proposed priors implementing EK improve the robustness of genomic modeling and the selection of the genetically best individuals in a breeding program. We observe this improvement in both variety selection on genetic values and parent selection on additive values; the variety selection benefited the most. In a real case study, EK increases phenotype prediction accuracy for cases in which the standard maximum likelihood approach did not find optimal estimates for the variance components. Finally, we discuss the importance of EK priors for genomic modeling and breeding, and point to future research areas of easy-to-use and parsimonious priors in genomic modeling.

The value of kidney injury molecule 1 in predicting acute kidney injury in adult patients: a systematic review and Bayesian meta-analysis

Introduction

The aim of the study was to systematically review relevant studies to evaluate the diagnostic value of urinary kidney injury molecule 1 (uKIM-1) for acute kidney injury (AKI) in adults.

Method

We searched PubMed and Embase for literature published up to November 1st, 2019 and used the Quality Assessment Tool for Diagnosis Accuracy Studies (QUADAS-2) to assess the quality. Then, we extracted useful information from each eligible study and pooled sensitivity, specificity, and area under the curve (AUC) values.

Results

A total of 14 studies with 3300 patients were included. The estimated sensitivity of urinary KIM-1 (uKIM-1) in the diagnosis of AKI was 0.74 (95% CrI 0.62–0.84), and the specificity was 0.84 (95% CrI, 0.76–0.90). The pooled diagnostic odds ratio (DOR) was 15.22 (95% CrI, 6.74–42.20), the RD was 0.55 (95% CrI 0.43–0.70), and the AUC of uKIM-1 in diagnosing AKI was 0.62 (95% CrI 0.41–0.76). The results of the subgroup analysis showed the influence of different factors.

Conclusion

Urinary KIM-1 is a good predictor for AKI in adult patients with relatively high sensitivity and specificity. However, further research and clinical trials are still needed to confirm whether and how uKIM-1 can be commonly used in clinical diagnosis.

Diagnostic Accuracy of Four Commercial Triplex Immunochromatographic Tests for Rapid Detection of Rotavirus, Adenovirus, and Norovirus in Human Stool Samples

Noroviruses (NoV), rotaviruses (RVA), and adenoviruses (AdV) are the main viral agents responsible for acute gastroenteritis (AGE) in humans. We aimed to determine the diagnostic accuracy of four commercial immunochromatographic tests (ICTs) intended for the rapid and simultaneous detection of these three pathogens. Diagnostic accuracy of bioNexia Noro/Rota-Adeno (bioMérieux), Immunoquick NoRotAdeno (Biosynex), Rota+Adeno+Noro combo card (CerTest Biotec), and Rida Quick Rota/Adeno/Noro Combi (R-Biopharm) ICTs was assessed retrospectively using a collection of 160 stool specimens (including 43 RVA-, 47 AdV-, and 42 NoV-positive samples) from French patients with AGE and using molecular methods as the reference standard. For RVA, the four ICTs demonstrated similar high sensitivity (93%) and excellent specificity (97.4 to 100%). For AdV, the four ICTs demonstrated similar poor sensitivity (54.3 to 58.7%) but excellent specificity (95.5 to 100%). They performed the best in AdV-F species (sensitivity, 80.8 to 84.6%) and worst in AdV non-F species (sensitivity, 22.2 to 27.8%). For NoV, the Rida Quick Rota/Adeno/Noro combi ICT exhibited high sensitivity (87.5%), but the sensitivity of the three others was poor (42.5 to 47.5%). The four ICTs exhibited high specificity (96.6 to 99.1%). Diagnostic accuracy was genogroup dependent. When we tested genogroup I NoV, the Rida Quick Rota/Adeno/Noro Combi ICT presented high sensitivity (90%), while the three other ICTs presented poor sensitivity (10 to 30%); when we tested genogroup II NoV, sensitivity was similar for the four ICTs (65 to 85%). In conclusion, the four ICTs are suitable first-line tests for the rapid diagnosis of RVA infections. The four ICTs are not suitable for the routine diagnosis of AdV infections but could provide a rapid response in case of positivity, notably in the context of AGE. Only the Rida Quick Rota/Adeno/Noro Combi ICT is suitable for the rapid detection of NoV, while the sensitivity for the detection of genogroup I NoV needs to be improved for the 3 other ICTs before being implemented in the routine diagnosis of NoV.

Diagnostic reliability of elastography in thyroid nodules reported as indeterminate at prior fine-needle aspiration cytology (FNAC): a systematic review and Bayesian meta-analysis

OBJECTIVES

To assess the diagnostic yields of elastography in thyroid nodules reported as indeterminate in FNAC according to guidelines.

METHODS

Databases of Medline, Embase, and Cochrane Central were searched till 31 October 2019. Two different reviewers check the studies and extracted the data. The diagnostic accuracy and yield were quantitatively synthesized using Bayesian bivariate model in R.

RESULTS

Twenty studies with 1734 indeterminate thyroid nodules undergoing elastography were included. The summary estimates of sensitivity and specificity were 0.766 (95% credible interval (CrI), 0.686-0.835) and 0.867 (95% CrI, 0.780-0.931), respectively. The summary estimate for diagnostic odds ratio (DOR) was 25.9 (95% CrI, 12.8-46.2). Summary receiver operating characteristic plots for elastography showed a right-diagonal curvilinear relationship, suggesting a trade-off between sensitivity and specificity, and the estimate of area under curve (AUC) was 0.743. The summary estimates for positive and negative likelihood ratios were 6.6 (95% CrI, 4.2-11.3) and 0.27 (95% CrI, 0.21-0.36), respectively.

CONCLUSIONS

Elastography had fair diagnostic yields in indeterminate thyroid nodules. Shear wave elastography and strain ratio elastography could be more efficient in diagnosis and should evolve in the next years while combing elastography with ultrasound would contribute more to sensitivity and specificity currently.

KEY POINTS

• Elastography has fair diagnostic yields in indeterminate thyroid nodules. • Shear wave elastography and strain ratio elastography are more efficient than real-time elastography. • Combining elastography and other ultrasound techniques improves evaluation of indeterminate thyroid nodules.

A decision-theoretic approach for model interpretability in Bayesian framework

A salient approach to interpretable machine learning is to restrict modeling to simple models. In the Bayesian framework, this can be pursued by restricting the model structure and prior to favor interpretable models. Fundamentally, however, interpretability is about users’ preferences, not the data generation mechanism; it is more natural to formulate interpretability as a utility function. In this work, we propose an interpretability utility, which explicates the trade-off between explanation fidelity and interpretability in the Bayesian framework. The method consists of two steps. First, a reference model, possibly a black-box Bayesian predictive model which does not compromise accuracy, is fitted to the training data. Second, a proxy model from an interpretable model family that best mimics the predictive behaviour of the reference model is found by optimizing the interpretability utility function. The approach is model agnostic—neither the interpretable model nor the reference model are restricted to a certain class of models—and the optimization problem can be solved using standard tools. Through experiments on real-word data sets, using decision trees as interpretable models and Bayesian additive regression models as reference models, we show that for the same level of interpretability, our approach generates more accurate models than the alternative of restricting the prior. We also propose a systematic way to measure stability of interpretabile models constructed by different interpretability approaches and show that our proposed approach generates more stable models.