How to prevent overdiagnosis.

Overdiagnosis is the diagnosis of an abnormality that is not associated with a substantial health hazard and that patients have no benefit to be aware of. It is neither a misdiagnosis (diagnostic error), nor a false positive result (positive test in the absence of a real abnormality). It mainly results from screening, use of increasingly sensitive diagnostic tests, incidental findings on routine examinations, and widening diagnostic criteria to define a condition requiring an intervention. The blurring boundaries between risk and disease, physicians' fear of missing a diagnosis and patients' need for reassurance are further causes of overdiagnosis. Overdiagnosis often implies procedures to confirm or exclude the presence of the condition and is by definition associated with useless treatments and interventions, generating harm and costs without any benefit. Overdiagnosis also diverts healthcare professionals from caring about other health issues. Preventing overdiagnosis requires increasing awareness of healthcare professionals and patients about its occurrence, the avoidance of unnecessary and untargeted diagnostic tests, and the avoidance of screening without demonstrated benefits. Furthermore, accounting systematically for the harms and benefits of screening and diagnostic tests and determining risk factor thresholds based on the expected absolute risk reduction would also help prevent overdiagnosis.

Computed tomography (CT) virtual autopsy and classical autopsy discrepancies: radiologist's error or a demonstration of post-mortem multi-detector computed tomography (MDCT) limitation?

Modern imaging technologies, such as computed tomography (CT) techniques, represent a great challenge in forensic pathology. The field of forensics has experienced a rapid increase in the use of these new techniques to support investigations on critical cases, as indicated by the implementation of CT scanning by different forensic institutions worldwide. Advances in CT imaging techniques over the past few decades have finally led some authors to propose that virtual autopsy, a radiological method applied to post-mortem analysis, is a reliable alternative to traditional autopsy, at least in certain cases. The authors investigate the occurrence and the causes of errors and mistakes in diagnostic imaging applied to virtual autopsy. A case of suicide by a gunshot wound was submitted to full-body CT scanning before autopsy. We compared the first examination of sectional images with the autopsy findings and found a preliminary misdiagnosis in detecting a peritoneal lesion by gunshot wound that was due to radiologist's error. Then we discuss a new emerging issue related to the risk of diagnostic failure in virtual autopsy due to radiologist's error that is similar to what occurs in clinical radiology practice.

Diagnostic performance of high-sensitive troponin T in patients with renal insufficiency

In the present study, we wanted to (1) evaluate whether high-sensitive troponin T levels correlate with the grade of renal insufficiency and (2) test the accuracy of high-sensitive troponin T determination in patients with renal insufficiency for diagnosis of acute myocardial infarction (AMI). In this cross-sectional analysis, all patients who received serial measurements of high-sensitive troponin T from August 1, 2010, to October 31, 2012, at the Department of Emergency Medicine were included. We analyzed data on baseline characteristics, reason for referral, medication, cardiovascular risk factors, and outcome in terms of presence of AMI along with laboratory data (high-sensitive troponin T, creatinine). A total of 1,514 patients (67% male, aged 65 ± 16 years) were included, of which 382 patients (25%) had moderate to severe renal insufficiency and significantly higher levels of high-sensitive troponin T on admission (0.028 vs 0.009, p <0.0001). In patients without AMI, high-sensitive troponin T correlated inversely with the estimated glomerular filtration rate (R = -0.12, p <0.0001). Overall, sensitivity of an elevated high-sensitive troponin for diagnosis of AMI was 0.64 (0.56 to 0.71) and the specificity was 0.48 (0.45 to 0.51). The area under the curve of the receiver operating characteristic for all patients was 0.613 (standard error [SE] 0.023), whereas it was 0.741 (SE 0.029) for patients with a Modification of Diet in Renal Disease estimated glomerular filtration rate >60 ml/min presenting with acute chest pain or dyspnea and 0.535 (SE 0.056) for patients with moderate to severe renal insufficiency presenting with acute chest pain or dyspnea. In conclusion, the diagnostic accuracy for presence of AMI of a baseline measurement of high-sensitive troponin in patients with renal insufficiency was poor and resembles tossing a coin.

Estimating Diagnostic Test Accuracies for Brachyspira hyodysenteriae Accounting for the Complexities of Population Structure in Food Animals

For swine dysentery, which is caused by Brachyspira hyodysenteriae infection and is an economically important disease in intensive pig production systems worldwide, a perfect or error-free diagnostic test ("gold standard") is not available. In the absence of a gold standard, Bayesian latent class modelling is a well-established methodology for robust diagnostic test evaluation. In contrast to risk factor studies in food animals, where adjustment for within group correlations is both usual and required for good statistical practice, diagnostic test evaluation studies rarely take such clustering aspects into account, which can result in misleading results. The aim of the present study was to estimate test accuracies of a PCR originally designed for use as a confirmatory test, displaying a high diagnostic specificity, and cultural examination for B. hyodysenteriae. This estimation was conducted based on results of 239 samples from 103 herds originating from routine diagnostic sampling. Using Bayesian latent class modelling comprising of a hierarchical beta-binomial approach (which allowed prevalence across individual herds to vary as herd level random effect), robust estimates for the sensitivities of PCR and culture, as well as for the specificity of PCR, were obtained. The estimated diagnostic sensitivity of PCR (95% CI) and culture were 73.2% (62.3; 82.9) and 88.6% (74.9; 99.3), respectively. The estimated specificity of the PCR was 96.2% (90.9; 99.8). For test evaluation studies, a Bayesian latent class approach is well suited for addressing the considerable complexities of population structure in food animals.

Functional error modeling for uncertainty quantification in hydrogeology

Approximate models (proxies) can be employed to reduce the computational costs of estimating uncertainty. The price to pay is that the approximations introduced by the proxy model can lead to a biased estimation. To avoid this problem and ensure a reliable uncertainty quantification, we propose to combine functional data analysis and machine learning to build error models that allow us to obtain an accurate prediction of the exact response without solving the exact model for all realizations. We build the relationship between proxy and exact model on a learning set of geostatistical realizations for which both exact and approximate solvers are run. Functional principal components analysis (FPCA) is used to investigate the variability in the two sets of curves and reduce the dimensionality of the problem while maximizing the retained information. Once obtained, the error model can be used to predict the exact response of any realization on the basis of the sole proxy response. This methodology is purpose-oriented as the error model is constructed directly for the quantity of interest, rather than for the state of the system. Also, the dimensionality reduction performed by FPCA allows a diagnostic of the quality of the error model to assess the informativeness of the learning set and the fidelity of the proxy to the exact model. The possibility of obtaining a prediction of the exact response for any newly generated realization suggests that the methodology can be effectively used beyond the context of uncertainty quantification, in particular for Bayesian inference and optimization.