Tumor architecture exerts no bias on nuclear grading in breast cancer diagnosis

We recently reported that nuclear grading in prostate cancer is subject to a strong confirmation bias induced by the tumor architecture. We now wondered whether a similar bias governs nuclear grading in breast carcinoma. An unannounced test was performed at a pathology conference. Pathologists were asked to grade nuclei in a PowerPoint presentation. Circular high power fields of 27 invasive ductal carcinomas were shown, superimposed over low power background images of either tubule-rich or tubule-poor carcinomas. We found (a) that diagnostic reproducibility of nuclear grades was poor to moderate (weighed kappa values between 0.07 and 0.54, 27 cases, 44 graders), but (b) that nuclear grades were not affected by the tumor architecture. We speculate that the categorized grading in breast cancer, separating tubule formation, nuclear pleomorphism, and mitotic figure counts in a combined three tier score, prevents the bias that architecture exerts on nuclear grades in less well-controlled situations.

What was I thinking? Eye-tracking experiments underscore the bias that architecture exerts on nuclear grading in prostate cancer

We previously reported that nuclear grade assignment of prostate carcinomas is subject to a cognitive bias induced by the tumor architecture. Here, we asked whether this bias is mediated by the non-conscious selection of nuclei that "match the expectation" induced by the inadvertent glance at the tumor architecture. 20 pathologists were asked to grade nuclei in high power fields of 20 prostate carcinomas displayed on a computer screen. Unknown to the pathologists, each carcinoma was shown twice, once before a background of a low grade, tubule-rich carcinoma and once before the background of a high grade, solid carcinoma. Eye tracking allowed to identify which nuclei the pathologists fixated during the 8 second projection period. For all 20 pathologists, nuclear grade assignment was significantly biased by tumor architecture. Pathologists tended to fixate on bigger, darker, and more irregular nuclei when those were projected before kigh grade, solid carcinomas than before low grade, tubule-rich carcinomas (and vice versa). However, the morphometric differences of the selected nuclei accounted for only 11% of the architecture-induced bias, suggesting that it can only to a small part be explained by the unconscious fixation on nuclei that "match the expectation". In conclusion, selection of « matching nuclei » represents an unconscious effort to vindicate the gravitation of nuclear grades towards the tumor architecture.

Is the allegiance effect an epiphenomenon of true efficacy differences between treatments? a meta-analysis

Many meta-analyses of comparative outcome studies found a substantial association of researcher allegiance (RA) and relative treatment effects. Therefore, RA is regarded as a biasing factor in comparative outcome research (RA bias hypothesis). However, the RA bias hypothesis has been criticized as causality might be reversed. That is, RA might be a reflection of true efficacy differences between treatments (true efficacy hypothesis). Consequently, the RA-outcome association would not be indicative of bias but an epiphenomenon of true efficacy differences. This meta-analysis tested the validity of the true efficacy hypothesis. This was done by controlling the RA-outcome association for true efficacy differences by restricting analysis to direct comparisons of treatments with equivalent efficacy. We included direct comparisons of different versions of trauma-focused therapy (TFT) in the treatment of posttraumatic stress disorder (PTSD). RA was measured from the research reports. Relative effect sizes for symptoms of PTSD were calculated. Random effects meta-regression was conducted. Twenty-nine comparisons of TFTs from 20 studies were identified. Initial heterogeneity among relative effect sizes was low. RA was a significant predictor of outcome and explained 12% of the variance in outcomes. The true efficacy hypothesis predicted the RA-outcome association to be zero; however, a substantial association was found. Thus, this study does not support the true efficacy hypothesis. Given findings from psychotherapy research and other fields that support a biasing influence of researcher preferences, RA should be regarded as a causal factor and conceptualized as a threat to the validity of conclusions from comparative outcome studies.

Diagnostic performance of line-immunoassay based algorithms for incident HIV-1 infection

Background Serologic testing algorithms for recent HIV seroconversion (STARHS) provide important information for HIV surveillance. We have previously demonstrated that a patient's antibody reaction pattern in a confirmatory line immunoassay (INNO-LIA™ HIV I/II Score) provides information on the duration of infection, which is unaffected by clinical, immunological and viral variables. In this report we have set out to determine the diagnostic performance of Inno-Lia algorithms for identifying incident infections in patients with known duration of infection and evaluated the algorithms in annual cohorts of HIV notifications. Methods Diagnostic sensitivity was determined in 527 treatment-naive patients infected for up to 12 months. Specificity was determined in 740 patients infected for longer than 12 months. Plasma was tested by Inno-Lia and classified as either incident (< = 12 m) or older infection by 26 different algorithms. Incident infection rates (IIR) were calculated based on diagnostic sensitivity and specificity of each algorithm and the rule that the total of incident results is the sum of true-incident and false-incident results, which can be calculated by means of the pre-determined sensitivity and specificity. Results The 10 best algorithms had a mean raw sensitivity of 59.4% and a mean specificity of 95.1%. Adjustment for overrepresentation of patients in the first quarter year of infection further reduced the sensitivity. In the preferred model, the mean adjusted sensitivity was 37.4%. Application of the 10 best algorithms to four annual cohorts of HIV-1 notifications totalling 2'595 patients yielded a mean IIR of 0.35 in 2005/6 (baseline) and of 0.45, 0.42 and 0.35 in 2008, 2009 and 2010, respectively. The increase between baseline and 2008 and the ensuing decreases were highly significant. Other adjustment models yielded different absolute IIR, although the relative changes between the cohorts were identical for all models. Conclusions The method can be used for comparing IIR in annual cohorts of HIV notifications. The use of several different algorithms in combination, each with its own sensitivity and specificity to detect incident infection, is advisable as this reduces the impact of individual imperfections stemming primarily from relatively low sensitivities and sampling bias.