Do we truly see what we think we see? The role of cognitive bias in pathological interpretation

Abstract

In the histomorphological grading of prostate carcinoma, pathologists have regularly assigned comparable scores for the architectural Gleason and the now-obsolete nuclear World Health Organization (WHO) grading systems. Although both systems demonstrate good correspondence between grade and survival, they are based on fundamentally different biological criteria. We tested the hypothesis that this apparent concurrence between the two grading systems originates from an interpretation bias in the minds of diagnostic pathologists, rather than reflecting a biological reality. Three pathologists graded 178 prostatectomy specimens, assigning Gleason and WHO scores on glass slides and on digital images of nuclei isolated out of their architectural context. The results were analysed with respect to interdependencies among the grading systems, to tumour recurrence (PSA relapse > 0.1 ng/ml at 48 months) and robust nuclear morphometry, as assessed by computer-assisted image analysis. WHO and Gleason grades were strongly correlated (r = 0.82) and demonstrated identical prognostic power. However, WHO grades correlated poorly with nuclear morphology (r = 0.19). Grading of nuclei isolated out of their architectural context significantly improved accuracy for nuclear morphology (r = 0.55), but the prognostic power was virtually lost. In conclusion, the architectural organization of a tumour, which the pathologist cannot avoid noticing during initial slide viewing at low magnification, unwittingly influences the subsequent nuclear grade assignment. In our study, the prognostic power of the WHO grading system was dependent on visual assessment of tumour growth pattern. We demonstrate for the first time the influence a cognitive bias can have in the generation of an error in diagnostic pathology and highlight a considerable problem in histopathological tumour grading.

The Use of Immunohistochemistry for Biomarker Assessment—Can It Compete with Other Technologies?

“…Any human evaluation is based on subjective visual assessment and is thus susceptible to bias and interobserver variability (Fandel et al 2008;Hamilton et al 2010). …”

Diagnosis at the microscope: a workplace study of histopathology

“…For example, microscopy imagery is interpreted by the histopathologist in the context of clinical and sample information (Hamilton et al 2009) and diagnoses are made within the context of grading and classification systems (Fandel et al 2008). Diagnostic work does not necessarily begin and end at the microscope.…”

Systematic review: Bias in imaging studies - the effect of manipulating clinical context, recall bias and reporting intensity

“…The secondary search (Table 2) identified 2828 publications with the full text retrieved for 34: ultimately 6 were included [6,13,[26][27][28][29] and 28 rejected because the research focused on case-specific information. The tertiary search (Table 3) identified 74 MeSH terms which were combined into 18 Boolean search strings: These identified 111 potential articles with a further 2 via snowballing; 5 articles were ultimately included [11,12,[30][31][32]. Overall, 11247 abstracts were reviewed, 201 full articles retrieved, and 12 ultimately included for systematic review ( Table 4).…”

“…The remaining 9 studies investigated observer performance in different situations with fixed prevalence: 4 compared performance in the laboratory to daily practice [10,12,32]; 3 investigated observer blinding to previous clinical investigations [29][30][31]; 1 investigated training [27]; 1 investigated varying reporting conditions [25]; 1 investigated recall bias [28]. The 4 studies that investigated interpretation in "the field" used retrospective data obtained from normal clinical practice [10,12,25,32].…”

“…Study characteristics and settings (Table 4) The following diagnostic tests were investigated by the 12 included studies: 9 studies were radiological (5 mammographic [10,12,25,28,29], 3 chest radiology [13,26,27], 1 angiographic [6]), 2 endoscopic [30,32], and 1 histopathological [31]. A single research group contributed 5 studies [10,13,[26][27][28].…”