LAUSR

Clinicians and pathologists alike have long struggled to differentiate early-stage mycosis fungoides (MF) from benign inflammatory dermatoses (ID). Conventional wisdom suggests that consideration of clinical information is invaluable for accurate histopathological diagnosis, yet the direct role this information plays is not well understood. Ramano et al previously investigated how inclusion of relevant clinical information can influence histopathological workup by the pathologist, and discovered mixed results. With a subset of inflammatory lesions, more clinical information prompted ordering of fewer additional stains and sections and decreased turnaround time. However, when lymphoproliferative cases were included these metrics reversed. A pilot study performed by our group surveyed dermatopathologists for diagnostic accuracy of MF vs ID using static images paired with concordant, discordant, or absent clinical information. 2 Interestingly, the percentage of correct diagnosis of MF was highest when clinical information was withheld from participants. This follow-up study aimed to expand upon prior work by better simulating practice-like conditions with digitally scanned glass slides in the place of static images. Following IRB approval, MF cases were selected from our institution's cutaneous lymphoma patient registry. Pathology accessions were selected for initial screening if the lesion was clinically suspicious for MF, was rendered a histopathologic diagnosis of MF, and the patient's skin condition was subsequently managed as MF without a change in diagnosis during follow up. Lesions consistent with advanced plaque and tumor stages were excluded. Of the remaining early stage MF cases, a numeric score was assigned to each case: (1) equivocal for MF vs ID on histopathology alone, (2) suggestive of MF, or (3) likely MF. This was to assure representation of a spectrum of diagnostic “difficulty”. Clonal T cell receptor gene rearrangements were recorded when available. ID cases were screened by HEDI search of the University of Iowa Hospital LIS using the following search terms: “spongiotic dermatitis”, “PLEVA”, “PLC”, “contact dermatitis”, “lichenoid drug reaction”, “arthropod bite”, “lichen striatus”, “lichen sclerosus”, “lichenoid pigmented purpura”, and “T cell pseudolymphoma”, comprising entities which can closely overlap with MF by light microscopy. A few additional clinicopathologically verified cases of these target entities were retrieved from the personal slide collection of one of the authors (V.L.). Characteristics of the included pathologies are summarized in Table 1. Slides were digitally scanned at ×20 magnification using an Olympus digital microscope camera and NIS-Elements microscope imaging software. Clinical scenarios were crafted, modeled off true clinical histories. A slide quiz of 30 cases (15 MF and 15 ID cases in random order) was created using a REDCap electronic survey which referenced a Biolucida slide library. Participants were asked to score each slide on a 5-point scale from diagnostic for ID to diagnostic for MF. Slides were paired with concordant, discordant, or absent clinical information using a random number generator. The Biolucida slide viewer was linked to the survey, so participants could view the history and slide, and record their responses prior to moving on to the next case. Dermatopathologists within the American Society of Dermatopathology (ASDP) membership were invited to participate. Participants completing the quiz were also asked to report demographic information including annual number of MF cases encountered. Of 800 eligible respondents, 8 finished the survey in its entirety. Descriptive statistics are summarized in Table 2. Training background for respondents included dermatology in three, pathology in four, and double training in dermatology and pathology in one. With 30 slide scans in the electronic survey assessed by each of the 8 respondents, a total of 240 images were tested, of which 123 images (51%) were correctly classified (65% ID slides and 38% MF slides). Multi-rater agreement (kappa) for slide classifications was 0.10 (SE: 0.02; P-value: .0003) overall. The statistical model for correct slide classification had an accuracy of 86% (95% CI: 82%, 91%) and controlled for slide and reviewer as random effects and type of information provided with the slide as a fixed effect.