Abstract
INTRODUCTION: Anemia is common in the emergency department (ED). Physicians often rely on inspecting conjunctival pallor or other body parts for gestalt estimates. We aimed to evaluate the validity and reliability of physician gestalt for anemia detection and examine the impact of clinical experience and incorporating images of multiple body parts on physician gestalt-based anemia detection. METHODS: Prospective observational study in the ED at an academic medical center between January-November 2023. Using convenience sampling, we included patients ≥ 18 years with recent laboratory hemoglobin (Hgb) measurements. We used a smartphone to capture the images of the patient's conjunctiva, palm, and fingernails. Five board-certified attending emergency physicians (two junior, two mid-level, and one senior) reviewed the patient images and provided gestalt predictions of Hgb levels and anemia likelihood on a 1-10 scale. Two pairs of physicians evaluated the same set of patient images to assess reliability. Anemia was defined as Hgb < 13.1 grams per deciliter (g/dL) for men and < 11.0 g/dL for women, according to our laboratory standard. RESULTS: We enrolled a total of 100 patients (mean age 67 years; 45% male). Of these, 59 (59%) had anemia and 41 (41%) did not. The correlation coefficients between physicians' predicted Hgb levels and actual Hgb levels were only moderate (0.31, 0.41, and 0.40 for junior, mid-level, and senior physicians, respectively; P < .05 for all). Although not statistically significant, the mid-level physicians' gestalt had the highest area under the receiver operating characteristic curve (0.78), followed by senior- (0.74) and junior physicians (0.72). The impact of incrementally adding images of other body parts to conjunctiva was small (mean changes in anemia likelihood <1 on a 1-10 scale). The agreement on predicted Hgb levels between the paired physicians was high (0.71 for junior physicians, 0.67 for mid-level physicians, P < .001 for both). CONCLUSION: Physician gestalt demonstrated moderate validity and moderate-to-high reliability for anemia detection. Adding images other than conjunctiva did not improve the performance of physician gestalt. However, the clinical experience did matter slightly in detecting anemia.