Objectives To evaluate the test characteristics of clinical gestalt for detecting the presence and severity of anaemia in emergency department patients at a tertiary referral hospital in Tanzania.

Methods This prospective study enrolled a convenience sample of emergency department patients who had a complete blood count ordered by the treating physician in the course of their clinical care. Physicians recorded their impression of the presence and severity of anaemia before viewing the laboratory results. To assess interobserver agreement, a second physician provided their blinded gestalt impression of the patient's haemoglobin level.

Results We enrolled 216 patients and complete data were available for 210 patients (97%), 59% male, median age 30 years. The range of measured haemoglobin values was 1.5–15.4 g/dL. The physicians rated anaemia mild or absent in 74 (35%), moderate in 72 (34%) and severe in 64 patients (30%). These estimates were significantly concordant with the laboratory haemoglobin measurements (Kendall's τ b=0.63, 95% CI 0.57 to 0.69, p<0.0001). The test characteristics of physician gestalt estimates for severe anaemia were: sensitivity 64% (95% CI 53% to 74%), specificity 91% (95% CI 85% to 96%), positive likelihood ratio of 7.4 (95% CI 4.2 to 13.3) and negative likelihood ratio of 0.40 (0.3 to 0.5). The weighted Cohen's κ for interobserver agreement between physicians on the gestalt estimate of the degree of anaemia was 0.87 (95% CI 0.76 to 0.98).

Conclusion Physicians’ estimates of the severity of anaemia were significantly concordant with laboratory haemoglobin measurements. Sensitivity of the gestalt estimate for severe anaemia was moderate. Interobserver agreement was ‘almost perfect’.