Evaluation of BCDE, a microcomputer program to analyze automated blood counts and differentials

Current blood analysis practices often rely on manual interpretation of automated counts, which can be time-consuming and subject to variability. Prior research has shown that automated systems can flag abnormal results but lack detailed diagnostic guidance. This gap motivated the development of tools that not only detect anomalies but also suggest potential diagnoses. No prior work had resolved how to integrate diagnostic suggestions with automated blood count data. Existing studies focus on accuracy of counts, not diagnostic relevance. This paper introduces a novel approach to combine data analysis with disease categorization. The knowledge gap lies in how to translate numerical blood data into actionable diagnostic insights. This paper addresses that by introducing a microcomputer-based system.

The study aimed to evaluate a microcomputer program called BCDE for its ability to interpret blood counts and differentials. The specific problem addressed is the lack of automated systems that can not only detect abnormalities but also prioritize likely diagnoses. The motivation stems from the need for faster, more consistent diagnostic support in hematology. The program was tested against both normal and abnormal datasets. The goal was to determine if BCDE could accurately rank diagnoses in line with known conditions. The study focused on triage accuracy and differential ranking. It also aimed to compare BCDE's performance with that of human experts. The ultimate purpose was to assess the program's utility in clinical settings.

The BCDE program was tested on datasets from normal and diseased subjects. Blood counts and differentials were input into the program, which then ranked possible diagnoses. The program compared results to 36 disease categories and normal values. Data from 50 normal subjects and 182 with known hematologic disorders were used. The program's output was compared to actual diagnoses for accuracy. Sensitivity and specificity metrics were calculated for each condition. A separate test compared BCDE's performance to that of 37 physicians. The study used a retrospective design with pre-existing blood data.

In normal subjects, BCDE listed the correct diagnosis first in 49 of 50 cases. For 44 cases, it was the only diagnosis provided. In 182 diseased subjects, the program listed the correct diagnosis first in 134 cases. An additional 40 had the correct diagnosis in second or third place. Conditions like iron deficiency and thalassemia had over 80% sensitivity and 98% specificity. Acute leukemia and folate deficiency had lower sensitivity. In 11 test cases, physicians identified the correct diagnosis only 72% of the time. BCDE listed the correct diagnosis first in 91% of those cases.

The BCDE program demonstrated strong diagnostic accuracy in both normal and abnormal blood data. It correctly ranked diagnoses in most cases, particularly for common hematologic disorders. The program outperformed a panel of physicians in identifying correct diagnoses in test cases. It appears useful for triage and initial differential analysis. The authors suggest it could support clinical decision-making in hematology. The program's performance varied by disorder, with some conditions showing lower sensitivity. These findings trace directly to the study's results. The authors propose further testing in real-world clinical settings.