Background/Purpose:  Recently, automated interpretation (AI) systems for anti-nuclear antibody (ANA) analysis have been introduced based on assessment of indirect immunofluorescence (IIF) patterns. However, the reliability of these systems has not been assessed in routine clinical practice. Therefore, the diagnostic performance of a novel automated IIF reading system was compared with visual interpretation (VI) of IIF in daily clinical practice.

Methods:  ANA-IIF tests of consecutive serum samples from patients with suspected connective tissue disease were carried out using HEp-2 cells according to routine clinical care. VI was performed first by two laboratory experts independently; discrepant results were resolved by collegial discussion with a third expert. Afterwards, AI of IIF findings were obtained using a visual analyzer (Zenit G-sight, Menarini, Germany) utilizing novel automated pattern recognition algorithms. VI and AI readings were stratified as negative (serum dilution of 1:80 with no fluorescence), ambiguous (dilution of 1:80 with slight fluorescence, but no defined pattern) and positive (dilution of ≥ 1:80 with definite IIF pattern). Agreement rates between ANA results by AI and VI were calculated.  IIF patterns were categorized as either being homogeneous, fine granular, coarse granular, nucleolar, centromeric, or mitochondrial by AI and VI, respectively.

Results:  Of the 340 samples investigated, VI yielded 205 (60%) negative, 42 (12%) ambiguous and 93 (27%) positive results, whereas 82 (24%) were determined negative, 176 (52%) ambiguous and 82 (24%) positive by AI. Of the 82 negative samples by AI, 78 (95%) were also negative, 3 (4%) were ambiguous and 1 (1%) was positive as judged by VI. Of the 176 ambiguous samples by AI, 125 (71%) were negative, 30  (17%) also ambiguous and 21 (12%) positive by VI. Of the 82 positive samples by AI, 2 (2%) were negative, 9 (11%) ambiguous and 71 (87%) also positive by VI. AI displayed a diagnostic accuracy of 179/340 samples (53%) with a kappa-coefficient of 0.35 compared to VI as the gold standard. Solely relying on AI with VI only performed for all ambiguous samples by AI would have missed 1 of 93 (1%) positive results by VI, and misclassified 2 of 205 (1%) negative results by VI as positive. Of the 93 positive samples by VI, AI identified 36 (39%) IIF patterns correctly (18 (95%) of 19 homogeneous, 1 (3%) of 3 fine granular, 13 (65%) of 20 coarse granular, 2 (18%) of 11 nucleolar, 1 (33%) of 3 centromeric, and 1 (100%) of 1 mitochondrial patterns).

Conclusion:  Utilizing novel algorithms, AI of ANA-IIF displayed reliable detection of positive and negative results as confirmed by VI. However, there were a substantial number of ambiguous results as well as inconsistencies in the identification of the correct pattern by AI requiring additional VI analysis. In contrast to previous studies utilizing well defined biobank samples, the use of AI in daily clinical practice resulted only in a moderate reduction of the VI workload (82 of 340 samples: 24%), which was predominantly due to a large proportion of ambiguous AI-results.

---

« Back to 2013 ACR/ARHP Annual Meeting

ACR Meeting Abstracts - https://acrabstracts.org/abstract/the-reliability-of-a-novel-automated-system-for-ana-immunofluorescence-analysis-in-daily-clinical-practice/