Performance of Gram Stain Characteristics for Predicting Culture Growth in Lower Respiratory Specimens Across Multiple Institutions

Abstract Data from Gram stains performed on clinical specimens can guide clinical decision-making and downstream laboratory workflows. However, the extent to which Gram stain results predict gold-standard culture results, and the generalizability of this performance across sites, warrants additional exploration. We sought to discover which Gram stain features would be associated with clinically significant culture growth (“growth”). To do so, 574 remnant specimens from the lower respiratory tract collected from four institutions were analyzed. A standardized Gram stain and culture protocol was performed across all study sites. Final reference organism identification was assigned using MALDI-ToF MS, and clinically significant growth was defined as any organism reported to the medical record. Gram stains with mixed microorganisms detected were considered concordant if ≥2 organisms of different Gram stain appearances were identified in culture. Growth was observed in 53% of specimens, including 25% of specimens in which Gram stain demonstrated no organisms. The 95% confidence intervals (CI) of the site-wise growth rates were 30-42%, 42-52%, 51-67%, and 60-77% (p < 0.001 by Chi-squared). By Gram stain morphology, 53-69% of specimens with Gram-positive cocci grew any clinically significant bacteria, while 31-49% grew a concordant Gram-positive cocci. 65-85% of specimens with Gram-negative bacilli grew any clinically significant bacteria, while 59-74% grew a concordant Gram-negative bacilli. 63-83% of specimens with mixed microorganisms grew 2 or more clinically relevant bacteria of differing Gram stain morphologies. Specimens from bronchoalveolar lavages (BALs) were less likely to have growth (30-42%) than endotracheal aspirates (66-77%). Polymorphonuclear leukocyte (PMNs) abundance was associated with culture growth, with Gram stains showing no, rare, few, moderate, or abundant PMNs demonstrating growth 28, 44, 52, 60, and 67% of the time (p < 0.001). Specimens with None and few squamous epithelial cells on Gram stain demonstrated growth 38-52% and 59-68% of the time (p < 0.001). No significant difference in growth was observed for red blood cells. Organism abundance on Gram stain was significantly associated with culture growth, with growth rate CIs being 20-32% for no organisms, 44-62% for rare, 58-77% for few, 70-93% for moderate, and 81-95% for abundant organisms seen. The extent to which these associations could guide the development of decision rules or machine learning models that demonstrate acceptable negative predictive value for pathogen growth to alleviate the operational burden of low-utility cultures is an exciting future direction to explore.