Abstract
Aim
The clinical relevance of microbial DNA detected via next-generation sequencing (NGS) remains unknown. This multicenter study was conceived to: 1) identify species on NGS that may predict periprosthetic joint infection (PJI), then 2) build a predictive model for PJI in a developmental cohort, and 3) validate predictive utility of the model in a separate multi-institutional cohort.
Method
Fifteen institutions prospectively collected samples from 194 revision TKA and 184 revision THA between 2017–2019. Synovial fluid, tissue and swabs were obtained intraoperatively and sent to MicrogenDx (Lubbock, TX) for NGS analysis. Reimplantations were excluded. Patients were classified per the 2018 ICM definition of PJI. DNA analysis of community similarities (ANCOM) was used to identify 17 bacterial species of 294 (W-value>50) for differentiating infected vs. noninfected cases. Logistic regression with LASSO selection and random-forest algorithms were then used to build a model for predicting PJI. ICM classification was the response variable (gold-standard) and species identified through ANCOM were predictors. Patients were randomly allocated 1:1 into training and validation sets. Using the training set, a model for PJI diagnosis was generated. The entire model-building procedure and validation was iterated 1000 times.
Results
The model's assignment accuracy was 75.9%. There was high accuracy in true-negative and false-negative classification using this model, which has previously been a criticism of NGS. Specificity was 97.1%, PPV 75.0% and NPV 76.2%. On comparison of abundance between ICM-positive and ICM-negative patients, Staphylococcus aureus was the strongest contributor (F=0.99) to model predictive power. In contrast, Cutibacterium acnes was less predictive (F=0.309) and abundant across infected and noninfected revisions.
Discussion
This is the first study to utilize predictive algorithms on a large multicenter dataset to transform analytic NGS data into a clinically relevant diagnostic model. Our collaborative findings suggest NGS may be an independent adjunct for PJI diagnosis, while also facilitating pathogen identification. Future work applying machine-learning will improve accuracy and utility of NGS.
