P054
Performance of open-access genomic drug resistance prediction tools for Mycobacterium tuberculosis: a systematic review and meta-analysis
K Dewaele(1,2) C G Jouego Tagne(1,4) C Meehan(2,3) L Laenen(1) E André(1)
1:KU Leuven; 2:Institute of Tropical Medicine Antwerp; 3:Nottingham Trent University; 4:University of Yaoundé
Whole-genome sequencing (WGS) accelerates and simplifies drug-susceptibility testing in Mycobacterium tuberculosis (Mtb). Open-access software tools have become available that offer sample-to-answer drug-susceptibility prediction. Their performance and clinical usability has not yet been summarized. We review and meta-analyse the performance of open-access prediction tools for Mtb drug-susceptibility prediction, compared to phenotypic testing as reference. We queried Medline, Embase and Web of Science databases on the 5th of July 2023. Publications that assess prediction performance of any of the following open-access tools were eligible: KvarQ, TGS-TB, CASTB, PhyResSE, ReSeqTB/ReSeqWHO, MTBseq, Mykrobe, TBProfiler, GenTB, Resistance Sniffer, SAM-TB and MycoVarP. We appraised study quality using the QUADAS-2 tool, and extracted summary performance data per tool. We obtained pooled sensitivity and specificity estimates per tool and per tuberculostatic using a random-effects model. Pooled performance was assessed in 20 studies. Of examined tools, only Mykrobe, TBProfiler, MTBseq and GenTB had been maintained in the three years before the search date. Considering recent versions of these tools, pooled sensitivity generally reaches 90% for isoniazid and ethambutol, and 95% for rifampicin (with specificity reaching 98%). This fulfils criteria set by the WHO. For streptomycin, pyrazinamide, and second-line drugs, sensitivity generally ranges between 70-90%, with specificity approaching or reaching 98%. The performance of open-access drug-susceptibility prediction tools reaches WHO-set criteria for isoniazid, rifampicin and ethambutol, justifying their use as a resistance rule-in and rule-out test for these drugs. For second-line drugs, genomic prediction tools may be used as resistance rule-in tests.