Joint Multidrug-Resistance Information in Indian AMR Surveillance: Bounds from Marginal and Pairwise Susceptibility Data and a Temporal Validation

Abstract: Antimicrobial-resistance surveillance summaries may report resistance separately by antibiotic while omitting the within-isolate co-occurrence structure needed to describe multidrug states. This study evaluated how much information about an all-five nonsusceptible state can be recovered from marginal antibiotic frequencies, whether pairwise information narrows that uncertainty, and whether relationships identified in an early period remain valid later. Historical Indian ATLAS records for Klebsiella pneumoniae and Escherichia coli were analysed; the source manuscript reports 641 target-species records, of which 544 had complete results for five prespecified antibiotics. For each organism, the observed all-five nonsusceptible frequency was compared with an independence estimate formed by multiplying the five separate frequencies and with feasible ranges constrained first by the five marginals alone and then by every drug-pair frequency. For K. pneumoniae, the observed all-five state was 16.54%, compared with 1.17% under the independence approximation; the marginal-only feasible range was 0–22.06%, and the pair-informed range was 14.34–19.12%. For E. coli, the corresponding values were 5.51%, 0.18%, 0–6.99%, and 5.15–5.88%. Temporal validation showed that a later K. pneumoniae burden of 25/72 (34.7%) exceeded the early-information upper bound of 16.5% and the reported 97.5th bootstrap percentile of 21.5%. These findings show that higher-order joint resistance is not identified by marginal frequencies alone and that pairwise information can materially narrow the feasible range in the analysed data, while temporal transportability of such relationships is not guaranteed. The study uses historical, non-nationally representative data and does not estimate current Indian prevalence, recommend treatment, or justify reduced clinical antimicrobial susceptibility testing. Keywords: Antimicrobial resistance; multidrug resistance; AMR surveillance; antimicrobial susceptibility testing; joint probability