Laboratory testing policies and their effects on routine surveillance of community antimicrobial resistance.

OBJECTIVE To investigate the effects of laboratory testing policies, particularly selective testing, rule-based reporting and isolate identification, on estimates of community antimicrobial resistance. MATERIALS AND METHODS Antibiotic resistance estimates were analysed from an all-Wales dataset for approximately 300 000 community isolates of common pathogens. RESULTS Selective testing policies were often associated with markedly increased resistance, particularly for second-line testing. Site-specific testing tended to yield variant resistance estimates for eye and ear isolates. Estimates from rule-based reporting deviated markedly from test-result-based reporting. Urinary isolates reported as Escherichia coli showed greater susceptibility than those reported as undifferentiated urinary 'coliforms'. The proportion of isolates tested for an antibiotic by a laboratory was a useful indicator of selective testing in this dataset. Selective testing policies had invariably been applied where the proportion of isolates of a species tested against an antibiotic was <90%. As this proportion fell with increasingly selective policies, divergence from pooled-all-Wales non-selective estimates tended to increase, with a bias to increased resistance. CONCLUSIONS Selective testing, rule-based reporting and urinary coliform identification policies all had significant effects upon resistance estimates. Triage based upon the proportion of isolates tested seemed a useful tool in assigning analysis resources. Where <20% of isolates were tested, selective policies with inherent bias to increased resistance were common, the low number of isolates gave high potential sampling errors, and little confidence could be placed in the resistance estimate. Where 20-90% of isolates were tested, detailed analysis sometimes revealed resistance estimates that might be usefully retrieved. Where >/=90% of isolates were tested, there was no evidence of selective testing, and inter-laboratory variation in estimates appeared to be safely ascribable to other effects, e.g. methodology or real variation in resistance levels.