Evaluating clinical decisions across different professional training profiles is essential to understand the limits and potential of clinical decision support systems in medical practice. This study aimed to compare the proportion of agreement of a deterministic algorithm created by infectious disease physicians with that of Internal Medicine residents at different training stages (first and second year), using as reference the choices of experienced infectious disease specialists who did not participate in building the decision system.

Cross-sectional study with 87 cases composed of real urine culture antibiograms associated with simulated clinical scenarios including relevant variables (severity, pregnancy/lactation, route of administration, infection site). Therapeutic decisions were defined blindly and independently by a first-year resident (R1), a second-year resident (R2), a rules-based algorithm with structured clinical rules, and two infectious disease specialists (E1 and E2, defined as the reference for agreement for the other comparisons). Agreement proportions with 95% confidence intervals (Wilson method) were calculated for outcomes: (1) agreement with E1, (2) with E2, and (3) with at least one of the two specialists, the latter defined as the primary outcome.

The algorithm’s agreement was statistically higher than the residents’ in all analyses. Against E1, Doxa achieved 78.16% (95% CI: 68.39%–85.55%), compared with 39.08% for R1 and 55.17% for R2. Against E2, the algorithm reached 90.80% (95% CI: 82.89%–95.27%), while R1 obtained 36.78% and R2 57.47%. For the primary outcome, the algorithm achieved 95.40% (95% CI: 88.77%–98.20%), versus 47.13% for R1 and 64.37% for R2, with no overlap of confidence intervals. R1 and R2 confidence intervals overlapped, so no performance difference between them could be inferred.

In realistic simulated scenarios for urinary tract infections, a rules-based digital clinical algorithm showed superior performance compared with physicians in training, matching or exceeding the agreement expected between specialists. Therefore, it has potential as a decision-support tool in clinical practice and as a clinical and educational aid.