Target Trial Emulation Cannot Randomize Clinical Judgment: A Pertussis Study Audit

What the Study Reported

The retrospective cohort included 196 infants aged 12 months or younger across 11 Italian health centres. Clinicians selected clarithromycin or azithromycin according to clinical judgment. The analysis used propensity-score weighting to address measured confounding.

The public abstract concludes that clarithromycin was associated with more favourable outcomes. That is appropriately association-shaped language. The reviewer's job is to determine how far the design can move the result toward a comparative treatment effect—and where it cannot.

The Blueprint Is Not the Assignment Mechanism

A target trial protocol aligns eligibility, treatment strategies, time zero, follow-up, outcomes, estimands, and analysis. This prevents avoidable design errors such as immortal time, inconsistent eligibility, or a vague comparator. Those gains are real.

But exchangeability remains a separate requirement. If a clinician chooses one macrolide for younger, frailer, more unstable, or differently presenting infants, the treatment groups can differ in prognosis before the first dose. Weighting can balance recorded variables. It cannot balance an undocumented bedside impression, an unrecorded referral history, or a center-specific preference that was never represented in the model.

Four Places the Causal Claim Can Still Break

Why Endpoint Definition Deserves Its Own Timeline

“Severe disease course” sounds clinically direct, but a score is a construction. The abstract identifies a Pertussis Severity Score above 5 as the primary outcome without listing its components. A full review should mark when each component becomes observable and whether treatment or subsequent care can affect it.

This does not prove endpoint contamination. It identifies the question the methods section must answer. A physiological state measured consistently after treatment is a legitimate outcome. A component driven by ICU admission, oxygen use, testing intensity, or another clinician response can partly reflect the care pathway. In a multicentre study, thresholds for those responses may also vary by hospital.

Draw the sequence: baseline presentation → antibiotic choice → evolving illness → care decisions → score assessment. Then decide whether the score estimates patient severity, a combined treatment-and-care strategy, or an inseparable mixture.

A Reviewer Checklist for Weighted Comparative Effectiveness

The fastest stress test is to ask what would make the two treatment groups exchange places. If one hospital strongly prefers clarithromycin, one clinician reserves azithromycin for a particular infant, or one pretreatment feature changes both drug choice and prognosis, that fact belongs in the design—not in a limitations paragraph after the estimate.

What the Estimate Can and Cannot Say

1. It can report the observed, weighted contrast. The point estimate and interval describe the analysis that was performed.
2. It cannot prove all relevant treatment drivers were measured. Covariate balance is limited to the covariates and representations shown.
3. It should not turn two deaths into a mortality verdict. Rare events can be clinically important and statistically unstable at the same time.
4. It does not automatically transport. Hospitalised Italian infants during one pertussis period may differ from infants in other systems, eras, or disease-severity ranges.

None of these cautions make the study useless. They make its contribution legible: an important, hypothesis-strengthening comparative effectiveness signal whose causal interpretation depends on measured treatment assignment, overlap, endpoint construction, and precision.

Where Aqrab Fits

Method labels are easy to reward. Design evidence is harder to inspect. Aqrab helps reviewers unpack a target-trial emulation into the exact questions that govern the claim: how treatment was chosen, what was known before time zero, whether both strategies had support, and whether the outcome sits downstream of treatment, care, or both.

If you are reviewing a real-world comparative effectiveness study, try Aqrab on the target-trial table and propensity-score methods together. For structured critique inside a research workflow, see the developer documentation.

Sources and Further Reading

• Lo Vecchio A, et al. Comparative effectiveness of clarithromycin and azithromycin for pertussis in hospitalised infants in Italy, the 2026 multicentre target-trial emulation used for this audit.
• Hernán MA, Robins JM. Using Big Data to Emulate a Target Trial, on specifying the target protocol and conditions needed for causal interpretation.
• Aqrab guide to target trial emulation, for the core protocol components.
• Aqrab guide to confounding by indication, for treatment assignment tied to prognosis.
• Aqrab guide to positivity and overlap, for weight and common-support diagnostics.

The Practical Bottom Line

Target trial emulation improves observational research by forcing the question into protocol form. That discipline is the beginning of causal credibility, not its completion.

When treatment follows clinical judgment, inspect what the clinician knew, what the dataset captured, where both strategies overlap, and when the outcome was created. The blueprint can clarify the experiment. It cannot randomize the past.