The lot number is a signal: PV needs batch-level detection as a standing discipline

The unit of risk is sometimes the batch

Most pharmacovigilance is built around the product: a drug, a reaction term, a reporting rate. This week's records suggest the batch is sometimes the better unit of analysis. Greece's EOF reports that the EMA's safety committee, PRAC, is examining rising allergic-reaction reports linked to batches of normal human immunoglobulin medicines. The phrase that matters is "batch-specific". A product-level view would dilute a cluster like that across everything on the market. The outcome of the review is not yet known, and so is the cause. I will not guess at either.

The same week supplies parallel cases from the quality side. AIFA published a DHPC after CSL Behring reported stopper defects in certain Respreeza batches that could prevent administration or allow microbial contamination. In the US, Compact Medical is removing nine ButterflyBVM lots and applying a pre-use test correction to 22 others after holes were found forming in the bellows. B. Braun recalled three lots of Streamline Bloodline Sets after a missed solvent step left a connector inadequately bonded. These are different products and different failure modes, but each is defined by a lot.

Why this is a PV problem and not only a QA one

When risk follows a lot, the first sign may be a handful of cases that look unremarkable one by one. They become meaningful only if someone asks which batch each patient received. That depends on whether lot numbers were captured at intake, whether they were captured accurately, and whether anyone looks at them across cases. Many safety databases hold batch data as an optional field, and follow-up for it is the first thing dropped under workload.

The reporting channel matters too. EMA has published QRD Appendix V, covering the standard statement and contact details for reporting suspected adverse reactions in product information. The records do not say what changed, so I draw no conclusion about its content. It is still a reminder that spontaneous reporting is the front door. If the front door does not prompt for a lot number, no downstream analytics can recover it. MHRA's push for Yellow Card reports on medicines and devices, including its note on low reporting for AI-enabled devices, makes the same point about under-capture.

The strongest counter-argument

The objection is fair. Batch fields in spontaneous reports are patchy, and counts per lot are often tiny. Clustering on thin data produces false alarms, and a PV team that escalates every lot-level wobble will exhaust its quality partners. Allergic reactions to biologicals also have background rates and many causes. A batch association is a hypothesis, not a finding, and PRAC's review has not yet said otherwise.

I accept all of that, and it argues for a better process, not for ignoring batches. Imperfect data is a reason to start capturing it properly now. The cost of a false lot-level alert is a conversation with quality. The cost of missing a true one is patients continuing to receive the affected product.

What to do

Make batch analysis a scheduled activity for products where manufacturing variability is a plausible risk, such as plasma-derived products, sterile injectables and combination products. Mandate lot-number follow-up in case-handling procedures and track the completeness rate as a KPI. Then give PV and quality a shared, documented trigger for when a lot-level pattern goes to joint review.

Do not wait for a regulator to open a review before the exercise is done once. Pull your own data for products named in this week's alerts and check how many cases carry a usable lot number. That figure will tell you how much a batch signal would currently reach you.

— The Editor