Styxis
Company
Knowledge

Hold incomplete ICSR loads before monthly timeliness reporting

Reconcile expected batches, senders, case versions and receipts before calculating a monthly ICSR indicator.

A good ratio can come from missing inputs

A high on-time percentage may be misleading when the input is incomplete. Missing or late transactions might have disappeared from the denominator. Truthound can sit before the monthly calculation, checking a declared batch and reference contract.

The load control EMA describes

EMA's October 20, 2026 pharmacovigilance industry-platform meeting is recorded in a document dated October 30, 2026. It describes a data-load control introduced after incomplete data affected a report for March and differences between internal submission counts and EudraVigilance records. The workflow here is a conditional design responding to that documented issue.

Connect expectation, receipt and load

An expected-batch register should connect sender, reporting period, expectation basis, arrival and load state. Keep the receiving response separate from internal processing. A file's arrival does not establish that every transaction was loaded.

A case is not a transaction count

One case can produce several follow-up transactions. Separate case count from transmission count before choosing a monthly denominator. Reconcile case, version, sender and receipt relationships. Removing legitimate follow-ups as duplicates can simplify the calculation while changing its meaning.

The pre-calculation Truthound hold

Truthound Core Python SDK 3.3.4, operating on a prepared Polars table, provides RowCountValidator for supplied bounds and ColumnPairValidator for specified comparisons between non-null fields. Owners establish the batch boundary and whether null means absent or unresolved. Matching row counts alone cannot establish content completeness or correct case relationships.

When the existing Spark checks should remain

An existing Spark workflow may keep Deequ 2.1.0 VerificationSuite and Check rules for size, completeness and uniqueness. Comparing it with a prepared Polars-table Truthound path requires aligned inputs and transfer boundaries. General capabilities on different runtimes do not establish the same ICSR admission result or a performance advantage.

Separate late data and revisions

Keep late transactions and corrections separate from the state used in the original monthly calculation. If recomputation changes the figure, preserve which batch or definition changed. Overwriting with the latest number loses what the earlier report actually observed.

Send the right question to each owner

Transmission, loading and case-processing owners may need different questions. Check sender identity, load completion and revised follow-up classification separately. Do not quietly remove unresolved rows from the denominator merely to produce a report sooner.

Admit the monthly input before its metric

Admit the monthly input when batch scope and case/transaction relationships are explainable and unresolved material is visible. Only then interpret the timeliness metric. Repeated calculation needs an equally repeatable input-admission boundary.

Conditions

Use approved batch and case/sender definitions within authorised access to sensitive information.

Alternatives

Keep Deequ 2.1.0 in an existing Spark workflow; moving to a prepared Polars table needs a separate operating justification.

Limitations

  • These proposed checks are not an execution on EMA data and do not determine drug safety or legal reporting compliance.

Evidence

Event checked
Subject checked
Modified