SystemAPI & Drug Substance

API Impurity Profiling & Investigation

An impurity profiling solution for drug substance makers and CDMOs. When an intermediate tests out of trend, agents read the process historian, balance the impurity across cake, mother liquor and wash, and predict what the next stage removes. Every specified and unspecified impurity is then trended across batches against the ICH Q3A thresholds, so a rising unknown peak is named before it reaches one.

See one impurity traced to its cause, screen by screen ↓
Watch with transcript →
The problem

Confirmed by the lab, the impurity's cause was still argued in a meeting

An isolated intermediate comes back with an impurity well above its usual range but still inside its in-process limit. The lab confirms the number, an investigation opens, and the argument starts: did the chemistry make more of it, did an incoming lot bring it, or did the crystals hold on to it? The answer sits in the historian, the mother-liquor results, the logbook and the lot certificates, each in its own system, so the first week goes on gathering.

Meanwhile the intermediate waits on hold, its next-stage slot waits, and the client's quality reviewer asks why. Across the plant, an unknown peak that sat below the reporting threshold climbs a little each batch, and nobody trends it until it is close to the identification threshold.

Built forProcess investigatorsPlant QA headsMSAT leadsQC lab heads

The Problem Today

A confirmed impurity result starts a debate in meetings, while the records that would settle it sit in separate systems.

164
times FDA cited drug investigations as not thorough, FY2025 published
0.10%
ICH Q3A identification threshold, daily dose of 2 g or less published

Pulled one system at a time

  • Lab results
  • Historian
  • Logbooks
  • Lot records
  • Formed more, or trapped more?Causes debated without the impurity's own mass balance
  • Unknown peaks climb unwatchedBelow the reporting threshold, nobody plots them

The Solution

  1. The run read beside the resultHistorian trace, logbook and lot records joined to the confirmed result
  2. Where the impurity wentCake, mother liquor and wash balanced, so trapped and formed are told apart
  3. Each impurity on its own trendSeries against the ICH Q3A thresholds, with a projection before a crossing
  4. People decideThe MSAT lead approves the experiment; the QA head decides the intermediate's path

Built-in controls

  • A prediction is labeled as one wherever it appears
  • Nobody approves a plan or path they proposed
  • Agents read; root cause and disposition stay with people

What You Get

Investigations start from the chemistry, and a rising unknown reaches the QC lab head before it reaches a threshold.

  • Clients read the same evidenceThe client's reviewer gets the pack the investigator works from
  • Crystallization know-how that staysRamp, mass balance, purge and past rulings on one board
  • One impurity record across productsEach result tied to stage, lot and method, comparable across clients
Features

Impurity profiling and investigation for one out-of-trend intermediate

The process investigator runs the board and the MSAT lead approves the experiment; the plant's QA head picks the intermediate's path, and the QC laboratory's head owns the rising unknown.

demoSample data throughout: the plant, its client, each batch, each person and each number.

  1. 01Fate board

    Impurity mass balance across cake, mother liquor and wash for three batches

    • Grams in solution, then where they went. Each batch shows the impurity before seeding and what landed in the cake, the mother liquor and the wash, with the closure.
    • Rejected less, not formed more. When the amount in solution matches the good batches but more of it stays in the cake, the case turns to the crystallization, away from the chemistry and the incoming lots.
    • Agents working in view. Each agent shows what it linked, read and computed, how sure it is, and who it handed to, and the investigator can steer it.
  2. 02Historian

    Process historian trace read against the cooling band in the instructions

    • Ramp measured against the written band. The peak ramp through the growth range is measured from the trace itself and set against the band the instructions allow.
    • Valve position beside temperature. The stuck valve and the operator's switch to manual sit on one record, which turns an argued fault into a read one.
  3. 03Next stage

    Impurity fate and purge across the next stage, predicted from plant history

    • Purge measured from the plant's own batches. The purge factor comes from the last eleven batches through the final stage, so the forecast rests on this process and its equipment.
    • A band, labeled as a prediction. The predicted final result carries its interval and says when the input sits outside the range the forecast learned from.
    • Hypotheses and their evidence state. Supported, refuted, closed by the lab and not yet tested sit in one strip, with root cause still pending.
  4. 04Causes

    Root cause hypotheses with the evidence for and against, each record cited

    • For, against and still missing. Each cause lists the passages that support or refute it, so the MSAT lead reviews the records themselves.
    • Rulings from earlier cases applied. A standing ruling keeps operator error off the list while a historian trace is unread, and a new ruling waits for the QA head.
  5. 05Confirmation plan

    Confirmation experiment with success criteria locked before it runs

    • Only the crystallization varies. Two scale-down runs at the normal and the fast ramp, plus a reslurry of the retained cake, test the cause without repeating the chemistry.
    • Criteria fixed in advance. What counts as trapping, and what points to the crystal surface instead, is written before the lab runs it.
    • Proposed by one, approved by another. The investigator proposes and the MSAT lead approves after opening the plan; root cause stays pending until the results are in.
  6. 06Paths

    Three paths for the held intermediate, rehearsed and decided by QA

    • Predicted impurity, yield and time per path. Process with added testing, reprocess, or reject each show what the intermediate becomes and who must agree under the quality agreement.
    • Assumptions the investigator can edit. Changing the purge factor or the repeat-crystallization loss shows how sensitive each path is before anyone recommends one.
    • The investigator recommends, the QA head signs. Lifting the hold needs a signature from QA at the plant, and blending to dilute an impurity is not offered as a path.
  7. 07Profiles

    Impurity profile of each API by stage, against the ICH Q3A thresholds

    • Specified and unspecified, stage by stage. Each impurity carries its limit, a sparkline of reviewed results and its place against the reporting, identification and qualification thresholds.
    • Thresholds set from the daily dose. The thresholds follow each product's maximum daily dose, so only the client's specification change can move them.
  8. 08Rising unknown

    Unknown impurity projected to the identification threshold, source linked

    • Projection, not results. A linear fit of approved results shows how many batches remain before the identification threshold, so the lab names the peak first.
    • What moved with the peak. The trend watcher ties the rise to a starting-material lot, and shows a recovered solvent that changed earlier with no peak is not a cause by itself.
    • Decisions listed with their owner. Identification, a watch on the supplier's lots, the next lot to charge and telling the client each name who decides.

The investigator reads the chemistry, the MSAT lead tests it, QA decides

Four jobs meet on one board, and each signs only what is theirs.

Process investigators

Run impurity investigations, build the evidence and propose experiments and paths

  • Every linked record within minutes of the lab's confirmation
  • The mass balance worked out from routine samples
  • Hypotheses with their passages cited
Plant QA heads

Approve investigations and decide what happens to an affected intermediate

  • Paths rehearsed with predicted impurity and yield
  • A pack for the client, approved before it goes
  • A signed evidence summary to export
MSAT leads

Approve hypotheses and confirmation experiments, and own campaign trending

  • Criteria locked before the lab runs
  • Past rulings applied to the next case
  • Supplier and solvent links on each trend
QC lab heads

Close lab investigations and own identification of unknown peaks

  • Rising unknowns flagged before the threshold
  • Identification requests with sample and method
  • Results confirmed by a second person

A challenger reading as the client caught the agents' own liquor volume

A nominal volume in the arithmetic

The first mass balance used 1,650 L from the instructions; the receiver read 1,610 L, which brings the closure to 97.6% from 99.2%.

A batch that seemed to contradict the cause

The batch before peaked at 0.47 °C/min, inside the band, and its 0.06% sat in the plant's normal 0.04–0.06% scatter.

Operator error with the trace unread

A standing ruling blocked that cause until the historian was read, and the trace showed a stuck valve.

A peak below anyone's limit

An unspecified peak rose from 0.03% to 0.07% after a supplier change, with the identification threshold about 5 batches away.

demoSample data, from the working solution.

More screens

The client's reviewer reads the pack the investigator works from

Past the one case: the client's view, the lab's identification queue, the open investigations, the signed summary, the audit record and the rules the agents follow.

An evidence pack the client can questionThe client's quality reviewer reads status, the intermediate's path, where the impurity went and the prediction; a typed question counts only once someone has read it.
A name for the rising peakThe QC lab head requests LC-MS identification on a retained sample and assigns it to analytical development, whose result needs the lab head's confirmation.
Eight open investigations on one clockEach open impurity investigation, showing who owns it, when it is due and its stage, the released hold beneath them, and trend flags that can be acknowledged only once opened.
Signed by the person who did not compile itPlan, path, identification, pack, ruling and the interim evidence summary, each with who signs; the summary exports into the investigation record only when signed.
Each record opened before each signatureThe trail shows who opened which record, who answered which challenger point with what reason, and what each signature means.
Two approved rulings steer the agentsThresholds from each product's daily dose, the trend rules, the closure band, the investigation and identification clocks, and two rulings the QA head approved.

demoSample data, from the working solution.

The purge forecaster labels each final-result forecast a prediction

Eight agents link, read, balance, forecast, test and challenge; people decide: the investigator, the MSAT lead, then QA.

Record linker

Pulls the deviation, lab investigation, batch record pages, logbook, work requests, certificates and earlier batches tied to a result.

  • Never edits a record
Historian reader

Reads temperature, valve and agitation traces and measures ramps against the instructions.

  • Never fills a gap in a trace
  • Never calls a fault human or equipment by itself
Mass-balance calculator

Computes the impurity in solution, cake, mother liquor and wash, and the closure.

  • Uses measured volumes over nominal ones
Purge forecaster

Measures purge across the next stage from plant history and predicts the final result with an interval.

  • Never shows a prediction as a result
Hypothesis tester

Weighs each cause with evidence for and against and drafts the confirmation experiment.

  • Never sets a root cause
Client's eye

Reads the case as the client's quality reviewer, including the agents' own arithmetic.

  • Hands any judgment call to a named person
Path rehearser

Plays forward each path for an affected intermediate: impurity, yield, time and who must agree.

  • Never releases a hold
  • Never offers blending
Trend watcher

Trends each impurity across stages, batches and campaigns and links a change to a lot, solvent, vessel or change control.

  • Uses reviewed results only
  • Never names a structure

Historian traces and mother-liquor results the plant already collects

Chromatography data and LIMSIn-process, mother-liquor, wash and release results, reviewed
Process historianTemperature, valve position and agitation for each stage
Batch records and logbooksExecuted steps, operator notes and equipment entries
Maintenance requests and work ordersReported faults, findings and parts replaced
Lot and incoming test recordsStarting materials, intermediates, seed lots and supplier changes
Development reports and instructionsSpiking studies, cooling bands and allowed reprocessing steps

What investigators and MSAT leads ask about API impurity profiling

What is impurity profiling for an API?

Impurity profiling is the record of each identified and unidentified impurity in a drug substance made by a controlled process: its identity or retention time, the range observed and its classification, compared over time to spot change. Here agents keep that profile for each product and stage, and trace a new or rising impurity back to its cause.

How does the mass balance tell trapped from formed?

The impurity in solution before seeding is compared with what ends up in the cake, the mother liquor and the wash, from the routine samples. If the amount in solution is unchanged but more of it stays in the cake, the crystals held on to it; the chemistry did not make more.

When must an unknown impurity be identified under ICH Q3A?

Above the identification threshold, which follows the maximum daily dose. The trend watcher projects each unspecified peak against that threshold from reviewed results, so identification can be requested while the peak is still below it.

Who decides what happens to the affected intermediate?

The plant QA head, after the investigator recommends a path. Reprocessing also waits for the client's written consent where the quality agreement requires it, while the medicine itself is the client's Qualified Person's call.

Does it set the root cause?

No. The agents rank hypotheses and draft the experiment that would confirm one. The investigator proposes, the MSAT lead approves the experiment, and QA approves the root cause once the results are in.

What does the client's quality reviewer see?

An evidence pack shared under the quality agreement: status, the intermediate's path, where the impurity went and the labeled prediction. Questions come back in fixed fields, and typed text changes nothing until someone reads it.

Bring one out-of-trend result
and the batch records behind it.

Send one confirmed impurity result with its historian export, mother-liquor results and batch record pages; we will show where the impurity went.