Any Report In.A Submission-Ready ICSR Out.
ThinkPV turns raw adverse event reports into submission-ready ICSRs. It reads reports from every channel in any format, confirms each one is a valid case, extracts and codes it, routes it by seriousness and deadline, and converts it to E2B(R3) or E2B(R2). Your safety team reviews and signs every case.
Inbound Call Summary
Possible adverse event, transcribed and summarized by the call center
CallerPharmacist, Elm Street PharmacyAI
ProductVeltrazine 20 mg tabletsAI
Lot numberVZ-24117
The pharmacist reports that a 58-year-old womanAI started Veltrazine for hypertension on March 2. On March 14 she noticed yellowing of her eyes and skinAI and dark urine.
She was admitted to Riverside GeneralAI on March 15 for evaluation. Veltrazine was stoppedAI at admission, and her symptoms were improving at the time of the call.
A sample call-center summary turned into a case in ThinkPV. All names, products and numbers are made up.
From Intake to Submission in Five Steps
Every report follows the same path. Each step below opens the screen your safety team works in, with a sample case moving through it.
- Step 1IngestEmail, calls, literature, partner and E2B files land in one queue.
- Step 2ExtractUnstructured documents become structured cases.
- Step 3CodeEvents coded to MedDRA, products to your dictionary.
- Step 4Triage & RouteSeriousness, expectedness and deadlines set, then routed.
- Step 5ConvertEach case exported as validated E2B(R3) or E2B(R2).
Four Agents. Every Handoff Evaluated.
Every report, whatever its format, passes through four specialist agents. Before any agent’s work moves on, an independent evaluator agent checks it against the source and your thresholds. Work that passes moves on in seconds. Work that does not goes to a person, with the reason.
Grounded in the Source
Every extracted value must point to the text it came from. A value with no source does not pass.
Valid Before It Moves
An identifiable patient, reporter, suspect product and event are confirmed before a case is created.
Coded and Consistent
MedDRA terms, dates, doses and seriousness are checked against each other and against the source.
People on Every Exception
Anything below your thresholds goes to a reviewer, and each decision becomes a test case for the next agent version.
Every Report, Every Channel, One Queue
Reports arrive by email, call center, literature, web forms, partner files and E2B gateways. ThinkPV ingests each one, reads it whatever its format, checks it against the cases you already have, and sorts the queue by seriousness and reporting deadline.
Every Intake Channel
Email and attachments, call-center summaries, literature, web forms, partner spreadsheets and incoming E2B(R2) or E2B(R3) files.
Duplicates Caught Early
Each new report is compared with existing cases, and likely duplicates or follow-ups are linked for a reviewer to confirm.
Volume Without Backlog
Reports are processed in parallel, so a spike in volume does not become a backlog.
Cholestatic Liver Injury With Veltrazine: Three Cases
R. Okafor, L. Chen, M. Duarte
We describe three patients treated with Veltrazine 20 mg daily. Patient A, a 61-year-old manAI, developed jaundice after 5 weeks.
Patient B, a 47-year-old womanAI, had pruritus and ALT 6× ULN at week 3. Patient C, aged 70AI, was hospitalized with cholestatic hepatitis.
All three improved after Veltrazine was stoppedAI. The authors judged the reactions probably relatedAI to Veltrazine.
Unstructured Documents Become Complete Cases
Narratives, letters, scanned faxes, emails and journal articles carry the case in free text. ThinkPV reads them the way a case processor does: it finds each patient, product and event, splits one document into as many cases as it describes, and fills every case field with a link back to the sentence it came from.
Any Document, Any Layout
Free-text narratives, scans, handwriting, tables and email threads are read with OCR where it is needed.
One Document, Many Cases
A case series or a partner listing is split into individual cases, each with its own patient and events.
Every Field Sourced
Each extracted value links to the page and sentence it came from, so review takes seconds.
Coded to MedDRA, With the Evidence
Every verbatim term is coded to MedDRA and every product to your dictionary. Clear terms are coded with high confidence, and ambiguous ones go to a coder with the candidates already ranked.
Verbatim to MedDRA
Reported terms are coded to the lowest level term and preferred term in the current MedDRA version.
Products to Your Dictionary
Products, strengths and formulations are matched to your product dictionary or WHODrug.
Coders Where It Matters
Low-confidence and ambiguous terms go to a coder with ranked candidates and the source sentence.
| Verbatim | MedDRA PT | Conf. |
|---|---|---|
| yellowing of eyes and skin | Jaundice | 96% |
| dark urine | Chromaturia | 94% |
| headache | Headache | 99% |
| itchy rash on arms | Rash pruritic | 93% |
| felt off for a few days | — | 41% |
| liver enzymes up | Hepatic enzyme increased | 92% |
| couldn’t sleep | Insomnia | 95% |
Triaged and Routed by Rules You Set
Each case is assessed for seriousness and expectedness against your reference safety information, given its reporting clock, and routed to the right queue and reviewer. Expedited cases go to the front.
Seriousness and Expectedness
Each case is checked against seriousness criteria and your reference safety information.
Day Zero Set Correctly
Day zero is set when the minimum criteria for a valid case are first met, and every case shows the time left to report.
Routed to the Right Reviewer
Expedited, non-serious, literature and follow-up cases each follow their own route.
Converted to E2B(R3) or E2B(R2), Validated
Every structured case is mapped to E2B data elements and exported as E2B(R3) or E2B(R2) XML, for each regulator and partner you report to. Schema, required elements and regional rules are checked before anything leaves.
Both E2B Formats
The same case exports as E2B(R3) or E2B(R2), with the element mapping kept for each.
Validated Before Submission
Schema, required elements, regional rules and MedDRA version are checked, and errors are fixed first.
Every Element Traced
Each E2B element links back to the source document and the sentence it came from.
This record is append-only. Nobody can change or delete an entry.
- Received from the call centercall-summary-0418.pdf, 2 pages
- Case extracted by ThinkPV6 fields, 1 event, serious
- Coded by ThinkPVMedDRA: Jaundice, Chromaturia
- Verified by Maya RuizIntake QC, 6 of 6 fields
- Medical review by Dr. Rhea PatelSerious, unexpected, possibly related
- Exported by ThinkPVE2B(R3), validated, 0 errors
- Submitted by Leo GrantGateway acknowledgment received
Every Case Decision on the Record
From the report arriving to the E2B file leaving, every extraction, edit, decision and signature is recorded with who or what made it and when. The audit trail cannot be edited by anyone, including administrators.
Inspection-Ready
Case histories, signatures and source links export for an inspection or an audit.
Agents and People Side by Side
Agent actions and human decisions are recorded together, each with its author.
Compliance Reporting
On-time submission rates, case volumes and turnaround by queue, ready for your safety committee.
Built for Pharmacovigilance
ThinkTrends runs in your environment, keeps a qualified person on every decision, and keeps your data out of model training.
A Person Signs Every Case
ThinkPV extracts, codes and triages. Case processors and safety physicians confirm, and every decision is recorded under a name.
Your Data Never Trains a Model
Cases, labels, trial data and documents are used to do the work and nothing else.
Runs in Your Environment
Deploy in your cloud or data center, with patient and product data kept inside your boundary.
Validation Evidence Included
Each agent version is tested against your acceptance criteria, and the results are ready for your validation file.
21 CFR Part 11 Aligned
An audit trail no one can edit, electronic signatures with their meaning, and access by role.
Configured, Not Coded
Rules, routes, checks and thresholds are settings your administrators can read and change.
ROI calculator
Compare the costs under
your billing model.
Enter your case volume, labor costs, and billing terms. The calculator starts with a 60% reduction in eligible processing and preparation time, then deducts added review effort and solution costs.
Case-by-case scenario
Year 1 estimateEstimated net financial benefit / year
- Preparation per case
- —
- Total effort incl. review
- —
- Annual case revenue, unchanged
- —
- Original-work billing change
- —
- Revenue from redeployed hours
- —
- Paid labor cost avoided
- —
- First-year solution & setup costs
- —
Enter your assumptions to model the value.
Capacity value is shown separately and is not added to financial benefit. Results exclude taxes, financing, and costs outside the entered scope.
Calculator assumptions and formulas
Time and capacity
Net hours released = annual cases × (preparation minutes × reduction % − added review minutes) ÷ 60.
- Total time before = preparation + existing review. Total time after = reduced preparation + existing review + added review.
- Labor-capacity value = net hours released × loaded hourly cost. This is an economic equivalent, not automatically cash saved.
- The example assumes 25 preparation minutes per case. A 60% reduction saves 15 minutes. The separate 15-minute review allowance stays unchanged.
Financial benefit and ROI
- Per case: contract revenue remains unchanged at the same volume. Paid cost avoided = positive released hours × cost-realization share × labor rate. No extra case sales are assumed.
- Hourly: original-work revenue change = −net hours released × billing rate. Redeployed revenue = positive released hours × redeployment share × billing rate. Paid cost avoided applies only to the remaining hours and the selected cost-realization share.
- If added review exceeds time saved, the extra hours incur full loaded labor cost. Hourly mode assumes those extra hours remain billable at the entered rate; check your contract.
- Net benefit = revenue change + paid cost avoided − annual solution cost − setup cost. ROI = net benefit ÷ (annual solution cost + setup cost). ROI is undefined when investment is zero.
The example assumes a full year at the entered run rate, a common loaded labor rate, and no change in case mix or quality. Add ramp-up, specialist review rates, new constraints, and other costs to your business case. Solution costs and rates are editable placeholders, not vendor pricing. Billing and reconciliation benefits are not included.
Related Modules
The platform modules these solutions are built on.
Frequently Asked Questions
Common questions about ThinkTrends ThinkPV.
What is ThinkPV?
+
ThinkPV is ThinkTrends’ agentic AI for pharmacovigilance case intake. It reads adverse event reports from every channel and format, confirms each is a valid case, extracts and codes it in MedDRA, routes it by seriousness and deadline, and converts it to E2B(R3) or E2B(R2). Your safety team reviews and signs every case.
How is ThinkPV different from the automation in safety databases?
+
Safety databases such as Argus, ArisGlobal LifeSphere or Veeva Vault Safety manage cases once they are entered. ThinkPV does the work before that: reading unstructured reports, extracting and coding every field with its source sentence, and sending structured cases to the safety database you already use.
Does ThinkPV replace our safety database?
+
No. ThinkPV handles intake, extraction, coding, triage and E2B conversion, then sends structured cases to your existing safety database.
Which E2B formats does ThinkPV export?
+
E2B(R3) and E2B(R2) XML, validated against the schema and the regional rules you configure, plus structured case data for your safety database.
Can ThinkPV process unstructured adverse event reports?
+
Yes. Free-text narratives, emails, scanned faxes, handwritten forms, call summaries and journal articles are read into structured cases, and each field links to the sentence it came from.
Who decides seriousness and causality in ThinkPV?
+
ThinkPV suggests seriousness, expectedness and deadlines from your rules. Case processors and safety physicians confirm them, and causality is always assessed by a person.
Is our data used to train AI models?
+
No. ThinkTrends uses case data only to do the work, inside your own environment. Your data is never used to train, tune or evaluate a model.