This workflow is designed for
Clinical / Regulatory Writers in pharmaceutical companies
CRO delivery teams
The real-world problem
A Clinical Study Report is a critical document that summarises a clinical trial and its results. Errors can lead to delays and losses in the millions of dollars.
Lack of predefined templates
Writers start drafting from scratch from empty template.
Time-consuming manual effort
There exists clear mapping between the different sections/tables of the source documents and the various sections of the CSR.
Despite this, drafting a CSR is challenging and a resource-intensive exercise. Manually rewriting, copy-paste, summarising, interpretation is subject to human errors.
Risk of incomplete information
Non-data-driven sections (sections 1-9 in the ICH template) have to be rewritten, summarised very carefully from the different parts of the trial protocol document.
Missing crucial information such as study objectives, inclusion & exclusion criteria can have huge consequences.
High-volume of data to parse and interpret consistently
Data-driven sections (sections 10 onwards) require mindful and consistent interpretation of the statistical outputs of the study. These sections convey the success or failure of the trial and therefore need to be accurate.
Writers spend a lot of time ensuring that they provide a clear and unbiased interpretation of the statistical outputs. Since studies might have multiple objectives and therefore as many statistical outputs, this puts a lot of strain on writers leading to errors.
How Asthra supports this document
Asthra supports CSR drafting
By converting approved source documents into structured section drafts, aligned to regulatory expectations, while keeping writers fully in control.
Asthra supports the writer
By generating the initial draft of the entire CSR from its source documents, Trial Protocol and the TFL (Tables Figures Listings). Writers can generate the content on MS Word section-by-section by just clicking the pre-configured prompts.
For each content output, Asthra understands the intent of the writer and intelligently retrieves appropriate granular (section/table/image) content from the right source document. Writers can transparently see which source content is being retrieved.
Writers can also get deeper insights by retrieving sentence level citations on-demand.
In case, writers are not happy with the output, then can immediately jump in and modify the prompt or select the right source content (file, section/table/image) to correct the output.
Inputs → Draft → Review
Template
Import the relevant CSR template that has been onboarded. It can either be either the standard ICH or Trancelerate templates or the customer’s custom template
Inputs
-
Upload Source documents for the CSR – Trial Protocol, Tables Figures Listings (TFL)
-
Documents can either be uploaded from the writer’s local machine or via integration with a DMS (Document Management System)
Drafting with Asthra
-
Writers can straightaway start generating content section-by-section by just clicking and running on the pre-configured prompts.
-
There is ZERO configuration or manual effort that the writer has to put before generating content
-
Writers can review the provenance and citations on the fly
2. Asthra generates the initial draft in 30 minutes
Evidence, traceability & controls
Asthra is purpose-built as a closed system. All content generated originates strictly from the source documents uploaded.
-
There has not been any hallucination so far. This is due to our approach on granular context engineering
-
LLMs are not allowed to use their own knowledge. Our guardrails and prompts ensure that is the case.
Asthra flags missing information with an appropriate response for example, “There is no relevant information in my context to answer this question.”
Asthra transparently shares the provenance for each output. That is, the writer can easily get information on which file and section/table/image in the file was referenced for the output.
In addition, writers can select any sentence generated and verify its citation i.e. which file, its page number, and relevant snippet of text from which the sentence was generated from.
Asthra enables writers to review and edit content generated.
-
Writers can directly modify content in the word document
-
Writers can easily modify the pre-configured prompts to generate alternate responses.
-
Writers can change the provenance i.e. change the file and section/table/image and regenerate alternate content
Inputs → Draft → Review
01
Template
Import the relevant CSR template that has been onboarded. It can either be either the standard ICH or Trancelerate templates or the customer’s custom template.
02
Inputs
-
Upload Source documents for the CSR – Trial Protocol, Tables Figures Listings (TFL).
-
Documents can either be uploaded from the writer’s local machine or via integration with a DMS (Document Management System).
03
Drafting with Asthra
-
Writers can straightaway start generating content section-by-section by just clicking and running on the pre-configured prompts.
-
There is ZERO configuration or manual effort that the writer has to put before generating content.
-
Writers can review the provenance and citations on the fly.
2. Asthra generates the initial draft in 30 minutes.
04
Human Review & QC
-
Writers are empowered to jump in and edit content at any point in the draft generation process.
-
If the provenance needs to be corrected for any output, writers can easily select the prompt and seamlessly select the file and then the section/table/image from that file and run the prompt to regenerate content.
-
Since Asthra is built as an Add-in on MS Word, writers can easily review and edit content during and after the initial draft is generated.
-
All transactions on Asthra are persistent and reviewers and auditors/inspectors can review anytime in the future.
What Asthra helps with —
and what it doesn’t
Where Asthra can help
All tasks related to retrieve or generate content where the context required is available. For example, copy-paste, change tense, summarisation, table & figure interpretation.
Asthra can also do simple calculation and math, and simple table manipulation with high accuracy. For example, o jJ
Asthra can compile information and present them in the form of tables.
Where Asthra cannot help
Complex calculation and math.
Retrieving information from huge tables or manipulation of huge tables.