Report Writing

You can use AI to help you write reports. Here is an overview of how I use BastionGPT, a HIPAA-compliant version of ChatGPT, to help write my reports. For Bastion, you can adjust the “temperature” or creativity of the chat, which ranges from 0 (precise) to 1 (creative). I have found that a temperature of 0.2 works best, but experiment with what works for you!

Developmental History: I give the GPT a series of prompts, starting with “Please act as an expert US child neuropsychologist. First, I am going to give you 3 developmental histories to read and analyze for structure, tone, syntax, style, and reading level. You do not need to output your analysis. Does this make sense?”

I then provide 3 examples of developmental histories I have written in the past so that it “learns” my style. Note that Bastion does not learn this style across chats. This “context” is specific to one conversation.

Finally, I give another prompt: “Thank you. Please create a new developmental history based on the following intake interview. Do not make anything up. Adhere closely to the structure of the previous examples. Sections should include: developmental history, academic history, family history, social-emotional history, medical history and "strengths and goals". Refer to the patient by first name. Write in a positive, professional tone, as in the previous examples.

I then copy and paste my typed notes from my intake appointment. Bastion GPT produces a developmental in my report writing style. I copy and paste that interview into my report and thoroughly edit it line by line.

Classroom Observations: I have tried using Bastion to turn my school observation notes into a narrative. So far, I’ve preferred to do this part myself. I like to have control over what I include and how it is phrased, and I haven’t cracked getting Bastion to write in the neutral, play-by-play style I prefer.

Behavioral Observations: I have a pretty good template in place for this part, so I don’t find benefit from AI tools here.

Tests Scores and Tables: Here, the formatting is pretty specific and not conducive to copy and pasting from Bastion. Plus, many of my scores are on a summary scoresheet I use during testing. So far, entering these in myself has made the most sense. However, I also write short interpretations next to the scores. For projective tasks like sentence completion or the Robert’s Apperception Test, BastionGPT actually does a pretty solid job at extracting themes from responses.

Summary: Similar to the developmental history, I have found that a series of prompts, rather than a single, long, detailed prompt has worked best for me. So, I start with: “Please act as an expert US child neuropsychologist. First, I am going to give you 3 summaries to read and analyze them for structure, tone, syntax, style, and reading level. You do not need to output your analysis.”

Then, I provide 3 examples of summaries from a range of cases (ages, presenting problems). I have saved 3 summaries I think are some of my best work, and saved them as a shortcut prompt I can use over and over.

Next, another prompt: “"Thank you. Please create a new summary based on the following assessment data. Do not make anything up. Adhere closely to the structure of the previous examples. For example, start the strengths paragraph with ([Name] is a highly intelligent child with many strengths), do not include much information from the developmental history, and start each diagnostic paragraph with the area being evaluated instead of the diagnosis (e.g., a key goal of this assessment was...). Use a friendly and professional tone, but avoid highly technical jargon so that the summary is accessible to educated parents who may be unfamiliar with learning disabilities, developmental conditions, or mental health terminology. Include one paragraph of strengths and one paragraph for each diagnosis. For this case, the diagnoses are [DIAGNOSIS] and [DIAGNOSIS]. Put the diagnosis in bold font. After each diagnosis, include one sentence about effective treatment or intervention recommendations to instill hope, as in the previous summary examples. Here is the assessment data:”

Here’s the cool part. I now copy and paste my entire developmental history narrative, behavioral observations, and test tables with results. BastionGPT writes the summary in my style. As always, I edit thoroughly! However, this has been extremely helpful. It writes closely to my style and structure because it has “learned” my writing style from the previous prompts and examples.

Key Strengths: A similar process to the above sections. Here is my final prompt: “Thank you. Please create a new "key strengths" section based on the case we have been discussing. Use a 3 paragraph structure with the following key strengths: [cognitive], [academic] [emotional]. Do not make anything up. Adhere closely to the writing style of the previous examples:”

What’s awesome about this part is that it remembers the whole context of the case so far, and the report we’ve been writing together.

Key Weaknesses: Same process as key strengths.

Recommendations: I still write my own recommendations. So far, my strategy has been to tell Bastion my recommendations and ask it if there is anything else I should consider as a recommendation.

That’s it. I am still getting used to these tools and exploring how they work best. Even with some experimentation as I go, Bastion has easily shaved 1-2 hours off each of my reports. I anticipate that this will become even more efficient for me in the future!

If you’ve tried this out for your self, let me know what you think in the Reddit community!