Train a wearable-data warning model for a thyroid flare

Use long-term Apple Watch and Whoop data to build a model that may flag a thyroid episode before lab results do.

Done with Claude Code· Advanced

What they did
A poster exported 9.5 years of Apple Health records from Apple Watch and paired them with Whoop data, then marked the stretches that lined up with hyperthyroid phases. They handed the dataset to Claude Code and asked it to find patterns that might predict a flare. In comments, they said Claude tested a bunch of models for more than an hour, settled on XGBoost, and pared 53 engineered features down to three resting-heart-rate features. They also kept the most recent episode out of training so they could check it by hand.
How it went
They said the held-out backtest would have raised concern in early August, before labs confirmed the episode at the end of the month. They described the output as a heads-up to request confirmatory bloodwork, not as a diagnosis or a reason to change medication automatically.
Tips
This approach depends on a very long personal data history and careful labeling of past flare periods. If you try it, keep the result as a screening aid and make sure a person reviews anything that could affect treatment.

Try it yourself with Claude Code

I’ve exported my wearable records from [Apple Watch/Whoop/other devices] into [file or folder path], covering [date range]. Use these data to look for patterns that show up before my past [health event] episodes, using the dates I provide as labels. Try several model types, choose the one that performs best, and report which features matter most. Hold out my most recent episode for a simple backtest so I can see whether the model would have warned me early. Before you do anything that would send, submit, delete, or change records, pause and ask for my approval. Finish with a plain summary of how well it worked and where it could fail, and treat it as a personal screening tool rather than a diagnosis.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on reddit.com ↗

2 more reports of this

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.