Fitbit Recovery Tracking for Post-Op Patients

ISEF Category: Translational Medical Science

Subcategory: Other  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

A smartwatch can catch tiny changes in movement, sleep, and heart rate long before a person says they feel worse. After surgery, those small changes may map to recovery speed, pain, or complications. That means your project can ask a real clinical question with data you can actually access. You do not need a hospital lab to start thinking like a medical data researcher.

What Is It?

This project asks whether wearable data can act like a recovery signal after surgery. A digital biomarker is a number pulled from sensor data that helps describe a health state. In this case, you would look at daily patterns like step count, sleep time, heart rate, or resting activity, then see how those patterns change after surgery.

Think of recovery like a road trip, not a single checkpoint. Some patients bounce back fast. Some move slowly. Some improve, then stall. Unsupervised clustering groups recovery paths that look similar without you telling the computer what the groups should be first. That can help you spot patterns that may match slower recovery or higher risk.

Why This Is a Good Topic

This is a strong science fair topic because the main question is measurable, and the data can come from public datasets instead of a wet lab. You can compare real recovery traces, test whether wearable features separate groups, and use statistics or machine learning to check your results. The project connects to a real medical problem, post-operative monitoring, but still lets you work with data you can handle as a high school student.

Research Questions

• How does post-operative step count recovery differ across surgery types in public wearable datasets?
• What is the effect of using sleep features versus activity features on clustering recovery trajectories?
• Does combining heart rate and activity data improve the separation of slow and fast recovery groups?
• To what extent can early post-surgery wearable data predict later recovery pattern membership?
• Which wearable-derived features change first after surgery, and which return to baseline last?
• How does the choice of clustering method change the number and stability of recovery groups?

Basic Materials

• Laptop with enough storage to handle CSV files.
• Spreadsheet software or Python with pandas and scikit-learn.
• Public surgery and Fitbit dataset downloads.
• Data dictionary or codebook for each dataset.
• Notebook for tracking feature definitions and cohort rules.
• Stable internet connection for downloading files and reading papers.

Advanced Materials

• Laptop or workstation with Python, Jupyter, and sufficient RAM for larger datasets.
• Access to public wearable and surgery datasets with linked outcome labels.
• Statistical software such as R or Python for survival or longitudinal analysis.
• Version control software such as Git for tracking code changes.
• Optional access to cloud computing or university servers for larger model runs.
• Graphics software for plotting longitudinal recovery clusters.

Software & Tools

• Python: Cleans wearable data, builds features, and runs clustering models.
• Jupyter Notebook: Keeps code, notes, and figures in one place while you iterate.
• pandas: Organizes time-stamped wearable records into usable tables.
• scikit-learn: Supports clustering, scaling, and basic model checks.
• R: Handles mixed-effects or longitudinal statistics if you want a second analysis path.

Experiment Steps

1. Define the recovery outcome you want to study, then decide how you will label time after surgery.
2. Select one public dataset and write clear inclusion rules for patients, procedures, and wearable data completeness.
3. Engineer a small set of recovery features from daily wearables, then decide how to normalize them across patients.
4. Choose an unsupervised method that groups similar trajectories, then plan how you will test cluster stability.
5. Compare cluster membership with known clinical outcomes or proxies for delayed recovery.
6. Build a figure set that shows individual trajectories, cluster summaries, and the meaning of each group.

Common Pitfalls

• Mixing surgery types without accounting for different baseline recovery speeds, which hides real patterns.
• Using raw wearable counts without normalizing for missing wear time, which makes low-activity days look worse than they are.
• Clustering on too many features at once, which can make groups look different even when they are not stable.
• Ignoring irregular sampling after discharge, which breaks trajectory comparisons across patients.
• Treating cluster labels as clinical truth instead of a pattern to validate against outcomes, which weakens your conclusions.

What Makes This Competitive

A stronger version of this project goes beyond one simple cluster plot. You would pre-register your feature choices, test multiple clustering methods, and check whether the same recovery groups appear in different datasets or patient subsets. You could also compare wearable-only models against models that use a few clinical variables, then ask what signal wearables add on their own. That kind of careful validation makes the work feel like real translational research, not just data mining.

Project Variations

• Focus on sleep recovery instead of activity, and compare how sleep duration, sleep regularity, and wake time shift after surgery.
• Use only the first week after discharge to test whether early wearable patterns can flag delayed recovery sooner.
• Compare different surgery categories, such as orthopedic, abdominal, or cardiac cases, to see whether recovery trajectories cluster differently.

Learn More

• PubMed: Search for review articles on digital biomarkers, post-operative monitoring, and wearable recovery studies.
• NIH NCCIH or NIH PubMed tutorials: Learn how to read human-subjects research and study design basics through free NIH resources.
• Stanford Medicine research pages: Search for public discussions of wearables, postoperative recovery, and clinical data science projects.
• scikit-learn documentation: Read the official clustering and model evaluation guides for Python-based analysis.
• R documentation and vignettes: Find free guides for longitudinal analysis, clustering, and data visualization in R.