📄 medicine

A Multi-site Dashboard for Evaluating HiGHmed Cardiovascular Sensor Data from Apple Watch

This paper presents a centralized R Shiny dashboard that integrates Apple Watch sensor data with clinical records from a multi-site HiGHmed study to enable cardiologists to systematically evaluate cardiovascular metrics and detect early signs of clinical worsening in heart failure patients.

Original authors: Khalid Olusola Yusuf, Tabea Agnes Steinbrinker, Katharina Joerss, Jendrik Richter, Sabine Hanss, Tanja Zeppernick, Kim Katrin Werle, Erik Tute, Robert Freund, Georg Fette, Udo Bavendiek, Dirk Grosse-M

Published 2026-07-07

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Khalid Olusola Yusuf, Tabea Agnes Steinbrinker, Katharina Joerss, Jendrik Richter, Sabine Hanss, Tanja Zeppernick, Kim Katrin Werle, Erik Tute, Robert Freund, Georg Fette, Udo Bavendiek, Dirk Grosse-Meininghaus, Stefan Störk, Dagmar Krefting

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a team of doctors trying to solve a mystery: Why do some heart failure patients suddenly get sick enough to need hospital care?

Usually, doctors only see these patients when they walk into the clinic or the hospital. It's like trying to understand a movie by only watching the scenes where the characters are in the hospital room; you miss the buildup, the small clues, and the slow decline that happened at home.

This paper introduces a new tool—a digital "mission control" dashboard—designed to fill in those missing scenes. Here is how it works, broken down into simple concepts:

1. The "Smart Watch" Spy

The study gave 93 patients Apple Watches to wear. Think of these watches as 24/7 spies that quietly report back to the doctors' headquarters.

What they spy on: How fast the heart is beating and how many steps the patient takes every day.
The extra homework: Patients also had to manually type in their weight, blood pressure, and answer a short questionnaire about how they feel (like "Do my feet feel swollen?" or "Can I breathe easily?").

2. The "Big Picture" Dashboard

The researchers built a special computer screen (a dashboard) that acts like a central command center. Instead of looking at piles of paper or separate spreadsheets, all the data from four different hospitals gets mashed together into one view.

The dashboard has three main "rooms" or tools:

Room 1: The Timeline Wall (Observation Overview)
Imagine a giant wall chart where you can see everything happening at once. It stacks the heart rate, step count, weight, and "feeling" scores on top of each other.
- What it shows: It can spot the "slow fade." For example, it might show that a patient's steps dropped below 2,000 a day and their "feeling" score got worse before they ever went to the hospital. It also highlights "blank spots" where the patient forgot to wear the watch or didn't write down their weight, acting like a quality control check.
Room 2: The Heartbeat Zoom (Rhythm Analyzer)
This tool lets doctors zoom in on the heart rate hour-by-hour.
- The Analogy: It's like looking at a weather map that shows the temperature every hour, rather than just the average for the day.
- What it shows: The dashboard can compare a "normal" day to the day a patient was admitted to the hospital. In one case study, the watch showed the heart rate dropping significantly on the day of hospitalization compared to the days before, giving a clear signal of what happened.
Room 3: The Connection Detective (Correlation Module)
This tool asks: "Do these two things move together?"
- The Analogy: It's like checking if the more you pedal a bike, the faster your heart beats.
- What it shows: The dashboard confirmed what we expect: when patients walked more steps, their heart rate went up (a strong connection). It also checked if weight changes were linked to heart rate or blood pressure, finding that these relationships were weak or non-existent in the short term.

3. What They Actually Found

The paper doesn't claim this tool cured anyone or that it will automatically save lives in the future. Instead, it proves the tool works as a way to look at data:

It sees the decline: It successfully showed that patients often get worse (lower steps, lower "feeling" scores) before they end up in the hospital.
It catches errors: It clearly showed gaps where patients didn't wear the watch or forgot to measure their blood pressure, helping doctors know when data is missing.
It connects the dots: It successfully combined data from Apple Watches, hospital records, and manual notes into one picture.

The Bottom Line

Think of this dashboard as a unified telescope for heart failure patients. Before, doctors had to look through four different telescopes (one for each hospital) and try to guess the story. Now, they have one powerful lens that shows the continuous story of the patient's heart, steps, and feelings, helping them see the "movie" of the patient's health from start to finish, rather than just the final scene.

Technical Summary: A Multi-site Dashboard for Evaluating HiGHmed Cardiovascular Sensor Data from Apple Watch

Problem Statement
The management of chronic heart failure requires the timely detection of clinical deterioration to prevent hospitalizations. While healthcare dashboards exist to visualize electronic health records (EHR) and population-level metrics, there is a significant gap in monitoring outpatients using continuous ambulatory data. The increasing volume of data generated by wearable devices (e.g., Apple Watch) presents challenges regarding data quality, adherence, and the integration of these streams with traditional clinical records. Furthermore, the multi-site nature of the HiGHmed cardiology study creates a need for a unified, collaborative tool that allows clinicians across different institutions to access and analyze integrated datasets without compromising data security or usability.

Methodology
The authors developed a centralized R Shiny dashboard designed to aggregate and visualize data from the HiGHmed use case cardiology sensor sub-study. The methodology encompasses four key phases:

Data Collection and Sources:
- Participants: Data was collected from 93 participants across four study sites (Göttingen, Hannover, Cottbus, and Würzburg). The initial analysis focused on 12 participants who experienced hospitalization events during the study period.
- Sensor Data: Apple Watches continuously recorded step counts (via accelerometer) and device heart rate (HR).
- Patient-Reported Outcomes (PROs): An integrated app (medPower®) captured manual measurements of resting HR, weight, blood pressure, and Kansas City Cardiomyopathy Questionnaire (KCCQ-12) scores.
- Clinical Data: Hospitalization records (admission/discharge dates and medical indication codes) were integrated to distinguish cardiovascular events from non-cardiovascular ones.
- Interoperability: Data was standardized using Fast Healthcare Interoperability Resources (FHIR). Two Medical Data Integration Centers (MeDICs) received data as bundled FHIR binaries, while two others mapped FHIR resources to openEHR compositions.
Data Processing Pipeline:
- A pre-processing pipeline in R (Version 4.5.0) harmonized data from FHIR documents and openEHR compositions.
- Aggregation: Device HR was aggregated into daily averages; step counts were summed for daily totals. Hourly HR was preserved for rhythm analysis. Manual measurements were kept as individual daily observations.
- Scoring: KCCQ-12 summary scores were calculated using standard algorithms at specific intervals (baseline, 4, 8, 12 weeks, and every 3 months thereafter).
- Integration: All datasets were merged with hospitalization timestamps to enable temporal alignment.
Dashboard Architecture:
- Developed using the R Shiny framework, the tool was refined through iterative feedback sessions with three cardiologists.
- Privacy: To ensure data protection, the dashboard was distributed as a bundled package with embedded dependencies for local setup. Patient IDs were limited to the last three characters.
- Modules: The interface features three primary analytical modules:
  - Observation Overview: Displays stacked trends for all metrics (device HR, steps, manual vitals, KCCQ-12) overlaid with hospitalization events. It allows toggling between a multi-metric daily view and detailed KCCQ-12 responses.
  - Heart Rhythm Analyzer: Presents hourly device HR patterns using a three-day rolling window (selected day plus two preceding days), with manual resting HR overlaid as reference points.
  - Correlation Module: Visualizes pairwise Pearson correlations between metrics using a 14-day sliding window, allowing clinicians to explore changing relationships over time.

Key Results

Visualization of Deterioration: The dashboard successfully identified pre-hospitalization patterns in the 12 hospitalized patients. These patterns included declining KCCQ-12 scores (hovering around 35–55) and reduced physical activity (daily steps frequently dropping below 2,000).
Heart Rhythm Analysis: Hourly analysis revealed a decline in both device and manual heart rates on the day of hospitalization (dropping below 50 bpm) compared to baseline levels (50–70 bpm).
Correlation Findings: The correlation module confirmed expected physiological relationships, such as a strong positive correlation between steps and device HR ( $r \approx 0.8$ ). It also highlighted weaker or moderate correlations between other parameters (e.g., weight and HR).
Data Quality Insights: The visualization exposed significant gaps in sensor recordings and inconsistent adherence to manual measurements (e.g., missing blood pressure and weight entries), demonstrating the dashboard's utility in identifying data quality issues in real-time.

Significance and Claims
The paper claims that this dashboard provides a practical, unified tool for multi-site cardiovascular data evaluation. Its primary significance lies in:

Integration: Bridging the gap between continuous ambulatory wearable data, manual clinical measurements, patient-reported outcomes, and hospitalization records.
Collaboration: Enabling clinicians across multiple sites to collaboratively assess patient trajectories rather than reviewing parameters on a case-by-case basis.
Pattern Recognition: Facilitating the transition from isolated parameter review to systematic multi-metric assessment, which aids in identifying patterns of clinical worsening (e.g., declining activity and KCCQ scores) prior to hospitalization.
Data Quality Monitoring: Serving as a mechanism to detect non-compliance or device failures promptly, preventing retrospective data loss.

The authors position the tool as an instrument for exploratory analysis and retrospective pattern recognition. They modestly note that while the dashboard supports current clinical oversight, its utility for prospective evaluation would require future integration of threshold-based alerting mechanisms and machine learning approaches to establish personalized baselines.

1. The "Smart Watch" Spy

2. The "Big Picture" Dashboard

3. What They Actually Found

The Bottom Line

Technical Summary: A Multi-site Dashboard for Evaluating HiGHmed Cardiovascular Sensor Data from Apple Watch

More like this