The End Stage Renal Disease (ESRD) program under Medicare is highly data-intensive and data-driven. My organization has a long history of leveraging clinical and operational data to improve patient outcomes. Our data scientists are exploring innovative ways to use all those data to enhance predictive and personalized models of care.
Traditional health care data includes objective clinical data such as blood pressure, heart rate, temperature, body weight and laboratory values. These data are commonly used to assess patient trends and to build predictive models. In this way, historical objective data are used to change therapy and provide interventions to improve patient outcomes.
The ESRD hemodialysis (HD) patient treatment time is rich with subjective data, such as symptom and patient-reported data, which can also add to care and outcome models. ESRD patients on HD receive in-center dialysis treatments 3 days a week. Patients are monitored and cared for throughout this HD treatment by a multi-disciplinary team, including a registered nurse (RN) and a patient care technician (PCT). During the 4-hour dialysis treatment time, these clinicians have the opportunity to document data including clinical assessments focused on patient symptoms.
What can subjective data, like patient symptoms, add to predictions about patient outcomes?
In a recent data innovations project, we looked beyond traditional objective clinical data and examined patient symptoms documented in the dialysis electronic health record (EHR).
- We mined treatment data for patients who started dialysis between January 2013 and June 2015 to learn about specific treatment-related clinical symptoms and complaints, such as shortness of breath, chest pain, and cramping during the first year of HD.
- We included multiple data sources, such as symptom check boxes recorded by RNs and PCTs in the point-of-care electronic system.
- We also used text mining to extract symptom data from free text nursing notes.
We then analyzed the relationship of the presence of these specific symptoms to clinical outcomes, such as patient mortality and hospitalization during the second year of HD. Using survival analysis, we found a strong correlation between the symptoms at the point-of-care and outcomes in patients.
Here are a few findings:
- Patients who have a documented complaint of shortness of breath during HD treatment were ten times more likely to die compared to patients who did not complain about shortness of breath even after taking into account a number of other patient related characteristics.
- Patients who complained about cramping during treatments were five times more likely to die compared to patients who did not complain about cramping.
- Patients who had more symptoms in general had higher mortality risk compared to patients who had no symptoms or lesser symptoms.
Including subjective data has provided many valuable insights and opportunities.
- Identification of patients at increased mortality risk creates the opportunity to design interventions to improve patient outcomes.
- Expanding the dataset to include nontraditional, subjective clinical symptoms provided a robust predictive model for patient outcomes.
- Symptom recognition and documentation by an experienced clinician adds a new data dimension that values clinical observation and judgment.
Finding innovative ways to include this clinical judgment in the analytical model should improve model performance.
We treat more than 180,000 patients with ESRD every year. This care includes over 250 million dialysis treatment records and a vast dataset. We have the opportunity to use traditional and novel data elements to gain insights that will help us and others to improve patient care and outcomes.
- What are you doing to leverage additional data points in your records to improve patient outcomes?
- How have you used new types of data to build your predictive models?
Join the HIMSS C&BI Community today to learn more about applying data and analytics to improve health, and other critical topics that will help you on your journey to Turn Data to Action.
*Authors of this blog post are: Ahmad Sharif, MD, MPH, SCPM, member, HIMSS Clinical & Business Intelligence Committee; vice president, clinical health information technology; Sheetal Chaudhuri, senior director, knowledge management and analytics; and Dugan Maddux, MD, vice president, kidney initiatives. The three bloggers are with Fresenius Medical Care.