Fully one-third of the money spent on healthcare in the United States – roughly $750 billion a year – is wasted on things of no value. That recognition is driving the most significant trend among healthcare providers today: A concerted effort to simultaneously improve quality and control cost. That dual goal marks a distinct change from past efforts, which focused only on controlling utilization and not improving clinical outcomes.
To get to the goal, providers must focus first on the population of patients that accounts for approximately 70 percent of the healthcare spend in the United States: those with established chronic disease. Chronic disease patients, such as those with congestive heart failure (CHF), chronic obstructive pulmonary disease (COPD), or diabetes, experience frequent hospitalizations and other expensive acute care interventions. Thus, targeting established chronic disease is an important effort in transforming healthcare, improving outcomes, and curbing expenditures.
Many of the hospital admissions for chronic disease patients occur after long intervals during which they had no contact with their clinicians. The idea of monitoring such patients at home, to close that communication gap, has been around for at least 20 years. By using devices to collect and record data on measurements such as weight, blood sugar, and blood pressure, the hope was that patient deterioration could be identified early, allowing an intervention that would reduce the need for hospitalization. Some pilot programs for small groups of patients were able to reduce avoidable hospitalizations but were so labor intensive and expensive that they could not be scaled to large numbers of patients. These programs often involved frequent phone calls or home visits to follow up on alerts, many of which were false positives. A sustainable program has to be effective both clinically and economically.
Early versions of home monitoring systems for patients – “first generation,” if you will – made numerous simplifying assumptions. They focused on patients with one disease, such as CHF, and ignored the patients’ comorbidities, even though older patients typically have multiple chronic diseases, such as CHF and COPD or diabetes. The early systems further assumed that the state of the one disease could be represented by a single measurement, such as patient weight for CHF, where it’s assumed that an increase in weight reflects fluid retention and a worsening of the condition. Additionally, a one-day change in the measure, such as a two- or three-pound weight gain from one day to the next, was the threshold for an alert that the patient was deteriorating. Those assumptions served to simplify the monitoring effort but resulted in many false positives (a measured weight could vary from one day to the next for many reasons) and missed opportunities to detect deterioration early enough to intervene, reverse the process, and keep the patient out of the hospital. The rate of false positives currently hovers around 70 percent, which means a substantial amount of staff time is spent chasing down false positives.
A program that treats all patients in the same way but benefits only a few patients may provide clinical value, but its sustainability would be threatened by high cost. An important component of effective prevention programs such as home monitoring of patients is identifying the patient for whom the program has value. Careful choices about target populations and the nature of the program will make the prevention strategy more effective clinically and economically.
An effective approach has several components:
- Determine the range of diagnoses in a target group of patients to maximize the potential benefit.
- Assess the patients for impactibility. Because monitoring patients who won’t benefit from monitoring adds costs with no value, choosing patients more likely to be helped by monitoring is important. For example, a patient with a chronic diagnosis such as COPD but without any acute care encounters related to that disease is not a prime candidate for data monitoring because there are no acute events to prevent.
- Track data points for multiple parameters, chosen based on the patient’s diagnoses. A patient with just COPD might be monitored for pulse, blood pressure, oxygen saturation, and peak flow rate, while a patient with COPD and CHF could also be monitored for weight.
- Limit the patient’s burden. Because patients may be reluctant to participate in data monitoring if they are required to work with too many devices, limit the number of devices requiring active participation on the part of the patient while still collecting sufficient data. Monitor device technology is changing rapidly, with new devices making multiple measurements passively, not requiring the patient to actively perform the measurements, which may improve participation on the part of the patient.
- Track longer-term trends as well as day-to-day changes in order to detect subtle progression that would otherwise be missed.
As in any big data initiative, collecting the data is relatively easy. But unless that data is transformed into actionable insights, they provide little value. Select an analytics engine that can detect patterns in the collected data and help you predict which patients are likely to deteriorate. That kind of advance warning allows clinicians to initiate a modest intervention to reduce the likelihood of serious deterioration.
Personalized healthcare, using data to produce insights about an individual patient’s condition, helps improve decision making and results in care more likely to be beneficial to the patient. The same principle, of course, applies to home monitoring. Not all CHF patients are the same. A patient with CHF and COPD is likely to have different predictive patterns than a patient with CHF alone.
Improving the home monitoring process requires learning from experience. Machine learning allows predictive power of home monitoring to improve over time. The outcome of each prediction, whether it was accurate or false, is fed back into the system and improves future predictions. Analytics systems can generate their own suggestions for improvement (unsupervised machine learning) or incorporate expert input (supervised or expert-enhanced machine learning). In this way, the longer machine learning systems works with a group of patients, the better they become at predicting deterioration. The result is an entire toolbox well-suited to the job of keeping each individual healthier at home.
Martin S. Kohn, MD, MS, FACEP, FACPE is Chief Medical Scientist of Sentrian, the Remote Patient Intelligence Company, and a world-renowned expert on healthcare population analytics and the role of expert systems in the clinical decision process.
Subscribe to Data Informed for the latest information and news on big data and analytics for the enterprise.