Designing Healthcare IoT Systems

The “Internet of Things” (IoT) has exciting near-term prospects in healthcare.  But what does that mean, and how can we most efficiently realize its potential?

Healthcare IoT can take many forms.  Here, we’re referring to sensors deployed onto or inside a human body, that send their data readings to the cloud, which then communicates processed data to clinicians for action.

It sounds straightforward, especially if you’re a technologist, because most of the words in the previous sentence are technology words: “sensor,” “data,” “cloud,” “communicate,” and “process.”

But notice that other word: “action.”  It’s the last word because it’s the system’s entire reason for being.  If you’re designing your IoT system, and you aren’t clear idea what the actions are, how well they work, and, crucially, how the data are tied to the actions, then pause.

What’s Being Tried?

Let’s take an example: the recently published BEAT-HF study of heart failure patients.  All patients got their usual care, but half were randomly selected to additionally get coaching telephone calls plus an IoT solution that acquired daily blood pressure, weight, and oxygen saturation – exactly the parameters cardiologists follow in their heart failure patients.

Unfortunately, the trial showed no benefit of the IoT solution.  Compared to the control group, the IoT patients died just as often, and they came into the hospital just as often.  This is not the first trial to show such failures, and it is fortunate that BEAT-HF did not harm the subjects by wasting physician time and distracting them from interventions that could actually benefit patients.

A Better Mouse-Trap

But now let’s look at a different system, also aimed at heart failure patients.  Here, a small Bluetooth-enabled pressure sensor is placed into the pulmonary artery via catheter.  (Pressure in the pulmonary artery is a key indicator of heart failure.)  Once a day the patients lies quietly in bed, near a Bluetooth receiver, and the sensor’s measurements of pulmonary artery pressure are sent to the cloud, and then to the cardiologist’s office.

In a randomized study of 550 patients, the patients who received the pressure sensor had their medications changed by the cardiologist 250% more times than the control group.  That is not a typo – 250% -- a remarkable change in the “action” step. But did all that extra “action” help? Yes!  Patients with the pressure system experienced 43% fewer deaths, and 57% fewer heart failure hospital admissions.  The word “spectacular” underestimates this accomplishment, especially given the statistics that, among fee-for-service Medicare enrollees, heart failure is responsible for 39% of all deaths, and for 42% of all hospital admissions.

Wrap-up

If you are designing an IoT system for healthcare, what lessons can you draw?

  • (1) Sensor choice matters.  A lot. Try to obtain data from the core of the disease process, not peripheral or indirect indicators.
  • (2) Merely increasing the data collection frequency, as BEAT-HF tried, may not be beneficial. “Big data” is not a panacea.  Data quantity may not make up for only marginal improvements in data quality.
  • (3) Patient choice matters.  BEAT-HF failed in its general population of heart failure patients, but might have succeeded with certain subgroups of patients.  For example, patients having both heart failure and depression might disproportionally benefit from the Hawthorne effect (increased attention) that telemonitoring can provide.
  • (4) Test your system with a randomized trial.  It is increasingly clear that other study designs are unreliable when evaluating tele-health systems.

Although technology terms may dominate the definition of a healthcare IoT system, the single clinical word dominates its success.