The “Crystal Ball” of Healthcare Information Technology
One of the exciting new frontiers in medicine is clinical analytics, sometimes called streaming analytics, or clinical decision support software. This is basically the science of importing all of the thousands of bits of data, both literally, and figuratively, and using them to cohesively construct clinical pathways based upon computer algorithms designed The concept of a computer algorithm looking at various clinical or lab parameters and suggesting a clinical pathway is not particularly new. Early simplistic algorithms include the Modified Early Warning Score (MEWS) which provided feedback on clinical deterioration based upon heart rate, blood pressure, respiratory rate, temperature, and their AVPU score. The concept was simple: aggregate, and database the parameters and write software algorithms that would look for particular patterns that would suggest, (in the case of the MEWS algorithm). In the simplest case of an algorithm, obviously HR (heart rate) = zero = bad. Hundreds of parameters are available via the streaming broadcast from the monitor or gateway, everything from cardiac output to Arterial Blood Gas data (via a Laboratory interface), to ST segment amplitude and slope data derived from the ECG. All of these measurements take place on the patient monitor and are passed over in their communications stream which is usually in their native ‘language’. It takes a sophisticated date integration platform to parse out the parameter data streams and moreover, the waveform data.
This communications stream from the patient monitoring network, is comprised of monitors, and switches or hubs, connected on a VLAN, or a protected, isolated, critical network connection. These networks are typically not bridgable, ie, “you can’t get there from here”, from a hacker’s perspective, and thus are extremely secure. The devices communicate via either TCP/IP or UDP packets with devices assigned IP addresses communicating data to a server on a specific IP address and port. Most, if not all modern intensive (or “acute”) care monitoring devices provide some method of exporting this parameter data out to a SQL database via an interface referred to as “HL7”, which stands for Health Level Seven, a venerable “etched-in-butter” standard, which takes the parameter data at an interval and stores it in an XML file. These results objects, or “ORU” files contain man-readable parameters such as heart-rate, blood pressure, SPO2, 12 lead ECG interpretations, any parameter you want to define. A typical ORU results message might look like the image on the right. If it sounds antiquated, it is to some degree but HL7 was invented back in the day of RS232, as is most device data transfer to this day, even USB is a UART or Universal Asynchronous Receiver Transmitter, which is the heart of both RS232 and USB.
This sort of slow-moving data computation led the way to aggregating multiple methodologies including waveforms from ECG, invasive pressures, and capnography, where you can identify patterns, and apply algorithms, and adaptive workflows to better manage the patient. As an example. Therapeutic Monitoring Systems (TMS) in Ottowa Canada developed an entire workflow for ventilator weaning that provides not only a checklist, ala Dr. Atul Gawande’s book “Checklist Manifesto“, but also a score based upon tiny changes in the heart rate and respiratory rate as measured via capnography (CO2 expiratory outflow measurement). The simple score calculates the likelihood that the patient will successfully be able to be extubated and weaned from the ventilator successfully. This is an example of a new breed of algorithms that look at subtle changes in waveforms. Waveforms are difficult to extract and capture for a number of reasons:
- Every manufacturer has their own way of coding waveforms, and transmitting the data out. Some do it well, some do it not so well. Most require a gateway server and separate ADT (Admit, Transfer, and Discharge) server to keep the association between medical record number and name straight, as well as updates to patient’s location.
- Waveforms stored in high fidelity, or at least the same fidelity that the manufacturer provides. Most sampling rates don’t exceed 500 samples per second for ECG. Without a patented lossless waveform compression algorithm, such as Excel Medical’s lossless Huffman encoding, (think PK-Zip for physiological waveforms), these files would be extremely large.
- Finding a waveform storage standard is difficult. IHE is working on a waveform standard recommendation as part of their larger Patient Care Device or PCD Profile.
The waveforms that you see displayed on a patient’s monitor in a hospital is basically a series of measurements occurring ~250 times per second that depicts the waveform in a signed-integer format. If you plot these sets of parameter points back together they will reform the waveforms (tutorial here). Play them back 10 seconds at a time and you can re-create those waveforms just like the patient was still connected to the monitor. The Excel Medical “Bedmaster Ex” or “BMEX” software sit’s on the patient monitoring network listening to UDP packets extracting and copying to a SQL database every 2 seconds. With the BMEX system in the hospital, the clinicians have a ‘socket’ connection to the realtime and retrospective data. This data can be used to, among other things:
- Store the data forever for forensic purposes as long as you have 200 MB/pt/day storage. You can play it back and forth like a DVD, or search specific times, for instance leading to a sentinel event.
- Feed Analytics Applications (Apps) such as TMS ventilator weaning algorithm, and other published scores such as Visensia.
- Use the data to perform their own research and develop their own algorithms which they can patent and commercialize.
Apps for Healthcare
With a stable source of data, the BedMaster Ex platform provides a socket connection to all of the patient information “in the pipeline”. This data in addition to data collected from any number of disparate devices such as ventilators is inserted into the database all in the same time domain, in other words, when an even happens on the external device, the coincidental data from the monitors is stored at exactly the same time. This ready access to waveforms has fueled a host of new algorithms from providers such as OBS Medical, inventors of the Visensia Index, and Clemetrics which uses some of the new research linking Heart Rate Variablity (HRV) to patient degradation, or organ failure. Visensia, on the other hand is marketed as a “Safety Index” that both hallmarks subtle changes in degradation but can also allow patients to be discharged earlier, thus saving hospital resources.
The Electronic Medical Record system of the future could pre-assign, engage particular algorithms either enterprise-wide or based upon ICD-10 diagnosis codes. What they all have in common is the need for a stable front-end that can capture enterprise-wide, granular data from any hospital, and provide an environment where the algorithm “apps” could be called upon, and to provide one stable base with which to re-display archive waveform as well as to provide streaming analytics. The future will also include more Machine Learning, and neural networks that can educate themselves and recommend differential diagnosis by aggregating data from lab, patient monitoring and even genetic information.