I would start looking at a couple of open-source architectures:
Edge X Foundry:
edgexfoundry.org/
Akarino:
lfedge.org/projects/akraino/

Entity Resolution has been discussed for more than a decade -- this is a "must-have" if explainable and Interpretable AI (especially one derived from stream data) is to become reality. Entity resolution is a fancy term for annotation and building ontologies on the fly.
Sanjay

The question sort of answers itself.

A little bit of context from my perspective: before I went back to school to re-tool and "re-invent" myself, I was in the aerospace / defense industry for a number of years. Have always held this comparison in the back of my mind between defense & healthcare. Aerospace / defense is (pretty much) governed by government and Federal contracting. As such, the MIL-SPECS and standards that are applied make it mandatory that any contractor bidding on a government contract must meet these requirements else they are not qualified to bid. Now, there are really only a handful of big defense contractors left. But, the point is that, regardless of contractor, they all must meet specs or they will be disqualified from bidding.

Let's now jump the tracks to healthcare as an industry. For the longest time, healthcare struggled with interoperability of multiple sorts, despite the advent of Health Level Seven back in the 1990s (or thereabouts). Medical devices have taken much longer to come of age from an interoperable and integrated perspective. MU, frankly, was the motivation behind that really accelerated a number of vendors towards better integration, because real money was on the table for health systems and real penalties for those that did not comply.

Not advocating for a "fascistic" approach here. Merely observing what has seemingly happened.

Now, Let's consider your question: "...System vendors are wary...unless mandated..."

Let's suppose that a Health System ABC wishes to adopt new mechanical ventilator technology, or anesthesia machine technology, or home-health spirometry to support its ambulatory patients, or what-have-you. If the health system made as part of its RFI and RFQ process the requirement that vendors must qualify based on specifics surrounding interoperability (e.g.: must support IHE DEC profiles; must be WiFi ready; must support cellular connectivity, must meet AES or DES encryption standards, ...

...or others that are important that I'm omitting or forgetting here), then this would reduce the possible list of potential vendors that could bid but may also motivate others as a wake-up call that if they wish to remain viable, they need to get with the program, from a basic existential business perspective.

Again, not advocating anything "fascistic" here... merely pointing out that the motivations in business come down to the profit motive and the size of the market.

To me, and I'm saying this from the perspective of an individual who has been operating inside of this health information technology and medical space for near 30 years, I have seen standards come and standards go. I have listened to the promise of "mandates" and seen them come and go. The only thing that I have seen really work -- I mean really work -- is motivation out of market driving forces and the recognition that if you want to play, you need to pay (that is, pay the price of adoption of that which is missing from the solution).

My 2 cents.

Such as? (Is there another context in mind?)

I think we have to look at how the regulators in the US classify them. FDA is a major player here. Here's an article about different classifications of device issues: californiahealthline.org/news/...

From a laboratory perspective, can be classified by methods (devices) which have undergone FDA 510(k) approval (most commercial systems) versus laboratory developed tests (LDTs) which do not and developed by a single lab usually. Laboratory professionals are required to validate them before offering the testing in their labs.

Further, CLIA regulations will have tests classified as Waived or non waived, which includes moderate complexity or high complexity with regulations indicating the type of clinical/laboratory training needed to perform tests in the classifications. For example, waived testing covers many point of care devices like glucose meters, PT/INR testing, etc. which can be performed at home by patients or by health professionals and have minimal risk for error. High complexity testing is performed by laboratory professionals in a clinical laboratory setting. .cdc.gov/clia/Resources/Waived...

Realize there are classifications for other areas of devices used in remote monitoring. There are different levels of trust with the devices/ health profesionals performing the remote monitoring vs. non health professionals. In a health setting, due to job requirements (i.e. clinical education) and training, there is more trust (although errors may still occur with either, hence why we have recalls).

I like this question. Too often we think about medical technology as being solely driven by healthcare professionals, overlooking the fact that people should be partners in their own care. When it comes to RPM technology, if there is a perceived benefit to an individual and they know that their insurer will reimburse for the technology, why shouldn't there be 'consumer' driven demand. For example, if there is a technology out there that can detect deteriorating quality in monitoring health signals (such as respiration, heart rate, activity, sleep etc) and notify both the patient and the treating provider, wouldn't that be something that a patient might want to have?

Today we're seeing emerging trends that I believe will be the forces that shape and drive the convergence of remote monitoring data with ever evolving and changing patient health record. Let's take for example what's happening with wearable health monitoring devices and services: (1) Wireless smart devices such as BP monitors, exercises monitors (e.g., Fitbit, Samsung, Apple, Garmin, etc), (2) cloud based integrating services that can aggregate and provide insights from such monitors (e.g., MyFitnessPal, Apple Health, Samsung Health, Endomondo), secure single sign-on (SSO) making integration and sharing of data much more consumer friendly, (3) emerging trust frameworks that enable sharing of personalized health data with health "communities". This trend in integration of consumer health information is an extremely powerful driving force that I believe will drive solutions to remote monitoring because it's driven by consumer demand. So what about certification of valid data , devices, regulation of such devices... can we trust the data from a healthcare/patient care perspective? Well, what if we were to solve 2 parallel problems that would ultimately lead to a practical solution? First, removing some of the barriers between the healthcare IT and the consumer health would create the first and most important bridge. Second, since adoption of consumer health products and services is a rapidly growing market, leveraging that data in ways that can provide incremental insights before attempting to place heavy regulatory barriers to leveraging consumer health may be a good way to start. For example, examining consumer health generated BP averages , weights, exercise level, dietary consumption, medication adherence patterns all can initially provide one-sided insights that can be extremely valuable in assessing the effectiveness of clinical interventions, wellness programs, and general population trends. I'd gladly discuss some real examples if there's interest.

Regarding is there another context, yes definitely. The consumer (not patient) context. I do agree that regulators need to provide consumer protection through standards, classification of devices, certification just as the FDA does today for medications. However, I believe there are ways what the industry can gracefully leverage trends in the consumer health market that can accelerate practical use of that data in parallel (please see my reply below around emerging trends).

Have been following this conversation and have had mixed thoughts, keeping an open mind.

Observations: the consumer devices with which I am most familiar support varying degrees of integration capability. Newer glucometers and blood pressure monitors provide cloud-based data collection services and their own software information platforms for tracking. Some will support integration with existing platforms (e.g.: Epic "MyChart" et al.) These capabilities are a bit different from in-hospital use where the integration is directly with the record either through network or gateway connection and there is no standalone presentation layer for patient interaction.

Concerns about consumer space: patients who are on a regimen involving post-hospital care, including visiting home nursing use consumer devices as well, including scales, blood pressure, glucose, spirometry, etc. The benefit in having a health provider oversee consumer use is to ensure accuracy, compliance, validation and review, among more general reasons. Many consumers monitor their parameters and are self-educated. This can be ok, as well, so long as they are diligent and understand what they are doing and seek guidance as required.

My soapbox: many who are not familiar with physiology, body chemistry and the workings of the various organs and systemic functions look at the numbers produced by devices (consumer or other) without knowing the complex interactions they can foretell and reveal inside the human "machine". The body is a system of systems with many interrelationships that do not necessarily reveal themselves in intuitive ways. Hence, a measurement of one parameter may foretell something completely unrelated. An example includes onset of high blood pressure and reduced heart rate (Cushing's reflex) related to increasing intracranial pressure that can result from a stroke. The bottom line is: know how to measure, know when to measure,... (see part 2)

(from part 1)...know how to report the value(s) together with their context, know what surrounding information may be pertinent and relevant, and know when to seek help and advice.

For instance, if blood glucose is low, what happened in the past? what did you eat? When did you eat? Has the glucose been trending low? Have you taken your insulin?

Front-line providers in the medical fields are trained in how to obtain information from patients. In the inpatient space, some refer to the gathering of this information using the term SOAP notes (acronym: Subjective, Objective, Assessment, Plan). In the EMS field a similar approach is taken involving the acronym SAMPLE (Signs & Symptoms, Allergies, Medications, Past medical history, Last oral intake, Events leading up to current emergency or behavior). These serve to establish the history of present illness and assist in developing a (crude) differential diagnosis for what may be going on. Having this history is important to placing objective measurements in context with what they can mean presently, and how they have trended in the past.

There is more to write, but I've already gone on too long.

I see the difference as clinical care vs self-care. I know for myself I try to diagnose myself based on symptoms and vitals I have but not being a clinician and being trained on physiology biology and all the other 'ologies I often misdiagnose myself or can't figure it out and go straight to the Dr's office. But, as AI evolves both in the consumer realm this will change and give the individual much more information and insight as to what is happening with a clearer direction for more information to review or action to take.

Undoubtedly, tons of potential that is stimulated by consumer demand. A slightly related topic - if we rely on consumer demand, we must also consider who composes the population of consumers to ensure that we are obtaining robust, generalizable insights from the remote monitoring data. For example, smartphones and wearable devices, while becoming increasingly common, are less likely to be affordable and, thus, utilized, by individuals who may have more limited socioeconomic means. We need to ensure that we are not leaving some populations behind, if we do not obtain or study their data, which may be different (e.g. different activity levels because of differential access to safe spaces for exercise; different diet because of differential access to supermarkets). Just mentioning to be considered with the potential on the consumer adoption side of things.

@Sanket: Precisely.

That's a very interesting question and sometimes implies an "or" rather than an "and". What I mean by this is that there's a lot of value in AI/ML solutions that leverage structured/coded data rather than replacing it. Learning systems can in turn enhance ontologies and models that can strike a nice balance between "valid" AI/ML inferences and practicality/usability. Let's take a look at a simple example, capturing structured/codified instructions for medications from physicians in the context of e-prescribing.
At first glance it seems like it would be relatively easy for a physician to create a structured instruction by entering a dose, selecting a frequency or interval , a duration if applicable and adding some additional instructions such as 'take with food" or take at bedtime. Today we have solutions that include precoordinated instructions to make "1-click" instructions readily accessible. However, that hasn't improved the situation much because of searching limitations and unintended errors due to wrong selections. What if we can leverage AI/ML to address 2 issues initially: (A) the user friendly common lexical representations of the instruction as the system learns more from inputs and (B) context in which the medication instruction applies (the restricted set of instructions that apply to specific medications for a specific use/condition for an adult patient for example. Today, most drug knowledgebases have data at this level of granularity. However, the challenge has been presentation to the user. I can envision in the not too distant future a learning system working on top of a well defined ontology for medication dosing instructions that can recognize using NLP a free text string as a physician is typing and suggest a reasonable, context specific short list of options that are specific to the condition and patient attributes while also learning from user input patterns of common expression that will continuously improve recognition safely.

What is the meme (mostly real now)? 70% of all clinical care decisions involve Radiology, Pathology (and now Genomics) data.
The issue of annotation and entity resolution comes in here as well. The scourge of clinical medicine is missing, incorrect or no annotations. This gives rise to dangerous biases (yes, there are more than 100 bias types).
If all data were "schemaless" and MLI (machine learning interpretability) as well as General Causal Inference became standard-of-practice (as compared to probabilistic systems with bionomial distributions which are not at all common) then ontologies could be built with any data type.

Some of the leading RPM vendors including Cloud DX have API's to pass the vital signs data to EMR/EHR's and other platforms. The challenge has been establishing a business relationship with the EMR/EHR vendors for them to accept the data and at what cost. If a business relationship can be agreed to then integration can in most cases be fairly easy to accomplish.

Yes. This has been my experience and understanding. The only other thing I might add is the creation of a workflow for data upload that is fairly straightforward (i.e., not "clunky") and repeatable that will meet the operational needs of the clinical users interacting with the product data. This, of course, is a tacit necessity and may be implied anyway. But, just making sure the idea is not lost.

The siloed data problem goes well beyond remote monitoring data. When the health care delivery system rushed toward digital solutions they created chaos for all stakeholders. From a consumer perspective we now have portals and passwords for every doctor, hospital, lab, pharmacy, insurance, an average of 7-21 point solutions for employer sponsored health and wellness programs, HSA/FSA, wearables, clinical devices, and what is already ~ 1/2 million digital health apps on the US cloud.

As a result, stakeholders including providers, insurers, pharmacists, caregivers, and individuals each only have a partial view of a consumers holistic health data. This has created an overall lack of trust in data around health in general as no one has the full picture. With the emergence of APIs, the modified regulatory environment, and platform aggregators like b.well this is rapidly changing.

If we had the full picture of an individual along with persistent and portable data to show trends and changes in health over time we could triangulate with much more confidence on all data.

My view is that when consumer wearable technology started to hit the 'peak of inflated expectations' on the technology hype cycle many insurers and some providers jumped in and started handing out unvalidated health trackers to members and patients, either to incentivize or because they thought it might encourage healthier behaviors. As we entered the 'trough of disillusionment' it became clear that these technologies did not provide the right outcome and that there was limited value to the data captured (and it was difficult to get this to the people who needed to see it most). The good news is that remote patient monitoring technologies are now in the slope of 'enlightenment' and will soon hit the 'plateau of productivity' - providers and insurers have become wiser as to how these technologies can add value and are being smarter and more selective about how they are deploying them.

Great points. One of the fundamentals of informatics is the generation of information whether paper or electronic. Taking a paper process, workflows, etc. and just automating them does not necessarily mean improved processes, but perhaps quicker ones that magnify issues more. One should think about what the goals are from a patient perspective to ensure new errors, safety issues or quality isn't degraded by switching to automated workflows.

To address the question, various departments had their own view of data and systems tailored to their workflows. They may work well in the specialty areas, but hen data in different formats, modeled differently, described differently all need to be shared, some of these differences are realized. In some cases, it may be a potatoe/potato aspect in other cases it could be clinically significant. All stakeholders need to be at the table with design to work on interoperable solutions. Otherwise only some voices will be represented and designs will evolve accordingly.

It used to be if you've seen one HL7 message, you've seen one HL7 message. Customer interfaces to exchange data were prevalent. Now folks want more of a plug and play approach. Standards have evolved or tightened up to reduce variability. New standards such as FHIR have emerged reducing cost and models to connect. Still evolving, but we're getting closer to interoperability with shared remote monitoring data.

At some level silos are unavoidable. Remote monitoring devices don't need to be integrated to have clinical impact. Manufacturers of remote monitoring devices should put their efforts behind the product, making it safe and easy to use.

When the product reaches a critical mass, it's much easier to move to integrate. It's much easier to say to a provider organization/IT team: "hey thousands of your patients use our device, let's integrate" vs "here's a product you've never heard of let's integrate".

Excellent observations and points all around. To do away with silos, there needs to be a motivation to integrate that includes both the bottom line and health care quality improvement. That integration has been an afterthought is not news to anyone working in the field of medical device integration. That the seeming benefits to clinical workflow, patient care and cost reduction have not provided motivation is not motivation enough continues to be astonishing, at least to me. Bear in mind that there has been improvement in data integration through the use of targeted standards and profiles, particularly over the past 5 years through the efforts of patient generated health data working groups, Integrating the Healthcare Enterprise, Health Level Seven, and others -- including recognition by and through the Office of the National Coordinator. Despite this recognition and action, healthcare is a peculiar "industry". It seems that silos are tolerated in healthcare where they would not be anywhere else. I'm thinking of banking, telecommunications, the cable industry, and others. The point of employing standards was to make medical device technology akin to plug-and-play. There are a number of devices that approach this capability but many (most?) still do not. Ergo, the rise and continuing prevalence of medical device integrators. The consumer space is improving but it is still a bastion of silos. Noted improvements include cloud-based data uploads that at least suggest improved access to the data. I agree with the hype-cycle (credit to Gartner), and it seems there is a trajectory towards generalized integration. I simply cannot say whether it will take 5, 10, 15, ... years to get to nirvana.

Thanks for the question posted to the group.

Semantic Interoperability (IXP) is only one element for data liquidity which infers that data is useful, safe and satisfying.

Vanderbilt's Dr. Stead maturity model made for holistically approaching the advancement of data liquidity at the Center for Medical Interoperability identifies the multiple domains of interoperability. Infrastructure, Syntactic, Orchestration, Contextual/dynamic AND semantic together are the elements that holistically address true benefits and value of interoperability.

Patient's, clinicians and "systems" of healthcare care coordination, innovation and transformation will only evolve from our current state of analog into the digital age once health care adopts a modern architecture that addresses Trust. The continuation of piecing together today's antiquated data for improving health outcomes is an opportunity of scale and impact.

Welcome discussion: reference includes:
1. Procuring Interoperability: Achieving High-Quality, Connected, and Person-Centered Care. A Special Publication from the National Academy of Medicine.
2. Center for Medical Interoperability medicalinteroperability.org/

Kelly Aldrich DNP, MS, RN-BC, FHIMSS
Chief Clinical Transformation Officer at CMI

Advocate for Clinicians and Patients to benefit from Data Liquidity instead of being victims of lost and useless data contributing to "medical errors" and missed diagnosis.

So where would you put remote monitoring devices in the IMM? How is the issue of trust accommodated in the IMM?

The Patient-Centered CDS Learning Network (pccds-ln.org/) Trusted Use Frameworks Working Group (pccds-ln.org/tfwg), with Blackford Middleton, Jody Platt, Barry Blumenfeld, et al recently completed their Recommendations for Building and Maintaining Trust in CDS Artifacts (pccds-ln.org/sites/default/fil...) which identifies actors, trust attributes, and recommendations with applicability that overlaps interoperability more generally, including remote monitoring devices.

To respond to Josh Phillips' question, there is probably an increasing recognition that medical device interoperability is an issue with impact on patient care and safety. With semantic (and pragmatic) interoperability of remote monitoring devices, new workflows and models of monitoring may be realized. If my home BP measurement device can talk to my wrist accelerometer, it might say "please sit for 5 minutes before taking your BP, since you just finished some strenuous activity". Thus closed-loop autonomous and semi-autonomous device systems might be realized with a greater potential for quality data, patient safety, fewer errors, less provider burden, and better quality and more efficient care.

The value to the patient, and treatment team, for the distribution of remote monitoring devices is to enable real-time insight into the most critical indicators that are relevant to a patients unique risks.

That is simple enough to understand, but if that information isn't correctly interpreted then you get anti-value instead of value.

Without the reliable and consistent ability to transport and correctly interpret this information in a predictable way, the belief that the information is providing 'truth' will result in unjustified complacent that truth is 'known' - when in fact it is not 'understood'.

Reliability and Certainty are critical during this moment when amazing devices are exploding on the scene. It will only take a few well publicized cases of 'lost or misinterpreted data' leads to a loss of life for this amazing opportunity to leverage the 'Internet of Medical Things' to get a black eye withing the clinical and patient communities.

The plethora of inexpensive and diverse remote monitoring devices represent and amazing opportunity, but it also poses a complex challenge to the systems that shoulder the responsibility to interpret that data and leverage it to inform care. Both sides of this partnership need to step up and organize collaborative efforts to insure that this communications boundary is carefully and fully understood.

As a patient who just had a knee replacement, my activity data should be available to my orthopedic surgeon as we monitor post-surgery recovery and the same data should be available to my physical therapist to monitor how I am doing with my exercises and the same date should be available to my PCP to discuss why I gained weight in the last 30 days due to lack of activity during recovery. My entire care team should be able to see the same data, standardized, regardless of the EHR/care management platform that is consuming the data.

Ultimately, if we can achieve semantic interoperability in remote monitoring, it removes siloed care, encourages a dialogue vs. a monologue, and would help both care providers and recipients live into a holistic approach to care.

Great points. The challenge is achieving semantic interoperability to achieve @michelle's points when some of the same data items conceptually are described differently, encoded differently or otherwise transformed in different systems (information systems, devices, etc.).

From a lab perspective, EHR's often name things differently than the performing lab, the source of truth. As data flow downstream, in addition to naming changes, coding may change resulting in inoperability. In some cases, the end to end (origination to downstream end users/use cases) of data is not a smooth flowing process semantically or clinically accurate (in some cases), even though syntactically the data are successfully moved from point a to point b. To facilitate, data need to be structured, encoded, without loss of meaning or fidelity.

Addressing the question, benefits include transparency of health data to all health professionals no matter their role as Michelle describes. In theory, reduced healthcare costs from elimination of duplicate testing/procedures/data by different providers. Elimination of lost time by care team members, patients, etc. in tracking down information in disparate systems (location, time, vendor, standards/structure/modeling). Improved coordination of care, as right data are available at right time in right place to right people.

Most data transmissions are a communication problem (when not successful), whether paper, CD/Disk based (images/large data); or electronic. Previously siloed depts., information/filing/data systems, providers/health systems, are expected now /soon to all "talk" to each other seamlessly. One thing that hasn't been mentioned are the cultural barriers that should be broken down as different roles now work together as the health care team. Some depts./roles/specialties collaborate well, while other roles/depts./expertise aren't recognized as much as the value they bring to the total patient care.

Undoubtedly achieving semantic interoperability, as Michelle states, can reduce silos and lead to patient-centered care. Physicians see patients just while in the hospital or for a 15-45 minute visit in the office/clinic; nearly the entirety of patient's lives take place outside of health care settings. Being able to obtain and interpret these data in a usable format can lead to a more holistic understanding of a patient's health - exactly what the goal of the doctor-patient relationship should be. In my field of cardiology, there is the potential to intervene sooner for patients experiencing symptoms that may be related to cardiovascular disease and, conversely, keeping patients from needing to come to the hospital or clinic if a visit is not necessary. True personalized care is not possible without achieving semantic interoperability in the continued growing field of remote monitoring.

Having been in the field of medical data interoperability and medical device integration as long as I have, It seems like the repeating of an old saw that siloing of data and lack of clear integrated data workflow are keys to the kingdom which are still being worked out. I remember standing in patient ICU rooms back in the mid-1990s before EMRs really were coming of age and noting that all of the data I was seeing from patients should be available for patient care rather than being "dropped on the floor". That was the inpatient space, of course. Similar challenges exist in the consumer space, of course. There is improvement - some vendors of electronic health records do provide capabilities to integrate (certain) consumer devices today and there seems to be an improving trend. Nowhere near "nirvana", of course.

I think interoperability of the data is the important thing and the device is just the mechanism. But again, is it trustable and accurate data being collected by the device. If so, then having a patient-centric repository of all patient data benefits the patient because all data is in one place. The data can be shared by those clinicians taking care of the patient and that data is the same for all involved. This is kinda the holy grail that everyone seems to want but the path to get there seems to be different for payers, providers, vendors and the gov't.

@Larry: yes. This topic gets me back on my soapbox about data accuracy, integrity and trustworthiness. Addressed in other threads related to this topic.

The ability to collect data from consumer devices has improved greatly over the past several years through the use of physical media such as Bluetooth and WiFi. Data standards related to patient generated health data (PGHD), patient care devices (PCDs) have also matured, providing improved and normalized representations of data through common semantic protocols reflected through standardized protocols. This is not to say that the challenges are all solved. But, improvements have taken place. An additional challenge, in my opinion, is validating patient data from consumer devices and ensuring data integrity and security in both collection and transmission. Other challenges include oversight and education to ensure data collection takes place correctly and associated observations are recorded together with objective measurements to provide context for correct interpretation.

The current issues are some of the same as the past issues. Those being how hard is it to deploy a RPM program, ensuring quality data and who is going to pay for it.

Developing a good strategy of what you want to accomplish with a RPM program and how it fits into the every day workflow is very important and the starting points for any organization. Otherwise it's the tail wagging the dog scenario rather than vice versa. If an organization doesn't have that expertise in-house to develop strategy and workflow there are a number of seasoned experts consulting out there.

Data accuracy is critical. There are a number of vendors out there touting bring your own device and/or manual input. How do we know that the device is accurate? How do we know what is being reported is accurate? Remember, garbage in garbage out. Make sure you vet the vendor(s) you are considering for the accuracy of their device(s) and how the data is transmitted (electronically vs manual). Also, has the technology been tested to scale. Nobody is looking to just do a pilot of some 25-50 patients. The tech needs to be able to scale to 100's or 1000's of patients. Ease of patient use of the device(s)/app and deployment are also critical.

Lastly, there are new RPM reimbursement codes effective as of January 1, 2019:CPT 99453 Remote monitoring of physiologic parameter(s) (eg, weight, blood pressure, pulse oximetry, respiratory flow rate), initial; set-up and patient education on use of equipment, CPT 99454 Remote monitoring of physiologic parameter(s) (eg, weight, blood pressure, pulse oximetry, respiratory flow rate), initial; device(s) supply with daily recording(s) or programmed alert(s) transmission, each 30 days. CPT 99457 Remote physiologic monitoring treatment management services, 20 minutes or more of clinical staff/physician/other qualified healthcare professional time in a calendar month requiring interactive communication with the patient/caregiver during the month.

At Redox I work with innovative companies in the remote monitoring space. EHR integration, or more broadly data integration can be a major hurdle. As @John Zaleski points out the data can represented through standards - that problem is mostly solved. What is hard is going the last mile; feeding the data back to a system where a clinician or care coordinator can make sense of it. Often that requires really understanding the needs of those looking at the data, rather than trying to blast them with observations.

In times past, a patient with CHF would call their physician and say "I gained 7 pounds" the last few days. They were comparing weighs using their own scale that was not calibrated since its purchase. To insist on "accurate" measures is not always (often?) necessary for CDS. So, in talking about quality, a clear definition is needed. Perfection will kill many plans.

I couldn't agree more with what has been stated above. The RPM codes introduced by CMS above are aimed at lowering the (financial) barriers for physicians to start capturing data. But whats the point in capturing data if there is nothing actionable or insightful about it? If we are going to capture data remotely and act upon it we should consider the following:

1. Can we be confident that the data has been captured correctly and is an accurate reflection of the truth? Has the capture device been validated against other devices that we trust? Is it approved by the FDA? Is there any chance that the data is corrupted in some way? What is the sensitivity and specificity of the data?

2. How do we ensure we capture enough data? Some variables like weight don't change from minute to minute, but heart rate, respiration, SpO2 etc fluctuate based on what the patient is doing around the time that they take the reading. Too often we capture a limited number of spot metrics which do not provide enough information upon which to take decisions. Lowering the patient barrier to capture data is an important consideration.

3. Once we have accurate and consistent data, how does it help us evaluate and manage a patient? What are the signals that we care about in a specific patient population? And what is our ability to respond to what we are seeing? Ideally, we want to recognize patterns of activity which are indicative of deteriorating health or imminent decompensation, which allow us to deploy health resources efficiently to intervene earlier and prevent more serious interventions.

There are several different challenges being discussed here. In my post, I was originally only focusing on the challenges of semantics and data. Those issues have had some energy directed at them over the years and stand, in my opinion, in good stead of being solved through recognition of data standards.

The other challenges that have been identified are still challenges. For example, cross device calibration and accuracy; how to verify whether measurements are accurate or accurately taken; how to take the data the final mile from the patient to the clinician, and other points brought up here are not fully resolved.

As a pre-hospital clinical provider, I am often presented with measurements from patient home devices, and many measurements taken using such devices by nursing facility or assisted living facility staff on patients in which the measurements are very questionable. These range from glucometer readings to pulse oximetry measurements to blood pressures. It is always necessary to assess the patient — that is, to look at, listen to, and palpate. There is no proper substitute for these.

For instance, if I am presented with a pulse oximetry reading of 75% by nursing or assisted living staff as the basis for an emergency call, and I look at the patient and see the patient is alert, oriented, skin is pink, dry and warm and the patient is not engaged in labored breathing, then it is very obvious the measurement is erroneous.

So, context is important and certain measurements mean more than others when managing a patient based in their history of present illness. For these reasons, having some type of active case management and regular visitation is necessary to prevent open-loop data collection run amok.

Discernment from clinical presentation details more about the patient than merely looking at numbers from a sensor as these can be just plain wrong.

To my mind, item 3 merits more detailed discussion. Clinical presentation and interpretation of differential diagnosis is the basic object from the observations. The question I will ask is whether the measured values are accurate and match with observations and in what context. I can tell much from patient presentation if I see, for example, a diaphoretic patient with altered mental status and rapid pulse. Combine this with declining blood pressure or with low blood glucose, and a picture becomes much clearer as to the emergent presentation. The question in my mind is whether I can trust the measurements. If I see a measurement that does not make sense, I will immediately attempt to verify through a different means, or will seek to validate the measurements on the basis of other independent observations.

Clinicians and care managers need to have confidence in the data presented. Different devices relay the same data sets in different measurements. Validic has spent the last 8 years standardizing data sets from hundreds of wearables and in-home medical devices. We standardize the endpoints to ensure actionable, meaningful data is delivered within the clinical workflow.

In our experience, both health systems and payers implementing a remote monitoring program tend to limit which devices can be used by patients, especially for programs that manage chronic diseases such as diabetes, hypertension, CHF, etc. However, as these programs expand to include not just blood pressure and/or blood glucose data but also activity data, a BYOD approach is compelling.

An additional challenge is the type of connectivity supported by each source/device. Some support a web API only while others support BLE connection only. As remote monitoring programs administrators look at how they wish to engage their populations and who should own the patient experience, often the devices sources themselves create a barrier.

Ultimately, for data to be utilized effectively in remote monitoring programs, the data has to be presented within the clinical user's workflow. When I started last year, I spent time interviewing stakeholders, particularly care managers. These stakeholders could not emphasize enough the importance of getting the data in front of clinicians in a consumable and actionable format. When that happens, patients see their clinicians, in addition to care managers, as engaged in their health journey and in turn, take a more proactive approach in using remote monitoring to inform their decisions.

That said, getting actual remote monitoring data sets embedded in the big 3 EHRs is not a task for the faint of heart. However, having successfully integrated a full remote monitoring dashboard complete with logbooks and data visualizations, the value-add for both the clin

Lots of great discussion here. In framing the challenges to a patient - as consumers we rely on the ability to change over time. Behavioral devices outside of a clinical setting can be helpful to consumers to learn how they are sleeping, moving, managing weight, etc. However, technology changes very quickly and as consumers we want the right to change those devices (ex: fitbit to apple watch) without losing their history. As we migrate further into connected health devices for RPM or for the purposes of digital therapeutic assignment we should know that we are asking consumers to either abandon or add on to the wearables or devices already in their every day routine. We encourage sponsors of these devices to provide choice on device to consumers as they are more likely to use them. Additionally, as change occurs between devices (due to advancement in tech or otherwise) we will need to re-educate the consumer and re-baseline the data.

I resonate with the need for full integration and the ability to support multiple types of consumer devices as outlined by both Michelle and Kristen above. My main concern, which I state at the risk of being labeled a Luddite, is that properly taken and validated measurements are key if ANY data are to be useful for remote monitoring and management. Thus, the only medical devices that I carry regularly, and in which I hold close to 100% confidence, are my Littmann cardiology stethoscope and my manual blood pressure cuff. I know what to listen for, how to take the measurements, and whether the measurements can be trusted on the basis of how the patient presents to me. I have seen far too many times the use of automated measurement equipment inaccurately on patients even by trained providers, or trusting in measurements when absurd conditions apply, such as pulse oximetry measurements in patients with poor perfusion or severe edema.

Again...at the risk of being labeled a Luddite... from the perspective of semantics and data, I think that, based on my knowledge and experience, the standards have come quite far over the past 5 years. Those working these standards are to be congratulated for the effort. I would like to see more effort placed on the validation, secure transfer, and methods for educating consumers on data collection as a continuing form of process improvement. This will aid in reinforcing the proper techniques for data collection and help the clinical end users in achieving the confidence required to trust the measurements, regardless of the consumer product.

John - I believe your points are valid but would ask you to consider two things -- devices are not being put into market to replace physicians methods at the site of care, they are to generate trends to improve health literacy in patients themselves to become more self-aware the 99% of the time they are outside the four walls of a doctors office or hospital and two, to help identify early risks for events (examples include air quality and temperature for asthmatics or seizures about to occur in patients with epilepsy) to get alerts to someone close by or to prompt a visit to the doctor. There are obviously a large number of use cases and the data must be trusted as valid by physicians receiving it but even if titration is off for an FDA approved device and it is consistently off there is value in the trend line. As an industry we need to get in front of events wherever we can and move toward true prevention. Without improvement in health literacy which involves (among many other things) consumers understanding their data and how their behaviors affect health on a more regular basis I do not believe we can get there.

yes, I think you make a great point - RPM techniques will not and should not replace or replicate what a trained healthcare professional could do - instead they should be confined to acting as an early indicator of increased patient risk. Focussing health resources on biological changes, which warrant further attention is exactly what we need if RPM is to be a successful adjunct to current evaluation and management practices. Example: Mrs Jones' sleep quality has deteriorated over the past 7 days OR Mr Smith's respiration rate has increased 20% from his historic norms over the past 48hours. Gaining an early indication of biological data trending towards aberration (again, assuming the data can be trusted) and then investigating through established outpatient diagnostic techniques, such as the biomarkers you mention above, can prevent crisis leading to a hospitalization.

Exactly.

You should not be labelled as a luddite to expect that data, upon which you are expected to make clinical interpretations against should be standardized and held up to regulatory scrutiny. What is important is to recognize is that there are emergent physiological parameters out there which we have not had the ability to capture to a clinical grade of accuracy historically, which technology is now making possible. A good example of this is longitudinal respiratory data, which Spire's Health Tags are able to capture - respiration rate, variability and breathing patterns are not data sources, which physicians have had access to before. That doesn't mean they aren't valuable - in fact we should invest behind conducting systematic research of these technologies to better understand how new data may be useful in managing patient populations.

Philip - Thanks. Having respiration data would be valuable in painting a picture of the patient in the pre-hospital environment by consumer devices. Accurate respiration measurements could be used to monitor not only for breathing trends but also for apnea and sleep disturbances as well as being part of the basic set of vitals normally obtained regularly on home-monitored patients.

Kristen - Thanks. I understand your points and it was not my intention to supplant clinical decision-making via use of consumer devices. The trending of data is a normal part of the assessment process as-is, and this requires multiple measurements over time to discern whether patients are stable, de-compensating or improving.

As to value in the trend line, that all depends on the accuracy and integrity of the measurements. If a patient consistently day-after-day reports normal oxygen saturation on room air of, say, 95% and then one afternoon a reading drops to, say, 80%, that is worthy of note and at least some action, if the intention is, as you write, to prompt a clinical action. Yet, there needs to be an assessment to determine whether the 80% is valid. The resultant measurement could trigger case management to contact the patient, or prompt a home visit, or even prompt EMS. This was my point earlier about being prompted on calls in which a patient was reported to be hypotensive or to have a low oxygen saturation only to find that the measurements were in error. Prompting action in the absence of other information is most probably necessary to avoid the possibility of a false negative, but will also promote many false positive events. To some degree, this will be acceptable in order to accommodate and remain in front of the possibility of a de-compensating patient, as you write.

My point was the validating of measurements is necessary to aid in reducing false positive events. Hence, one way of doing this is to ensure that patients are properly trained to use their devices.

Great discussion, responses and points!

A few keys items are: trust of the data reported by the user (especially when not obtained by a trained health professional, but rather the patient); trust the point of care (POC)/personal device is accurate (as compared to health system devices); connectivity aspects (how are the data safely/securely accurately transmitted from device to the EHR/Information System); Integration into EHR/Information System (need to distinguish non health system/health professional generated data due to different methods/trust/decision making aspects); Integration into clinical workflows (needs to be structured, encoded, actionable information not buried in a clinical note/text blob, etc.).

As a laboratory professional, one issue I've seen gets back to the semantics question asked. Most EHRs don't show clinicians the same name of the test performed by the performing laboratory as indicated in their CLIA Specimen Collection Manual/Laboratory Compendium. There can be "lost in translation" issues and lost information/meaning and other semantic impacts.

Realize one aspect that drives folks crazy about laboratory testing is the same test may be called different names by different labs (i.e. Automated Blood Count or Complete Blood Count), but really performed the same way as reflected by their mapping to the same LOINC code. The opposite is true too, same name for test, but differently performed and thus different LOINCs.

Been having a discussion on the FHIR Zulip chat for folks globally developing apps and devices to generate PGHD, SDH, etc., either to integrate within EHR or on phones/fitness devices. With lab data it is similar to Point of Care Testing and EHR/LIS test build, naming conventions, etc. However, for device/apps that are remote/independent, each vendor needs to determine what they are going to name the data (conceptually the thing) they collect. Suspect we may have quite a variety... end part 1

Part 2- Suspect vendors may each name something slightly differently resulting in a variety of naming conventions. Using standardized terminology will help reduce the variety and encoding the names with standardized codesystems will aid folks in ensuring they are computer processible and comparing apples to apples and apples to oranges. Vital with health data for clinical decision making so health professionals know which things are conceptually the same or different and which items can be comingled together in apps, or EHRs or other information systems.

I think we all agree PGHD should be separately distinguished from health professional generated health data. However, given the case where I have a fitness device and apple watch both monitoring the same and different data of my workout, how are these data uniquely distinguished, collected at their pinot of origin, then shared, integrated and utilized? They will also be treated differently than data such as how many minutes exercised or patient survey type questions such as to assess how they are feeling (great, not so great), etc. Recommend LOINCs are requested for all these data points, to facilitate distinctions.

Would love if say exercise metrics feed into EHR and physicians can see patients are meeting or exceeding guidelines for exercise minutes, in the EHR and thus no longer have to answer said questions on pre-visit questionnaires (unless they are not sharing such SDH/PGHD). Would love to have smarter, more efficient computer systems to automagically use the data for clinical decision support, especially in between visits, perhaps notifying patient their exercise goals have not been achieved in a week. Is this an indicator they are sick or injured or they didn't have time due to job/family/other constraints?

Am not comfortable with insurance knowing some of this information though. Some insurance companies are receiving CPAP usage metrics. see part 3

Part 3- Insurance is incorrectly deducing from cpap data that non use means it is patient non compliance, when in fact the patient may totally want to be compliant, but is unable to do so due to equipment failure, breakage, malfunction, repair, etc. While waiting for a part, and then resuming usage, insurance has notified providers of patient non use/non compliance and refusing to pay for costs at a later point indicating the patient didn't use properly (in their opinion). This gets into some ethical questions too.

pausing as there is plenty of food for thought....

Your point, Andrea, takes us back to the point about validating data in context and not simply trusting what the machine issues. This remains a basic yet essential issue to resolve as it manifests in terms of real dollar costs to the patient.

Not sure if this belongs in it's own thread... In light of the NPRM on March 4, 2019 concerning information blocking, APIs and FHIR, and certification of the APIs, a few points come to mind related to the discussion. 1. Will certification also cover clinical validation criteria for the content in addition to technical certification aspects? 2. Who will determine clinical validation criteria (to assess whether an app meets or does not meet said criteria)? 3. Will medical specialty professional societies chime in as to what they see as minimum/best practices for said criteria?

Asking as there have been Connectathons testing the technical connections (i.e. getting data from point a to b) for the last 5+ years, but I haven't seen much in the clinical content capability testing except for Clinicians on FHIR and perhaps with vendor beta testing with clinicians, etc. Also FDA does regulate many types of devices and some software. How is quality and safety for the patient ensured throughout the information flow/integration process?

Thank you for expressing this. Exactly what’s been in my head.

The NQF Model for Coordination of Care incorporated tasks with outcomes and accountabilities. These tasks can be arranged into workflow. Since the model development, it is clear that another element is needed ... actionable events. We need curated, agreed upon, measurable parameters for each task or set of tasks that would trigger an event detector. If we think in this way, then the required device and its capabilities would flow from that knowledge. A device is search of a use is not the same as a clinical decision in a search for data. In many cases, the data triggering events is a change rather than a threshold; this does not require calibrated accuracy but rather consistency over time (same devise; consistent values). Much of this discussion puts the device before the decision ... it's a chicken & egg problem. It will be a much easier path to implementation if the patient comes first and the device second. A patient's crude uncalibrated scale can deliver changes in weight over time ... and that may be all we need to make a decision.

Agreed. So long as the patient's crude uncalibrated scale is not nonsense. A simple assessment such as "is this normal for you?" may be all that is needed, if the patient or his/her immediate family can attest to what the definition of normal is.

Indeed, despite what studies show and common guidelines/practice, we are each a N of 1 and sometimes don't fit into the "norms." Agree that for patients where their norm is high, low or off scale, @john's question applies to level set expectations and so providers can tailor care accordingly.