Free EMR Newsletter Want to receive the latest news on EMR, Meaningful Use, ARRA and Healthcare IT sent straight to your email? Join thousands of healthcare pros who subscribe to EMR and HIPAA for FREE!!

Alexa Can Truly Give Patients a Voice in Their Health Care (Part 1 of 3)

Posted on October 16, 2017 I Written By

Andy Oram is an editor at O'Reilly Media, a highly respected book publisher and technology information provider. An employee of the company since 1992, Andy currently specializes in open source, software engineering, and health IT, but his editorial output has ranged from a legal guide covering intellectual property to a graphic novel about teenage hackers. His articles have appeared often on EMR & EHR and other blogs in the health IT space. Andy also writes often for O'Reilly's Radar site (http://oreilly.com/) and other publications on policy issues related to the Internet and on trends affecting technical innovation and its effects on society. Print publications where his work has appeared include The Economist, Communications of the ACM, Copyright World, the Journal of Information Technology & Politics, Vanguardia Dossier, and Internet Law and Business. Conferences where he has presented talks include O'Reilly's Open Source Convention, FISL (Brazil), FOSDEM, and DebConf.

The leading pharmaceutical and medical company Merck, together with Amazon Web Services, has recently been exploring the potential health impacts of voice interfaces and natural language processing (NLP) through an Alexa Diabetes Challenge. I recently talked to the five finalists in this challenge. This article explores the potential of new interfaces to transform the handling of chronic disease, and what the challenge reveals about currently available technology.

Alexa, of course, is the ground-breaking system that brings everyday voice interaction with computers into the home. Most of its uses are trivial (you can ask about today’s weather or change channels on your TV), but one must not underestimate the immense power of combining artificial intelligence with speech, one of the most basic and essential human activities. The potential of this interface for disabled or disoriented people is particularly intriguing.

The diabetes challenge is a nice focal point for exploring the more serious contribution made by voice interfaces and NLP. Because of the alarming global spread of this illness, the challenge also presents immediate opportunities that I hope the participants succeed in productizing and releasing into the field. Using the challenge’s published criteria, the judges today announced Sugarpod from Wellpepper as the winner.

This article will list some common themes among the five finalists, look at the background about current EHR interfaces and NLP, and say a bit about the unique achievement of each finalist.

Common themes

Overlapping visions of goals, problems, and solutions appeared among the finalists I interviewed for the diabetes challenge:

  • A voice interface allows more frequent and easier interactions with at-risk individuals who have chronic conditions, potentially achieving the behavioral health goal of helping a person make the right health decisions on a daily or even hourly basis.

  • Contestants seek to integrate many levels of patient intervention into their tools: responding to questions, collecting vital signs and behavioral data, issuing alerts, providing recommendations, delivering educational background material, and so on.

  • Services in this challenge go far beyond interactions between Alexa and the individual. The systems commonly anonymize and aggregate data in order to perform analytics that they hope will improve the service and provide valuable public health information to health care providers. They also facilitate communication of crucial health data between the individual and her care team.

  • Given the use of data and AI, customization is a big part of the tools. They are expected to determine the unique characteristics of each patient’s disease and behavior, and adapt their advice to the individual.

  • In addition to Alexa’s built-in language recognition capabilities, Amazon provides the Lex service for sophisticated text processing. Some contestants used Lex, while others drew on other research they had done building their own natural language processing engines.

  • Alexa never initiates a dialog, merely responding when the user wakes it up. The device can present a visual or audio notification when new material is present, but it still depends on the user to request the content. Thus, contestants are using other channels to deliver reminders and alerts such as messaging on the individual’s cell phone or alerting a provider.

  • Alexa is not HIPAA-compliant, but may achieve compliance in the future. This would help health services turn their voice interfaces into viable products and enter the mainstream.

Some background on interfaces and NLP

The poor state of current computing interfaces in the medical field is no secret–in fact, it is one of the loudest and most insistent complaints by doctors, such as on sites like KevinMD. You can visit Healthcare IT News or JAMA regularly and read the damning indictments.

Several factors can be blamed for this situation, including unsophisticated electronic health records (EHRs) and arbitrary reporting requirements by Centers for Medicare & Medicaid Services (CMS). Natural language processing may provide one of the technical solutions to these problems. The NLP services by Nuance are already famous. An encouraging study finds substantial time savings through using NLP to enter doctor’s insights. And on the other end–where doctors are searching the notes they previously entered for information–a service called Butter.ai uses NLP for intelligent searches. Unsurprisingly, the American Health Information Management Association (AHIMA) looks forward to the contributions of NLP.

Some app developers are now exploring voice interfaces and NLP on the patient side. I covered two such companies, including the one that ultimately won the Alexa Diabetes Challenge, in another article. In general, developers using these interfaces hope to eliminate the fuss and abstraction in health apps that frustrate many consumers, thereby reaching new populations and interacting with them more frequently, with deeper relationships.

The next two parts of this article turn to each of the five finalists, to show the use they are making of Alexa.

Eliminate These Five Flaws to Improve Asset Utilization in Healthcare

Posted on October 4, 2017 I Written By

The following is a guest blog post by Mohan Giridharadas, Founder and CEO, LeanTaaS.

The passage of the Health Information Technology for Economic and Clinical Health (HITECH) Act accelerated the deployment of electronic health records (EHRs) across healthcare. The overwhelming focus was to capture every patient encounter and place it into an integrated system of records. Equipped with this massive database of patient data, health systems believed they could make exponential improvements to patient experiences and outcomes.

The pace of this migration resulted in some shortcuts being taken — the consequences of which are now apparent to discerning CFOs and senior leaders. Among these shortcuts was the use of resources and capacity as the basis of scheduling patients; this concept is used by hundreds of schedulers in every health system. While simple to grasp, the definition is mathematically flawed.

Not being able to offer a new patient an appointment for at least 10 days negatively impacts the patient experience. Likewise, exceeding capacity by scheduling too many appointments results in long wait times for patients, which also negatively impacts their experience. The troubling paradox is that the very asset creating long wait times and long lead times for appointments also happens to perform at ~50 percent utilization virtually every day. The impact of a mathematically flawed foundation results in alternating between overutilization (causing long patient wait times and/or long delays in securing an appointment) and under-utilization (a waste of expensive capital and human assets).

Here are five specific flaws in the mathematical foundation of health system scheduling:

1. A medical appointment is a stochastic — not deterministic — event.

Every health system has some version of this grid — assets across the top, times of the day for each day of the week along the side — on paper, in electronic format or on a whiteboard. The assets could be specific (e.g., the GE MRI machine or virtual MRI #1, #2, etc.). As an appointment gets confirmed, the appropriate range of time on the grid gets filled in to indicate that the slot has been reserved.

Your local racquet club uses this approach to reserve tennis courts for its members. It works beautifully because the length of a court reservation is precisely known (i.e., deterministic) to be exactly one hour in duration. Imagine the chaos if club rules were changed to allow players to hold their reservation even if they arrive late (up to 30 minutes late) and play until they were tired (up to a maximum of two hours). This would make the start and end times for a specific tennis appointment random (i.e., stochastic). Having a reservation would no longer mean you would actually get on the court at your scheduled time. This happens to patients every day across many parts of a health system. The only way to address the fact that a deterministic framework was used to schedule a stochastic event is to “reserve capacity” either in the form of a time buffer (i.e., pretend that each appointment is actually longer than necessary) or as an asset buffer (i.e., hold some assets in reserve).

2. The asset cannot be scheduled in isolation; a staff member has to complete the treatment.

Every appointment needs a nurse, provider or technician to complete the treatment. These staff members are scheduled independently and have highly variable workloads throughout the day. Having an asset that is available without estimating the probability of the appropriate staff member also being available at that exact time will invariably result in delays. Imagine if the tennis court required the club pro be present for the first 10 and last 10 minutes of every tennis appointment. The grid system wouldn’t work in that case either (unless the club was willing to have one tennis pro on the staff for every tennis court).

3. It requires an estimation of probabilities.

Medical appointments have a degree of randomness — no-shows, cancellations and last-minute add-ons are a fact of life, and some appointments run longer or shorter than expected. Every other scheduling system faced with such uncertainty incorporates the mathematics of probability theory. For example, airlines routinely overbook their flights; the exact number of overbooked seats sold depends on the route, the day and the flight. They usually get it right, and the cancellations and no-shows create enough room for the standby passengers. Occasionally, they get it wrong and more passengers hold tickets than the number of seats on the airplane. This results in the familiar process of finding volunteers willing to take a later flight in exchange for some sort of compensation. Nothing in the EHR or scheduling systems used by hospitals allows for this strategic use of probability theory to improve asset utilization.

4. Start time and duration are independent variables.

Continuing with the airplane analogy: As a line of planes work their way toward the runway for departure, the controller really doesn’t care about each flight’s duration. Her job is to get each plane safely off the ground with an appropriate gap between successive takeoffs. If one 8-hour flight were to be cancelled, the controller cannot suddenly decide to squeeze in eight 1-hour flights in its place. Yet, EHRs and scheduling systems have conflated start time and appointment duration into a single variable. Managers, department leaders and schedulers have been taught that if they discover a 4-hour opening in the “appointment grid” for any specific asset, they are free to schedule any of the following combinations:

  • One 4-hour appointment
  • Two 2-hour appointments
  • One 2-hour appointment and two 1-hour appointments in any order
  • One 3-hour appointment and one 1-hour appointment in either order
  • Four 1-hour appointments

These are absolutely not equivalent choices. Each has wildly different resource-loading implications for the staff, and each choice has a different probability profile of starting or ending on time. This explains why the perfectly laid out appointment grid at the start of each day almost never materializes as planned.

5. Setting appointments is more complicated than first-come, first-served.

Schedulers typically make appointments on a first-come, first-served basis. If a patient were scheduling an infusion treatment or MRI far in advance, the patient would likely hear “the calendar is pretty open on that day — what time would you like?” What seems like a patient-friendly gesture is actually mathematically incorrect. The appointment options for each future day should be a carefully orchestrated set of slots of varying durations that will result in the flattest load profile possible. In fact, blindly honoring patient appointment requests just “kicks the can down the road”; the scheduler has merely swapped the inconvenience of appointment time negotiation for excessive patient delays on the day of treatment. Instead, the scheduler should steer the patient to one of the recommended appointment slots based on the duration for that patient’s specific treatment.

In the mid-1980s, Sun Microsystems famously proclaimed that the “network is the computer.” The internet and cloud computing were not yet a thing, so most people could not grasp the concept of computers needing to be interconnected and that the computation would take place in the network and not on the workstation. In healthcare scheduling, “the duration is the resource” — the number of slots of a specific duration must be counted and allocated judiciously at various points throughout the day. Providers should carefully forecast the volume and the duration mix of patients they expect to serve for every asset on every day of the week. With that knowledge the provider will know, for example, that on Mondays, we need 10 1-hour treatments, 15 2-hour treatments and so on. Schedulers could then strategically decide to space appointments throughout the day (or cluster them in the morning or afternoon) by offering up two 1-hour slots at 7:10 a.m., one 1-hour slot at 7:40 a.m., etc. The allocation pattern matches the availability of the staff and the underlying asset to deliver the most level-loaded schedule for each day. In this construct, the duration is the resource being offered up to patients one at a time with the staff and asset availability as mathematical constraints to the equation (along with dozens of other operational constraints).

Health systems need to re-evaluate the mathematical foundation used to guide their day-to-day operations — and upon which the quality of the patient experience relies. All the macro forces in healthcare (more patients, older patients, higher incidence of chronic illnesses, lower reimbursements, push toward value-based care, tighter operating and capital budgets) indicate an urgent need to be able to do more with existing assets without upsetting patient flow. A strong mathematical foundation will enable a level of operational excellence to help health systems increase their effective capacity for treating more patients while simultaneously improving the overall flow and reducing the wait time.

About Mohan Giridharadas
Mohan Giridharadas is an accomplished expert in lean methodologies. During his 18-year career at McKinsey & Company (where he was a senior partner/director for six years), he co-created the lean service operations practice and ran the North American lean manufacturing and service operations practices and the Asia-Pacific operations practice. He has helped numerous Fortune 500 companies drive operational efficiency with lean practices. As founder and CEO of LeanTaaS, a Silicon Valley-based innovator of cloud-based solutions to healthcare’s biggest challenges, Mohan works closely with dozens of leading healthcare institutions including Stanford Health Care, UCHealth, NewYork-Presbyterian, Cleveland Clinic, MD Anderson and more. Mohan holds a B.Tech from IIT Bombay, MS in Computer Science from Georgia Institute of Technology and an MBA from Stanford GSB. He is on the faculty of Continuing Education at Stanford University and UC Berkeley Haas School of Business and has been named by Becker’s Hospital Review as one of the top entrepreneurs innovating in healthcare. For more information on LeanTaaS, please visit http://www.leantaas.com and follow the company on Twitter @LeanTaaS, Facebook at https://www.facebook.com/LeanTaaS and LinkedIn at https://www.linkedin.com/company/leantaas.

Translating Social Determinants of Health Into Clinical Action

Posted on September 25, 2017 I Written By

The following is a guest blog post by Anton Berisha, MD, Senior Director, Clinical Analytics and Innovation, Health Care, LexisNexis Risk Solutions.
The medical community recognizes the importance of social determinants of health (SDOH) – social, economic and environmental conditions in which people are born, grow, live, work and age that impact their health – as significant and direct risk factors for a large number of health care outcomes.

The negative outcomes include stress, mental health and behavioral disorders, alcoholism and substance abuse, to name a few. Negative SDOH worsen a slew of major chronic conditions, from hypertension and Coronary Artery Disease to obesity; they also lead to lower patient engagement and medication adherence while increasing low-intensity ER visits and hospital admissions and readmissions.

In fact, a study shows that medical care determines only 20% of overall health outcomes while social, economic and environmental factors determine about 50% of overall health. The National Quality Forum, Centers for Disease Control and Prevention and World Health Organization have all acknowledged the importance of addressing SDOH in health care.

Not all SDOH are “created equal”

When it comes to SDOH, there is a misconception that all data regarding a person’s lifestyle, environment, situation and behaviors relate to their health. Although there is a myriad of basic demographic data, survey data and other Electronic Health Records (EHR) data available to providers today, much of it has a limited potential for identifying additional health costs and risks.

The key to addressing SDOH is to use current, comprehensive and longitudinal data that can be consistently linked to specific patient populations and provided in a standardized format. One example is attributes derived from public records data such as proximity to relatives, education, income, bankruptcy, addresses and criminal convictions.

Moreover, each SDOH attribute has to be clinically validated against actual healthcare outcomes. Clinically validating attributes is critical to successful predictive analytics because some attributes do not correlate strongly to health outcomes.

For example, while knowing how close an individual’s nearest relative or associate lives to the patient does correlate to health outcomes; knowing how many of those relatives or associates have registered automobiles does not. Even when attributes are clinically validated, different attributes correlate to different outcomes with different accuracy strengths.

Translating SDOH into actionable intelligence

After SDOH have been correlated to healthcare outcomes, providers have two implementation options. One is to use relevant individual SDOH attributes per outcome in clinical and analytic models to better assess and predict risk for patients. Another is to use SDOH as part of risk scores estimating specific healthcare risks; for e.g., to estimate an individual’s total health care risk over the next 12 months based on cost; a 30-day readmission risk; or a patient engagement score.

Risk estimation can be done either in combination with other types of legacy healthcare data, such as claims, prescription and EHR data or with SDOH alone, in the absence of medical claims.

Recently, a client of LexisNexis® Health Care did an independent study to evaluate the impact and usefulness of Socioeconomic Health Score (SEHS) in risk assessment for several key chronic conditions, when no other data are available. Findings proved that the top decile of SEHS captures significantly more members with given conditions than the bottom decile. The study concluded that the difference was important and very helpful in estimating risks in a newly acquired population without legacy healthcare data.

Integrating SDOH into clinical workflows and care recommendations

Validated SDOH can be presented in a form of risk drivers or reason codes directing the clinician toward the most important factors influencing a given negative outcome for each patient: income, education, housing or criminal records.

The risk drivers and reason codes can then be integrated into workflows within the clinician’s IT systems, such as the EHR or care and case management, in the form of an easy-to-understand presentation. It could be a data alert that is customizable to patients, treatments and conditions, helping the provider make score-based decisions with greater accuracy and confidence. At this point, the SDOH information becomes actionable because it has the following characteristics:

  • It is based on hard facts on every individual.
  • It is based on correlation and statistical significance testing of large pools of patients with similar behavior.
  • It provides clear and understandable reason codes driving the negative outcomes.
  • It can be tied to intervention strategies (outlined below) that have demonstrated positive results.

Clinicians empowered with actionable SDOH information can modify their interventions and follow-up strategies accordingly. Based on resources at hand, patients living in negative SDOH could be either properly managed by clinicians themselves or other medical staff, social workers and newly created roles such as health coaches. Sub-populations at risk could benefit from access to community resources to get help with housing (permanent supportive housing for homeless), transportation, education, childcare and employment assistance.

Moreover, SDOH are particularly effective in helping providers develop a population health management strategy fueled by prioritized tactics for preventive care. Tactics can range from promotion of healthy food to free screening services. For patients with chronic diseases (who can typically be managed appropriately when they adhere to therapy and healthy lifestyle choices), SDOH-informed interventions can help keep them under control and potentially reduce severity. For patients recently released from the hospital, aftercare counseling could prevent complications and readmissions.

To sum it up

Socioeconomic data is a vital force for healthcare risk prediction as it provides a view into the otherwise hidden risks that cannot be identified through traditional data sources. When SDOH are clinically validated and correlated to healthcare outcomes, they help providers better understand an individual’s risk level and address it through appropriate intervention strategies.

Searching EMR For Risk-Related Words Can Improve Care Coordination

Posted on September 18, 2017 I Written By

Anne Zieger is a healthcare journalist who has written about the industry for 30 years. Her work has appeared in all of the leading healthcare industry publications, and she's served as editor in chief of several healthcare B2B sites.

Though healthcare organizations are working on the problem, they’re still not as good at care coordination as they should be. It’s already an issue and will only get worse under value-based care schemes, in which the ability to coordinate care effectively could be a critical issue for providers.

Admittedly, there’s no easy way to solve care coordination problems, but new research suggests that basic health IT tools might be able to help. The researchers found that digging out important words from EMRs can help providers target patients needing extra care management and coordination.

The article, which appears in JMIR Medical Informatics, notes that most care coordination programs have a blind spot when it comes to identifying cases demanding extra coordination. “Care coordination programs have traditionally focused on medically complex patients, identifying patients that qualify by analyzing formatted clinical data and claims data,” the authors wrote. “However, not all clinically relevant data reside in claims and formatted data.”

For example, they say, relying on formatted records may cause providers to miss psychosocial risk factors such as social determinants of health, mental health disorder, and substance abuse disorders. “[This data is] less amenable to rapid and systematic data analyses, as these data are often not collected or stored as formatted data,” the authors note.

To address this issue, the researchers set out to identify psychosocial risk factors buried within a patient’s EHR using word recognition software. They used a tool known as the Queriable Patient Inference Dossier (QPID) to scan EHRs for terms describing high-risk conditions in patients already in care coordination programs.

After going through the review process, the researchers found 22 EHR-available search terms related to psychosocial high-risk status. When they were able to find nine or more of these terms in the patient’s EHR, it predicted that a patient would meet criteria for participation in a care coordination program. Presumably, this approach allowed care managers and clinicians to find patients who hadn’t been identified by existing care coordination outreach efforts.

I think this article is valuable, as it outlines a way to improve care coordination programs without leaping over tall buildings. Obviously, we’re going to see a lot more emphasis on harvesting information from structured data, tools like artificial intelligence, and natural language processing. That makes sense. After all, these technologies allow healthcare organizations to enjoy both the clear organization of structured data and analytical options available when examining pure data sets. You can have your cake and eat it too.

Obviously, we’re going to see a lot more emphasis on harvesting information from structured data, tools like artificial intelligence and natural language processing. That makes sense. After all, these technologies allow healthcare organizations to enjoy both the clear organization of structured data and analytical options available when examining pure data sets. You can have your cake and eat it too.

Still, it’s good to know that you can get meaningful information from EHRs using a comparatively simple tool. In this case, parsing patient medical records for a couple dozen keywords helped the authors find patients that might have otherwise been missed. This can only be good news.

Yes, there’s no doubt we’ll keep on pushing the limits of predictive analytics, healthcare AI, machine learning and other techniques for taming wild databases. In the meantime, it’s good to know that we can make incremental progress in improving care using simpler tools.

Analytics Take an Unusual Turn at PeraHealth

Posted on August 17, 2017 I Written By

Andy Oram is an editor at O'Reilly Media, a highly respected book publisher and technology information provider. An employee of the company since 1992, Andy currently specializes in open source, software engineering, and health IT, but his editorial output has ranged from a legal guide covering intellectual property to a graphic novel about teenage hackers. His articles have appeared often on EMR & EHR and other blogs in the health IT space. Andy also writes often for O'Reilly's Radar site (http://oreilly.com/) and other publications on policy issues related to the Internet and on trends affecting technical innovation and its effects on society. Print publications where his work has appeared include The Economist, Communications of the ACM, Copyright World, the Journal of Information Technology & Politics, Vanguardia Dossier, and Internet Law and Business. Conferences where he has presented talks include O'Reilly's Open Source Convention, FISL (Brazil), FOSDEM, and DebConf.

Data scientists in all fields have learned to take data from unusual places. You’d think that monitoring people in a hospital for changes in their conditions would be easier than other data-driven tasks, such as tracking planets in far-off solar systems, but in all cases some creativity is needed. That’s what PeraHealth, a surveillance system for hospital patients, found out while developing alerts for clinicians.

It’s remarkably hard to identify at-risk patients in hospitals, even with so many machines and staff busy monitoring them. For instance, a nurse on each shift may note in the patient’s record that certain vital signs are within normal range, and no one might notice that the vital signs are gradually trending worse and worse–until a crisis occurs.

PeraHealth identifies at-risk patients through analytics and dashboards that doctors and nurses can pull up. They can see trends over a period of several shifts, and quickly see which patients in the ward are the most at risk. PeraHealth is a tool for both clinical surveillance and communication.

Michael Rothman, co-founder and Chief Science Officer, personally learned the dangers of insufficient monitoring in 2003 when a low-risk operation on his mother led to complications and her unfortunate death. Rothman and his brother decided to make something positive from the tragedy. They got permission from the hospital to work there for three weeks, applying Michael’s background in math and data analysis (he has worked in the AI department of IBM’s Watson research labs, among other places) and his brother’s background in data visualization. Their goal, arguably naive: to find a single number that summarizes patient risk, and expose that information in a usable way to clinicians.

Starting with 70 patients from the cardiac unit, they built a statistical model that they tested repeatedly with 1,200 patients, 6,000 patients, and finally 25,000 patients. At first they hoped to identify extra data that the nurse could enter into the record, but the chief nurse laid down, in no uncertain terms, that the staff was already too busy and that collecting more data was out of the question. It came time to get creative with data that was already being collected and stored.

The unexpected finding was that vital signs were not a reliable basis for assessing a patient’s trends. Even though they’re “hard” (supposedly objective) data, they bounce around too much.

Instead of relying on just vital signs, PeraHealth also pulls in nursing assessments–an often under-utilized source of information. On each shift, a nurse records information on a dozen different physical systems as well as essential facts such as whether a patient stopping eating or was having trouble walking. It turns out that this sort of information reliably indicates whether there’s a problem. Many of the assessments are simple, yes/no questions.

Rothman analyzed hospital data to find variables that predicted risk. For instance, he compared the heart rates of 25,000 patients before they left the hospital and checked who lived for a year longer. The results formed a U-shaped curve, showing that heart rates above a certain level or below a certain level predicted a bad outcome. It turns out that this meaure works equally well within the hospital, helping to predict admission to the ICU, readmission to the ICU, and readmission after discharge.

The PeraHealth team integrated their tool with the hospital’s EHR and started producing graphs for the clinicians in 2007. Now they can point to more than 25 peer-reviewed articles endorsing their approach, some studies comparing before-and-after outcomes, and others comparing different parts of the hospital with some using PeraHealth and others not using it. The service is now integrated with major EHR vendors.

PeraHealth achieved Rothman’s goal of producing a single meaningful score to rate patient risk. Each new piece of data that goes into the EHR triggers a real-time recalculation of the score and a new dot on a graph presented to the nurses. In order to save the nurses from signing into the EHR, PeraHealth put a dashboard on the nurse’s kiosk with all the patients’ graphs. Color-coding denotes which patients are sickest. PeraHealth also shows which patients to attend to first. In case no one looks at the screen, at some hospitals the system sends out text alerts to doctors about the most concerned patients.

PeraHealth is now expanding. In an experiment, they did phone interviews with people in a senior residential facility, and identified many of those who were deteriorating. So the basic techniques may be widely applicable to data-driven clinical decision support. But without analytics, one never knows which data is most useful.

Tips on Implementing Text Analytics in Healthcare

Posted on July 6, 2017 I Written By

Anne Zieger is a healthcare journalist who has written about the industry for 30 years. Her work has appeared in all of the leading healthcare industry publications, and she's served as editor in chief of several healthcare B2B sites.

Most of us would agree that extracting clinical data from unstructured physician notes would be great. At present, few organizations have deployed such tools, nor have EMR vendors come to the rescue en masse, and the conventional wisdom holds that text analytics would be crazy expensive. I’ve always suspected that digging out and analyzing this data may be worth the trouble, however.

That’s why I really dug a recent article from HealthCatalyst’s Eric Just, which seemed to offer some worthwhile ideas on how to use text analytics effectively. Just, who is senior vice president of product development, made a good case for giving this approach a try. (Note: HealthCatalyst and partner Regenstrief Institute offer solutions in this area.)

The article includes an interesting case study explaining how healthcare text analytics performed head-to-head against traditional research methods.

It tells the story of a team of analysts in Indiana that set out to identify peripheral artery disease (PAD) patients across two health systems. At first gasp, things weren’t going well. When researchers looked at EMR and claims data, they found that failed to identify over 75% of patients with this condition, but text analytics improved their results dramatically.

Using ICD and CPT codes for PAD, and standard EMR data searches, team members had identified less than 10,000 patients with the disorder. However, once they developed a natural language processing tool designed to sift through text-based data, they discovered that there were at least 41,000 PAD patients in the population they were studying.

To get this kind of results, Just says, there are three key features a medical text analytics tool should have:

  • The medical text analytics software should tailor results to a given user’s needs. For example, he notes that if the user doesn’t have permission to view PHI, the analytics tool should display only nonprivate data.
  • Medical text analytics tools should integrate medical terminology to improve the scope of searches. For example, when a user does a search on the term “diabetes” the search tool should automatically be capable of displaying results for “NIDDM,” as this broadens the search to include more relevant content.
  • Text analytics algorithms should do more than just find relevant terms — they should provide context as well as content. For example, a search for patients with “pneumonia,” done with considering context, would also bring up phrases like “no history of pneumonia.” A better tool would be able to rule out phrases like “no history of pneumonia,” or “family history of pneumonia” from a search for patients who have been treated for this illness.

The piece goes into far more detail than I can summarize here, so I recommend you read it in full if you’re interested in leveraging text analytics for your organization.

But for what it’s worth, I came away from the piece with the sense that analyzing your clinical textual information is well worth the trouble — particularly if EMR vendors being to add such tools to their systems. After all, when it comes to improving outcomes, we need all the help we can get.

Hands-On Guidance for Data Integration in Health: The CancerLinQ Story

Posted on June 15, 2017 I Written By

Andy Oram is an editor at O'Reilly Media, a highly respected book publisher and technology information provider. An employee of the company since 1992, Andy currently specializes in open source, software engineering, and health IT, but his editorial output has ranged from a legal guide covering intellectual property to a graphic novel about teenage hackers. His articles have appeared often on EMR & EHR and other blogs in the health IT space. Andy also writes often for O'Reilly's Radar site (http://oreilly.com/) and other publications on policy issues related to the Internet and on trends affecting technical innovation and its effects on society. Print publications where his work has appeared include The Economist, Communications of the ACM, Copyright World, the Journal of Information Technology & Politics, Vanguardia Dossier, and Internet Law and Business. Conferences where he has presented talks include O'Reilly's Open Source Convention, FISL (Brazil), FOSDEM, and DebConf.

Institutions throughout the health care field are talking about data sharing and integration. Everyone knows that improved care, cost controls, and expanded research requires institutions who hold patient data to safely share it. The American Society of Clinical Oncology’s CancerLinQ, one of the leading projects analyzing data analysis to find new cures, has tackled data sharing with a large number of health providers and discovered just how labor-intensive it is.

CancerLinQ fosters deep relationships and collaborations with the clinicians from whom it takes data. The platform turns around results from analyzing the data quickly and to give the clinicians insights they can put to immediate use to improve the care of cancer patients. Issues in collecting, storing, and transmitting data intertwine with other discussion items around cancer care. Currently, CancerLinQ isolates the data from each institution, and de-identifies patient information in order to let it be shared among participating clinicians. CancerLinQ LLC is a wholly-owned nonprofit subsidiary of ASCO, which has registered CancerLinQ as a trademark.

CancerLinQ logo

Help from Jitterbit

In 2015, CancerLinQ began collaborating with Jitterbit, a company devoted to integrating data from different sources. According to Michele Hazard, Director of Healthcare Solutions, and George Gallegos, CEO, their company can recognize data from 300 different sources, including electronic health records. At the beginning, the diversity and incompatibility of EHRs was a real barrier. It took them several months to figure out each of the first EHRs they tackled, but now they can integrate a new one quickly. Oncology care, the key data needed by CancerLinQ, is a Jitterbit specialty.

Jitterbit logo

One of the barriers raised by EHRs is licensing. The vendor has to “bless” direct access to EHR and data imported from external sources. HIPAA and licensing agreements also make tight security a priority.

Another challenge to processing data is to find records in different institutions and accurately match data for the correct patient.

Although the health care industry is moving toward the FHIR standard, and a few EHRs already expose data through FHIR, others have idiosyncratic formats and support older HL7 standards in different ways. Many don’t even have an API yet. In some cases, Jitterbit has to export the EHR data to a file, transfer it, and unpack it to discover the patient data.

Lack of structure

Jitterbit had become accustomed to looking in different databases to find patient information, even when EHRs claimed to support the same standard. One doctor may put key information under “diagnosis” while another enters it under “patient problems,” and doctors in the same practice may choose different locations.

Worse still, doctors often ignore the structured fields that were meant to hold important patient details and just dictate or type it into a free-text note. CancerLinQ anticipated this, unpacking the free text through optical character recognition (OCR) and natural language processing (NLP), a branch of artificial intelligence.

It’s understandable that a doctor would evade the use of structured fields. Just think of the position she is in, trying to keep a complex cancer case in mind while half a dozen other patients sit in the waiting room for their turn. In order to use the structured field dedicated to each item of information, she would have to first remember which field to use–and if she has privileges at several different institutions, that means keeping the different fields for each hospital in mind.

Then she has to get access to the right field, which may take several clicks and require movement through several screens. The exact information she wants to enter may or may not be available through a drop-down menu. The exact abbreviation or wording may differ from EHR to EHR as well. And to carry through a commitment to using structured fields, she would have to go through this thought process many times per patient. (CancerLinQ itself looks at 18 Quality eMeasures today, with the plan to release additional measures each year.)

Finally, what is the point of all this? Up until recently, the information would never come back in a useful form. To retrieve it, she would have to retrace the same steps she used to enter the structured data in the first place. Simpler to dump what she knows into a free-text note and move on.

It’s worth mentioning that this Babyl of health care information imposes negative impacts on the billing and reimbursement process, even though the EHRs were designed to support those very processes from the start. Insurers have to deal with the same unstructured data that CancerLinQ and Jitterbit have learned to read. The intensive manual process of extracting information adds to the cost of insurance, and ultimately the entire health care system. The recent eClinicalWorks scandal, which resembles Volkswagon’s cheating on auto emissions and will probably spill out to other EHR vendors as well, highlights the failings of health data.

Making data useful

The clue to unblocking this information logjam is deriving insights from data that clinicians can immediately see will improve their interventions with patients. This is what the CancerLinQ team has been doing. They run analytics that suggest what works for different categories of patients, then return the information to oncologists. The CancerLinQ platform also explains which items of data were input to these insights, and urges the doctors to be more disciplined about collecting and storing the data. This is a human-centered, labor-intensive process that can take six to twelve months to set up for each institution. Richard Ross, Chief Operating Officer of CancerLinQ calls the process “trench warfare,” not because its contentious but because it is slow and requires determination.

Of the 18 measures currently requested by CancerLinQ, one of the most critical data elements driving the calculation of multiple measures is staging information: where the cancerous tumors are and how far it has progressed. Family history, treatment plan, and treatment recommendations are other examples of measures gathered.

The data collection process has to start by determining how each practice defines a cancer patient. The CancerLinQ team builds this definition into its request for data. Sometimes they submit “pull” requests at regular intervals to the hospital or clinic, whereas other times the health care provider submits the data to them at a time of its choosing.

Some institutions enforce workflows more rigorously than others. So in some hospitals, CancerLinQ can persuade the doctors to record important information at a certain point during the patient’s visit. In other hospitals, doctors may enter data at times of their own choosing. But if they understand the value that comes from this data, they are more likely to make sure it gets entered, and that it conforms to standards. Many EHRs provide templates that make it easier to use structured fields properly.

When accepting information from each provider, the team goes through a series of steps and does a check-in with the provider at each step. The team evaluates the data in a different stage for each criterion: completeness, accuracy of coding, the number of patients reported, and so on. By providing quick feedback, they can help the practice improve its reporting.

The CancerLinQ/Jitterbit story reveals how difficult it is to apply analytics to health care data. Few organizations can afford the expertise they apply to extracting and curating patient data. On the other hand, CancerLinQ and Jitterbit show that effective data analysis can be done, even in the current messy conditions of electronic data storage. As the next wave of technology standards, such as FHIR, fall into place, more institutions should be able to carry out analytics that save lives.

Scenarios for Health Care Reform (Part 2 of 2)

Posted on May 18, 2017 I Written By

Andy Oram is an editor at O'Reilly Media, a highly respected book publisher and technology information provider. An employee of the company since 1992, Andy currently specializes in open source, software engineering, and health IT, but his editorial output has ranged from a legal guide covering intellectual property to a graphic novel about teenage hackers. His articles have appeared often on EMR & EHR and other blogs in the health IT space. Andy also writes often for O'Reilly's Radar site (http://oreilly.com/) and other publications on policy issues related to the Internet and on trends affecting technical innovation and its effects on society. Print publications where his work has appeared include The Economist, Communications of the ACM, Copyright World, the Journal of Information Technology & Politics, Vanguardia Dossier, and Internet Law and Business. Conferences where he has presented talks include O'Reilly's Open Source Convention, FISL (Brazil), FOSDEM, and DebConf.

The first part of this article suggested two scenarios that could promote health care reform. We’ll finish off the scenarios in this part of the article.

Capitalism Disrupts Health Care

In the third scenario, reform is stimulated by an intrepid data science firm that takes on health care with greater success than most of its predecessors. After assembling an impressive analytics toolkit from open source software components–thus simplifying licensing–it approaches health care providers and offers them a deal they can’t refuse: analytics demonstrated to save them money and support their growth, all delivered for free. The data science firm asks in return only that they let it use deidentified data from their patients and practices to build an enhanced service that it will offer paying customers.

Some health care providers balk at the requirement to share data, but their legal and marketing teams explain that they have been doing it for years already with companies whose motives are less commendable. Increasingly, the providers are won over. The analytics service appeals particularly to small, rural, and safety-net providers. Hammered by payment cuts and growing needs among their populations, they are on the edge of going out of business and grasp the service as their last chance to stay in the black.

Participating in the program requires the extraction of data from electronic health records, and some EHR vendors try to stand in the way in order to protect their own monopoly on the data. Some even point to clauses in their licenses that prohibit the sharing. But they get a rude message in return: so valuable are the analytics that the providers are ready to jettison the vendors in a minute. The vendors ultimately go along and even compete on the basis of their ability to connect to the analytics.

Once stability and survival are established, the providers can use the analytics for more and more sophisticated benefits. Unlike the inadequate quality measures currently in use, the analytics provide a robust framework for assessing risk, stratifying populations, and determining how much a provider should be rewarded for treating each patient. Fee-for-outcome becomes standard.

Providers make deals to sign up patients for long-term relationships. Unlike the weak Medicare ACO model, which punishes a provider for things their patients do outside their relationship, the emerging system requires a commitment from the patient to stick with a provider. However, if the patient can demonstrate that she was neglected or failed to receive standard of care, she can switch to another provider and even require the misbehaving provider to cover costs. To hold up their end of this deal, providers find it necessary to reveal their practices and prices. Physician organizations develop quality-measurement platforms such as the recent PRIME registry in family medicine. A race to the top ensues.

What If Nothing Changes?

I’ll finish this upbeat article with a fourth scenario in which we muddle along as we have for years.

The ONC and Centers for Medicare & Medicaid Services continue to swat at waste in the health care system by pushing accountable care. But their ratings penalize safety-net providers, and payments fail to correlate with costs as hoped.

Fee-for-outcome flounders, so health care costs continue to rise to intolerable levels. Already, in Massachusetts, the US state that leads in universal health coverage, 40% of the state budget goes to Medicaid, where likely federal cuts will make it impossible to keep up coverage. Many other states and countries are witnessing the same pattern of rising costs.

The same pressures ride like a tidal wave through the rest of the health care system. Private insurers continue to withdraw from markets or lose money by staying. So either explicitly or through complex and inscrutable regulatory changes, the government allows insurers to cut sick people from their rolls and raise the cost burdens on patients and their employers. As patient rolls shrink, more hospitals close. Political rancor grows as the public watches employer money go into their health insurance instead of wages, and more of their own stagnant incomes go to health care costs, and government budgets tied up in health care instead of education and other social benefits.

Chronic diseases creep through the population, mocking crippled efforts at public health. Rampant obesity among children leads to more and earlier diabetes. Dementia also rises as the population ages, and climate change scatters its effects across all demographics.

Furthermore, when patients realize the costs they must take on to ask for health care, they delay doctor visits until their symptoms are unbearable. More people become disabled or perish, with negative impacts that spread through the economy. Output decline and more families become trapped in poverty. Self-medication for pain and mental illness becomes more popular, with predictable impacts on the opiate addiction crisis. Even our security is affected: the military finds it hard to recruit find healthy soldiers, and our foreign policy depends increasingly on drone strikes that kill civilians and inflame negative attitudes toward the US.

I think that, after considering this scenario, most of us would prefer one of the previous three I laid out in this article. If health care continues to be a major political issue for the next election, experts should try to direct discussion away from the current unproductive rhetoric toward advocacy for solutions. Some who read this article will hopefully feel impelled to apply themselves to one of the positive scenarios and bring it to fruition.

Scenarios for Health Care Reform (Part 1 of 2)

Posted on May 16, 2017 I Written By

Andy Oram is an editor at O'Reilly Media, a highly respected book publisher and technology information provider. An employee of the company since 1992, Andy currently specializes in open source, software engineering, and health IT, but his editorial output has ranged from a legal guide covering intellectual property to a graphic novel about teenage hackers. His articles have appeared often on EMR & EHR and other blogs in the health IT space. Andy also writes often for O'Reilly's Radar site (http://oreilly.com/) and other publications on policy issues related to the Internet and on trends affecting technical innovation and its effects on society. Print publications where his work has appeared include The Economist, Communications of the ACM, Copyright World, the Journal of Information Technology & Politics, Vanguardia Dossier, and Internet Law and Business. Conferences where he has presented talks include O'Reilly's Open Source Convention, FISL (Brazil), FOSDEM, and DebConf.

All reformers in health care know what the field needs to do; I laid out four years ago the consensus about patient-supplied data, widespread analytics, mHealth, and transparency. Our frustration comes in when trying to crack the current hide-bound system open and create change. Recent interventions by US Republicans to repeal the Affordable Care Act, whatever their effects on costs and insurance coverage, offer no promise to affect workflows or treatment. So this article suggests three potential scenarios where reform could succeed, along with a vision of what will happen if none of them take hold.

Patients Forge Their Own Way Forward

In the first scenario, a tiny group of selfer-trackers, athletes, and empowered patients start a movement that ultimately wins over hundreds of millions of individuals.

These scattered enthusiasts, driven to overcome debilitating health problems or achieve extraordinary athletic feats, start to pursue self-tracking with fanaticism. Consumer or medical-grade devices provide them with ongoing data about their progress, and an open source platform such as HIE of One gives them a personal health record (PHR).

They also take charge of their interactions with the health care system. They find that most primary care providers aren’t interested in the data and concerns they bring, or don’t have time to process those data and concerns in the depth they need, or don’t know how to. Therefore, while preserving standard relationships with primary care providers and specialists where appropriate, the self-trackers seek out doctors and other providers to provide consultation about their personal health programs. A small number of providers recognize an opportunity here and set up practices around these consultations. The interactions look quite different from standard doctor visits. The customers, instead of just submitting themselves to examination and gathering advice, steer the conversation and set the goals.

Power relationships between doctors and customers also start to change. Although traditional patients can (and often do) walk away and effectively boycott a practice with which they’re not comfortable, the new customers use this power to set the agenda and to sort out the health care providers they find beneficial.

The turning point probably comes when someone–probabaly a research facility, because it puts customer needs above business models–invents a cheap, comfortable, and easy-to-use device that meets the basic needs for monitoring and transmitting vital signs. It may rest on the waist or some other place where it can be hidden, so that there is no stigma to wearing it constantly and no reason to reject its use on fashion grounds. A beneficent foundation invests several million dollars to make the device available to schoolchildren or some other needy population, and suddenly the community of empowered patients leaps from a miniscule pool to a mainstream phenomenon.

Researchers join the community in search of subjects for their experiments, and patients offer data to the researchers in the hope of speeding up cures. At all times, the data is under control of the subjects, who help to direct research based on their needs. Analytics start to turn up findings that inform clinical decision support.

I haven’t mentioned the collection of genetic information so far, because it requires more expensive processes, presents numerous privacy risks, and isn’t usually useful–normally it tells you that you have something like a 2% risk of getting a disease instead of the general population’s 1% risk. But where genetic testing is useful, it can definitely fit into this system.

Ultimately, the market for consultants that started out tiny becomes the dominant model for delivering health care. Specialists and hospitals are brought in only when their specific contributions are needed. The savings that result bring down insurance costs for everyone. And chronic disease goes way down as people get quick feedback on their lifestyle choices.

Government Puts Its Foot Down

After a decade of cajoling health care providers to share data and adopt a fee-for-outcome model, only to witness progress at a snail’s pace, the federal government decides to try a totally different tack in this second scenario. As part of the Precision Medicine initiative (which originally planned to sign up one million volunteers), and leveraging the ever-growing database of Medicare data, the Office of the National Coordinator sets up a consortium and runs analytics on top of its data to be shared with all legitimate researchers. The government also promises to share the benefits of the analytics with anyone in the world who adds their data to the database.

The goals of the analytics are multi-faceted, combining fraud checks, a search for cures, and everyday recommendations about improving interventions to save money and treat patients earlier in the disease cycle. The notorious 17-year gap between research findings and widespread implementation shrinks radically. Now, best practices are available to any patient who chooses to participate.

As with the personal health records in the previous scenario, the government database in this scenario creates a research platform of unprecedented size, both in the number of records and the variety of participating researchers.

To further expand the power of the analytics, the government demands exponentially greater transparency not just in medical settings but in all things that make us sick: the food we eat (reversing the rulings that protect manufacturers and restaurants from revealing what they’re putting in our bodies), the air and water that surrounds us, the effects of climate change (a major public health issue, spreading scourges such as mosquito-borne diseases and heat exhaustion), disparities in food and exercise options among neighborhoods, and more. Public awareness leads to improvements in health that lagged for decades.

In the next section of this article, I’ll present a third scenario that achieves reform from a different angle.

What’s a Patient?

Posted on May 10, 2017 I Written By

John Lynn is the Founder of the HealthcareScene.com blog network which currently consists of 10 blogs containing over 8000 articles with John having written over 4000 of the articles himself. These EMR and Healthcare IT related articles have been viewed over 16 million times. John also manages Healthcare IT Central and Healthcare IT Today, the leading career Health IT job board and blog. John is co-founder of InfluentialNetworks.com and Physia.com. John is highly involved in social media, and in addition to his blogs can also be found on Twitter: @techguy and @ehrandhit and LinkedIn.

For quite a while I’ve been pushing the idea that healthcare needs to move beyond treating patients. Said another way, we need to move beyond just helping people who have health problems which are causing them to complain and move into treating patients that otherwise feel healthy.

Said another way, Wanda Health once told me “The definition of a healthy patient is someone who’s not been studied long enough.”

If you look long enough and hard enough, we all have health issues or we’re at risk for health issues. There’s always something that could be done to help all of us be healthier. That’s a principle that healthcare hasn’t embraced because our reimbursement models are focused on treating a patients’ chief complaint.

In another conversation with NantHealth, they suggested the idea that we should work towards knowing the patient so well that you know the treatment they need before you even physically see the patient.

These two ideas go naturally together and redefine our current definition of patient. In the above context, all of us would be considered patients since I have little doubt that all of us have health issues that could be addressed if we only knew the current state of our health better.

While NantHealth’s taken a number of stock hits lately for overpromising and under delivering, the concept I heard them describe is one that will become a reality. It could be fair to say that their company was too early for such a big vision, but it’s inspiring to think about creating technology and collecting enough data on a patient that you already know how to help the patient before they even come into the office. That would completely change the office visit paradox that we know today.

This is an ambitious vision, but it doesn’t seem like a massive stretch of the imagination either. That’s what makes it so exciting to me. Now imagine trying to do something like this in the previous paper chart world. Yeah, it’s pretty funny to just even think about it. Same goes with what we call clinical decision support today.