Veda solutions

For technology

0
0
0
technology
Artificial Intelligence, ChatGPT, and the Relationship Between Humans and Machines
Optimal and Proven Provider Data from Veda
Star Systems Meet Star Ratings: Using Science and Imagination to Solve Healthcare’s Most Complex Data Problems
Why Healthcare is Behind in AI and How The Industry Can Catch Up

Artificial Intelligence, ChatGPT, and the Relationship Between Humans and Machines

By: Dr. Bob Lindner, Chief Science & Technology Officer, Co-Founder

If the explosive launch of ChatGPT has taught us anything, it’s that there is a growing appetite for engaging with AI. According to a recent UBS study, the chatbot from OpenAI reached 100 million monthly active users in January— only two months after its launch. By comparison, it took TikTok about nine months to reach that milestone and Instagram two-and-a-half years.

While ChatGPT and the generative AI that powers it represent the latest advancements in AI and machine learning, the fact is that organizations and individuals have been trying to harness the power of AI for years. Some see it as the wave of the future. Others are scared of what it portends for the complicated relationships between humans and machines.

Many people are so afraid of being displaced by the automation that artificial intelligence brings that they overlook the benefits of this amazing technology. But the fear of “robots replacing humans” isn’t the only thing that gives people pause. There’s also concern that machines will make unacceptable errors. Of course, when people make the occasional mistake, we’re used to giving them the benefit of the doubt, but we struggle to do the same for machines because we don’t know how to contextualize their errors.

Why do we react so emotionally to AI? How can we shift our perspectives? And how can we actually score recommendations in AI systems? The hope is that with greater understanding, we can apply AI to more business settings and drive greater success.

Digging deeper into our fears and hesitations

Behaviorally, people tend to fear things we don’t understand or that seem out of our control. When it comes to risk, specifically, we struggle to comprehend how to assess it in an objective—rather than emotional—way.

For example, think about self-driving cars. The thought of a car without a driver makes many of us uneasy. Even though more than 75% of us will be in at least one major car accident during our driving lifetime, we’re afraid to put autonomous cars with this type of driving record on the road. While the probability of an accident is likely not higher than for a human driving a car, the combination of not knowing the exact percentage of risk and not being in control makes it harder to accept. We’re just not used to making our decisions based on probability; we are used to listening to our gut.

In order to process the data with a probabilistic AI system, we have to score it and set a threshold for “good” data; anything with a score below our threshold is discarded and anything higher is deemed an acceptable level of risk and included in the data set.

In my experience, the best way to get comfortable with objective assessment of risk is practice. Over time, it becomes more natural to look at the numbers as opposed to looking at our emotional response. Of course, understanding exactly how AI works helps too.

Understanding how to assess risk associated with AI

AI acts on two types of systems: deterministic and probabilistic. With a deterministic system, an outcome can be determined with relative certainty. This includes apps like Amazon, Doordash, and Venmo, which generate predictable types of data within a confined system. These are usually not considered “mission-critical,” and as a result, we’re willing to tolerate some level of inaccuracy in their algorithms. For example, when Netflix recommends a movie that doesn’t actually interest us, we don’t cancel our subscription to the service. We just look at the next recommendation in the queue or scan the top 10 titles of the week. We’re forgiving.

Probabilistic systems have built-in uncertainty. The exact output is not known. Think about the difficulty of forecasting the weather. It’s hard for us to understand the uncertainty of probabilistic systems and the stakes get even higher when we’re dealing with “mission critical” data, like we are in healthcare technology. In order to process the data with a probabilistic AI system, we have to score it and set a threshold for “good” data; anything with a score below our threshold is discarded and anything higher is deemed an acceptable level of risk and included in the data set.

The first step is to understand how these systems work, and the second is to set thresholds to score data that matches your risk tolerance.

Take a risk

With machine learning models, we are training a system to learn and adapt in order to improve—so it’s necessary to make assessments on an ongoing basis, rather than measuring an automation system’s performance once and only once. Because of that, it’s essential to have patience, as data can and will change, depending on many factors.

While risk makes people feel uncomfortable regardless of the setting, it’s time to address those fears and reluctance to move forward. Once we have tangible examples and parallels we often relate and tolerate it better.

As for ChatGPT and its generative AI brethren, the key will be for each person who engages with these tools to determine what level of risk they are willing to take. For most of us, a simple chat about something mundane or unimportant is likely acceptable. For some, the exchange of critical data or asking it to perform an important function will be a bridge too far. For now.

Dr. Bob Lindner is the Chief Science & Technology Officer and Co-Founder of Veda. More about Veda’s science and technology: Automation, Machine Learning, and the Universe: Q&A with Bob Lindner.

Learn more

Optimal and Proven Provider Data from Veda

What makes Veda’s data so great?

Healthcare provider data can be riddled with inaccuracies—just ask anyone who uses network directories to find an in-network specialist or view clinics in a 10-mile radius. The Centers for Medicare & Medicaid Services (CMS) Medicare Advantage (MA) online provider directory reviews between September 2016 and August 2017 found that 52.2% of the provider directory locations listed had at least one inaccuracy.  

Health tech companies have attempted to solve provider data inaccuracy problems with a number of products, platforms, and integrations. No solutions have been able to ultimately offer a better experience for members where it matters: the ability to easily book an appointment armed with accurate information.

Many solutions in the market focus on gathering all data sources available to identify providers but don’t have the ability to clean up those databases so they have only current and accurate information. A patient might find a doctor in a directory but if the location and coverage information was wrong, they still can’t make an appointment.

Enter Veda’s latest offering: Vectyr Data Curation. Vectyr offers the most up-to-date, comprehensive, and accurate source of provider data on the market. Vectyr’s database uses more than 100,000 unique sources to create an optimal collection of provider information. The data is continuously monitored, validated daily, and backed by our accuracy guarantees.

Prove It

How does Veda back up claims of accuracy and completeness? For one, our team of data scientists behind the development of Vectyr has the clout and expertise needed for intensive data modeling. From creating ground-breaking machine learning code to researching at the largest particle physics laboratory in the world, the best in science and technology are found at Veda. Here is how Veda employs a different approach than other data companies on the market:

  • Automation: Veda fully automates static and temporal data, boosting accuracy and reducing provider barriers. This validation process is automated in real-time, a fundamental advantage for healthcare companies seeking effective data structure.
  • Performance Measurement: Veda’s team of scientists carefully monitors the data’s success rate, creating statistical models, sample sizes, and methodologies to consistently guarantee accuracy. This process ensures specialty and data demands are evaluated and performing at the highest level.
  • Data Reconciliation: As temporal data evolves, Veda’s entity resolution process follows. Our technology accounts for data drifts over time, so our entity resolution is calibrated to recognize correct data from the abundant sources available today. New data is always cleansed and standardized, then consolidated within a database to eliminate duplicates.
  • Test Outcomes: Even with 95%+ accuracy, Veda doesn’t rely on automation to do all the work. The Veda team inspects all aspects of delivered data, including quality, delivery methods, bugs, and errors with a continuous monitoring process. By continually auditing and testing our data fields to confirm they are the competitive, current, and optimal quality we know reasonable coverage is reached.

Coverage, Precision, and Recall are numbers reported and recorded by the science team.

Coverage: What is the fraction of the data that isn’t blank?

Precision: When we do have an answer, how often is it right?

Recall: If we should have an answer, how often do we actually have it?

“Anyone can make an API. They are flashy, they can help operations, they can automate processes. But if your API is pulling in duplicative, inaccurate, and just plain bad information it’s useless,” says Dr. Robert Lindner, Chief Science & Technology Officer at Veda. “With our science backing, Veda’s data is guaranteed accurate and with flexible query so data delivery is where, when, and how users need it.”

Veda’s data is currently being used by top health plans for the correction and cleansing of their directories. Now, customers, new prospects, and new channel partners have direct access to Veda’s best-in-class provider information based on their nuanced business use case. 

Vectyr has profiles on more than 3.5 million providers who have an NPI 1 number—including MDs, DOs, RNs, social workers, DDS, and pharmacists.

What can health plans do with Veda’s data?
Staying atop changing information ensures provider directories are always accurate. This is no small feat as 20-30% of all provider directory information changes annually. With Vectyr, health plans can offer a better experience for members and providers by:

  • Expanding network offerings: Members need both provider options and location access to get the care they need. Using Veda’s data can help health plans identify providers they aren’t currently contracted with and fill geographic or provider gaps in their network. 
  • Sourcing correct providers for referrals: Providing accurate and on-the-spot information for in-network referrals relieves administrative burdens and eliminates frustrating hours spent searching for answers.
  • Quick credentialing: Credential providers faster and deliver faster onboarding and credentialing support with data that’s updated every 24 hours and guaranteed accurate.

What’s possible with optimal provider data?
There are immediate benefits to using Veda’s data. Health plan members will no longer wonder if their doctor of choice accepts their insurance or where the closest allergist to their home is. Hours of phone calls and administrative burdens are eliminated for both the member and the health plan. And, most importantly, health plans can trust Veda’s rigorous scientific validation methodology to ensure they have the optimal data for every provider in the country, on-demand, every day.

When health plans have access to optimal data, it means members have access to optimal data and that results in a markedly better customer experience.

More about Vectyr Data Curation

Learn more

Star Systems Meet Star Ratings: Using Science and Imagination to Solve Healthcare’s Most Complex Data Problems

What the heck does astrophysics have to do with provider data quality?

With an entire science department dedicated to solving complex data issues, science is at the very core of Veda’s existence. After all, our Chief Science & Technology Officer and Co-founder, Dr. Bob Lindner, began his career in astrophysics.

After taking a leap from the academic world and into political data analysis, Bob and co-founder Meghan Gaffney realized the potential of provider data automation. [READ Q&A WITH DR. BOB LINDNER]

The commitment to the scientific method and investment in science is what sets Veda apart from other data and healthcare tech companies—and what led to a robust science department with an impressive five IP and automation patents.

But you might be thinking: what exactly does this background in galaxy-staring, particle-measuring, and the expansive universe have to do with ensuring health plans’ provider directories are accurate?

The answer lies in wholeheartedly embracing the scientific method and Veda’s mission: We blend science and imagination to arrive at solutions for our customers. In fact, Bob argues one would not be able to tackle provider data problems accurately and reliably without a science department.

In the healthcare industry, data changes rapidly, some sources of data claim to be sources of truth but may in fact not be accurate, and data can be a heavily manual process. The only way to uncover the truth is with a careful and accurate measurement process.

Science meets imagination with Veda’s Science Team

Here is an expert from Dr. Lindner on problem-solving at Veda:

There are two kinds of main prediction problems. One where the answer to any given problem is self-evident. You can look at it and immediately know what the answer is. You give this problem a fast feedback loop and design your system to get the right answer based on immediate feedback from engineers. Outside of healthcare, an example is image classification. Is there a smiling person in this picture or not? You can look at an image and immediately tell.

A different kind of problem that we’re faced with every day at Veda is if the answer you’re trying to predict is not self-evident by a trained user in the field. For example, does this provider work at this address? It may look like a reasonable address and provider name but you don’t know if it’s accurate just by looking at it.

The only way to know if the system is working is to be very disciplined with the art of measurement and calibration. You must have a good set of test data that you trust that was collected in a way that was very tightly controlled. And you have to trust you are training your models on the data in a way that’s not overfitting because when your system gets used in production you don’t know—aside from that measurement in comparison to your training data set—if it’s working or not. You have to trust in science fully because if you do that part wrong, by using a biased training set or too narrow of a sample, there are errors that are invisible until you actually try to use the data. It can be a devastating effect. If you have a 10-digit number that says it’s the phone number of a provider, and you can’t call every phone number, how do you know it’s correct? You must have faith in the process. And the process to have faith in is the scientific method.

Dr. Robert Lindner

Provider data is complex and vast just like data in the field of astrophysics. However, provider data is nuanced and complicated in ways that even monitoring billions of stars is not.

The challenges with provider data are more complex than say, finding the largest thing in the universe, because the information included in directories changes often and the scope of required information keeps expanding. Practices move locations, physicians change practices, and contracts between practices and health plans expire. Multiple industry reports state between 20% and 30% of directory information changes annually.

Yet, no single party is the exclusive keeper of this information. Some of the information is governed and controlled by the practice, such as contact information and the roster of clinicians who practice there. Other data, such as whether a clinician is accepting new patients under a specific plan, can be owned by the practice, the health plan, or in some instances, shared by both parties.

Veda’s Science Team

Having different authoritative sources depending on the data contributes to the difficulty for health plans and practices in keeping information accurate.

So yes, provider data is more complicated to monitor than the stars but the Veda science department, using the scientific method day in and day out, can solve complex provider data problems faster and more accurately than anyone else. We start by understanding problems deeply before pairing them with an appropriate model and AI technology.

Before you select a healthcare data vendor, ask yourself, why don’t they have a science department backed by patented IP?



Get your provider data assessed by Veda.

Learn more

Why Healthcare is Behind in AI and How The Industry Can Catch Up

Artificial intelligence (AI) and machine learning have proven their worth in numerous industries—social media platforms that are perfectly curated to your tastes, the ability to shop online for clothes, groceries, and even real estate and cars (not to mention cars that drive themselves). The healthcare industry however, lags behind others. In this post, we’ll discuss why this happened, how automation solutions can help process and surface insights from the masses of data flooding the healthcare system, and what the future will look like for patients and plans alike when healthcare catches up and embraces automation.

WHY HEALTHCARE IS BEHIND WHEN IT COMES TO AI

There’s an understandable extreme level of caution around embedding automation in healthcare systems and technology; lives are on the line, and if there were ever an industry where it’s critical that humans make major decisions, healthcare is it. That being said, many of the decision-makers in healthcare lack an in-depth understanding of the current capabilities of these kinds of tools, the use cases for them (many of which are administrative rather than clinical), or the mechanisms put in place to ensure humans remain in control of patient care.

 A holistic view of a patient’s health is just out of reach in the absence of tools that make data processing efficient.

A second reason AI hasn’t achieved deep penetration in healthcare is the state of the industry’s technology. It wasn’t too long ago that hospitals housed huge document storage rooms and hired file clerks to sort, alphabetize, and distribute medical documents into physical patient folders. Although electronic health records (EHRs) are now the standard, every hospital has customized its installation, making it difficult for these systems (even those from the same manufacturer) to “talk” to one another. There are many examples of technology not standardized across the industry. The typical national payer, for instance, uses up to 15 technology tools and platforms to meet the needs of its members. But interoperability is an issue—only a few of these systems can communicate with each other.

Further complicating the picture, is the very nature of healthcare data. There is not one standard way of recording and translating data between healthcare institutions or corporations, or even systems within the same corporation. Because of that, it makes it very challenging for an automation algorithm to predict and understand errors in the data (…but not impossible, as we’ll elaborate on below). It’s much easier to leverage automation for Uber, DoorDash, or Amazon, because the data is generated by machines, and therefore inherently controlled and clean. The humans who run healthcare are anything but standard, on the other hand. Each has their own way of understanding and organizing data points (language, phrasing, punctuation, emojis, and shorthand). It takes incredibly sophisticated algorithms to process an Excel spreadsheet created by a person.

HOW AUTOMATION SOLUTIONS CAN PROCESS AND SURFACE INSIGHTS FROM THE MASSES OF DATA FLOODING THE HEALTHCARE SYSTEM

Given all these barriers—particularly the “messy data” issue—some question whether it’s even possible to successfully leverage AI and machine learning in healthcare. The answer is a resounding, “Yes.” As tech platforms intended to advance care continue to proliferate, so do the data they generate. The problem in healthcare today isn’t a lack of data; it’s actually the inverse. There’s so much data that neither administrators nor clinicians can successfully process all of it and extract value. A holistic view of a patient’s health is just out of reach in the absence of tools that make data processing efficient.

A smart solution like Veda’s can step in as a “Rosetta stone” to translate this messy data and process it in just hours and with 98% accuracy.

Luckily, in the past few years, automation algorithms have become more sophisticated, with a “next generation” of solutions that are capable of parsing the messy, human-generated data that permeate healthcare now emerging. There are almost endless use cases for putting such sophisticated solutions to use, but one that’s very easy to understand is using AI to make the search for in-network care simpler for patients.

Health plans are constantly receiving updates from providers in their networks, such as where they are located, who has joined or left a practice, and more. Currently, most plans have staff manually inputting these updates from Excel spreadsheets into their unique systems. As a result, updates take up to six weeks to show in the patient-facing portals, and the accuracy of the entries can be as low as 60%, despite payors’ best efforts.

A smart solution like Veda’s can step in as a “Rosetta stone” to translate this messy data and process it in just hours and with 98% accuracy. Veda’s AI understands human-generated data points, in all their diversity, and makes it possible for healthcare organizations to exchange data seamlessly. The provider directory use case is just one of many ways that automation can be used to organize and cleanse data, making it possible to extract insights that previously remained locked.

A FUTURE WHERE HEALTHCARE CATCHES UP AND PATIENTS BENEFIT

The pandemic created a huge influx of patient data that overwhelmed healthcare organizations, creating the final push that many needed to finally test the automated solutions they had been wary of for so long. The outcomes of these “tests” conducted out of pure necessity were overwhelmingly positive; patients were receiving the care they needed in a more timely manner, reduced administrative costs and errors, and health plan readiness for compliance with the provision of the No Surprises Act that requires them to make provider directory updates in just 48 hours starting January 1, 2022.

What do we have to look forward to in the future as more and more healthcare organizations adopt automation? We’ll continue to see the $1 trillion annual administrative spend in healthcare go down. We’ll continue to see patients accessing care more easily. And best of all, we’ll see more resources dedicated to what really matters to all stakeholders in the system—patient care.

Veda’s AI understands human-generated data points, in all their diversity, and makes it possible for healthcare organizations to exchange data seamlessly.

Veda’s AI automation solution helps health plans leverage machine learning to process data efficiently and effectively, so you can continuously maintain compliance and improve ROI. Schedule a demo to see what Veda can do for you.

Learn more
You know your business. We know data.

One Simplified Platform

Veda’s provider data solutions help healthcare organizations reduce manual work, meet compliance requirements, and improve member experience through accurate provider directories. Select your path to accurate data.

Velocity
ROSTER AUTOMATION

Standardize and verify unstructured data with unprecedented speed and accuracy.

See how it works
Vectyr
PROFILE
SEARCH

The most up-to-date, comprehensive, and accurate data source of healthcare providers, groups, and facilities on the market.

See how it works
Quantym
DIRECTORY ANALYSIS

Review and refresh your network directory to identify areas that affect your quality metrics.

See how it works

Resources & Insights

VIEW ALL
Provider Data Solution Veda Automates Over 59 Million Hours of Administrative Healthcare Tasks Since 2019
October 21, 2024
Read more
HealthX Ventures Blog: How Veda Is Aiming to Fix Healthcare’s Broken Provider Directories
October 17, 2024
Read more
AI accountability in healthcare for business success
September 16, 2024
Read more
Contact Veda Today

Take your healthcare data to the next level

Let’s transform your healthcare data. Contact Veda to learn how our solutions can help your organization improve efficiency and data accuracy.

Linkedin Instagram

©2024 – Veda  |  All rights reserved