Multiple blog posts, entertainment pieces, and quantitative research exist on why healthcare data is uniquely challenging. Maintaining provider directories costs physician practices in the United States roughly $2.76 billion annually. One main culprit is healthcare provider data inaccuracies.
A recent piece by John Oliver on Last Week Tonight highlighted how these inaccuracies cause real implications on people’s lives. Aside from high annual costs to physician practices, these data issues are impacting people’s welfare and causing delays or access to care. But what does “accurate” look like? Why is it hard for healthcare organizations to achieve? Let’s go over the four major reasons healthcare provider data accuracy is so difficult.
1. The overall accuracy of an entire data source is likely skewed
The use of a provider directory changes depending on your team’s goal. Meaningful data requires an understanding of the population dynamics you are working within instead of the accuracy of the overall data points. But take caution. A provider directory contains a wide range of different provider types. It’s very likely many providers you don’t care about in your data set are skewing the overall “accuracy” of what you want to measure. For example, if you are trying to measure the accuracy of provider address entries, you wouldn’t want to include providers that only see patients virtually. In doing so, your data becomes skewed and makes it harder to measure what matters for your end goal.
2. Accuracy may not be the evaluation metric you need
Today, many organizations conflate accuracy with the goal of reliability and data integrity. The word “accuracy” is a bloated term and often overused. More descriptive statistical terms require data analysts or experts to interpret the data (and take action on it). Instead, we fall to accuracy as our metric of choice simply because it’s familiar. It means something to us.
However, we often miss addressing the most actionable metrics when we are focusing on provider data accuracy as the main goal. What do we mean by this? Think about your end goal first and work backward from there. What is:
- The outcome most valuable to your business?
- A problem are you working to solve?
- The result you are trying to measure?
Let’s say you want to understand how many call attempts are being made before patients connect with a doctor. In this case, correlating phone accuracy to a user’s ability to connect with a doctor is an end goal. Then, you will set metrics and measurements that expand beyond the accuracy of the overall provider directory. Whatever metrics or measurements you decide to set up need to result in meaningful action.
3. Attestation is complicated at best
Attestation is one of the best ways healthcare has to verify its data. It is the act or process of verifying the data with a primary source. For example, call centers may call a statistically significant sampling of providers throughout your provider directory. They are attesting various data points throughout your data. This human-in-the-loop solution brings pitfalls with it.
Multiple people work through a backlog of calls. How each person asks the questions and interprets the responses varies. Risk of human errors when recording values are high. Additionally, after an attestation call has been completed, a different person must then interpret those results. This type of cognitive bias affects reliability and accuracy. Because a manual process exists, it creates risks and complications in the data verification process: Wide error margins, inconsistencies, and unactionable results.
4. Single number statistics reported on provider data accuracy is the best guess estimate
In the world of data science, we use a common set of techniques when pulling data for statistics. Tools such as point estimates and confidence intervals help estimate population parameters (means, variances, etc.) from a sample of data. Point estimates are a single number and confidence intervals are a range of numbers. Confidence intervals provide much more information to the statistics story and are commonly used in published research or reports.
Let’s look at this in an example relevant to provider data statistics:
- The reported statistic: In 2016, research from West Virginia University found that about 30 percent of the cases with specialties listed in the provider directory did not match the one stated by the receptionist at the practice.
- The breakdown of this statistic: The “30 percent” is the point estimate. But the confidence interval is 16 to 43 percent.
- The conclusions of this statistic: The data accuracy range is actually 16 to 43 percent from the sample data calculated. The “30 percent” is the researcher’s best guess as to what level of provider data accuracy was achieved.
Therefore, the most accurate statement we can make is: The real value of provider data accuracy of these cases with specialties listed lies somewhere between 16 and 43 percent. However, that’s a mouthful, so the statistic restates to a point estimate. But we lose much of our fidelity and falsely increase our confidence in the value itself. By reaching for certainty, we must sacrifice the nuance and reality of the data.
What to takeaway about achieving provider data accuracy
Accuracy is hard because it is a term often seen as a panacea, but it’s often skewed. What’s more, dependency on manual processes, human interpretation, and single number statistics make it difficult to rely on an overall accuracy statistic alone. Given all this, accuracy is important but must be taken in context to be put to its fullest use.
As new drugs, devices, and digitally-driven medicine solutions continue to develop, data is pushing healthcare toward creating a digital healthcare ecosystem. This means the healthcare industry is making huge strides in resolving manual, human-dependent processes and looking at more ways to improve patient care with health IT solutions. Provider data accuracy is complex but doesn’t have to be complicated. When properly scoped, it is possible to measure (and achieve) data accuracy. But to do so you need to remove the risks from manual processes and aggregate reliable and the right sources of data. Read on in this article about why data automation for provider directories is worth the investment.
About Our Guest Author:
Jordan Hagan is the VP of Data Science at Orderly Health. Jordan is a tech leader who focuses on helping really smart people doing really cool things. In Jordan’s spare time she does Pro-Bono tech work for minority groups as a mentor and advocate. Jordan’s sincere passion for continued learning and expertise contributes to Orderly Health’s customers and its mission to create a more connected healthcare ecosystem through data and technology.