Better Patient Care Starts with Better Data

After a decade and a half of healthcare digitization and data collection, we’ve entered an era of seemingly limitless data. Around 30% of the world’s total data is now generated by healthcare. Every year, the average hospital produces twice as much data as contained in the Library of Congress.

Managing data of this magnitude has changed the operation and business of healthcare dramatically in just the last five years. 

Recently, I spoke with Joshua Stephens, CEO of Stephens Insight Group, and Dr. Mohammad Ghassemi, professor of computer science at Michigan State University, about those changes, healthcare’s quest for data quality and how data integrity is critical for powering downstream analytics. 

1)   When does data quality really matter?

Healthcare is shifting from the EHR era of “How do we collect more data?” to “Which data matters most?” 

The vast majority of every organization’s data can be classified as “Dark Data” – or data that is not used or needed to influence practices, generate insights, or make decisions. There’s little reason to worry about the quality of that data given its lack of utility. 

But what about the portion of the organization’s data that is leveraged for analytics and information sharing? In theory, data quality can be judged by its accuracy, completeness, consistency, timeliness, and accessibility. In practice, the importance of data quality is completely dependent on its use cases. Not all data is equally valuable. One of the fundamental challenges in achieving the right data quality standards is knowing how much data quality matters to the downstream applications.

Basic patient information, for example, could include ages that are wildly inaccurate (490 instead of 49) or only off by a few years (49 instead of 51). If the use case depends more on gender or race than age, then those inaccuracies might not impact data quality. But if age affects when certain tests must be conducted, for instance, then even a small (and less obvious) inaccuracy could have a big impact on data quality.

When evaluating data quality initiatives, it’s critical that you take the use case and end goals into account. 

2)   How do you know when your data quality is good enough?  

The cost of data quality depends on the downstream use cases relying on that data. When deciding which data quality issues are worth that cost, it’s important to understand the source of the data quality issue in the first place. 

A data quality problem can be due to underlying issues related to process, technology, or people that may also need to be addressed. In general, issues with technology are cheaper to fix than issues of process, which are cheaper to fix than data quality issues that arise from people.

Because 100% data quality is not achievable, it’s also important to know when your data – and its quality – is fit for its intended purpose. This will help you avoid devoting critical resources to solving problems with very small returns.

Business stakeholders will sometimes insist IT teams fix a data quality problem right away without understanding the total cost of that process. According to Scaled Agile principles, data quality costs can be calculated by assessing three components: (1) the business value to the end user; (2) the time criticality; and (3) the compliance risk.

By collaborating with business stakeholders, you can assign value to each of those variables and identify the appropriate level of investment. 

3)   How does one use case affect data quality in another?

There is always a primary purpose behind why data is collected. Often, that primary purpose does not align with a different downstream use case. 

For example, EHR data is collected primarily for billing and accounting purposes. The quality of that data might be perfectly acceptable when processing claims for fee-for-service reimbursement. However, there may be a need to use EHR data for other reasons, such as assessing chronic disease risk in a patient population. Overlooking the primary purpose of the data (i.e. to perform billing) may lead to data quality problems affecting the results of a secondary question (i.e. assessing disease risk).

A chronically ill patient who develops a cough is less likely to visit the health system (and therefore get a billing entry created) than a patient from the same population who has a dramatic event (e.g. heart attack). The EHR, therefore, may not be a good way to assess incidences of mild health issues for chronically ill populations but will be a reliable source of information for more acute health events.

Using data across use cases will require some prework to identify how valid that data will be for secondary purposes. By considering the source and purpose of the data collected, you can determine the best opportunities to use the data for other business initiatives. 

4)   Will AI fix or exacerbate your data quality challenges?

Given the exciting potential of AI to process and derive insights from data, it’s tempting to assume that AI tools will be well-suited for overcoming data quality challenges. But the value and validity of AI insights is also highly dependent on data quality. 

AI has a well-documented problem with hallucinations - producing inaccurate or even fantastical patterns, conclusions, and research findings because of unexpected problems with the sources of information. AI also has a problem with biases that can become embedded in processes, conclusions, or findings.

If you can’t trust your data, you can’t trust the AI that’s using your data. There’s a growing view in the AI community that the best way to train an AI model is not to give it more data but to learn on smaller, extremely high-quality data sets. Organizations that want to integrate AI into healthcare technology systems and decision-making must ensure their data quality is high or risk compounding errors, biases, and inaccurate insights. 

Conclusion: The data imperative has changed

A recent McKinsey report on data and analytics best practices across industries observed that even leading organizations have lots of room to grow on data quality. But those leading companies have a strategy to build a powerful data culture because they see data capabilities as fundamental to higher levels of performance. Healthcare is no different. Indeed, there may be few industries where data is more important. 

Data quality issues can drain financial and human resources. But more importantly, they can create significant opportunity costs in terms of administrative savings, performance, innovation, patient/member engagement, and health outcomes. Knowing when data quality matters, how much data quality costs, and how to evaluate the problems that underlie data quality issues will be critical as healthcare shifts more of its focus from data collection to important use cases that are only possible because of data analytics.

Watch now to hear the full conversation.