Amid the rush of daily operations, it’s easy to overlook the nuances of data quality. Yet, if your data isn’t accurate, reliable, or up to date, even the best reports lose their value. Before you dive deeper into analytics, let’s talk about data quality and how to clean up your data flows without a lot of stress.
Define Clean…
Data that is not actively cleaned up will produce reports that are wrong. Sometimes the error rates are low and the purpose of that data can tolerate them. This is where knowing the purpose of reports (remember building those maps of flows?) comes in handy. This is also part of why a change in purpose is not a small thing. If you collect data and report on that data for one purpose, then think “Oh, this could help me do this other thing, too”, it’s important to stop and make sure that data is ok to use for the new job. You can start making bad decisions or producing false reports pretty quickly when you change the purpose.
But how do you define the quality of data?
In its simplest form, data quality refers to how well your data meets the needs of your organization. The phrase you’ll hear often from data professionals is “fit for purpose”. There are a lot of different definitions out there, but a short list list says high-quality data must be:
- Accurate (Reflect reality)
- Complete (Not half entered or blank)
- Timely (Arrives when you need it)
- Consistent (Always entered the same)
Without these attributes, meaningful reports and metrics are impossible to build. That said, your purpose is the ultimate determination. If your purpose isn’t impacted by timeliness, then don’t worry about that attribute. This is where that flow map can come in handy. Knowing the Purpose, Parts, Place, and Pace of your data makes this process much, much, much simpler.
Quality Improvement Made Easy
Any process or method you use to improve the quality of care at your clinic, can probably be used to work on data quality, too. In fact, some of the most successful organizations rarely have separate data quality processes, they instead make sure that data quality is part of all their improvement efforts!
If you’re working on improving diabetic care or patient outcomes, establish guidelines for entering A1C labs, patient home testing records, education efforts, etc. Put a process in place to check and make sure those guidelines are followed. If establishing those guidelines and making sure they’re followed is your first process improvement, every additional step will be easier to implement!
Quality in Detail
The Data Management Association recommends considering a longer list of quality dimensions. So if you’re a “dive in head first” type, here’s your expanded list:
- Validity: data conforms to a format, range, or precision. For example, if the desired entry method for an A1C is 9.02 (uses percentage, not decimal; two decimal places, and is within the range of possible A1Cs), an A1C entry of .01 would not be valid.
- Completeness: all required data is present. For most clinics, a patient record that doesn’t contain any contact methods is not complete.
- Consistency: can get complicated very quickly. There’s consistency over time (did you change all your BP machines and now get slightly higher readings on average?), consistency between systems (does your bookkeeping system contain the same daily amounts as your EMR payments?), and consistency inside a field (does a particular appointment type get used consistently to describe the same expected services?).
- Integrity: duplicate records, references to records that no longer exist, or nonsensical combinations of entries (like if a home address was listed as Nashville, NY 90210).
- Timeliness: if you need a lab result available for an appointment and the data is there in time, it was timely.
- Currency: refers to the idea that data doesn’t age well. Values entered a year ago probably don’t reflect current situations.
- Reasonableness: does the data match real world expectations? If I drew a map of your patient home addresses, you would validate that against what you knew of the area. If I drew them coming from affluent neighborhoods, that wouldn’t be reasonable.
- Uniqueness: If you have the same person listed as a patient 3 times due to name changes, record outages, and/or address changes, that would mean your data does not reflect the uniqueness of that person.
- Accuracy: how well does the data reflect the real-world?
Don’t Just Treat Symptoms.
Finding and fixing errors is good. But any good healthcare clinic knows prevention is the most effective cure. Here’s how to manage data quality for long-term success.
Start by looking for one problem: usually looking for blank values that shouldn’t be blank is an easy starting place. Audits can usually be automated using Excel and that is a major time saver! Figure out how to automate a check and then put it on a schedule. Once a month, once a week, or once a quarter, run a prebuilt report in your EHR, financial system, etc. Them put it through your Excel process and find the errors. By starting with one easy check, you’ll make it easier to build, easier to run, and easier to find the underlying problem. And then, when that check is running smoothly and simply (it probably won’t ever be perfect), then – and only then – you can add another problem to your validation check.
Always make sure to take a look at how those errors happen. Have you unnecessarily complicated a data entry task and now folks are getting it wrong? Are there places you need to establish standards of entry? Was there confusion about a process change (did people know there was a process?). Keep track of how many errors you find each time. By focusing on one issue at a time, you should quickly see those numbers drop.
And then brag about that process, to your staff, to your donors, to your board… that you’ve made a measurable increase in the quality of data available for decision-making!
The Impact of Poor Quality
Quantifying the impact of data quality issues can be a powerful tool to drive action. Start by asking, “What decisions are being made based on this data?” The higher the stakes, the more urgent it is to resolve quality problems.
For example, estimate how much TN Safety Net funding you lose if your patient counts are missing 10 patients per quarter. Then, as you’re working on improving related data, you can estimate the cost of the errors you are finding quickly. Having a dollar amount attached to those errors helps everyone involved understand the impact of quality. Money isn’t all that matters, though. You can measure impact in terms of patients: outcomes, patients who received improved treatments, reduced wait times… Connect the impact of poor quality directly to your mission and values!
Data Owners and Stewards
You can’t do everything. So it’s important to differentiate between two different roles in data quality efforts. In industry lingo, they’re data owners and data stewards. Your data owner(s) can have many different roles, but your data steward will always be a front-line worker or manager.
Data Owner: High-level strategist. Owners decide what the data is for, who can use it, and how it should be handled. They’re responsible for writing policies and are responsible for making sure quality issues are addressed.
Data Steward: Keep the data on course. They implement the policies, go through the results of quality checks to identify problems, and also work with data owners (or other users) to make sure that changes in data purpose or needs are communicated. If you start feeling like you have a lot of data stewards, you’re doing something right. In small organizations like our clinics are, most of your staff will be stewards of some piece of data.
Both roles are essential for data quality, but they function at different levels. Owners set the direction, while stewards keep the data on course.
Now look at the data flow map you’ve chosen to tackle.
What attributes of data quality matter the most in your flow? Where are the places in your flow where problems are the most likely to creep in? Where can you put an audit step in your flow to identify those problems?
Add these to your map and start implementing them – either on their own or as part of a larger quality improvement project. If resources are minimal, set a time limit on the QI effort: “We’re going to identify and solve as many problems as we can in three months.” That way everyone feels confident this won’t drag out and be a time drain forever.
Related Resources to Explore
Data Quality explained by IBM
Short video going through the basics of data quality.
Automating with Power Query
A less than 10 minute video walking you through the basics of Power Query data cleaning in Excel.
Coming Soon
A detailed video walk-through of implementing quality metrics with Excel.
