Welcome back. In this lesson, we'll look at improving quality from the baseline. According to Tom Burton, president of Professional Services at Health Catalyst, a health consulting firm, data that's tracked for establishing a baseline and then compared to the baseline to determine whether or not an improvement occurred is data for learning. In order to learn from data and improve data quality, you've got to start with a baseline. For the purposes of our discussion, a baseline is a measurement of what the data is saying when we begin to measure it, and a place where we can come back to as a comparison for moving forward. After this lesson, you'll be able to identify a baseline of data quality, discuss data quality improvement programs and explain how to improve data quality from that baseline. Let's get started. When we start to collect a piece of data or piece of information or start to look at it and consider its use for a particular analysis, we want to be able to see how good the data is. That is, how effective is it at measuring what we want to measure. We're trying to determine what pieces of data we can count on, and count on over time. Often in healthcare systems, we have data quality improvement programs where we routinely look at the data for differences as it comes in day-to-day, where we continually look for differences in the data, and then based on those differences, ask ourselves the question, is this a systematic difference? Is this a real difference? Or is it a difference that's an artifact of the way that we're collecting data? Based on that information, we're able to make adjustments and ensure we can move forward. So, we can look at data in a couple of different ways. Initially, there are a few things to determine the quality of the data. We may be looking to basically identify different kinds of deficiencies in the data. These deficiencies may include missing or duplicate data, inconsistencies, problems with patient identification, things changing across systems. So, we need to start by evaluating the deficiencies. Let's look at some of these areas in more detail. Common deficiencies would be things like missing and duplicate data. Usually, if we're finding missing data, we can track that back to a place in the workflow where the data isn't being collected properly or isn't being entered properly. We can also see that it could be a problem with technology, where an interface is broken for some reason. If we identify our problem with technology, we can either address the process by fixing the technology if in some cases an interface, or by going and looking at the workflow and working to improve that workflow. Another deficiency we can identify is data when it's inconsistent. We may see that workflows change over time and the data which was looking one way has now started to look another way. A change to who records a piece of information or when in a visit, the information is recorded maybe the kinds of things which could impact the consistency of data. In such questions, ask the question, are they really the same? Is this just a normal variation in the data? If we decide that it is something other than normal variation, we can examine our workflows and our data sources, and we can either transform them to make them consistent or we can adjust the workflow to capture them in a way that's more consistent. Another thing we may see is that we are actually observing different phenomenon in different data. We might find that data collected one way gives us one set of results and data collected another way gives us another set, perhaps two different lab tests that have different normal values. If we don't recognize that their values are supposed to be different, we don't know how to interpret them. If we put them together, we're not going to be able to differentiate what differences come from the measurement methodology and what differences are actually phenomenon that we want to understand. An important change we can make here based on changes from the baseline, might be to differentiate the dataset to recognize the different sources of data, and then being able to transform them so that they are the same or comparable and recognize they need to be interpreted differently. Another deficiency we often run into as we go through looking at data quality or have data changing from baseline is difficulties in consistently identifying patients from one system to another. It may be that we are capturing identities of patient differently. Sometimes, we create a new version of a patient's record if we can't find the patients old record. Many systems have groups or have functions for connecting patients that have had their records split in half, and join them into a single record when they can determine that that has happened. We also see issues in consistency connecting patients when we're doing matches between different systems. If there are inconsistent identifiers, sometimes we need to make improvements to the ability of our systems to identify a patient with identifier from another system. Occasionally, this is done manually. Occasionally, this is also done by improvements to our master patient index that we've talked about earlier. By improving the way we load data and the way we connect patients, we can recognize a data quality issue that can be resolved. To wrap up this lesson, we've talked a little bit about data quality improving programs. How most large health systems have created data quality monitoring programs that keep track of how data is coming in, looks for consistency day-to-day and week-to-week and month-to-month. If that consistency is not seen, we look for problems like missing and duplicate data. We look for inconsistencies in data, and we look to ensure that identities are properly aligned in the way that we're connecting patients and providers, visit to visit, incident to incident, care episode to care episode, and so on. By doing this on a consistent basis and expecting it routinely, we improve quality from the baseline that will result in improved healthcare for the people we serve. In the next section, we'll talk a little bit more about managing consistency over time and what kinds of things we expect to see as we look at data over time.
