How to Turn Disparate Patient Data into Actionable, Timely Insights
April 1, 2024
Imagine you’re accessing a patient record from a hospital stay to find information relevant to their current diagnosis. You open the file, and discover a mass of unindexed, unstructured notes and data.
Unfortunately, this is the reality for most healthcare providers.
No one has the time to sift through all that data for one patient, let alone hundreds or thousands. But contained within those hundred pages are vital data on previous diagnoses, treatments, medications, discharge instructions—all of which are vital to providing better treatments and improving the patient experience.
The key is to adopt systems and platforms that take that raw information and turn it into something actionable. Let’s walk through the process you need to make that happen.
The problem with turning patient data actionable
Most providers face an unprecedented volume of data at their disposal, but too few resources to use it. That’s because most patient data isn’t formatted for ready retrieval at the point of need.Sometimes the inaccessibility is a bit more subtle than a mass of notes from previous hospital stays. For example, differently configured and customized EMR systems along different points of care, result in a network that, while technically interoperable, exchanges data that’s unreadable and unusable by the end user.
So say a provider is seeing a patient for a current GI issue, but knows they were in the hospital six weeks ago recovering from a knee replacement. Curious to see if any GI issues came up during the stay, the provider looks up those hospital notes.
Best case scenario, the provider has an AI/ML-powered system that can scrape the hospital records for anything GI-related. But even then, there are going to be errors because the data aren’t standardized. Worst case scenario, they don’t even try, because the cost-benefit just isn’t there.
Providers, then, need actionable insights delivered at the point of need. So when they’re making a diagnosis or prescribing treatment, all the information needed to make the right decision is there.
How to parse out usable patient data from noise
The key is to parse out usable data from noise. This is a dynamic process, as information relevant to one diagnosis or treatment plan won’t necessarily be relevant to another. As such, EMRs and related systems need to evolve to curate data, rather than simply capture and store it.There are two dimensions to going about this process. The first is defining the outcome, the second is building a system with functionality that will drive that outcome. If you try to do the second before the first, you’ll build a system that may or may not be a value add.
Defining the outcome in this case means figuring out which data providers need to make relevant decisions:
- Which data are necessary for a provider to make a diagnosis or treatment plan?
- Which data are most likely to contribute to a positive or negative health outcome?
- Which data indicate non-medical factors that can either aggravate the patient’s condition or influence the likelihood of their sticking to their treatment plan?
Natural Language Processing (NLP)
Let’s go back to those hospital notes and data. Natural Language Processing (NLP) techniques can automatically extract relevant clinical information from notes, identify gaps and inconsistencies, and even suggest amendments to enhance the quality of the data.However, NLPs can sometimes deliver inconsistent and inaccurate results, requiring manual verification of their outputs. This particular technology, while helpful, should be used in concert with other capabilities.
Query optimization
Often the difficulty in retrieving relevant information comes from the query. Rather than place additional pressure on providers to try multiple query formats if one fails to retrieve the necessary information, there are a number of methods for optimizing query performance.Simple solutions include applying filters to narrow datasets based on specific criteria, reducing the dataset to be queried and controlling the input. More advanced solutions include implementing semantic querying techniques, which understand user intent rather than exact phrasing of a request.
Such techniques can include ontology-based query expansion or semantic similarity search, helping providers retrieve related information within the same dataset.
Dimensionality reduction
Without getting too far into the weeds, dimensionality reduction can reduce the number of features existing in a dataset while preserving important information. This makes it easier to analyze and retrieve relevant data. Common dimensionality reduction techniques include principal component analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE).Exploratory Data Analytics (EDA)
Sometimes the problem with the data isn’t retrieval, but analysis. One example is with wearable technology. Wearables capture health data constantly. This data can be helpful to providers, but the raw datasets of over 10,000 heart rate measurements per month is unmanageable. Systems that can surface trends from those data, on the other hand, can be valuable.Exploratory Data Analysis (EDA) involves examining a dataset to understand its structure, map relationships between variables, identify patterns or trends, and make predictions.
How to make patient data more actionable today
Turning patient data into a more actionable format isn’t something that happens overnight. As I mentioned above, these are complex data management practices that are involved. And that’s not even mentioning the actual management and structuring of the data.To avoid costly errors, it’s best to work with a partner who not only has the necessary technical expertise, but experience leveraging technical functions to drive specific outcomes. Learn more about 3Pillar’s approach and healthcare expertise here.
About the author
BY
SHARE
Stay in Touch
Keep your competitive edge – subscribe to our newsletter for updates on emerging software engineering, data and AI, and cloud technology trends.