Vendors love the phrase "predictive analytics." In long-term care, it usually means a risk score: a number that estimates how likely a resident is to fall, return to the hospital or decline. The idea is appealing. If you could see trouble coming, you could prevent it. The reality is more modest and more useful than the sales pitch, if you set expectations correctly.
This post offers a grounded look at what risk scores can and cannot do, and how a small operator can use them without overinvesting.
A risk score combines information such as age, diagnoses, medications, recent assessments, vitals and past events, and produces an estimate of the chance of an outcome. Some scores are simple point systems that clinicians have used for years. Others come from statistical or machine learning models trained on large datasets.
Either way, a score is a probability, not a prediction of any one resident's future. A "high risk" label means a resident belongs to a group in which the event is more common. It does not mean it will happen, and a "low risk" label does not mean it cannot.
A model built on data from other facilities or populations can perform differently in yours. Differences in case mix, documentation habits and staffing change results. Ask vendors how their model was developed and tested, and whether it has been checked on communities like yours.
If assessments are late or incomplete, the score reflects that. A model fed inconsistent charting produces unreliable output, however sophisticated.
A list of fifty residents flagged high risk, with no plan for what to do, becomes background noise. A prediction only matters if it triggers a response.
Identifying risk does not guarantee a way to reduce it. Be skeptical of claims about dramatic results, and ask for evidence you can verify.
Familiar fall-risk and pressure injury screening tools already live in many EHRs. Used consistently, they are valuable, and improving the consistency of their use is often worth more than buying a new product.
Counts and trends you can understand often beat opaque scores: recent falls, weight changes, new antipsychotic or high-risk medication starts, repeated infections, missed meals, and increased call light use. A dashboard of these gives clinicians a prompt to look closer.
If your EHR offers risk or early-warning features, ask your vendor to explain what data they use, how they were validated and how they fit into clinical workflows. Do not assume what is or is not available; confirm it.
Reviewing your own transfers back to hospitals, together with the clinicians involved, often reveals patterns that no model needs to find.
What data does the score use, and where does it come from?
How was it built, and how was it tested?
How does it perform for facilities like ours?
Can we see why a resident was flagged?
What is the recommended action when someone is flagged?
How is resident data protected, and is a business associate agreement in place?
What happens to our data, and can it be used to train models for others?
Scores should support clinical judgment, never replace it. Nurses and physicians know things no model sees, such as a resident's mood on a given morning. Make sure staff can override or disregard a score, and track whether the tool actually changed what you do.
Pick one outcome, such as falls, and one clear response, such as a huddle review. Look at results over several months. If the process helps, expand. If it does not, stop.
UnityCare IT helps operators clean up their data, connect reporting to their EHR and evaluate vendor claims about analytics. Good prediction starts with reliable data and a plan for action, and we can help with both.
An outsourced IT department with proactive maintenance and one number to call.
Call or text: 405-285-3845
New customers: start@unitycareit.com
Existing customers: support@unitycareit.com
Address: UnityCare Technologies, 2524 N Broadway Ste 554, PMB 947974, Edmond, Oklahoma 73034-4172