For clinical operations leaders evaluating predictive analytics in post-acute care.
Not everything marketed as analytics is predictive. In home health and hospice, the distinction matters because the clinical value of a tool depends on what it is actually doing with data — whether it is reporting on the past, flagging present thresholds, or generating forward-looking risk estimates. This article provides a categorical framework for what counts as a genuine example of predictive analytics and what the examples look like in practice.
An example of predictive analytics has three distinguishing characteristics. First, it produces a forward-looking output — a probability estimate, a risk category, or a ranked prioritization — rather than a summary of what has already occurred. Second, that output is generated by a model that has learned patterns from historical data, not by applying a fixed rule to current data. Third, the output is specific to an individual patient or defined patient population.
By that definition, the following are not examples of predictive analytics, even when they appear in analytics platforms:
Understanding these distinctions gives clinical leaders a filter when vendors use the word 'predictive.' The question is not whether the platform generates risk flags. The question is whether those outputs come from a trained model predicting future events, or from rules applied to current data.
A model trained on longitudinal home health patient records identifies, for each active patient, the probability of hospitalization within a defined near-term window. The output reflects the patient's current clinical trajectory — not their diagnosis alone, and not a single threshold that has been crossed. This is predictive because the model is pattern-matching against historical populations with similar trajectories, not responding to a documented event.
A model trained on hospice patient visit data assigns each active hospice patient to a risk category reflecting the likelihood of mortality within approximately ten days of the most recent skilled visit. This is predictive because it generates a probability estimate for a future event — patient transition — that the care team can respond to before it occurs.
A model trained on historical hospice revocation data identifies current patients at elevated likelihood of disenrolling from hospice services within thirty days. This is predictive because the output is a forward-looking risk estimate for a specific patient behavior, generated from learned patterns rather than a single observed indicator.
When every analytics platform claims to be predictive, the label stops being meaningful. The categorical framework above gives clinical and operational leaders a filter: does the platform's output reflect a trained model generating forward-looking probability estimates from longitudinal data, or does it reflect rules applied to current data producing present-state alerts?
Both tools have clinical utility. Only one is predictive analytics. Buying the wrong category for the clinical problem you are trying to solve — paying for a reporting platform when you need a risk model — is a common and expensive misalignment. The examples above provide a reference point for evaluating which category a platform's outputs actually belong to.
Download Mosai's clinical outcomes data to see these examples in practice: Mosai | Success Stories