The four types of data analytics in healthcare — descriptive, diagnostic, predictive, and prescriptive — are often described in terms of what they tell you. But understanding what data each type requires, and what analytical methods it uses to produce its outputs, is more useful when evaluating whether a platform is actually doing what it claims.
This is especially relevant in home health and hospice, where vendors routinely describe their platforms as AI-powered or analytics-driven without clearly distinguishing between reporting on historical data and predicting future clinical outcomes. The methodology behind the output determines what the output is actually worth.
Descriptive Analytics: Structured Data, Aggregated Backward
Descriptive analytics works with structured historical data — documented clinical events, visit records, outcome codes, billing data — and aggregates it into summaries and trend reports. The data inputs are well-defined and standardized: OASIS fields, visit completion records, discharge dispositions, star rating components.
The analytical method is straightforward: aggregation, averaging, and trend visualization across defined time periods. The output is a report — accurate, legible, and backward-looking. It tells a clinical or operational leader what their agency's data looked like over a past period.
Most home health and hospice agencies have robust descriptive analytics infrastructure. Their EHR systems generate these reports automatically. The limitation is not data availability — it is the analytical ceiling. Descriptive analytics cannot process longitudinal patient trajectories or generate forward-looking probability estimates.
managing PDGM performance and value-based care contracts, that limitation matters.
Diagnostic Analytics: Correlation and Causation in Historical Data
Diagnostic analytics uses the same structured data as descriptive analytics but applies more sophisticated methods: correlation analysis, segmentation, and root cause investigation. It looks for relationships between variables across a historical patient population — which clinical factors appear together in episodes that ended in hospitalization, which documentation patterns correlate with benchmark misses, which patient characteristics predict shorter hospice length of stay.
The data inputs remain historical and structured. What changes is the analytical question: rather than 'what were our outcomes,' the question is 'which factors in our data appear to explain those outcomes.' The output is insight into patterns, not predictions about individuals.
In practice, diagnostic analytics in home health informs quality improvement initiatives and care protocol reviews. It identifies system-level patterns that can be addressed through policy or training. It does not identify which specific patient in the current active census is at risk today.
Predictive Analytics: Longitudinal Data, Machine Learning, Individual Risk
Predictive analytics requires a fundamentally different data foundation. Rather than structured historical snapshots, it requires longitudinal clinical records — data that tracks how individual patients change over time across multiple clinical touchpoints. In home health, that means OASIS assessments at start of care, recertification, and discharge; skilled nursing visit documentation across an episode; vital sign trends; functional status trajectories; and, in advanced platforms, clinician narrative notes processed through natural language processing.
The analytical method is machine learning: algorithms trained on large volumes of historical patient records to identify which combinations of variables and trajectory patterns reliably preceded specific outcomes. The model learns to recognize risk before thresholds are crossed, not by applying a rule, but by pattern-matching against thousands of historical cases where similar trajectories led to similar outcomes.
The output is a patient-level risk estimate — a probability score, a risk category, or a ranked census view — that reflects each patient's individual clinical trajectory rather than a population average. The accuracy of that output depends entirely on whether the training data came from the same care setting the model is being applied to.
Prescriptive Analytics: Optimization Across Multiple Variables
Prescriptive analytics uses predictive outputs as inputs and applies optimization methods to recommend specific actions. The data requirements are the most complex of the four types: real-time or near-real-time clinical data, outcome probability estimates from predictive models, and operational constraints like staff availability, visit capacity, and patient geography.
The analytical methods are drawn from operations research and optimization: constraint-based modeling, simulation, and decision trees that weigh competing resource and outcome variables simultaneously. The output is a recommendation: visit this patient today, assign this care team, adjust this care plan.
In home health and hospice, prescriptive capability is still emerging. The clinical complexity of individual patient care — where physician judgment, patient preference, and care context shape every decision — means that fully automated prescriptive outputs carry adoption challenges. The more immediate and achievable value for most agencies is closing the gap between descriptive reporting and predictive analytics: replacing backward-looking reports with forward-looking risk intelligence that clinical supervisors can act on each day.
See how Mosai's data architecture powers predictive analytics built specifically for home health and hospice: mosai.com/platform/clinical-management