12 Encounters
A patient goes to the emergency department, receives imaging, and is admitted to the hospital. The hospital submits a facility claim. The emergency physician, radiologist, and hospitalist submit professional claims. A long stay may produce several hospital bills.
How many times did the patient receive care? Counting claims will not answer that question.
In this chapter, we’ll discuss encounter grouping: the work of organizing billed services into visits and stays. We’ll begin with the difference between claims and encounters, then examine inpatient continuity, outpatient grouping, and the checks that make encounter-based analytics trustworthy.
12.1 Choose the unit of care
An encounter represents a defined interaction between a patient and the healthcare system. An inpatient encounter generally represents a stay. An outpatient encounter might represent an emergency department visit, an office visit, or an imaging service. The definition depends on the care setting and the question we are trying to answer.
Three units appear throughout claims analytics:
| Unit | Example | Main purpose |
|---|---|---|
| Claim or claim line | A hospital bill or a physician service line | Describes billed services and financial transactions |
| Encounter | A hospital stay or office visit assembled from relevant claims | Measures visits, stays, and their associated services |
| Episode | A joint replacement and related care over a defined period | Studies a course of treatment that may cross settings |
One encounter can contain many claims. One claim can contain lines associated with different service dates or encounters. An episode may include a hospitalization, rehabilitation, and follow-up office visits. These relationships are why the claim-line-to-encounter crosswalk is as important as the final encounter summary.
An encounter derived from claims is also different from an encounter recorded in an electronic health record. The EHR reflects the provider’s operational workflow. Claims reflect billing across organizations. Neither identifier can be assumed to map one-to-one to the other.
12.2 Build encounters from anchor services
A common approach starts with an anchor: a claim or service that supplies evidence for a particular encounter. An institutional acute inpatient claim can anchor a hospitalization. An office evaluation can anchor an office visit. Other lines are then linked to the encounter using setting-specific rules.
The main ingredients are patient identity, data source, service category, dates, and provider or facility information. Tuva’s encounter models implement different rules for different settings rather than treating every same-day service as a single visit.
A source identifier matters. Two payers can use the same claim number for unrelated claims. The same patient can also appear in multiple datasets containing overlapping care. Preserve the source’s claim identity and resolve cross-source duplication deliberately; an identical date is not enough to establish that two records describe the same event.
Dates need similar attention. An admission date, line service date, and billing statement start date have different meanings. Tuva’s claims staging logic uses admission and discharge dates when present, followed by line dates and claim dates. This provides a usable fallback, but it does not make a billing-period boundary equivalent to a clinical admission or discharge. Missing-date patterns remain part of the interpretation.
12.3 Reconstruct an inpatient stay
The first inpatient task is to distinguish multiple bills for one stay from multiple stays. Long stays are especially likely to cross billing periods.
Consider these illustrative facility claims for one patient at one hospital:
| Claim | Start | End | Discharge status | Interpretation |
|---|---|---|---|---|
| A | January 20 | January 31 | Still a patient | Interim bill |
| B | February 1 | February 10 | Discharged home | Continuation and discharge |
| C | February 18 | February 21 | Discharged home | Later admission |
Claims A and B support a single stay from January 20 through February 10. Claim C supports another stay. Counting facility claims would report three admissions instead of two.
Tuva’s acute inpatient grouping logic combines patient, facility, dates, and discharge disposition. For example, consecutive billing periods at the same facility can be joined when the earlier claim reports that the patient remains admitted. The algorithm also handles specified overlapping-date patterns. The exact comparisons matter at the boundaries, particularly for same-day events.
A discharge followed by another admission should not be merged merely because the dates are close. Doing so can erase readmissions. Conversely, failing to join an interim bill can create a false readmission and shorten the apparent length of stay.
Transfers introduce another choice. A patient moving from Hospital A to Hospital B may have two facility stays within one continuous course of acute care. A facility-level encounter definition should preserve that distinction. A separate transfer-chain analysis can combine the stays when appropriate. It should state how transfer gaps and discharge dispositions are handled rather than changing the meaning of an admission halfway through the analysis.
12.3.1 Attach professional and supporting services
Once a facility stay is established, the grouper can attach relevant professional claims and lower-priority services. This makes it possible to analyze both the hospital’s bill and the clinicians’ services associated with the stay.
Matching is an inference. A professional line can fall within a hospitalization’s dates without sharing a reliable facility identifier. A strict facility match may lose valid services; a broad date match may attach unrelated ones. Tuva’s acute inpatient professional matching uses the patient and encounter date span for eligible lines and resolves competing matches deterministically. That describes the implemented association, not proof that every matched line occurred at the anchor hospital.
Keep a distinction between facility spending, professional spending, and the total attributed to the stay. “Cost per admission” is ambiguous unless the reader knows which of these amounts is included.
12.4 Group outpatient and office-based care
Outpatient grouping often begins with a patient and service date, but the appropriate additional keys depend on the event.
| Family | Examples | Main grouping question |
|---|---|---|
| Emergency and urgent care | Emergency department, urgent care | Do the claims describe one visit, a visit spanning midnight, or separate visits? |
| Procedures | Ambulatory surgery center, outpatient surgery, office surgery | Which facility and professional services belong to the same procedure event? |
| Recurring treatment | Dialysis, therapy, injections | Are we counting treatment dates, individual services, or billed periods? |
| Visits | Office, hospital clinic, psychiatric, substance use, telehealth | Which same-day services represent one visit under the chosen definition? |
| Imaging | Office and outpatient radiology | Which procedure and billing components represent the examination? |
| Care spanning periods | Home health and hospice | Does the claims-derived record represent a visit day, billing period, or course of care? |
A same-day rule is practical when claims lack timestamps, but it has limits. A patient can visit two different practices on the same day or return to an emergency department after an earlier discharge. Without enough detail to separate them, the grouper may produce a coarser event than the clinical history contains.
The opposite problem occurs around midnight. One emergency department visit can generate claims on two calendar dates. Treating every date as a new visit overcounts that care. Continuity rules help, but their assumptions should be visible in the method.
12.4.1 Imaging needs a more specific unit
Suppose a patient receives a CT examination and an MRI on the same day. A patient-date-only rule would combine two different examinations. Tuva’s outpatient radiology anchors distinguish services by patient, source, start date, and HCPCS code.
This is useful for associating facility and professional components of the same coded imaging service. It does not guarantee a count of individual examinations. Repeated services with the same code on the same day, different code choices, and modifiers can require additional analysis.
Before comparing imaging prices, establish whether an encounter includes the technical component, the professional interpretation, or both. A $100 interpretation and a $600 combined examination are different units of purchase.
12.5 Preserve secondary features
An encounter needs a primary type, but that type cannot tell the whole story. An acute inpatient encounter may include emergency department care, observation, laboratory work, or a delivery. Preserve these as additional features instead of expecting one label to describe everything.
Tuva’s encounter outputs include indicators for services and characteristics such as emergency department involvement, observation, ambulance, laboratory, equipment, pharmacy, delivery, newborn care, and neonatal intensive care. Consult the Core encounter model and its dictionary for the fields in the version being used.
An indicator usually means that the relevant evidence was found among the linked claims. It does not establish the exact sequence or duration of that care. An observation indicator on an inpatient encounter, for example, does not mean the patient was simultaneously an inpatient and an outpatient throughout the stay.
This distinction also affects counting. The number of standalone emergency department encounters may differ from the number of all encounters involving emergency department care. Both can be useful metrics if they are labeled clearly.
12.6 Keep unmatched services visible
Some claims do not create an anchor and do not match an existing encounter. An independent laboratory test or equipment rental may be legitimate care without an associated office or hospital claim in the dataset. An ambulance trip may fall outside the date range required for attachment to a hospitalization.
Represent these services explicitly, whether as separate ancillary encounters or as unmatched claim lines. Do not discard them to improve a match rate. A large unmatched share can reveal missing facility claims, invalid dates, or missing identifiers, but it can also reflect the services a particular source covers.
Investigate unmatched dollars as well as unmatched lines. A large number of inexpensive laboratory lines and a handful of expensive unlinked hospital claims have different implications.
12.7 Validate the reconstructed care
A useful encounter output should let us move in both directions: from an encounter to its contributing claim lines, and from a claim line to its assigned encounter or unmatched status. For an exclusive assignment, a line should not contribute its dollars to two encounters.
Reconcile amounts before and after grouping using the same included population. Then examine a small set of event histories that exercise the difficult cases: an interim hospital bill, a readmission, a transfer, an emergency visit crossing midnight, and imaging with separate billing components. These examples reveal errors that aggregate totals can hide.
Also examine length-of-stay distributions, encounters per patient, and changes by source and month. Very long stays may be real, but they can also result from date fallbacks or over-merging. Many short stays may reflect fragmented bills.
Encounter grouping turns billing detail into a usable account of care. Its value comes from choosing a clear unit, preserving the underlying evidence, and understanding where the available claims cannot fully resolve the patient’s history.