Updated November 2026 · Written and maintained by the Progression Agency strategy team
A hospital’s numbers rarely come from one place. Visits sit in the EHR, payments in claims, appointments in scheduling and lab values in their own system, and each arrives on its own clock. Healthcare data analytics turns those sources into answers a manager can act on: what happened last month, why, what is likely next, and what to do about it. Each of those questions needs different data, different freshness and a different privacy tier. Progression’s team is based in New York and builds analytics pipelines and dashboards for hospitals, clinics, health plans and digital health companies across the US.
On this page · 14 sections
- A readmission rate that improved because the claims had not arrived
- Healthcare data analytics, rung by rung
- Healthcare data and analytics sources, and when each one arrives
- A pipeline that holds each source apart until it agrees
- Population health: registries, care gaps and outreach lists
- Readmission analytics against the six CMS measures
- HEDIS and quality reporting as NCQA describes it
- Dashboards by role, one decision per screen
- Privacy tiers for every report: full record, limited set, de-identified
- Predictive models in operations, with people deciding
- How to tell the analytics program is working
- What healthcare data analytics costs to build
- Ninety days from source inventory to the first trusted dashboard
- Before the first workshop: what to gather
Building a clinical application? A team that needs a patient portal, a care coordination tool or another clinical application built gets that work on our healthcare software development page.
Examples marked “Illustrative example.” describe hypothetical organizations, and no real hospital, health plan, patient or person is shown.
A readmission rate that improved because the claims had not arrived
Picture a hypothetical community hospital whose monthly dashboard shows the 30 day readmission rate falling sharply in the latest month. The quality committee is pleased. Then the claims for that month arrive, and the rate climbs back, because patients who returned to a different hospital were never in the EHR at all.
Nothing was wrong with the chart. The data behind it simply had two clocks: the hospital’s own records, available the next morning, and claims showing care elsewhere, which arrive weeks later. A sound analytics program says which months are still open and fills them in when the slower source lands.
- Every measure names the sources it depends on.
- Months that are still filling in are marked provisional.
- The report says when it was last refreshed.
- Final numbers replace provisional ones automatically.
Healthcare data analytics, rung by rung
Analytics in a hospital, clinic or health plan answers four questions, and they get harder in order. What happened is a count. Why it happened is a breakdown. What is likely next is a model. What to do about it is a recommendation someone is willing to act on. The table sets out what each rung needs.
| Rung | The question | A typical output | The data it needs | Who acts on it | The guardrail |
|---|---|---|---|---|---|
| Descriptive | What happened? | Monthly readmission, no show and denial rates | Encounters, claims, schedules | Leaders and managers | One written definition per measure |
| Diagnostic | Why did it happen? | Rates broken down by unit, payer, condition and day | The same data, joined and coded consistently | Quality and operations teams | Enough cases in each slice to mean something |
| Predictive | What is likely next? | A risk score or a forecast of census and staffing | History deep enough to learn from, and fresh inputs | Care managers and planners | Checked for drift and for uneven results across patients |
| Prescriptive | What should we do? | A ranked worklist or a short list of staffing options | Predictions plus capacity and rules | Front line teams | A person decides; the tool suggests |
| How the work is quoted | Not a rung | Scope set by sources, measures and audiences | An inventory of systems | The sponsor | Price set in the written proposal |
Where most programs should start
At the bottom. A descriptive layer that everyone believes is worth more than a predictive model built on numbers nobody trusts. Get five measures defined, traced to their systems and reconciled with the reports people already use, then climb. Diagnostic breakdowns follow naturally once the base counts hold up.
- Pick five measures people already argue about.
- Write each definition down and name its owner.
- Reconcile the new numbers with the old report before switching.
- Add breakdowns only where each slice has enough cases.
When does a prediction earn a place on a worklist?
When three things are true: the score is better than the rule of thumb staff use today, someone has time to act on the list it produces, and the action is something the patient would welcome, such as a reminder call or a follow up visit. A score that nobody has capacity to act on adds noise to a busy day.
Healthcare data and analytics sources, and when each one arrives
Most of the work in a program is getting the right data to the right place on time. Clinical systems, claims, operations tools and patient facing apps each deliver on their own schedule, in their own format, and each misses something the others see.
| Feed | What it holds | How it arrives | How fresh | What it misses |
|---|---|---|---|---|
| EHR | Encounters, diagnoses, orders, results, notes | FHIR APIs, scheduled extracts or messages | Near current for most fields | Care the patient received elsewhere |
| Payer APIs | Claims and encounter data the plan holds | FHIR APIs required by CMS for many payers | Set by each payer | Clinical values such as lab results |
| Medicare claims for ACOs (BCDA) | Enrollee identifiers, diagnosis codes, dates and times of service | API with NDJSON files in FHIR format | Available 2 to 14 days after submission, updated at least weekly | Care not yet billed |
| Medicare claims for patients (Blue Button) | Part A, B and D data for a person who connects an app | Standards based API, with the person’s permission | As the person authorizes | Anyone who has not connected |
| Admit, discharge, transfer feeds | Arrivals, transfers and discharges | HL7 messages from the hospital system | Within the hour for most events | Diagnoses and costs |
| Scheduling and billing | Appointments, no shows, charges, denials | Database extracts or vendor APIs | Daily is common | Clinical reasons for a visit |
| Surveys and devices | Patient reported outcomes, readings from home devices | App and device platforms | Varies by program | People who never respond or connect |
FHIR APIs under the ONC and CMS rules
FHIR is HL7’s standard for health care data exchange, and it is now the default way to pull data from EHRs and payers. The CMS Interoperability and Patient Access final rule, published May 1, 2020, requires or encourages Medicare Advantage, Medicaid, CHIP and exchange plans to share data through APIs with patients, with providers and with other payers, using the content and vocabulary standards from ONC’s 21st Century Cures Act rule.
The Interoperability and Prior Authorization final rule, released January 17, 2024, builds on it with a Provider Access API, a Payer-to-Payer API and a Prior Authorization API. CMS gives affected payers until January 1, 2027, for most of the API requirements. For analytics, that means more claims and encounter data reachable by a standard request rather than a custom file.
Population level pulls use the FHIR Bulk Data Access implementation guide, published as STU 3 on FHIR R4, which exports many patients’ records as files instead of one patient at a time. Integration with the record itself is covered on our EHR software development page.
Claims that arrive in days, not hours
Claims tell you about care that happened anywhere, which clinical data cannot. CMS’s Beneficiary Claims Data API serves organizations in its alternative payment models, including the Shared Savings Program and ACO REACH. Claims are available 2 to 14 days after submission, the data refreshes at least once a week, and files come as NDJSON in FHIR format with enrollee identifiers, diagnosis codes and dates and times of service.
For individuals, Blue Button is CMS’s API that delivers Medicare Part A, B and D data for over 60 million people with Medicare, to the apps they choose to connect. Digital health products use it to show members their own history; analytics teams use what members share with them, under that permission.
What does USCDI cover?
The United States Core Data for Interoperability is a standardized set of health data classes and data elements for nationwide exchange, maintained by ASTP/ONC and published in numbered versions; version 6 came out on July 24, 2025. When an EHR exposes USCDI data over FHIR, allergies, problems, medications, results and vital signs arrive in predictable shapes, which shortens the mapping work in every analytics build.
A pipeline that holds each source apart until it agrees
Raw feeds should never flow straight into a dashboard. Each one lands untouched first, so that when two systems disagree there is a record of what each said and when. Only rows that pass their checks move into the shared model that measures are computed from.
- Land every feed raw, with arrival time and source system.
- Map local codes to shared vocabularies and convert units.
- Join patients, encounters and claims in one model.
- Hold back rows that fail checks and show the count held back.
- Compute each measure from its written definition.
- Publish by role and privacy tier.
Warehouse design, streaming and storage choices are covered on our big data software development services page. The patient matching and metric dictionary work that keeps numbers consistent is on our healthcare software page.
Why a common data model helps
A common model lets the same query run on data from different systems and sites. The OMOP Common Data Model, maintained by the OHDSI community, is an open standard for the structure and content of observational data, with standardized vocabularies at its center. Health systems with research or real world evidence goals often keep an OMOP layer beside the operational model that feeds daily dashboards.
How fresh does each dashboard need to be?
As fresh as the decision it supports, and no fresher. A bed board needs minutes, a unit’s discharge plan needs hours, a quality committee needs a month that has closed, and a board report needs a quarter. Building everything to real time costs money and adds failure points that a monthly measure does not need.
- Minutes: beds, arrivals, waiting rooms.
- Hours: discharges expected, open orders, staffing.
- Days: denials, scheduling gaps, outreach lists.
- Closed months: quality measures and readmissions.
Numbers nobody quite trusts?Send a list of the reports people use today and the systems behind them. You get a free proposal for the healthcare data analytics build, source by source.
Get a free proposalTell us about a project you’d like help with.
Population health: registries, care gaps and outreach lists
Population health work starts with a registry: everyone the organization is responsible for who has a given condition or is due a given service. From the registry come care gaps, such as a screening not done or a medication not refilled, and from the gaps come outreach lists that care managers can work through.
- Define who is in the population and who is attributed to each clinician.
- Build one registry per condition or program.
- Show each open gap with the date it opened.
- Rank outreach lists by need and by the team’s capacity.
- Close the loop: record what was done and whether the gap closed.
Who belongs in the denominator?
The answer decides every rate on the screen. A diabetes measure might count patients seen twice in two years with the diagnosis, members continuously enrolled for a year, or anyone with a qualifying lab result. Each choice gives a different rate. Pick the definition the measure’s owner uses for reporting, write it down, and show it on the screen beside the number.
Readmission analytics against the six CMS measures
Medicare’s Hospital Readmissions Reduction Program has reduced payments to hospitals with excess readmissions since October 1, 2012. It uses six condition or procedure specific 30 day risk standardized unplanned readmission measures, and since fiscal year 2019 it compares each hospital with others serving a similar share of patients dually eligible for Medicare and full Medicaid.
| CMS measure | What an internal EHR view can see | What it can miss |
|---|---|---|
| Acute myocardial infarction (AMI) | Index stays and returns to the same hospital | Returns to other hospitals until claims arrive |
| Chronic obstructive pulmonary disease (COPD) | Index stays and returns to the same hospital | Risk standardization CMS applies |
| Heart failure (HF) | Index stays and returns to the same hospital | Patients outside the Medicare population CMS uses |
| Pneumonia | Index stays and returns to the same hospital | Exclusions written into the measure specification |
| Coronary artery bypass graft (CABG) surgery | Surgical stays and returns | Returns elsewhere and the rolling performance period |
| Elective primary total hip or knee arthroplasty (THA/TKA) | Surgical stays and returns | Returns elsewhere and the rolling performance period |
The reduction is capped at 3 percent of base operating payments. Each year CMS sends a confidential hospital specific report and allows 30 days to question the calculation, though corrections to the underlying claims are not accepted in that window. An internal readmission view earns its keep in the months before that report, by showing where returns are happening while there is still time to change discharge planning and follow up.
Why a hospital’s own count differs from the CMS result
Three reasons usually explain the gap. The hospital’s own data sees only returns to its own facilities, while the CMS result draws on claims from everywhere. CMS risk standardizes each measure, so a raw internal rate is not comparable. And CMS measures over a rolling performance period, not the calendar month a dashboard shows. Label the internal view as an early signal, never as the official rate.
HEDIS and quality reporting as NCQA describes it
The Healthcare Effectiveness Data and Information Set is, in NCQA’s words, one of health care’s most widely used performance improvement tools. NCQA says more than 235 million people are enrolled in plans that report HEDIS results, across more than 90 measures in six domains of care, and audits protect the validity and reliability of what plans submit.
One domain covers measures reported using electronic clinical data systems, which is where clinical data from providers matters most. A health plan’s analytics program pulls claims for every member, adds clinical results that claims cannot carry, and tracks each open gap by measure and by practice.
What do health plans need from provider data?
Clinical facts that claims leave out: lab values, blood pressure readings, screening results and dates. Providers who can send those facts in a clean, standard format help plans report accurately and spend less time chasing charts. On the provider side, the same data powers a practice’s own view of its open gaps for each plan.
- Results with dates, codes and units.
- Screenings recorded as structured data, not only in notes.
- Member identifiers that match the plan’s.
- A regular feed rather than a year end scramble.
Dashboards by role, one decision per screen
A single dashboard for everyone ends up serving no one. Each role makes different decisions on a different clock, and needs different detail and a different privacy tier. The cards below show six role views a program commonly builds; the visual design side is covered on our dashboard development page.
Every screen should answer one question at a glance, show when its data was refreshed, and let a reader click from a rate to the rows behind it if their role allows. Charts that need a legend to read usually belong on a different screen.
Should patient names ever appear on a dashboard?
Yes, on screens built for the people treating or following up with those patients, and nowhere else. A care manager’s worklist needs names and phone numbers. A quality committee’s trend chart does not, and showing them there adds risk without helping anyone decide anything.
Privacy tiers for every report: full record, limited set, de-identified
HIPAA does not treat every report the same, and an analytics program should not either. The minimum necessary standard asks covered entities and business associates to limit protected health information to what a purpose needs, but it does not apply to disclosures to a provider for treatment, to the patient, or under an authorization.
| Tier | What it holds | The rule | Who usually uses it | Paperwork |
|---|---|---|---|---|
| Full record for treatment | Everything a clinician needs for the patient in front of them | Minimum necessary does not apply to disclosures to a provider for treatment | Treating clinicians and care teams | Normal access controls and audit logs |
| Role limited | Only the categories of information each role needs | Minimum necessary, with workforce roles and categories defined | Operations, scheduling, billing | A written access matrix by role |
| Limited data set | Health information with 16 direct identifiers removed; dates, town, State and zip code may remain | Research, public health or health care operations only | Analysts, researchers, outside collaborators | A data use agreement |
| De-identified | Information that no longer identifies anyone | Expert determination or removal of the listed identifiers | Wider sharing and product analytics | The method documented |
| Aggregate counts | Totals and rates with no row level detail | Built from any tier above, with small counts reviewed | Boards and public reports | A rule for small cells |
In practice, the rule at 45 CFR 164.514 asks an organization to identify which people or roles in its workforce need access and which categories of information each needs, then limit access to match. That is a dashboard permission matrix written in legal language. Broader controls and hosting are covered on our HIPAA compliant software development page.
A de-identified extract sits outside HIPAA’s definition of identifiable health information altogether. The regulation gives two ways to reach that point: a qualified expert applies statistical methods, finds the risk of identification very small and documents the analysis, or the organization strips a fixed list of identifiers and knows of nothing that could still point to a person.
What a data use agreement has to say
A limited data set may be used or shared only for research, public health or health care operations, and only under a data use agreement. The agreement sets the permitted uses, names who may use or receive the data, and binds the recipient to use it only as agreed, keep appropriate safeguards and report any use outside the agreement.
- Permitted uses, tied to research, public health or operations.
- Who may use or receive the data.
- No further use or disclosure beyond the agreement.
- Appropriate safeguards.
- A duty to report any use the agreement does not allow.
Small areas and very old patients in de-identified extracts
Two details catch analytics teams out. Under the identifier removal route, only the first three digits of a zip code may stay, and only where the area those digits cover holds more than 20,000 people; otherwise they become 000. Ages over 89 are folded into a single category of 90 or older. Maps and age charts built from such extracts need to respect both.
Predictive models in operations, with people deciding
The most useful models in health care operations tend to be modest: which appointments are likely to be missed, how many beds tomorrow will need, which discharges may need extra planning. Each produces a list or a forecast, and a person decides what to do with it. Models that read images or lab results sit closer to the diagnosis, and our guide to AI diagnostics software development covers that side.
- Write down what the score is for and what it is not for.
- Compare results across age, language, payer and other patient characteristics before launch.
- Keep inputs to fields the users of the list are allowed to see.
- Track performance monthly against the launch baseline.
- Set a retraining date whenever inputs or workflows change.
Model work inside clinical and operational systems draws on our AI integration services, and broader strategy questions on AI consulting.
Which signals show a model needs review?
A drop in performance against the baseline, a shift in the mix of patients or inputs, a change in how the source system records a field, or staff overriding the list more often than before. Any one of those is reason to pause and look before the next list goes out.
Claims and clinical data in separate places?Tell us which feeds you receive and how often. The free proposal covers the pipeline, the measures and the privacy tier for each report.
Get a free proposalTell us about a project you’d like help with.
How to tell the analytics program is working
Usage counts are easy to collect and say little. The better test is whether decisions changed: a committee that stopped arguing about whose number is right, a care team that works from the list instead of around it, a measure owner who signs off on the definition each quarter.
- Measures with a named owner and a signed definition.
- Reports retired because the new screen replaced them.
- Rows held back by checks, and how fast they were fixed.
- Lists acted on, and what share of items were completed.
- Requests for new measures, which show people trust the old ones.
What belongs in the quarterly review?
Each measure’s owner, its current definition and any change, the data quality checks that failed and how they were fixed, which screens are used and which are ignored, and the next rung to climb. Screens nobody opens for a quarter should be retired rather than kept for completeness.
What healthcare data analytics costs to build
Cost follows the number of sources, the number of measures, the audiences and their privacy tiers, and how fresh each view must be. The bands below run from the typical low to the published high in the US market.
| Work | Typical low to published high | Notes |
|---|---|---|
| Multi-source reporting suite | $40,000 to $100,000 | Several sources, shared measures, dashboards by role |
| Embedded analytics in a product | $40,000 to $120,000 | Charts and reports inside a patient or provider application |
| Analytics platform with pipelines | $75,000 to $400,000 | Feeds, warehouse, measure logic, governed access and dashboards |
| Senior US developer time | $100 to $200 an hour | For extensions and support after launch |
Progression sets its price in the written proposal, after scope, and it is subject to change. Cost bands for broader clinical builds are on our healthcare software development page.
What moves the price?
The number and condition of source systems, whether claims and clinical data both need to land, the count of measures and how complex their definitions are, embedded versus internal dashboards, real time needs, and how many privacy tiers the reports must respect. Clean FHIR access lowers the cost; files sent by email raise it.
Pricing depends on scope and is confirmed in a written proposal. Published prices are subject to change.
Ninety days from source inventory to the first trusted dashboard
A first release with two or three sources and a handful of measures commonly fits in about a quarter, when access to the systems arrives on time. The timeline shows how those days usually divide.
- Inventory systems, reports and feeds.
- Secure access and sign the agreements each source needs.
- Write the first five measure definitions with their owners.
- Land the feeds and map codes.
- Conform the model and compute the measures.
- Reconcile with the reports people use today.
- Release to a small set of users and take feedback.
- Plan the next rung.
Related builds for patient facing products are on our healthcare app development and telemedicine app development pages.
Before the first workshop: what to gather
A first working session goes further when the reports, systems and decisions are on the table. The card lists what to bring.
- The reports and spreadsheets people use now.
- A list of systems and who administers each.
- How each feed arrives and how often.
- The five measures that matter most, with owners.
- Agreements already signed with vendors and other organizations.
- Who reads each report and what they may see.
- The decision each measure should change.
Teams that market a health system or plan alongside this work will find related pages on hospital marketing, marketing analytics and AEO for healthcare. Life sciences data products are covered under pharmaceutical software development.
What this page leaves out: the Progression Method, proprietary and reserved for clients. Reach out to learn more
Getting found in search
AI, AEO and what is changing
Paid media and lead generation
Websites and design
Choosing and working with an agency
Software and app development
Website design by industry and type
Web development, platforms and hosting
Social, content and brand
By industry and by situation
Frequently asked questions
What does healthcare data analytics cover for a provider or plan?
How do descriptive, diagnostic, predictive and prescriptive analytics differ in a hospital?
Where does healthcare data and analytics work usually get its data?
What is USCDI and why does it matter for reporting?
Can claims data reach a dashboard quickly enough to use?
Which payer APIs did the CMS interoperability rules create?
Which conditions does the CMS readmissions program measure?
How large can a readmission payment reduction be?
Why does our internal readmission rate differ from the CMS figure?
What is HEDIS, and who reports it?
Does the minimum necessary rule apply to a care team’s worklist?
When can a report use a limited data set?
What does a data use agreement for analytics have to include?
When is a de-identified extract the right tier?
Is OMOP a good fit for a health system’s research layer?
How fresh should a clinical operations dashboard be?
Who should own each measure definition?
Can a predictive model decide which patients get outreach?
What does a healthcare analytics build cost?
How long until the first dashboard is trusted?
Can Progression build our healthcare data analytics program?
Numbers nobody quite trusts?Send a list of the reports people use today and the systems behind them. You get a free proposal for the healthcare data analytics build, source by source.
Get a free proposalTell us about a project you’d like help with.
Get a free marketing proposal
Tell us what you are trying to grow and we will come back with a plan, not a pitch deck. Same-day reply on weekdays.
