Healthcare Claims and Denials

• Denial prediction

  before submission, with the true cause and a preventable flag as labels; features are the payer, plan type, facility, clinician, service lines, diagnosis, authorization and charge lag.

• Denial management and root cause dashboards

  denial rate by payer, facility, service and CARC category; preventable share; rework success by category; dollars at risk and recovered.

• Accounts receivable and cash forecasting

  days to remit by payer, days in A/R, pending and not-yet-submitted claims at year end, the late-filing tail.

• Patient responsibility

  deductibles that reset on 1 January and fill during the year (here and with other providers), copays that stand in for the deductible on office and telehealth visits, coinsurance; the patient share of allowed falls from 34% in January to 17% in the autumn.

• Payer contract analysis

  billed against contracted against allowed, contractual adjustments, commercial above Medicare above Medicaid.

• Teaching and demos

  an RCM course, a healthcare analytics portfolio, a Power BI/Tableau dashboard, or a billing-system demo database, with no PHI anywhere.

• claims.cause_coverage_inactiveOne of nine cause_* flags: every problem present on the claim, denied or not.true on 2.8% of rows
• claims.denial_causeThe cause the denial was really for, behind the CARC code the payer sent.10 values: coverage inactive, payer policy review, provider not enrolled…, 86.4% empty
• claims.preventableWhether the denial could have been prevented at the front end or in coding.true on 80.7% of rows, 86.4% empty
• encounters.expected_dx_categoryThe diagnosis the visit should have been coded with.6 values: chronic condition, musculoskeletal, preventive or screening…
• encounters.dx_coded_wrongWhether the coded diagnosis departs from it.true on 1.8% of rows
• encounters.needs_prior_authWhether the service needed prior authorisation, next to whether it was obtained.true on 13.1% of rows
• claim_lines.modifier_missingA line that needed a modifier and went without one.true on 1.9% of rows
• claims.recoveredWhether a reworked denial was paid in the end, with the amount recovered.true on 58.5% of rows, 93.6% empty
• patients.outside_allowed_monthlyWhat the plan allows a month for the patient's care elsewhere; it fills the same deductible, out of the provider's sight.2.48 to 5,512.1, mean 220.5

• 12.3% of answered claims were denied in full on first pass; 81% of denials were preventable at the front end or in coding. Billed $52,533,920, allowed $21,180,120, paid by the plans $16,497,244.
• Imaging, procedures and therapy are denied more than office visits; traditional Medicare pays fastest and Medicaid managed care slowest.
• Median charge lag is 3 days with a tail that waits months, which is where late-filing denials come from. Median days in A/R: 23.
• Of reworked denials resolved by year end, 74% were recovered; coding and coordination-of-benefits denials come back far more often than late filing.
• At the cut-off (31 December 2025) 7,252 claims were still with the payer and 1,964 not yet submitted.

• every key resolves and every primary key is unique; one claim per visit; each claim's patient, plan, clinician and facility are the visit's; a patient's plan type is their plan's;
• every service suits its visit type; line charges are base rate x facility markup x units and contracted amounts base rate x plan fee level x units; claim totals re-add from their lines;
• allowed, deductible, copay, coinsurance, patient share and paid re-add exactly; the deductible accumulator is rebuilt in service order from 1 January (with the care the patient had elsewhere) and no patient pays past their plan's deductible; on a copay plan an office or telehealth visit costs the copay alone;
• every cause flag is recomputed from the rows it comes from (coverage end date, enrollment date, charge lag against the plan's limit, authorization, the coded diagnosis); every denial's cause is present on its claim and its CARC code and category match it; every ICD-10-CM code belongs to its diagnosis group;
• no visit before its clinician joined; nothing submitted or answered after the cut-off; rework after the answer; days in A/R recompute.

• claim_linesService lines: units, modifier present or missing, charge, contracted and allowed amounts, line-level denials161,097 rows · 17 cols
• claimsOne claim per visit through its whole life: charge lag, submission, every cause that was present, the denial with CARC code and category, remittance, deductible, copay, coinsurance, paid, adjustments, rework and recovery, days in A/R85,659 rows · 54 cols
• encountersEvery visit: type, time (clinic hours or around the clock), the diagnosis expected and the ICD-10-CM code actually coded85,659 rows · 15 cols
• patientsAge, sex, plan, member id, annual deductible, what the plan allows a month for their care with other providers (it fills the same deductible), the date their coverage ended if it did, and other coverage24,000 rows · 14 cols
• providersClinicians by facility and specialty; the ones who joined during the year, with the date and the day their payer enrollment took effect140 rows · 9 cols
• services36 billable services with an illustrative base rate, whether they bill in timed units and whether they need a modifier (own codes, not CPT)36 rows · 7 cols
• payersTen health plans across commercial PPO and HMO, Medicare Advantage, traditional Medicare and Medicaid managed care: timely-filing limit, fee level against a base rate, days to pay, prior-auth policy, coinsurance and copay10 rows · 10 cols
• facilitiesClinics, hospital outpatient departments, an emergency department, imaging and therapy sites: chargemaster markup, front-desk registration error rate, billing backlog8 rows · 8 cols

Is this real data?

No. This is synthetic data generated by software. No row describes a real person, company, patient, store, machine or transaction. The patterns are modelled to be realistic and the statistics quoted are measured on these files, but they do not describe any real population or market. Use it for learning, testing, demos, benchmarks and prototyping, not as evidence about the real world. Provided as is, without warranty.

What do I get?

One zip of 8.9 MB: 8 linked tables and 356,609 rows as CSV, a README of what each table holds and how the data behaves, INTEGRITY.json with the 87 audit checks, and RECIPE.json, the Misata blueprint that made these exact rows (seed 20250101).

Can I try it before buying?

Yes. The free preview (117 KB) is a slice of the same data with the keys intact, plus the README, so you can load it and check it fits before you pay.

How was it checked?

87 of 87 checks pass, re-run on the delivered files by an independent script with plain pandas: every key resolves and every primary key is unique; one claim per visit; each claim's patient, plan, clinician and facility are the visit's; a patient's plan type is their plan's; every service suits its visit type; line charges are base rate x facility markup x units and contracted amounts base rate x plan fee level x units; claim totals re-add from their lines; allowed, deductible, copay, coinsurance, patient share and paid re-add exactly; the deductible accumulator is rebuilt in service order from 1 January (with the care the patient had elsewhere) and no patient pays past their plan's deductible; on a copay plan an office or telehealth visit costs the copay alone; every cause flag is recomputed from the rows it comes from (coverage end date, enrollment date, charge lag against the plan's limit, authorization, the coded diagnosis); every denial's cause is present on its claim and its CARC code and category match it; every ICD-10-CM code belongs to its diagnosis group; no visit before its clinician joined; nothing submitted or answered after the cut-off; rework after the answer; days in A/R recompute.

Can I use it commercially?

Yes: in any project, course, benchmark, demo or product, commercial or not. You may not resell or redistribute the dataset itself as a dataset.

What should I know before using it?

CARC codes (Claim Adjustment Reason Codes) and ICD-10-CM codes are the real public code sets. Services use this dataset's own codes and illustrative base rates, not CPT (which is licensed) or any real fee schedule. A claim can have several causes present; the one recorded is the first a payer's edits reach, and a present cause does not always fire (a payer may back-date an enrollment). The cause_ columns are every cause present. One claim per visit; secondary claims to the other payer and appeals beyond one rework are not modelled. Payers, facilities, clinicians and patients are invented; no real plan, organization or person is represented, and no record is derived from real patient data.

Can I get a bigger or different version?

Yes. RECIPE.json runs in Misata Studio or through Misata's MCP server to make a variant, or ask us to build one to your spec.

Want it bigger, in another setting, or with your own columns? Have us build it or make it in Studio.