File size: 8,508 Bytes
eb785f1
 
9ccc295
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
018bd7e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9ccc295
 
018bd7e
9ccc295
 
018bd7e
eb785f1
9ccc295
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
018bd7e
 
 
 
 
 
 
9ccc295
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
018bd7e
9ccc295
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
---
license: cc-by-4.0
pretty_name: x12sdk synthetic health care claims and remittances
language:
- en
tags:
- healthcare
- x12
- edi
- hipaa
- claims
- synthetic
- tabular
size_categories:
- 10K<n<100K
task_categories:
- tabular-classification
- tabular-regression
dataset_info:
- config_name: claims
  features:
  - name: claim_type
    dtype: string
  - name: pair
    dtype: int64
  - name: patient_control_number
    dtype: string
  - name: service_date
    dtype: string
  - name: claim_charge
    dtype: float64
  - name: line_count
    dtype: int64
  - name: is_dependent
    dtype: bool
  - name: relationship_code
    dtype: string
  - name: patient_gender
    dtype: string
  - name: patient_birth_year
    dtype: int64
  - name: diagnosis_code
    dtype: string
  - name: payer_name
    dtype: string
  - name: billing_provider_name
    dtype: string
  - name: billing_provider_npi
    dtype: string
  - name: claim_status_code
    dtype: string
  - name: denied
    dtype: bool
  - name: paid_amount
    dtype: float64
  - name: contractual_adjustment
    dtype: float64
  - name: patient_responsibility
    dtype: float64
  - name: denial_reason_code
    dtype: string
  - name: payer_claim_control_number
    dtype: string
- config_name: service_lines
  features:
  - name: claim_type
    dtype: string
  - name: pair
    dtype: int64
  - name: patient_control_number
    dtype: string
  - name: line_number
    dtype: int64
  - name: procedure_qualifier
    dtype: string
  - name: procedure_code
    dtype: string
  - name: revenue_code
    dtype: string
  - name: units
    dtype: int64
  - name: line_charge
    dtype: float64
  - name: tooth_number
    dtype: string
  - name: tooth_surfaces
    dtype: string
  - name: line_paid
    dtype: float64
  - name: contractual_adjustment
    dtype: float64
  - name: patient_responsibility
    dtype: float64
  - name: adjustment_reasons
    dtype: string
configs:
- config_name: claims
  data_files: data/claims.jsonl
  default: true
- config_name: service_lines
  data_files: data/service_lines.jsonl
---

# x12sdk synthetic health care claims and remittances

Synthetic ASC X12 5010 health care claims, the remittances that answer them,
and the same data as two flat tables. **Every record is synthetic.** Nothing
here is, or is derived from, real patient, provider or payer data.

Real claims and remittances are protected health information, so there is no
public set of them to build or test against. This dataset fills that gap for
anyone who needs X12 in the real format: engineers testing a claims pipeline,
people learning the transactions, and anyone wanting a claim and its outcome
side by side.

It was generated with [x12sdk](https://github.com/owgreen-dev/x12sdk)
2.3.0 and can be rebuilt byte for byte from the seed below.

## What is in it

| Path | Contents |
|---|---|
| `x12/837p/`, `x12/837i/`, `x12/837d/` | 300 X12 files: for each claim type, 50 claim submissions (`pair-NNN.837`) each paired with the remittance that answers it (`pair-NNN.835`), 200 claims per file |
| `data/claims.jsonl` | 30000 rows, one per claim: as billed, and what the payer did |
| `data/service_lines.jsonl` | 64831 rows, one per service line, likewise |
| `data/claims.csv`, `data/service_lines.csv` | the same two tables as CSV, for spreadsheets |

The JSON Lines files carry types: codes such as revenue code `0510` stay
text, and amounts are numbers. Tools that guess column types from a CSV may
read `0510` as the number 510, so read the codes as text if you use the CSVs.

Claim types: **837P** professional (005010X222A2), **837I** institutional
(005010X223A3), **837D** dental (005010X224A2). Remittances are **835**
(005010X221A1).

Every claim on an 837 is answered by exactly one claim payment on its 835,
linked as real files are, by the patient control number (CLM01 = CLP01), with
the same charge, the same procedure on each line, and the same patient, payer
and payee. Every file parses and validates against the implementation guide
rules the library models, including the 835 balancing rules.

The tables were made by parsing the X12 files with x12sdk, so they are a check
on the files as well as a convenience: `patient_control_number` (with
`claim_type` and `pair`) joins `claims` to `service_lines`.

### `claims` columns

| Column | Meaning |
|---|---|
| `claim_type` | `837p`, `837i` or `837d` |
| `pair` | which file pair the claim is in |
| `patient_control_number` | CLM01 / CLP01, the link between the two files |
| `service_date` | the claim's date of service (or statement date, 837I) |
| `claim_charge` | CLM02, total billed |
| `line_count` | number of service lines |
| `is_dependent` | the patient is a dependent of the subscriber, not the subscriber |
| `relationship_code` | the patient's relationship code (18 = self) |
| `patient_gender`, `patient_birth_year` | from the patient's DMG segment |
| `diagnosis_code` | the principal ICD-10-CM code (none on dental claims) |
| `payer_name`, `billing_provider_name`, `billing_provider_npi` | synthetic placeholders; NPIs carry a valid check digit but are not real |
| `claim_status_code` | CLP02: `1` processed as primary, `4` denied |
| `denied` | `claim_status_code` is `4` |
| `paid_amount` | CLP04 |
| `contractual_adjustment` | total CO (contractual obligation) adjustments |
| `patient_responsibility` | total PR adjustments: deductible, coinsurance, copay |
| `denial_reason_code` | for a denied claim, the claim adjustment reason code |
| `payer_claim_control_number` | CLP07 |

### `service_lines` columns

| Column | Meaning |
|---|---|
| `line_number` | position on the claim |
| `procedure_qualifier`, `procedure_code` | `HC` with a CPT/HCPCS code, or `AD` with a CDT dental code |
| `revenue_code` | institutional claims only |
| `units` | units of service (procedure count on dental claims) |
| `line_charge`, `line_paid` | billed, and paid on the 835 |
| `tooth_number`, `tooth_surfaces` | dental claims: Universal tooth number and surfaces (e.g. `M:O`), where the procedure is done to a tooth |
| `contractual_adjustment`, `patient_responsibility` | the line's CO and PR totals |
| `adjustment_reasons` | every adjustment on the line as `GROUP-REASON`, e.g. `CO-45;PR-2` |

## Outcomes are random

**Read this before using the data for machine learning.** Whether a claim is
paid in full, reduced or denied is decided at random, not learned from any
payer's behaviour. About 15% of claims are denied outright; most of the rest
are reduced by a contractual write-off (CO 45), some also leave a share to the
patient, and some are paid in full. Nothing in a claim predicts its outcome.

That makes the data right for testing that a pipeline reads, joins and totals
claims and remittances correctly, for teaching the transactions, and for
exercising a model's training code, and wrong for training a model you expect
to learn real-world denial patterns. A later version is planned with
published, administrative denial rules (a claim filed too late, missing
required information, lacking a required authorization, or for a service the
plan does not cover), so that outcomes follow documented causes.

Outright denials use administrative reason codes only (16, 27, 29, 96, 197).
Codes appear bare; no code-list descriptions are included.

## Limitations

- Codes come from a short fixed list of common procedure, dental and
  diagnosis codes, so their distribution is not realistic. Fees fall in
  plausible ranges but are random.
- Each file pair has one payer and one billing provider; there is no
  coordination of benefits between payers.
- Names, addresses and identifiers are obvious placeholders. They are kept in
  the X12 files, where the format requires them, and left out of the tables.
- The 835's claim filing indicator is always `CH`, which does not match the
  dental claims' `CI`.

## Rebuilding it

```shell
pip install "x12sdk==2.3.0"
git clone https://github.com/owgreen-dev/x12sdk && cd x12sdk
python scripts/build_dataset.py --out dataset --seed 2026 --pairs 50 --claims-per-pair 200
```

Or generate your own with any seed and size:

```python
from x12sdk.generate import generate_claim_pair

claim, remittance = generate_claim_pair(seed=7, claims=25, claim_type="837d")
```

## License and citation

Released under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/).
x12sdk itself is Apache-2.0. If you use this data, please credit
"x12sdk synthetic claims and remittances" and link
https://github.com/owgreen-dev/x12sdk.