ALOTT-MEDS 0.0.1__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -0,0 +1,21 @@
1
+ MIT License
2
+
3
+ Copyright (c) 2025 Robin van de Water
4
+
5
+ Permission is hereby granted, free of charge, to any person obtaining a copy
6
+ of this software and associated documentation files (the "Software"), to deal
7
+ in the Software without restriction, including without limitation the rights
8
+ to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
9
+ copies of the Software, and to permit persons to whom the Software is
10
+ furnished to do so, subject to the following conditions:
11
+
12
+ The above copyright notice and this permission notice shall be included in all
13
+ copies or substantial portions of the Software.
14
+
15
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16
+ IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17
+ FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
18
+ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19
+ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20
+ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21
+ SOFTWARE.
@@ -0,0 +1,296 @@
1
+ Metadata-Version: 2.4
2
+ Name: ALOTT-MEDS
3
+ Version: 0.0.1
4
+ Summary: An ETL pipeline to extract the ALOTT (Carescape) EMR data into the MEDS format.
5
+ Author-email: Robin van de Water <rpv2114@cumc.columbia.edu>
6
+ Project-URL: Homepage, https://github.com/rvandewater/ALOTT_MEDS
7
+ Project-URL: Issues, https://github.com/rvandewater/ALOTT_MEDS/issues
8
+ Classifier: Programming Language :: Python :: 3
9
+ Classifier: License :: OSI Approved :: MIT License
10
+ Classifier: Operating System :: OS Independent
11
+ Requires-Python: <3.14,>=3.11.4
12
+ Description-Content-Type: text/markdown
13
+ License-File: LICENSE
14
+ Requires-Dist: meds-extract[download]<0.8,>=0.7.1
15
+ Dynamic: license-file
16
+
17
+ # ALOTT-MEDS
18
+
19
+ <!-- Placeholders until available: the PyPI badges resolve once ALOTT-MEDS is published; replace
20
+ CODECOV_TOKEN_PLACEHOLDER with the repo's Codecov badge token and zenodo.XXXXXXX with its Zenodo DOI. -->
21
+
22
+ [![PyPI - Version](https://img.shields.io/pypi/v/ALOTT-MEDS)](https://pypi.org/project/ALOTT-MEDS/)
23
+ [![codecov](https://codecov.io/gh/rvandewater/ALOTT_MEDS/graph/badge.svg?token=CODECOV_TOKEN_PLACEHOLDER)](https://codecov.io/gh/rvandewater/ALOTT_MEDS)
24
+ [![tests](https://github.com/rvandewater/ALOTT_MEDS/actions/workflows/synthetic-etl.yaml/badge.svg)](https://github.com/rvandewater/ALOTT_MEDS/actions/workflows/synthetic-etl.yaml)
25
+ [![code-quality](https://github.com/rvandewater/ALOTT_MEDS/actions/workflows/code-quality-main.yaml/badge.svg)](https://github.com/rvandewater/ALOTT_MEDS/actions/workflows/code-quality-main.yaml)
26
+ ![python](https://img.shields.io/badge/python-%3E%3D3.11.4%2C%20%3C3.14-blue?logo=python&logoColor=white)
27
+ ![Static Badge](https://img.shields.io/badge/MEDS-0.4.1-blue)
28
+ [![license](https://img.shields.io/badge/License-MIT-green.svg?labelColor=gray)](https://github.com/rvandewater/ALOTT_MEDS#license)
29
+ [![PRs](https://img.shields.io/badge/PRs-welcome-brightgreen.svg)](https://github.com/rvandewater/ALOTT_MEDS/pulls)
30
+ [![contributors](https://img.shields.io/github/contributors/rvandewater/ALOTT_MEDS.svg)](https://github.com/rvandewater/ALOTT_MEDS/graphs/contributors)
31
+ [![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.XXXXXXX.svg)](https://doi.org/10.5281/zenodo.XXXXXXX)
32
+
33
+ ALOTT (ALarms, Outcomes Telemetry with Timing) is a pilot bedside-EMR database from two
34
+ hospitals, collected to support algorithms that predict emergency response team events. It
35
+ pairs bedside telemetry from 270 beds (alarms, monitor measurements and waveforms) with
36
+ electronic medical record data: admissions, diagnoses, ICU stays and bed moves, labs,
37
+ orders, medications and nursing flowsheets.
38
+ This pipeline extracts the EMR tables of ALOTT (from PhysioNet,
39
+ https://physionet.org/content/alott/) into the
40
+ [MEDS](https://github.com/Medical-Event-Data-Standard/meds) format. It is built the same way
41
+ as [MIMIC_IV_MEDS](https://github.com/Medical-Event-Data-Standard/MIMIC_IV_MEDS): one
42
+ declarative [MEDS-Extract](https://github.com/mmcdermott/MEDS_extract) spec, no Python.
43
+
44
+ ## Usage:
45
+
46
+ ```bash
47
+ pip install git+https://github.com/rvandewater/ALOTT_MEDS
48
+ export DATASET_DOWNLOAD_USERNAME=$PHYSIONET_USERNAME
49
+ export DATASET_DOWNLOAD_PASSWORD=$PHYSIONET_PASSWORD
50
+
51
+ meds-extract-run spec=ALOTT output_dir=$MEDS_OUTPUT_DIR
52
+ ```
53
+
54
+ That downloads only the `emr-datasets/` folder of the ALOTT 1.0.0 release (the telemetry is
55
+ skipped) into `$MEDS_OUTPUT_DIR/.meds_extract_run/raw_input` and runs the full MEDS-Extract
56
+ pipeline, writing `data/` and `metadata/` to `$MEDS_OUTPUT_DIR`. Files that already exist and
57
+ verify against PhysioNet's checksums are skipped, so an interrupted download resumes. Pass
58
+ `download_dest_dir=$RAW_INPUT_DIR` to keep the raw data somewhere durable.
59
+
60
+ ALOTT is a credentialed release: the PhysioNet account must have signed the ALOTT data use
61
+ agreement.
62
+
63
+ To reuse a download you already have, point `input_dir` at the directory that *contains*
64
+ `emr-datasets/`:
65
+
66
+ ```bash
67
+ meds-extract-run spec=ALOTT output_dir=$MEDS_OUTPUT_DIR do_download=false input_dir=$RAW_INPUT_DIR
68
+ ```
69
+
70
+ > [!NOTE]
71
+ > The folder the EMR CSVs sit in on PhysioNet, assumed to be `emr-datasets/`, has not been
72
+ > checked against a download. If your copy is a flat folder of the ten CSVs
73
+ > (`admission.csv`, `labs.csv`, ...), point `input_dir` at a directory holding an
74
+ > `emr-datasets` link to it.
75
+
76
+ Run `meds-extract-run --help` for the full set of arguments.
77
+
78
+ ## Configuration
79
+
80
+ The whole pipeline is one file,
81
+ [`src/ALOTT_MEDS/event_configs.yaml`](https://github.com/rvandewater/ALOTT_MEDS/blob/main/src/ALOTT_MEDS/event_configs.yaml), written in MESSY
82
+ (MEDS-Extract Specification Syntax YAML) and registered under the `MEDS_extract.pipelines`
83
+ entry-point group as `ALOTT`. That registration is what makes `spec=ALOTT` work from anywhere
84
+ and what stamps the output's `metadata/dataset.json` with
85
+ `<ALOTT release>:<this package's version>`. The file holds:
86
+
87
+ - a `sources:` block pointing at the PhysioNet release, restricted to `emr-datasets/*`;
88
+ - an `etl:` block with the run's few knobs;
89
+ - one block per raw EMR table describing the events extracted from it.
90
+
91
+ ### Timestamps
92
+
93
+ ALOTT shifts each patient's dates by a random -180..180 days but keeps the time of day. Most
94
+ tables store each timestamp as one shifted `%Y-%m-%d %H:%M:%S` column. Labs and orders carry
95
+ only a shifted **date**, and medication start and end times are always `00:00:00`, i.e. dates
96
+ too. Those events are placed at the **end of that day (23:59:59)**: midnight would make a
97
+ result or a new medication visible before it existed, which leaks the future into any
98
+ prediction made later that day. Same-day lab orders and results therefore share a time.
99
+ Medication administrations keep their recorded time of day. A row with a null timestamp
100
+ produces no event.
101
+
102
+ Absolute dates are shifted per patient, so only differences *within* a subject are
103
+ meaningful.
104
+
105
+ ### IDs
106
+
107
+ `aMRN` is the MEDS `subject_id`; the encounter key `aCSN` is carried as `hadm_id`. Lab and
108
+ order results use the resulting encounter, and orders the ordering one. The release
109
+ writes the string `nan` for an unresulted encounter; it is cast leniently, so it becomes a
110
+ null and `hadm_id` stays `Int64`.
111
+
112
+ ### Codes
113
+
114
+ Every code is `FAMILY//...`, from general to specific, so a prefix selects a group (all
115
+ labs, one lab panel, one drug). Where a table records both an order and its outcome, the
116
+ second component says which: `ORDERED` or `RESULT`, `START` or `STOP`, or `MAR`
117
+ (Medication Administration Record: the bedside log of each dose given, held or refused).
118
+ `UNK` stands for a null component that must not drop the row (see
119
+ [Null handling](#null-handling)).
120
+
121
+ | Code | Table | Time | Value and extra columns |
122
+ | ------------------------------------------------ | ------------ | --------------------- | -------------------------------------------------------------------------------------- |
123
+ | `HOSPITAL_ADMISSION//<admission type>//<source>` | admission | admission | reason for visit in `text_value` |
124
+ | `HOSPITAL_DISCHARGE//<destination>` | admission | discharge | |
125
+ | `DIAGNOSIS//ICD//10//<ICD-10-CM code>` | admission | discharge | diagnosis text; `diagnosis_rank` `PRIMARY` or `SECONDARY` |
126
+ | `<FIELD>//<value>` | demographics | static | `FIELD` is `GENDER`, `MARITAL_STATUS`, `RACE`, `ETHNICITY`, `LANGUAGE` or `BLOOD_TYPE` |
127
+ | `ALLERGY//<allergen>` | allergy | static | reaction in `text_value` |
128
+ | `FLOWSHEET//<FLO_MEAS_ID>` | flowsheet | recorded | reading or score; `flow_type` |
129
+ | `ICU_ADMISSION`, `ICU_DISCHARGE` | icu | start, end | |
130
+ | `TRANSFER` | location | start | |
131
+ | `MEDICATION//START//<drug>` | medications | start date, 23:59:59 | ordered dose (raw text if a range); `dose_unit`, `frequency` |
132
+ | `MEDICATION//STOP//<drug>` | medications | end date, 23:59:59 | |
133
+ | `MEDICATION//MAR//<drug>//<action>` | medications | administration | given dose; `dose_unit`, `route` |
134
+ | `LAB//ORDERED//<PROC_ID>` | labs | order date, 23:59:59 | |
135
+ | `LAB//RESULT//<PROC_ID>//<component>` | labs | result date, 23:59:59 | result; `order_status` |
136
+ | `ORDER//ORDERED//<order type>//<PROC_ID>` | orders | order date, 23:59:59 | `order_status` |
137
+ | `ORDER//RESULT//<order type>//<PROC_ID>` | orders | result date, 23:59:59 | result; `order_status` |
138
+
139
+ `codes.parquet` carries a description for flowsheet, lab and order codes, taken from the
140
+ release's display names.
141
+
142
+ ### Null handling
143
+
144
+ A null component of a composite code drops the row under MEDS-Extract 0.7. Components are
145
+ therefore coalesced with `?? 'UNK'` wherever a null must not drop the row (admission type and
146
+ source, discharge destination, lab component, order type, medication administration action).
147
+
148
+ The rest are deliberately left uncoalesced. In the release, the code columns of
149
+ flowsheets, labs, orders and allergies are never null, and `MEDICATION` is null in 10 of
150
+ 12,698,867 rows, so a coalesce would add nothing. A missing ICD-10 code or demographic
151
+ produces no event rather than an `UNK` category, because a missing value is not an observed
152
+ one.
153
+
154
+ ### Diagnoses
155
+
156
+ The primary and secondary ICD-10 codes on each visit become `DIAGNOSIS//ICD//10//<code>`
157
+ events at discharge time, with `diagnosis_rank` set to `PRIMARY` or `SECONDARY`. About 4% of
158
+ cells list up to four codes separated by a comma and a space; each code becomes its own event, carrying the
159
+ cell's full diagnosis text. No
160
+ `parent_codes` metadata is attached: the same code occurs as both primary and secondary, and
161
+ MEDS-Extract rejects one code produced by two code templates.
162
+ `DIAGNOSIS//ICD//10//<code>` maps to `ICD10CM/<code>` directly.
163
+
164
+ ### Other tables
165
+
166
+ - **ICU stays and bed moves** carry only their start and end times in the release (no unit,
167
+ room or bed), so they become bare `ICU_ADMISSION`, `ICU_DISCHARGE` and `TRANSFER`
168
+ events.
169
+ - **Demographics and allergies** carry no time in the release and are static events.
170
+ - **Flowsheet, lab and order-result values** are free text. The numeric reading is kept where
171
+ it parses; scored flowsheet items charted as `<score>--><label>` (e.g. `-1-->Drowsy`) keep
172
+ their score. The raw string goes to `text_value` unless it is just the number. Epic's
173
+ nonnumeric sentinel `9999999` never becomes a numeric value.
174
+ - **Medications:** each row of the release is one administration and repeats its order's
175
+ start and end, so an order yields one `START`, one `STOP` and one `MAR` event per
176
+ administration.
177
+ - **Orders** hold every order type except labs; roughly half carry a single result.
178
+ - **Birth and death:** the release has no birth date, age or death time, so there are no
179
+ `MEDS_BIRTH` or `MEDS_DEATH` events.
180
+ - **Not extracted:** the telemetry (alarms, 2-second measurements, waveforms; per-admission
181
+ files keyed by file name, far larger than the EMR), `social_hx` (no event times and
182
+ unlabeled Epic category codes), and the medication order date.
183
+
184
+ ### Validation against the release
185
+
186
+ The spec was first written from the authors' release code
187
+ ([Carescape](https://github.com/johnclawrence/Carescape)) and did not run on the published
188
+ files. Profiling the 1.0.0 EMR CSVs (value *shapes* and null counts only, never values)
189
+ showed what had to change:
190
+
191
+ - Tables are lowercase (`admission`, `labs`, ...) rather than `Patient`, `Labs`, ...; the
192
+ keys are `aMRN` and `aCSN`, not `csID` and `CS_PAT_ENC_CSN_ID`.
193
+ - Each timestamp is one already-shifted column rather than a date plus an `HH:MM:SS`
194
+ column. Labs and orders keep only the date (see [Timestamps](#timestamps)).
195
+ - `icu` and `location` have no department, event type, room or bed columns.
196
+ - `resulting_aCSN` holds the string `nan` for unresulted rows.
197
+ - 1,536 ICD-10 codes occur as both primary and secondary diagnoses, and some cells list
198
+ several codes (see [Diagnoses](#diagnoses)). The synthetic test covers both.
199
+ - Medication start and end times are always midnight, and `orders` carries results of its
200
+ own (it holds no lab orders).
201
+
202
+ All ten CSVs are valid UTF-8. On the full 1.0.0 EMR tables (6.3 GB of CSVs read),
203
+ `meds-extract-run` with MEDS-Extract 0.7.1 finishes in about 90 seconds on a laptop with about
204
+ 6.5 GB peak memory, producing 52.6M events for 18,472 subjects. Fewer lab, order and
205
+ medication start/stop events come out than raw rows because identical events are merged: lab
206
+ and order times have day resolution, and the medication table repeats an order's start and
207
+ end on every administration row.
208
+
209
+ ## MEDS-transforms settings
210
+
211
+ If you want to convert a large dataset, you can parallelize the MEDS-transforms stages
212
+ (the MEDS-transformation step that takes the longest).
213
+
214
+ Using local parallelization with the `hydra-joblib-launcher` package:
215
+
216
+ ```bash
217
+ pip install hydra-joblib-launcher --upgrade
218
+ ```
219
+
220
+ Then write a stage-runner file and pass it to the run:
221
+
222
+ ```yaml
223
+ # runner.yaml
224
+ parallelize:
225
+ n_workers: 8
226
+ launcher: joblib
227
+ ```
228
+
229
+ ```bash
230
+ meds-extract-run spec=ALOTT output_dir=$MEDS_OUTPUT_DIR stage_runner_fp=runner.yaml
231
+ ```
232
+
233
+ The number of subjects per shard is `n_subjects_per_shard` in the spec's `etl:` block
234
+ (1000 by default); raise it to cut the parallelization overhead.
235
+
236
+ ## Development
237
+
238
+ ```bash
239
+ uv sync --group dev
240
+ uv run pytest tests/
241
+ uv run pre-commit run --all-files
242
+ ```
243
+
244
+ A release bumps the version in `pyproject.toml` with `uv version --bump`, commits it and pushes
245
+ a tag of that exact version, which publishes to PyPI and creates the GitHub release. See [`RELEASING.md`](https://github.com/rvandewater/ALOTT_MEDS/blob/main/RELEASING.md) for choosing the
246
+ version, the one-time PyPI setup and recovering from a failed release.
247
+
248
+ ALOTT has no open demo, so CI runs the spec end to end over a small **synthetic** input
249
+ generated in [`tests/e2e_synthetic_test.py`](https://github.com/rvandewater/ALOTT_MEDS/blob/main/tests/e2e_synthetic_test.py). That checks the
250
+ spec's mechanics, not its agreement with the real release. Never commit raw ALOTT data or
251
+ MEDS output derived from it.
252
+
253
+ ## Citation
254
+
255
+ Please cite the following original dataset when using ALOTT:
256
+
257
+ ```bibtex
258
+ @article{PhysioNet-alott-1.0.0,
259
+ author = {Lawrence, John and Rayo, Mike and Huerta, Timothy},
260
+ title = {{ALarms, Outcomes Telemetry with Timing (ALOTT): a Bedside-EMR Database}},
261
+ journal = {{PhysioNet}},
262
+ year = {2025},
263
+ month = mar,
264
+ note = {Version 1.0.0},
265
+ doi = {10.13026/sbq5-dy17},
266
+ url = {https://doi.org/10.13026/sbq5-dy17}
267
+ }
268
+ ```
269
+
270
+ Also cite the ETL framework used to extract the dataset:
271
+
272
+ ```bibtex
273
+ @software{van_de_Water_ALOTT_MEDS_ETL_2025,
274
+ author = {van de Water, Robin Philippus},
275
+ license = {MIT},
276
+ month = apr,
277
+ title = {{ALOTT\_MEDS ETL}},
278
+ url = {https://github.com/rvandewater/ALOTT_MEDS},
279
+ year = {2025}
280
+ }
281
+ ```
282
+
283
+ For citing MEDS in general:
284
+
285
+ ```bibtex
286
+ @article{mcdermott2026meds,
287
+ title={MEDS—An Emerging Data Standard and Ecosystem for Health AI Research},
288
+ author={McDermott, Matthew BA and Steinberg, Ethan and Fries, Jason A and van de Water, Robin P and Pang, Chao and Rockenschaub, Patrick and Renc, Pawel and Oh, Jungwoo and Stankevi{\v{c}}i{\=u}t{\.e}, Kamil{\.e} and Xu, Justin and others},
289
+ journal={NEJM AI},
290
+ volume={3},
291
+ number={6},
292
+ pages={AIra2501253},
293
+ year={2026},
294
+ publisher={Massachusetts Medical Society}
295
+ }
296
+ ```
@@ -0,0 +1,280 @@
1
+ # ALOTT-MEDS
2
+
3
+ <!-- Placeholders until available: the PyPI badges resolve once ALOTT-MEDS is published; replace
4
+ CODECOV_TOKEN_PLACEHOLDER with the repo's Codecov badge token and zenodo.XXXXXXX with its Zenodo DOI. -->
5
+
6
+ [![PyPI - Version](https://img.shields.io/pypi/v/ALOTT-MEDS)](https://pypi.org/project/ALOTT-MEDS/)
7
+ [![codecov](https://codecov.io/gh/rvandewater/ALOTT_MEDS/graph/badge.svg?token=CODECOV_TOKEN_PLACEHOLDER)](https://codecov.io/gh/rvandewater/ALOTT_MEDS)
8
+ [![tests](https://github.com/rvandewater/ALOTT_MEDS/actions/workflows/synthetic-etl.yaml/badge.svg)](https://github.com/rvandewater/ALOTT_MEDS/actions/workflows/synthetic-etl.yaml)
9
+ [![code-quality](https://github.com/rvandewater/ALOTT_MEDS/actions/workflows/code-quality-main.yaml/badge.svg)](https://github.com/rvandewater/ALOTT_MEDS/actions/workflows/code-quality-main.yaml)
10
+ ![python](https://img.shields.io/badge/python-%3E%3D3.11.4%2C%20%3C3.14-blue?logo=python&logoColor=white)
11
+ ![Static Badge](https://img.shields.io/badge/MEDS-0.4.1-blue)
12
+ [![license](https://img.shields.io/badge/License-MIT-green.svg?labelColor=gray)](https://github.com/rvandewater/ALOTT_MEDS#license)
13
+ [![PRs](https://img.shields.io/badge/PRs-welcome-brightgreen.svg)](https://github.com/rvandewater/ALOTT_MEDS/pulls)
14
+ [![contributors](https://img.shields.io/github/contributors/rvandewater/ALOTT_MEDS.svg)](https://github.com/rvandewater/ALOTT_MEDS/graphs/contributors)
15
+ [![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.XXXXXXX.svg)](https://doi.org/10.5281/zenodo.XXXXXXX)
16
+
17
+ ALOTT (ALarms, Outcomes Telemetry with Timing) is a pilot bedside-EMR database from two
18
+ hospitals, collected to support algorithms that predict emergency response team events. It
19
+ pairs bedside telemetry from 270 beds (alarms, monitor measurements and waveforms) with
20
+ electronic medical record data: admissions, diagnoses, ICU stays and bed moves, labs,
21
+ orders, medications and nursing flowsheets.
22
+ This pipeline extracts the EMR tables of ALOTT (from PhysioNet,
23
+ https://physionet.org/content/alott/) into the
24
+ [MEDS](https://github.com/Medical-Event-Data-Standard/meds) format. It is built the same way
25
+ as [MIMIC_IV_MEDS](https://github.com/Medical-Event-Data-Standard/MIMIC_IV_MEDS): one
26
+ declarative [MEDS-Extract](https://github.com/mmcdermott/MEDS_extract) spec, no Python.
27
+
28
+ ## Usage:
29
+
30
+ ```bash
31
+ pip install git+https://github.com/rvandewater/ALOTT_MEDS
32
+ export DATASET_DOWNLOAD_USERNAME=$PHYSIONET_USERNAME
33
+ export DATASET_DOWNLOAD_PASSWORD=$PHYSIONET_PASSWORD
34
+
35
+ meds-extract-run spec=ALOTT output_dir=$MEDS_OUTPUT_DIR
36
+ ```
37
+
38
+ That downloads only the `emr-datasets/` folder of the ALOTT 1.0.0 release (the telemetry is
39
+ skipped) into `$MEDS_OUTPUT_DIR/.meds_extract_run/raw_input` and runs the full MEDS-Extract
40
+ pipeline, writing `data/` and `metadata/` to `$MEDS_OUTPUT_DIR`. Files that already exist and
41
+ verify against PhysioNet's checksums are skipped, so an interrupted download resumes. Pass
42
+ `download_dest_dir=$RAW_INPUT_DIR` to keep the raw data somewhere durable.
43
+
44
+ ALOTT is a credentialed release: the PhysioNet account must have signed the ALOTT data use
45
+ agreement.
46
+
47
+ To reuse a download you already have, point `input_dir` at the directory that *contains*
48
+ `emr-datasets/`:
49
+
50
+ ```bash
51
+ meds-extract-run spec=ALOTT output_dir=$MEDS_OUTPUT_DIR do_download=false input_dir=$RAW_INPUT_DIR
52
+ ```
53
+
54
+ > [!NOTE]
55
+ > The folder the EMR CSVs sit in on PhysioNet, assumed to be `emr-datasets/`, has not been
56
+ > checked against a download. If your copy is a flat folder of the ten CSVs
57
+ > (`admission.csv`, `labs.csv`, ...), point `input_dir` at a directory holding an
58
+ > `emr-datasets` link to it.
59
+
60
+ Run `meds-extract-run --help` for the full set of arguments.
61
+
62
+ ## Configuration
63
+
64
+ The whole pipeline is one file,
65
+ [`src/ALOTT_MEDS/event_configs.yaml`](https://github.com/rvandewater/ALOTT_MEDS/blob/main/src/ALOTT_MEDS/event_configs.yaml), written in MESSY
66
+ (MEDS-Extract Specification Syntax YAML) and registered under the `MEDS_extract.pipelines`
67
+ entry-point group as `ALOTT`. That registration is what makes `spec=ALOTT` work from anywhere
68
+ and what stamps the output's `metadata/dataset.json` with
69
+ `<ALOTT release>:<this package's version>`. The file holds:
70
+
71
+ - a `sources:` block pointing at the PhysioNet release, restricted to `emr-datasets/*`;
72
+ - an `etl:` block with the run's few knobs;
73
+ - one block per raw EMR table describing the events extracted from it.
74
+
75
+ ### Timestamps
76
+
77
+ ALOTT shifts each patient's dates by a random -180..180 days but keeps the time of day. Most
78
+ tables store each timestamp as one shifted `%Y-%m-%d %H:%M:%S` column. Labs and orders carry
79
+ only a shifted **date**, and medication start and end times are always `00:00:00`, i.e. dates
80
+ too. Those events are placed at the **end of that day (23:59:59)**: midnight would make a
81
+ result or a new medication visible before it existed, which leaks the future into any
82
+ prediction made later that day. Same-day lab orders and results therefore share a time.
83
+ Medication administrations keep their recorded time of day. A row with a null timestamp
84
+ produces no event.
85
+
86
+ Absolute dates are shifted per patient, so only differences *within* a subject are
87
+ meaningful.
88
+
89
+ ### IDs
90
+
91
+ `aMRN` is the MEDS `subject_id`; the encounter key `aCSN` is carried as `hadm_id`. Lab and
92
+ order results use the resulting encounter, and orders the ordering one. The release
93
+ writes the string `nan` for an unresulted encounter; it is cast leniently, so it becomes a
94
+ null and `hadm_id` stays `Int64`.
95
+
96
+ ### Codes
97
+
98
+ Every code is `FAMILY//...`, from general to specific, so a prefix selects a group (all
99
+ labs, one lab panel, one drug). Where a table records both an order and its outcome, the
100
+ second component says which: `ORDERED` or `RESULT`, `START` or `STOP`, or `MAR`
101
+ (Medication Administration Record: the bedside log of each dose given, held or refused).
102
+ `UNK` stands for a null component that must not drop the row (see
103
+ [Null handling](#null-handling)).
104
+
105
+ | Code | Table | Time | Value and extra columns |
106
+ | ------------------------------------------------ | ------------ | --------------------- | -------------------------------------------------------------------------------------- |
107
+ | `HOSPITAL_ADMISSION//<admission type>//<source>` | admission | admission | reason for visit in `text_value` |
108
+ | `HOSPITAL_DISCHARGE//<destination>` | admission | discharge | |
109
+ | `DIAGNOSIS//ICD//10//<ICD-10-CM code>` | admission | discharge | diagnosis text; `diagnosis_rank` `PRIMARY` or `SECONDARY` |
110
+ | `<FIELD>//<value>` | demographics | static | `FIELD` is `GENDER`, `MARITAL_STATUS`, `RACE`, `ETHNICITY`, `LANGUAGE` or `BLOOD_TYPE` |
111
+ | `ALLERGY//<allergen>` | allergy | static | reaction in `text_value` |
112
+ | `FLOWSHEET//<FLO_MEAS_ID>` | flowsheet | recorded | reading or score; `flow_type` |
113
+ | `ICU_ADMISSION`, `ICU_DISCHARGE` | icu | start, end | |
114
+ | `TRANSFER` | location | start | |
115
+ | `MEDICATION//START//<drug>` | medications | start date, 23:59:59 | ordered dose (raw text if a range); `dose_unit`, `frequency` |
116
+ | `MEDICATION//STOP//<drug>` | medications | end date, 23:59:59 | |
117
+ | `MEDICATION//MAR//<drug>//<action>` | medications | administration | given dose; `dose_unit`, `route` |
118
+ | `LAB//ORDERED//<PROC_ID>` | labs | order date, 23:59:59 | |
119
+ | `LAB//RESULT//<PROC_ID>//<component>` | labs | result date, 23:59:59 | result; `order_status` |
120
+ | `ORDER//ORDERED//<order type>//<PROC_ID>` | orders | order date, 23:59:59 | `order_status` |
121
+ | `ORDER//RESULT//<order type>//<PROC_ID>` | orders | result date, 23:59:59 | result; `order_status` |
122
+
123
+ `codes.parquet` carries a description for flowsheet, lab and order codes, taken from the
124
+ release's display names.
125
+
126
+ ### Null handling
127
+
128
+ A null component of a composite code drops the row under MEDS-Extract 0.7. Components are
129
+ therefore coalesced with `?? 'UNK'` wherever a null must not drop the row (admission type and
130
+ source, discharge destination, lab component, order type, medication administration action).
131
+
132
+ The rest are deliberately left uncoalesced. In the release, the code columns of
133
+ flowsheets, labs, orders and allergies are never null, and `MEDICATION` is null in 10 of
134
+ 12,698,867 rows, so a coalesce would add nothing. A missing ICD-10 code or demographic
135
+ produces no event rather than an `UNK` category, because a missing value is not an observed
136
+ one.
137
+
138
+ ### Diagnoses
139
+
140
+ The primary and secondary ICD-10 codes on each visit become `DIAGNOSIS//ICD//10//<code>`
141
+ events at discharge time, with `diagnosis_rank` set to `PRIMARY` or `SECONDARY`. About 4% of
142
+ cells list up to four codes separated by a comma and a space; each code becomes its own event, carrying the
143
+ cell's full diagnosis text. No
144
+ `parent_codes` metadata is attached: the same code occurs as both primary and secondary, and
145
+ MEDS-Extract rejects one code produced by two code templates.
146
+ `DIAGNOSIS//ICD//10//<code>` maps to `ICD10CM/<code>` directly.
147
+
148
+ ### Other tables
149
+
150
+ - **ICU stays and bed moves** carry only their start and end times in the release (no unit,
151
+ room or bed), so they become bare `ICU_ADMISSION`, `ICU_DISCHARGE` and `TRANSFER`
152
+ events.
153
+ - **Demographics and allergies** carry no time in the release and are static events.
154
+ - **Flowsheet, lab and order-result values** are free text. The numeric reading is kept where
155
+ it parses; scored flowsheet items charted as `<score>--><label>` (e.g. `-1-->Drowsy`) keep
156
+ their score. The raw string goes to `text_value` unless it is just the number. Epic's
157
+ nonnumeric sentinel `9999999` never becomes a numeric value.
158
+ - **Medications:** each row of the release is one administration and repeats its order's
159
+ start and end, so an order yields one `START`, one `STOP` and one `MAR` event per
160
+ administration.
161
+ - **Orders** hold every order type except labs; roughly half carry a single result.
162
+ - **Birth and death:** the release has no birth date, age or death time, so there are no
163
+ `MEDS_BIRTH` or `MEDS_DEATH` events.
164
+ - **Not extracted:** the telemetry (alarms, 2-second measurements, waveforms; per-admission
165
+ files keyed by file name, far larger than the EMR), `social_hx` (no event times and
166
+ unlabeled Epic category codes), and the medication order date.
167
+
168
+ ### Validation against the release
169
+
170
+ The spec was first written from the authors' release code
171
+ ([Carescape](https://github.com/johnclawrence/Carescape)) and did not run on the published
172
+ files. Profiling the 1.0.0 EMR CSVs (value *shapes* and null counts only, never values)
173
+ showed what had to change:
174
+
175
+ - Tables are lowercase (`admission`, `labs`, ...) rather than `Patient`, `Labs`, ...; the
176
+ keys are `aMRN` and `aCSN`, not `csID` and `CS_PAT_ENC_CSN_ID`.
177
+ - Each timestamp is one already-shifted column rather than a date plus an `HH:MM:SS`
178
+ column. Labs and orders keep only the date (see [Timestamps](#timestamps)).
179
+ - `icu` and `location` have no department, event type, room or bed columns.
180
+ - `resulting_aCSN` holds the string `nan` for unresulted rows.
181
+ - 1,536 ICD-10 codes occur as both primary and secondary diagnoses, and some cells list
182
+ several codes (see [Diagnoses](#diagnoses)). The synthetic test covers both.
183
+ - Medication start and end times are always midnight, and `orders` carries results of its
184
+ own (it holds no lab orders).
185
+
186
+ All ten CSVs are valid UTF-8. On the full 1.0.0 EMR tables (6.3 GB of CSVs read),
187
+ `meds-extract-run` with MEDS-Extract 0.7.1 finishes in about 90 seconds on a laptop with about
188
+ 6.5 GB peak memory, producing 52.6M events for 18,472 subjects. Fewer lab, order and
189
+ medication start/stop events come out than raw rows because identical events are merged: lab
190
+ and order times have day resolution, and the medication table repeats an order's start and
191
+ end on every administration row.
192
+
193
+ ## MEDS-transforms settings
194
+
195
+ If you want to convert a large dataset, you can parallelize the MEDS-transforms stages
196
+ (the MEDS-transformation step that takes the longest).
197
+
198
+ Using local parallelization with the `hydra-joblib-launcher` package:
199
+
200
+ ```bash
201
+ pip install hydra-joblib-launcher --upgrade
202
+ ```
203
+
204
+ Then write a stage-runner file and pass it to the run:
205
+
206
+ ```yaml
207
+ # runner.yaml
208
+ parallelize:
209
+ n_workers: 8
210
+ launcher: joblib
211
+ ```
212
+
213
+ ```bash
214
+ meds-extract-run spec=ALOTT output_dir=$MEDS_OUTPUT_DIR stage_runner_fp=runner.yaml
215
+ ```
216
+
217
+ The number of subjects per shard is `n_subjects_per_shard` in the spec's `etl:` block
218
+ (1000 by default); raise it to cut the parallelization overhead.
219
+
220
+ ## Development
221
+
222
+ ```bash
223
+ uv sync --group dev
224
+ uv run pytest tests/
225
+ uv run pre-commit run --all-files
226
+ ```
227
+
228
+ A release bumps the version in `pyproject.toml` with `uv version --bump`, commits it and pushes
229
+ a tag of that exact version, which publishes to PyPI and creates the GitHub release. See [`RELEASING.md`](https://github.com/rvandewater/ALOTT_MEDS/blob/main/RELEASING.md) for choosing the
230
+ version, the one-time PyPI setup and recovering from a failed release.
231
+
232
+ ALOTT has no open demo, so CI runs the spec end to end over a small **synthetic** input
233
+ generated in [`tests/e2e_synthetic_test.py`](https://github.com/rvandewater/ALOTT_MEDS/blob/main/tests/e2e_synthetic_test.py). That checks the
234
+ spec's mechanics, not its agreement with the real release. Never commit raw ALOTT data or
235
+ MEDS output derived from it.
236
+
237
+ ## Citation
238
+
239
+ Please cite the following original dataset when using ALOTT:
240
+
241
+ ```bibtex
242
+ @article{PhysioNet-alott-1.0.0,
243
+ author = {Lawrence, John and Rayo, Mike and Huerta, Timothy},
244
+ title = {{ALarms, Outcomes Telemetry with Timing (ALOTT): a Bedside-EMR Database}},
245
+ journal = {{PhysioNet}},
246
+ year = {2025},
247
+ month = mar,
248
+ note = {Version 1.0.0},
249
+ doi = {10.13026/sbq5-dy17},
250
+ url = {https://doi.org/10.13026/sbq5-dy17}
251
+ }
252
+ ```
253
+
254
+ Also cite the ETL framework used to extract the dataset:
255
+
256
+ ```bibtex
257
+ @software{van_de_Water_ALOTT_MEDS_ETL_2025,
258
+ author = {van de Water, Robin Philippus},
259
+ license = {MIT},
260
+ month = apr,
261
+ title = {{ALOTT\_MEDS ETL}},
262
+ url = {https://github.com/rvandewater/ALOTT_MEDS},
263
+ year = {2025}
264
+ }
265
+ ```
266
+
267
+ For citing MEDS in general:
268
+
269
+ ```bibtex
270
+ @article{mcdermott2026meds,
271
+ title={MEDS—An Emerging Data Standard and Ecosystem for Health AI Research},
272
+ author={McDermott, Matthew BA and Steinberg, Ethan and Fries, Jason A and van de Water, Robin P and Pang, Chao and Rockenschaub, Patrick and Renc, Pawel and Oh, Jungwoo and Stankevi{\v{c}}i{\=u}t{\.e}, Kamil{\.e} and Xu, Justin and others},
273
+ journal={NEJM AI},
274
+ volume={3},
275
+ number={6},
276
+ pages={AIra2501253},
277
+ year={2026},
278
+ publisher={Massachusetts Medical Society}
279
+ }
280
+ ```
@@ -0,0 +1,74 @@
1
+ [build-system]
2
+ requires = ["setuptools>=64", "wheel"]
3
+ build-backend = "setuptools.build_meta"
4
+
5
+ [project]
6
+ name = "ALOTT-MEDS"
7
+ # The last released version. A release bumps it with `uv version --bump` and tags the commit with
8
+ # the result (see RELEASING.md); CI refuses a tag that differs from it. 0.0.0 = nothing released yet.
9
+ version = "0.0.1"
10
+ authors = [
11
+ { name="Robin van de Water", email="rpv2114@cumc.columbia.edu" },
12
+ ]
13
+ description = "An ETL pipeline to extract the ALOTT (Carescape) EMR data into the MEDS format."
14
+ readme = "README.md"
15
+ # Mirrors MEDS-Extract's own floor and ceiling; a looser range admits interpreters it
16
+ # cannot be installed on, and uv refuses to resolve at all.
17
+ requires-python = ">=3.11.4,<3.14"
18
+ classifiers = [
19
+ "Programming Language :: Python :: 3",
20
+ "License :: OSI Approved :: MIT License",
21
+ "Operating System :: OS Independent",
22
+ ]
23
+ # This package ships no Python: the whole ETL is `event_configs.yaml`, run by
24
+ # MEDS-Extract's own `meds-extract-run`. So there is exactly one dependency — the
25
+ # `[download]` extra adds `meds-extract-download`, which the runner spawns to stage the
26
+ # raw data from PhysioNet. MEDS-transforms, polars and hydra all arrive through it at the
27
+ # versions it fences; duplicating those pins here would only create drift.
28
+ dependencies = [
29
+ "meds-extract[download]>=0.7.1,<0.8",
30
+ ]
31
+
32
+ [dependency-groups]
33
+ # `shfmt-py` bundles the `shfmt` binary that the `mdformat-shfmt` pre-commit hook shells
34
+ # out to, keeping local + CI behavior in lock step.
35
+ dev = [
36
+ "pre-commit~=3.8",
37
+ "ruff>=0.11,<1.0",
38
+ "pytest~=9.0",
39
+ "shfmt-py==3.12.0.2",
40
+ ]
41
+
42
+ [project.entry-points."MEDS_extract.pipelines"]
43
+ ALOTT = "ALOTT_MEDS:event_configs.yaml"
44
+
45
+ [project.urls]
46
+ Homepage = "https://github.com/rvandewater/ALOTT_MEDS"
47
+ Issues = "https://github.com/rvandewater/ALOTT_MEDS/issues"
48
+
49
+ [tool.setuptools.package-data]
50
+ # The MESSY spec is the whole ETL. It is not a Python file, so it has to be listed, or installed
51
+ # wheels lack it.
52
+ ALOTT_MEDS = ["event_configs.yaml"]
53
+
54
+ [tool.ruff]
55
+ target-version = "py311"
56
+ line-length = 110
57
+
58
+ [tool.ruff.lint]
59
+ select = [
60
+ "E", # pycodestyle
61
+ "F", # pyflakes
62
+ "I", # isort
63
+ "UP", # pyupgrade
64
+ "B", # flake8-bugbear
65
+ "C4", # flake8-comprehensions
66
+ "RUF", # Ruff-specific rules
67
+ "SIM", # flake8-simplify
68
+ ]
69
+ ignore = [
70
+ "SIM108", # ternary operator suggestions
71
+ ]
72
+
73
+ [tool.ruff.lint.per-file-ignores]
74
+ "__init__.py" = ["F401"]
@@ -0,0 +1,4 @@
1
+ [egg_info]
2
+ tag_build =
3
+ tag_date = 0
4
+
@@ -0,0 +1,12 @@
1
+ """The ALOTT MEDS ETL — a configuration package, not a library.
2
+
3
+ This distribution ships no code. Its entire content is ``ALOTT_MEDS/event_configs.yaml``, a MESSY
4
+ (MEDS-Extract Specification Syntax YAML) document declaring where the raw ALOTT EMR tables come from
5
+ and how every event is extracted from them. ``pyproject.toml`` registers that file with MEDS-Extract's
6
+ ``MEDS_extract.pipelines`` entry-point group under the name ``ALOTT``, so the ETL runs as::
7
+
8
+ meds-extract-run spec=ALOTT output_dir=$MEDS_OUTPUT_DIR
9
+
10
+ This package exists so the YAML is addressable as package data — ``importlib.resources.files("ALOTT_MEDS")``
11
+ — from an installed wheel. Nothing here is importable API; deliberately, there is nothing to import.
12
+ """
@@ -0,0 +1,243 @@
1
+ # The complete ALOTT MEDS ETL. This one file *is* the package: `sources:` says where the raw
2
+ # data comes from, `etl:` supplies the handful of run knobs, and every other top-level key is
3
+ # a raw table whose events are extracted from it. It is registered with `meds-extract-run`
4
+ # under the name `ALOTT` (see this repo's `pyproject.toml`), so the whole ETL is one command:
5
+ #
6
+ # meds-extract-run spec=ALOTT output_dir=$MEDS_OUTPUT_DIR
7
+ #
8
+ # Only the EMR tables (`emr-datasets/`) are extracted. The telemetry folders (alarms,
9
+ # 2-second measurements, WFDB waveforms) are per-admission files whose subject and encounter
10
+ # IDs live in the *file name*, not in a column, and they are several orders of magnitude
11
+ # larger; they are out of scope for this spec.
12
+ #
13
+ # SCHEMA PROVENANCE. The table and column names below were checked against the ten EMR CSVs
14
+ # of the 1.0.0 release (`admission.csv`, `labs.csv`, ...). They differ from the dataset
15
+ # authors' release code (github.com/johnclawrence/Carescape): the published files use
16
+ # `aMRN`/`aCSN` as keys and carry each timestamp in a single already-shifted column. The
17
+ # name of the folder the EMR CSVs sit in on PhysioNet (`emr-datasets/` here) has not been
18
+ # checked against a download.
19
+ #
20
+ # Date shifting: every patient's dates are shifted by a random -180..180 days, time of day is
21
+ # not. Most timestamps are `%Y-%m-%d %H:%M:%S`; labs, orders and medication start/end carry
22
+ # only a shifted *date*, so their events are placed at the end of that day (23:59:59).
23
+ # Midnight would make a result visible before it existed (future leakage for any prediction
24
+ # made later that day).
25
+
26
+ etl:
27
+ dataset_name: ALOTT
28
+ n_subjects_per_shard: 10000
29
+
30
+ # `aMRN` is the de-identified patient key; `aCSN` is the de-identified encounter key, carried
31
+ # as `hadm_id` wherever a table has one.
32
+ _defaults:
33
+ subject_id: $aMRN
34
+
35
+ emr-datasets/admission:
36
+ # One row per hospital visit.
37
+ admission:
38
+ code: f"HOSPITAL_ADMISSION//{$Admission_Type ?? 'UNK'}//{$Admit_Source ?? 'UNK'}"
39
+ time: '$admsnOffset::"%Y-%m-%d %H:%M:%S"'
40
+ hadm_id: $aCSN
41
+ text_value: $REASON_VISIT_NAME
42
+ discharge:
43
+ code: f"HOSPITAL_DISCHARGE//{$Discharge_Destination ?? 'UNK'}"
44
+ time: '$dischOffset::"%Y-%m-%d %H:%M:%S"'
45
+ hadm_id: $aCSN
46
+ # Diagnoses are only known at the end of the visit, so they are timed at discharge (as in
47
+ # MIMIC-IV MEDS). The free-text name rides in `text_value` rather than `description`: it is
48
+ # per-visit text, not guaranteed to be a function of the code. No `parent_codes` metadata:
49
+ # the same ICD code occurs in both columns, and MEDS-Extract refuses a code produced by two
50
+ # code templates. The codes are already dotted ICD-10-CM, so `DIAGNOSIS//ICD//10//<code>`
51
+ # maps to `ICD10CM/<code>` directly.
52
+ #
53
+ # About 4% of cells hold a `, `-separated list of up to four codes. MESSY cannot explode a
54
+ # list into rows, so each position is its own block: `_k` extracts the k-th code and yields
55
+ # no event when the list is shorter. The full diagnosis text rides on every one. The blocks
56
+ # run from the 4th position down because MESSY-synth builds its synthetic value from the
57
+ # first pattern it meets, and only a 4-code list satisfies all four.
58
+ primary_diagnosis_4: &primary_diagnosis
59
+ code: f"DIAGNOSIS//ICD//10//{extract group 1 of /^[^,]+, [^,]+, [^,]+, ([^,]+)/ from $Primary_ICD10}"
60
+ time: '$dischOffset::"%Y-%m-%d %H:%M:%S"'
61
+ hadm_id: $aCSN
62
+ text_value: $Primary_Diagnosis
63
+ diagnosis_rank: "'PRIMARY'"
64
+ primary_diagnosis_3:
65
+ <<: *primary_diagnosis
66
+ code: f"DIAGNOSIS//ICD//10//{extract group 1 of /^[^,]+, [^,]+, ([^,]+)/ from $Primary_ICD10}"
67
+ primary_diagnosis_2:
68
+ <<: *primary_diagnosis
69
+ code: f"DIAGNOSIS//ICD//10//{extract group 1 of /^[^,]+, ([^,]+)/ from $Primary_ICD10}"
70
+ primary_diagnosis:
71
+ <<: *primary_diagnosis
72
+ code: f"DIAGNOSIS//ICD//10//{extract group 1 of /^([^,]+)/ from $Primary_ICD10}"
73
+ secondary_diagnosis_4: &secondary_diagnosis
74
+ code: f"DIAGNOSIS//ICD//10//{extract group 1 of /^[^,]+, [^,]+, [^,]+, ([^,]+)/ from $Secondary_ICD10}"
75
+ time: '$dischOffset::"%Y-%m-%d %H:%M:%S"'
76
+ hadm_id: $aCSN
77
+ text_value: $Secondary_Diagnosis
78
+ diagnosis_rank: "'SECONDARY'"
79
+ secondary_diagnosis_3:
80
+ <<: *secondary_diagnosis
81
+ code: f"DIAGNOSIS//ICD//10//{extract group 1 of /^[^,]+, [^,]+, ([^,]+)/ from $Secondary_ICD10}"
82
+ secondary_diagnosis_2:
83
+ <<: *secondary_diagnosis
84
+ code: f"DIAGNOSIS//ICD//10//{extract group 1 of /^[^,]+, ([^,]+)/ from $Secondary_ICD10}"
85
+ secondary_diagnosis:
86
+ <<: *secondary_diagnosis
87
+ code: f"DIAGNOSIS//ICD//10//{extract group 1 of /^([^,]+)/ from $Secondary_ICD10}"
88
+
89
+ emr-datasets/demographics:
90
+ # Subject-level properties, so static (null time). The table is keyed per encounter, so a
91
+ # subject contributes one row per visit; identical static events are deduplicated
92
+ # downstream. Nulls are deliberately not coalesced: a missing demographic produces no
93
+ # event rather than a `RACE//UNK` code that would read as an observed category.
94
+ gender:
95
+ code: f"GENDER//{$Gender}"
96
+ time: null
97
+ marital_status:
98
+ code: f"MARITAL_STATUS//{$Marital_Status}"
99
+ time: null
100
+ race:
101
+ code: f"RACE//{$Patient_Race}"
102
+ time: null
103
+ ethnicity:
104
+ code: f"ETHNICITY//{$Ethnicity}"
105
+ time: null
106
+ language:
107
+ code: f"LANGUAGE//{$Preferred_Language}"
108
+ time: null
109
+ blood_type:
110
+ code: f"BLOOD_TYPE//{$ABO_RH_Blood_Type}"
111
+ time: null
112
+
113
+ emr-datasets/allergy:
114
+ # The release carries no time for allergies, so they are static.
115
+ allergy:
116
+ code: f"ALLERGY//{$Allergy}"
117
+ time: null
118
+ text_value: $Allergy_Reaction
119
+
120
+ emr-datasets/flowsheet:
121
+ # Provider-charted vitals, risk scores and rapid-response events. `MEAS_VALUE` is free
122
+ # text; `::?float64` keeps the numeric reading where there is one and null otherwise.
123
+ # Scored scale items are charted as `<score>--><label>` (e.g. `-1-->Drowsy`), so their
124
+ # leading score is the numeric reading. The raw string is kept in `text_value` unless it is
125
+ # just the number (MEDS-Extract drops a `text_value` equal to `numeric_value`).
126
+ flowsheet:
127
+ code: f"FLOWSHEET//{$FLO_MEAS_ID}"
128
+ time: '$recordedOffsetDate::"%Y-%m-%d %H:%M:%S"'
129
+ hadm_id: $aCSN
130
+ numeric_value: coalesce($MEAS_VALUE::?float64, (extract group 1 of /^(-?[0-9]+)-->/ from $MEAS_VALUE::str)::?float64)
131
+ text_value: $MEAS_VALUE
132
+ flow_type: $Flow_Type
133
+ _metadata:
134
+ _self:
135
+ description: coalesce($disp_name, $flo_meas_name)
136
+
137
+ emr-datasets/icu:
138
+ # The release keeps only the ICU stay's times, not which unit it was.
139
+ icu_admission:
140
+ code: "'ICU_ADMISSION'"
141
+ time: '$startOffsetDate::"%Y-%m-%d %H:%M:%S"'
142
+ hadm_id: $aCSN
143
+ icu_discharge:
144
+ code: "'ICU_DISCHARGE'"
145
+ time: '$endOffsetDate::"%Y-%m-%d %H:%M:%S"'
146
+ hadm_id: $aCSN
147
+
148
+ emr-datasets/location:
149
+ # Bed-level movement. The release keeps only each stay's start and end, not the department,
150
+ # room or bed, so a transfer is just the moment the patient moved.
151
+ transfer:
152
+ code: "'TRANSFER'"
153
+ time: '$startOffsetDate::"%Y-%m-%d %H:%M:%S"'
154
+ hadm_id: $aCSN
155
+
156
+ emr-datasets/medications:
157
+ # One row per administration, each repeating its order's start and end. Those two are
158
+ # datetimes in form but always 00:00:00 in the release, i.e. dates, so they are placed at
159
+ # the end of the day like labs and orders. The ordered dose is free text (`1-2` for a
160
+ # range); its numeric reading is kept where there is one and the raw string otherwise.
161
+ medication_start:
162
+ code: f"MEDICATION//START//{$MEDICATION}"
163
+ time: '$startOffsetDate::"%Y-%m-%d %H:%M:%S" @ 23:59:59'
164
+ hadm_id: $aCSN
165
+ numeric_value: $HV_DISCRETE_DOSE::?float64
166
+ text_value: $HV_DISCRETE_DOSE
167
+ dose_unit: $HV_DISCRET_DOSE_UNIT
168
+ frequency: $FREQUENCY
169
+ medication_stop:
170
+ code: f"MEDICATION//STOP//{$MEDICATION}"
171
+ time: '$endOffsetDate::"%Y-%m-%d %H:%M:%S" @ 23:59:59'
172
+ hadm_id: $aCSN
173
+ # The Medication Administration Record (MAR), the bedside log of each dose, is the only
174
+ # medication time with a time of day. The recorded action is part of the code, since a held
175
+ # or refused dose is not a given one; a row without an action is kept as `UNK`.
176
+ medication_administration:
177
+ code: f"MEDICATION//MAR//{$MEDICATION}//{$MAR_ACTION ?? 'UNK'}"
178
+ time: '$OffsetMAR_TAKEN_TIME::"%Y-%m-%d %H:%M:%S"'
179
+ hadm_id: $aCSN
180
+ numeric_value: $MAR_DOSE::?float64
181
+ dose_unit: $MAR_DOSE_UNIT
182
+ route: $MAR_ROUTE
183
+
184
+ emr-datasets/labs:
185
+ # Ordering and resulting encounters can differ; the resulting one is the visit the value
186
+ # belongs to. Unresulted rows carry the string `nan` as their resulting encounter, which the
187
+ # lenient cast turns into a null (and keeps `hadm_id` an integer across tables). The end-of-day
188
+ # time is attached with `date @ time` rather than an f-string so MESSY-synth can type the column.
189
+ lab_ordered:
190
+ code: f"LAB//ORDERED//{$PROC_ID}"
191
+ time: '$orderingOffsetDate::"%Y-%m-%d" @ 23:59:59'
192
+ hadm_id: $ordering_aCSN
193
+ _metadata:
194
+ _self:
195
+ description: $DESCRIPTION
196
+ lab_result:
197
+ code: f"LAB//RESULT//{$PROC_ID}//{$COMPONENT_NAME ?? 'UNK'}"
198
+ time: '$resultsOffsetDate::"%Y-%m-%d" @ 23:59:59'
199
+ hadm_id: $resulting_aCSN::?int64
200
+ numeric_value: "$ORD_NUM_VALUE::?float64 if $ORD_NUM_VALUE::?float64 != 9999999"
201
+ text_value: $ORD_VALUE
202
+ order_status: $order_status
203
+ _metadata:
204
+ _self:
205
+ description: f"{$DESCRIPTION} - {$COMPONENT_NAME}"
206
+
207
+ emr-datasets/orders:
208
+ # Every order type except labs. About half the rows carry a single result, with the same
209
+ # `nan` resulting-encounter and 9999999 nonnumeric sentinels as labs.
210
+ order:
211
+ code: f"ORDER//ORDERED//{$Order_type ?? 'UNK'}//{$PROC_ID}"
212
+ time: '$orderingOffsetDate::"%Y-%m-%d" @ 23:59:59'
213
+ hadm_id: $ordering_aCSN
214
+ order_status: $order_status
215
+ _metadata:
216
+ _self:
217
+ description: $DESCRIPTION
218
+ order_result:
219
+ code: f"ORDER//RESULT//{$Order_type ?? 'UNK'}//{$PROC_ID}"
220
+ time: '$resultsOffsetDate::"%Y-%m-%d" @ 23:59:59'
221
+ hadm_id: $resulting_aCSN::?int64
222
+ numeric_value: "$ORD_NUM_VALUE::?float64 if $ORD_NUM_VALUE::?float64 != 9999999"
223
+ text_value: $ORD_VALUE
224
+ order_status: $order_status
225
+ _metadata:
226
+ _self:
227
+ description: $DESCRIPTION
228
+
229
+ # `emr-datasets/social_hx` (tobacco / alcohol / drug use) is not extracted: it has no event
230
+ # time other than quit dates, and its `*_C` columns are Epic category codes whose labels are
231
+ # not in the release.
232
+
233
+ sources:
234
+ # Stamped by `meds-extract-run` into the output's `etl_metadata.dataset_version` as
235
+ # `<release>:<this package's version>`.
236
+ dataset_version: "1.0.0"
237
+ dataset:
238
+ - type: physionet
239
+ base_url: https://physionet.org/files/alott/${sources.dataset_version}
240
+ username: ${oc.env:DATASET_DOWNLOAD_USERNAME}
241
+ password: ${oc.env:DATASET_DOWNLOAD_PASSWORD}
242
+ # Only the EMR tables; the telemetry is out of scope (see the header).
243
+ include: ["emr-datasets/*"]
@@ -0,0 +1,296 @@
1
+ Metadata-Version: 2.4
2
+ Name: ALOTT-MEDS
3
+ Version: 0.0.1
4
+ Summary: An ETL pipeline to extract the ALOTT (Carescape) EMR data into the MEDS format.
5
+ Author-email: Robin van de Water <rpv2114@cumc.columbia.edu>
6
+ Project-URL: Homepage, https://github.com/rvandewater/ALOTT_MEDS
7
+ Project-URL: Issues, https://github.com/rvandewater/ALOTT_MEDS/issues
8
+ Classifier: Programming Language :: Python :: 3
9
+ Classifier: License :: OSI Approved :: MIT License
10
+ Classifier: Operating System :: OS Independent
11
+ Requires-Python: <3.14,>=3.11.4
12
+ Description-Content-Type: text/markdown
13
+ License-File: LICENSE
14
+ Requires-Dist: meds-extract[download]<0.8,>=0.7.1
15
+ Dynamic: license-file
16
+
17
+ # ALOTT-MEDS
18
+
19
+ <!-- Placeholders until available: the PyPI badges resolve once ALOTT-MEDS is published; replace
20
+ CODECOV_TOKEN_PLACEHOLDER with the repo's Codecov badge token and zenodo.XXXXXXX with its Zenodo DOI. -->
21
+
22
+ [![PyPI - Version](https://img.shields.io/pypi/v/ALOTT-MEDS)](https://pypi.org/project/ALOTT-MEDS/)
23
+ [![codecov](https://codecov.io/gh/rvandewater/ALOTT_MEDS/graph/badge.svg?token=CODECOV_TOKEN_PLACEHOLDER)](https://codecov.io/gh/rvandewater/ALOTT_MEDS)
24
+ [![tests](https://github.com/rvandewater/ALOTT_MEDS/actions/workflows/synthetic-etl.yaml/badge.svg)](https://github.com/rvandewater/ALOTT_MEDS/actions/workflows/synthetic-etl.yaml)
25
+ [![code-quality](https://github.com/rvandewater/ALOTT_MEDS/actions/workflows/code-quality-main.yaml/badge.svg)](https://github.com/rvandewater/ALOTT_MEDS/actions/workflows/code-quality-main.yaml)
26
+ ![python](https://img.shields.io/badge/python-%3E%3D3.11.4%2C%20%3C3.14-blue?logo=python&logoColor=white)
27
+ ![Static Badge](https://img.shields.io/badge/MEDS-0.4.1-blue)
28
+ [![license](https://img.shields.io/badge/License-MIT-green.svg?labelColor=gray)](https://github.com/rvandewater/ALOTT_MEDS#license)
29
+ [![PRs](https://img.shields.io/badge/PRs-welcome-brightgreen.svg)](https://github.com/rvandewater/ALOTT_MEDS/pulls)
30
+ [![contributors](https://img.shields.io/github/contributors/rvandewater/ALOTT_MEDS.svg)](https://github.com/rvandewater/ALOTT_MEDS/graphs/contributors)
31
+ [![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.XXXXXXX.svg)](https://doi.org/10.5281/zenodo.XXXXXXX)
32
+
33
+ ALOTT (ALarms, Outcomes Telemetry with Timing) is a pilot bedside-EMR database from two
34
+ hospitals, collected to support algorithms that predict emergency response team events. It
35
+ pairs bedside telemetry from 270 beds (alarms, monitor measurements and waveforms) with
36
+ electronic medical record data: admissions, diagnoses, ICU stays and bed moves, labs,
37
+ orders, medications and nursing flowsheets.
38
+ This pipeline extracts the EMR tables of ALOTT (from PhysioNet,
39
+ https://physionet.org/content/alott/) into the
40
+ [MEDS](https://github.com/Medical-Event-Data-Standard/meds) format. It is built the same way
41
+ as [MIMIC_IV_MEDS](https://github.com/Medical-Event-Data-Standard/MIMIC_IV_MEDS): one
42
+ declarative [MEDS-Extract](https://github.com/mmcdermott/MEDS_extract) spec, no Python.
43
+
44
+ ## Usage:
45
+
46
+ ```bash
47
+ pip install git+https://github.com/rvandewater/ALOTT_MEDS
48
+ export DATASET_DOWNLOAD_USERNAME=$PHYSIONET_USERNAME
49
+ export DATASET_DOWNLOAD_PASSWORD=$PHYSIONET_PASSWORD
50
+
51
+ meds-extract-run spec=ALOTT output_dir=$MEDS_OUTPUT_DIR
52
+ ```
53
+
54
+ That downloads only the `emr-datasets/` folder of the ALOTT 1.0.0 release (the telemetry is
55
+ skipped) into `$MEDS_OUTPUT_DIR/.meds_extract_run/raw_input` and runs the full MEDS-Extract
56
+ pipeline, writing `data/` and `metadata/` to `$MEDS_OUTPUT_DIR`. Files that already exist and
57
+ verify against PhysioNet's checksums are skipped, so an interrupted download resumes. Pass
58
+ `download_dest_dir=$RAW_INPUT_DIR` to keep the raw data somewhere durable.
59
+
60
+ ALOTT is a credentialed release: the PhysioNet account must have signed the ALOTT data use
61
+ agreement.
62
+
63
+ To reuse a download you already have, point `input_dir` at the directory that *contains*
64
+ `emr-datasets/`:
65
+
66
+ ```bash
67
+ meds-extract-run spec=ALOTT output_dir=$MEDS_OUTPUT_DIR do_download=false input_dir=$RAW_INPUT_DIR
68
+ ```
69
+
70
+ > [!NOTE]
71
+ > The folder the EMR CSVs sit in on PhysioNet, assumed to be `emr-datasets/`, has not been
72
+ > checked against a download. If your copy is a flat folder of the ten CSVs
73
+ > (`admission.csv`, `labs.csv`, ...), point `input_dir` at a directory holding an
74
+ > `emr-datasets` link to it.
75
+
76
+ Run `meds-extract-run --help` for the full set of arguments.
77
+
78
+ ## Configuration
79
+
80
+ The whole pipeline is one file,
81
+ [`src/ALOTT_MEDS/event_configs.yaml`](https://github.com/rvandewater/ALOTT_MEDS/blob/main/src/ALOTT_MEDS/event_configs.yaml), written in MESSY
82
+ (MEDS-Extract Specification Syntax YAML) and registered under the `MEDS_extract.pipelines`
83
+ entry-point group as `ALOTT`. That registration is what makes `spec=ALOTT` work from anywhere
84
+ and what stamps the output's `metadata/dataset.json` with
85
+ `<ALOTT release>:<this package's version>`. The file holds:
86
+
87
+ - a `sources:` block pointing at the PhysioNet release, restricted to `emr-datasets/*`;
88
+ - an `etl:` block with the run's few knobs;
89
+ - one block per raw EMR table describing the events extracted from it.
90
+
91
+ ### Timestamps
92
+
93
+ ALOTT shifts each patient's dates by a random -180..180 days but keeps the time of day. Most
94
+ tables store each timestamp as one shifted `%Y-%m-%d %H:%M:%S` column. Labs and orders carry
95
+ only a shifted **date**, and medication start and end times are always `00:00:00`, i.e. dates
96
+ too. Those events are placed at the **end of that day (23:59:59)**: midnight would make a
97
+ result or a new medication visible before it existed, which leaks the future into any
98
+ prediction made later that day. Same-day lab orders and results therefore share a time.
99
+ Medication administrations keep their recorded time of day. A row with a null timestamp
100
+ produces no event.
101
+
102
+ Absolute dates are shifted per patient, so only differences *within* a subject are
103
+ meaningful.
104
+
105
+ ### IDs
106
+
107
+ `aMRN` is the MEDS `subject_id`; the encounter key `aCSN` is carried as `hadm_id`. Lab and
108
+ order results use the resulting encounter, and orders the ordering one. The release
109
+ writes the string `nan` for an unresulted encounter; it is cast leniently, so it becomes a
110
+ null and `hadm_id` stays `Int64`.
111
+
112
+ ### Codes
113
+
114
+ Every code is `FAMILY//...`, from general to specific, so a prefix selects a group (all
115
+ labs, one lab panel, one drug). Where a table records both an order and its outcome, the
116
+ second component says which: `ORDERED` or `RESULT`, `START` or `STOP`, or `MAR`
117
+ (Medication Administration Record: the bedside log of each dose given, held or refused).
118
+ `UNK` stands for a null component that must not drop the row (see
119
+ [Null handling](#null-handling)).
120
+
121
+ | Code | Table | Time | Value and extra columns |
122
+ | ------------------------------------------------ | ------------ | --------------------- | -------------------------------------------------------------------------------------- |
123
+ | `HOSPITAL_ADMISSION//<admission type>//<source>` | admission | admission | reason for visit in `text_value` |
124
+ | `HOSPITAL_DISCHARGE//<destination>` | admission | discharge | |
125
+ | `DIAGNOSIS//ICD//10//<ICD-10-CM code>` | admission | discharge | diagnosis text; `diagnosis_rank` `PRIMARY` or `SECONDARY` |
126
+ | `<FIELD>//<value>` | demographics | static | `FIELD` is `GENDER`, `MARITAL_STATUS`, `RACE`, `ETHNICITY`, `LANGUAGE` or `BLOOD_TYPE` |
127
+ | `ALLERGY//<allergen>` | allergy | static | reaction in `text_value` |
128
+ | `FLOWSHEET//<FLO_MEAS_ID>` | flowsheet | recorded | reading or score; `flow_type` |
129
+ | `ICU_ADMISSION`, `ICU_DISCHARGE` | icu | start, end | |
130
+ | `TRANSFER` | location | start | |
131
+ | `MEDICATION//START//<drug>` | medications | start date, 23:59:59 | ordered dose (raw text if a range); `dose_unit`, `frequency` |
132
+ | `MEDICATION//STOP//<drug>` | medications | end date, 23:59:59 | |
133
+ | `MEDICATION//MAR//<drug>//<action>` | medications | administration | given dose; `dose_unit`, `route` |
134
+ | `LAB//ORDERED//<PROC_ID>` | labs | order date, 23:59:59 | |
135
+ | `LAB//RESULT//<PROC_ID>//<component>` | labs | result date, 23:59:59 | result; `order_status` |
136
+ | `ORDER//ORDERED//<order type>//<PROC_ID>` | orders | order date, 23:59:59 | `order_status` |
137
+ | `ORDER//RESULT//<order type>//<PROC_ID>` | orders | result date, 23:59:59 | result; `order_status` |
138
+
139
+ `codes.parquet` carries a description for flowsheet, lab and order codes, taken from the
140
+ release's display names.
141
+
142
+ ### Null handling
143
+
144
+ A null component of a composite code drops the row under MEDS-Extract 0.7. Components are
145
+ therefore coalesced with `?? 'UNK'` wherever a null must not drop the row (admission type and
146
+ source, discharge destination, lab component, order type, medication administration action).
147
+
148
+ The rest are deliberately left uncoalesced. In the release, the code columns of
149
+ flowsheets, labs, orders and allergies are never null, and `MEDICATION` is null in 10 of
150
+ 12,698,867 rows, so a coalesce would add nothing. A missing ICD-10 code or demographic
151
+ produces no event rather than an `UNK` category, because a missing value is not an observed
152
+ one.
153
+
154
+ ### Diagnoses
155
+
156
+ The primary and secondary ICD-10 codes on each visit become `DIAGNOSIS//ICD//10//<code>`
157
+ events at discharge time, with `diagnosis_rank` set to `PRIMARY` or `SECONDARY`. About 4% of
158
+ cells list up to four codes separated by a comma and a space; each code becomes its own event, carrying the
159
+ cell's full diagnosis text. No
160
+ `parent_codes` metadata is attached: the same code occurs as both primary and secondary, and
161
+ MEDS-Extract rejects one code produced by two code templates.
162
+ `DIAGNOSIS//ICD//10//<code>` maps to `ICD10CM/<code>` directly.
163
+
164
+ ### Other tables
165
+
166
+ - **ICU stays and bed moves** carry only their start and end times in the release (no unit,
167
+ room or bed), so they become bare `ICU_ADMISSION`, `ICU_DISCHARGE` and `TRANSFER`
168
+ events.
169
+ - **Demographics and allergies** carry no time in the release and are static events.
170
+ - **Flowsheet, lab and order-result values** are free text. The numeric reading is kept where
171
+ it parses; scored flowsheet items charted as `<score>--><label>` (e.g. `-1-->Drowsy`) keep
172
+ their score. The raw string goes to `text_value` unless it is just the number. Epic's
173
+ nonnumeric sentinel `9999999` never becomes a numeric value.
174
+ - **Medications:** each row of the release is one administration and repeats its order's
175
+ start and end, so an order yields one `START`, one `STOP` and one `MAR` event per
176
+ administration.
177
+ - **Orders** hold every order type except labs; roughly half carry a single result.
178
+ - **Birth and death:** the release has no birth date, age or death time, so there are no
179
+ `MEDS_BIRTH` or `MEDS_DEATH` events.
180
+ - **Not extracted:** the telemetry (alarms, 2-second measurements, waveforms; per-admission
181
+ files keyed by file name, far larger than the EMR), `social_hx` (no event times and
182
+ unlabeled Epic category codes), and the medication order date.
183
+
184
+ ### Validation against the release
185
+
186
+ The spec was first written from the authors' release code
187
+ ([Carescape](https://github.com/johnclawrence/Carescape)) and did not run on the published
188
+ files. Profiling the 1.0.0 EMR CSVs (value *shapes* and null counts only, never values)
189
+ showed what had to change:
190
+
191
+ - Tables are lowercase (`admission`, `labs`, ...) rather than `Patient`, `Labs`, ...; the
192
+ keys are `aMRN` and `aCSN`, not `csID` and `CS_PAT_ENC_CSN_ID`.
193
+ - Each timestamp is one already-shifted column rather than a date plus an `HH:MM:SS`
194
+ column. Labs and orders keep only the date (see [Timestamps](#timestamps)).
195
+ - `icu` and `location` have no department, event type, room or bed columns.
196
+ - `resulting_aCSN` holds the string `nan` for unresulted rows.
197
+ - 1,536 ICD-10 codes occur as both primary and secondary diagnoses, and some cells list
198
+ several codes (see [Diagnoses](#diagnoses)). The synthetic test covers both.
199
+ - Medication start and end times are always midnight, and `orders` carries results of its
200
+ own (it holds no lab orders).
201
+
202
+ All ten CSVs are valid UTF-8. On the full 1.0.0 EMR tables (6.3 GB of CSVs read),
203
+ `meds-extract-run` with MEDS-Extract 0.7.1 finishes in about 90 seconds on a laptop with about
204
+ 6.5 GB peak memory, producing 52.6M events for 18,472 subjects. Fewer lab, order and
205
+ medication start/stop events come out than raw rows because identical events are merged: lab
206
+ and order times have day resolution, and the medication table repeats an order's start and
207
+ end on every administration row.
208
+
209
+ ## MEDS-transforms settings
210
+
211
+ If you want to convert a large dataset, you can parallelize the MEDS-transforms stages
212
+ (the MEDS-transformation step that takes the longest).
213
+
214
+ Using local parallelization with the `hydra-joblib-launcher` package:
215
+
216
+ ```bash
217
+ pip install hydra-joblib-launcher --upgrade
218
+ ```
219
+
220
+ Then write a stage-runner file and pass it to the run:
221
+
222
+ ```yaml
223
+ # runner.yaml
224
+ parallelize:
225
+ n_workers: 8
226
+ launcher: joblib
227
+ ```
228
+
229
+ ```bash
230
+ meds-extract-run spec=ALOTT output_dir=$MEDS_OUTPUT_DIR stage_runner_fp=runner.yaml
231
+ ```
232
+
233
+ The number of subjects per shard is `n_subjects_per_shard` in the spec's `etl:` block
234
+ (1000 by default); raise it to cut the parallelization overhead.
235
+
236
+ ## Development
237
+
238
+ ```bash
239
+ uv sync --group dev
240
+ uv run pytest tests/
241
+ uv run pre-commit run --all-files
242
+ ```
243
+
244
+ A release bumps the version in `pyproject.toml` with `uv version --bump`, commits it and pushes
245
+ a tag of that exact version, which publishes to PyPI and creates the GitHub release. See [`RELEASING.md`](https://github.com/rvandewater/ALOTT_MEDS/blob/main/RELEASING.md) for choosing the
246
+ version, the one-time PyPI setup and recovering from a failed release.
247
+
248
+ ALOTT has no open demo, so CI runs the spec end to end over a small **synthetic** input
249
+ generated in [`tests/e2e_synthetic_test.py`](https://github.com/rvandewater/ALOTT_MEDS/blob/main/tests/e2e_synthetic_test.py). That checks the
250
+ spec's mechanics, not its agreement with the real release. Never commit raw ALOTT data or
251
+ MEDS output derived from it.
252
+
253
+ ## Citation
254
+
255
+ Please cite the following original dataset when using ALOTT:
256
+
257
+ ```bibtex
258
+ @article{PhysioNet-alott-1.0.0,
259
+ author = {Lawrence, John and Rayo, Mike and Huerta, Timothy},
260
+ title = {{ALarms, Outcomes Telemetry with Timing (ALOTT): a Bedside-EMR Database}},
261
+ journal = {{PhysioNet}},
262
+ year = {2025},
263
+ month = mar,
264
+ note = {Version 1.0.0},
265
+ doi = {10.13026/sbq5-dy17},
266
+ url = {https://doi.org/10.13026/sbq5-dy17}
267
+ }
268
+ ```
269
+
270
+ Also cite the ETL framework used to extract the dataset:
271
+
272
+ ```bibtex
273
+ @software{van_de_Water_ALOTT_MEDS_ETL_2025,
274
+ author = {van de Water, Robin Philippus},
275
+ license = {MIT},
276
+ month = apr,
277
+ title = {{ALOTT\_MEDS ETL}},
278
+ url = {https://github.com/rvandewater/ALOTT_MEDS},
279
+ year = {2025}
280
+ }
281
+ ```
282
+
283
+ For citing MEDS in general:
284
+
285
+ ```bibtex
286
+ @article{mcdermott2026meds,
287
+ title={MEDS—An Emerging Data Standard and Ecosystem for Health AI Research},
288
+ author={McDermott, Matthew BA and Steinberg, Ethan and Fries, Jason A and van de Water, Robin P and Pang, Chao and Rockenschaub, Patrick and Renc, Pawel and Oh, Jungwoo and Stankevi{\v{c}}i{\=u}t{\.e}, Kamil{\.e} and Xu, Justin and others},
289
+ journal={NEJM AI},
290
+ volume={3},
291
+ number={6},
292
+ pages={AIra2501253},
293
+ year={2026},
294
+ publisher={Massachusetts Medical Society}
295
+ }
296
+ ```
@@ -0,0 +1,11 @@
1
+ LICENSE
2
+ README.md
3
+ pyproject.toml
4
+ src/ALOTT_MEDS/__init__.py
5
+ src/ALOTT_MEDS/event_configs.yaml
6
+ src/ALOTT_MEDS.egg-info/PKG-INFO
7
+ src/ALOTT_MEDS.egg-info/SOURCES.txt
8
+ src/ALOTT_MEDS.egg-info/dependency_links.txt
9
+ src/ALOTT_MEDS.egg-info/entry_points.txt
10
+ src/ALOTT_MEDS.egg-info/requires.txt
11
+ src/ALOTT_MEDS.egg-info/top_level.txt
@@ -0,0 +1,2 @@
1
+ [MEDS_extract.pipelines]
2
+ ALOTT = ALOTT_MEDS:event_configs.yaml
@@ -0,0 +1 @@
1
+ meds-extract[download]<0.8,>=0.7.1
@@ -0,0 +1 @@
1
+ ALOTT_MEDS