README.md

---
license: cc-by-nc-4.0
task_categories:
- tabular-classification
- tabular-regression
- time-series-forecasting
language:
- en
tags:
- tsfile
- timeseries
- synthetic
- healthcare
- icu
- vital-signs
- continuous-monitoring
- mimic-iv
- eicu-crd
- time-series
- apache-ii
- sofa
- news2
- qsofa
- charlson
- alarm-fatigue
- joint-commission
- iec-60601
- ventilation
- vasopressor
- sepsis
- early-warning-score
- deterioration-prediction
- signal-quality
- monitoring-devices
- philips-intellivue
- ge-carescape
- masimo
- nihon-kohden
- format:tsfile
pretty_name: HLT-009 Synthetic Continuous Vital Sign Monitoring Dataset (Sample Preview) (TsFile)
size_categories:
- 10K<n<100K
configs:
- config_name: default
  data_files:
  - split: alarm_events
    path: alarm_events.tsfile
  - split: episode_summary
    path: episode_summary.tsfile
  - split: interventions
    path: interventions.tsfile
  - split: vitals_timeseries
    path: vitals_timeseries.tsfile
---

# HLT-009 Synthetic Continuous Vital Sign Monitoring Dataset (Sample Preview) (TsFile)

Apache TsFile version of [`xpertsystems/hlt009-sample`](https://huggingface.co/datasets/xpertsystems/hlt009-sample).

## Overview

A **fully synthetic** ICU continuous vital sign monitoring dataset combining 12-stream time-series vitals (HR/SpO2/RR/NBP/IBP/Temp/EtCO2/CVP/CO), alarm event logs with true/false labels, intervention logs (medication boluses, ventilator adjustments, code events), and 53-column episode-level summary data — calibrated to MIMIC-IV / eICU-CRD benchmarks with APACHE-II, SOFA, NEWS2, qSOFA, and CCI severity scoring.

- **Tables:** 4 TsFile table(s), 26,888 rows in total.
  - `alarm_events`: 166 rows, 2022-03-19 15:20:00 → 2023-09-19 22:15:00 (UTC)
  - `episode_summary`: 25 rows, 1970-01-01 00:00:00 → 1970-01-01 00:00:00.024000 (UTC)
  - `interventions`: 20 rows, 1970-01-01 00:00:00 → 2023-09-19 14:23:03.165000 (UTC)
  - `vitals_timeseries`: 26,677 rows, 2022-03-15 00:00:00 → 2023-12-07 06:30:00 (UTC)

## Schema (TsFile structure)

### `alarm_events.tsfile`

- **Time** (INT64, milliseconds) — the source `alarm_onset_ts` column.
- **TAG**: none — the table is stored as a single device.
- **FIELD** (14): `alarm_id` (string), `episode_id` (string), `alarm_type` (string), `alarm_priority` (string), `alarm_duration_sec` (int64), `true_alarm_flag` (int64), `false_alarm_subtype` (string), `response_time_min` (double), `intervention_triggered` (int64), `override_flag` (int64), `limit_at_alarm_low` (double), `limit_at_alarm_high` (double), `alarm_cascade_id` (int64), `shift` (string)

### `episode_summary.tsfile`

- **Time** (INT64, milliseconds) — synthetic row index over the stable source order; this table has no clock column.
- **TAG**: none — the table is stored as a single device.
- **FIELD** (54): `episode_id` (string), `monitoring_setting` (string), `icu_unit_type` (string), `admit_dt` (string), `discharge_dt` (string), `episode_duration_days` (double), `bed_id` (string), `age` (int64), `sex` (string), `apache2_score` (int64), `sofa_score` (int64), `sofa_at_discharge` (int64), `primary_dx_group` (string), `cci_score` (int64), `ventilation_status` (int64), `vasopressor_flag` (int64), `rrt_flag` (int64), `has_arterial_line` (int64), `has_central_line` (int64), `has_pa_catheter` (int64), `trajectory` (string), `monitor_manufacturer` (string), `rpm_device_type` (double), `lead_configuration` (string), `device_uptime_pct` (double), `connectivity_drops` (int64), `true_alarm_rate` (double), `total_alarms` (int64), `alarms_per_patient_day` (double), `actionable_alarm_rate` (double), `alarm_override_rate` (double), `median_response_time_min` (double), `alarm_limit_modification_count` (int64), `alarm_cascade_count` (int64), `fatigue_index_score` (double), `news2_max` (double), `news2_mean` (double), `qsofa_max` (int64), `deterioration_6h_label` (int64), `in_hospital_mortality` (int64), `readmission_30d` (int64), `rapid_response_event` (int64), `sqi_hr_bpm` (double), `sqi_spo2_pct` (double), `sqi_rr_bpm` (double), `sqi_nbp_sys_mmhg` (double), `sqi_nbp_dia_mmhg` (double), `sqi_nbp_map_mmhg` (double), `sqi_ibp_sys_mmhg` (double), `sqi_ibp_dia_mmhg` (double), `sqi_temp_c` (double), `sqi_etco2_mmhg` (double), `sqi_cvp_mmhg` (double), `sqi_cardiac_output_lpm` (double)

### `interventions.tsfile`

- **Time** (INT64, milliseconds) — the source `intervention_ts` column.
- **TAG**: none — the table is stored as a single device.
- **FIELD** (10): `intervention_id` (string), `episode_id` (string), `intervention_type` (string), `alarm_trigger_id` (string), `latency_min` (double), `drug_class` (string), `dose_delta` (double), `vent_param_changed` (string), `rosc_flag` (double), `rapid_response_team_id` (double)

### `vitals_timeseries.tsfile`

- **Time** (INT64, milliseconds) — the source `timestamp` column.
- **TAG**: none — the table is stored as a single device.
- **FIELD** (18): `episode_id` (string), `hr_bpm` (double), `spo2_pct` (double), `rr_bpm` (double), `nbp_sys_mmhg` (double), `nbp_dia_mmhg` (double), `nbp_map_mmhg` (double), `ibp_sys_mmhg` (double), `ibp_dia_mmhg` (double), `temp_c` (double), `etco2_mmhg` (double), `cvp_mmhg` (double), `cardiac_output_lpm` (double), `artifact_flag` (int64), `news2_score` (double), `qsofa_score` (double), `news2_roll_max_4h` (double), `news2_rate_of_rise` (double)

## Conversion notes

- One TsFile per source CSV (4 tables). Every table is stored as a single device, so all source columns (episode/acility ids, alarm ids, device vendors) are FIELDs.
- Time = the source timestamp column parsed as INT64 milliseconds: `timestamp` (5-minute vitals grid), `alarm_onset_ts`, `intervention_ts`. `episode_summary` is an episode-level table with no per-row clock, so its Time is a synthetic row index (0..N-1).
- No column is dropped.

## Usage

Install the Apache TsFile Python SDK (`pip install tsfile`) and read a converted file:

```python
from pathlib import Path
from tsfile import TsFileReader

path = Path("alarm_events.tsfile")
with TsFileReader(str(path)) as reader:
    schemas = reader.get_all_table_schemas()
    print("tables:", list(schemas))
    table_name = next(iter(schemas))
    table = schemas[table_name]
    columns = [column.get_column_name() for column in table.get_columns()]
    print("columns:", columns)
    field_names = [
        column.get_column_name()
        for column in table.get_columns()
        if column.get_column_name() not in {"Time", "time"}
    ]
    if field_names:
        with reader.query_table(table_name, field_names[:3], batch_size=1024) as result:
            batch = result.read_arrow_batch()
            if batch is not None:
                print(batch.to_pandas().head())
```

## Source & license

- Original dataset: [`xpertsystems/hlt009-sample`](https://huggingface.co/datasets/xpertsystems/hlt009-sample)
- Source card title / publisher: HLT-009 Synthetic Continuous Vital Sign Monitoring Dataset (Sample Preview)
- License: cc-by-nc-4.0 (as declared by the original dataset).