MemoA06
ClientDr. ABC
Client OrganizationABC Department
Version2.0
StatusDraft
ClassificationInternal use only
Note: Shared synthetic release 2.0. Real hospital names are reference labels only; all patient records and outcomes are simulated.

1 Overview

This summary now reads the current unified data release 2.0-2026-10-09. Counts and model summaries are computed during rendering, so results copied from a previous release cannot linger unnoticed after a data update. It uses the same primary formulas and linked cohort as A03 and A05.

Training data notice. Patient records, clinical measurements and outcomes are entirely simulated. Hospital names and official ED listings are public reference metadata. No result evaluates a real facility.

2 One source for the project

Cohort Records
NACRS ED visits 30000
DAD inpatient abstracts 60000
Admitted ED visits 5810
Linked ED admissions 4633
Adult ED visits 24926
Adult asthma visits 8170

The source workbooks are pseudo_NACRS.xlsx and pseudo_DAD.xlsx. A00 derives the RDS caches. All memos A00–A09 use these sources or explicitly defined subsets. The former resampled data and separate 12,000-visit asthma cohort have been archived. In the new release, each ED patient has one visit, and each hospital serves many patients.

The two source workbooks share one public hospital reference table. Hospital IDs are study codes, names are selected from actual ED facilities, and case volumes are simulated. Calgary and Edmonton groups include surrounding towns using the referenced geographic convention; patient residence remains a separate field.

3 Research questions and methods

  1. A03 uses logistic regression to describe how pre-disposition ED stay and CTAS are associated with admission. Total ED stay is an appendix comparison because it includes time after disposition.
  2. A04 links admitted ED visits to DAD using an exact patient-ID match and a six-hour window on the absolute time difference, keeping the closest unique match. It does not use the answer-key sheet.
  3. A05 models inpatient calendar length of stay on total ED stay and CTAS in the linked cohort, with acute stay as a sensitivity outcome.
  4. A07 compares adult asthma and other visits using within-hospital propensity-score matching with hospital-clustered uncertainty.
  5. A08 applies outcome regression and g-computation to A07’s fixed pairs, allowing the asthma effect to vary by region and including hospital fixed effects.
  6. A09 compares asthma admissions across hospital regions using logistic GLMMs, with an asthma-by-region secondary analysis.

4 Current descriptive results

4.1 Admission

5,810 of 30,000 ED visits (19.4%) have an admitted disposition. The primary A03 logistic model is reproduced below; duration coefficients are per minute.

Odds ratios: pre-disposition ED stay and CTAS
Term Estimate Low High P
pre_disposition_los_min 1.0004 1.0000 1.0007 0.0269
ctas_level2 0.4413 0.4060 0.4797 0.0000
ctas_level3 0.2309 0.2115 0.2521 0.0000
ctas_level4 0.1092 0.0978 0.1219 0.0000
ctas_level5 0.0564 0.0475 0.0670 0.0000

4.2 Linkage

4,633 of 5,810 admitted visits (79.7%) are linked under the prespecified ID/time rule. The synthetic answer key contains every true continuation, including records with deliberately missing/mistyped IDs or an admission more than six hours from the ED discharge. Unlinked visits are not assigned an imputed inpatient stay.

Status Freq
Linked 4633
No PHN match in DAD 888
PHN match outside the 6-hour window 289

4.3 Inpatient length of stay

The linked cohort has a mean calendar stay of 4.79 days and a median of 4 days. The primary A05 specification is reproduced below, with ED duration measured per minute.

Linear regression coefficients in inpatient days
Term Estimate Low High P
los_min -0.0002 -0.0009 0.0006 0.622
ctas_level2 -1.1411 -1.3202 -0.9620 0.000
ctas_level3 -1.6909 -1.8883 -1.4936 0.000
ctas_level4 -2.2640 -2.5285 -1.9995 0.000
ctas_level5 -2.7786 -3.2143 -2.3429 0.000

5 Interpretation and limitations

A03 and A05 retain their original descriptive model specifications and conventional standard errors. They do not account for hospital clustering introduced in this release, so their significance tests are illustrative. A07–A09 explicitly model or account for hospital clustering for the asthma questions. None of the descriptive associations proves that waiting longer causes admission or a longer inpatient stay.

The new release uses one patient per ED visit, but it still omits many features of real healthcare data, including realistic hospital volume weights, complex transfer pathways, diagnostic misclassification and follow-up reconstruction. The repeat-event flags remain generated labels. A09’s risk-adjusted admission rates should not be read as quality rankings.

6 Memos and reproducibility

A00 defines and builds variables; A01 and A02 profile the files. A03, A04, and A05 provide full descriptive analyses. A07, A08, and A09 cover the asthma analyses and validation.

The local rebuild script is code/rebuild_unified_project.R. Working source code, synthetic patient files and analysis bundles remain excluded from Git; rendered HTML and the public hospital reference/dictionary can be shared. The GitHub copy alone does not contain the patient data needed to rerun the full project.

Back to project