# Predicting Missed Newborn Postnatal Care in Ethiopia
Analysis code and figures for an interpretable data-mining model of missed
newborn postnatal care (PNC), using the 2019 Ethiopia Mini Demographic and
Health Survey (EMDHS) Births Recode.
The manuscript is not kept in this repository. Only analysis code, figures and
aggregate results are versioned here.
## Data
The analysis reads `ETBR81FL.SAV`, the 2019 Ethiopia Mini-EDHS Births Recode.
**It is not included in this repository and must not be committed** — the DHS
Program licence prohibits redistribution. Register at
and request the Ethiopia 2019 Mini-DHS Births
Recode, then place the file where the scripts expect it:
- R: `../raw-data/ETBR81SV/ETBR81FL.SAV` relative to the `.Rmd`
- Python: set `PNC_SAV` or edit the `SAV` constant at the top of
`reproduction-python/run_pipeline.py`
`.gitignore` blocks `*.SAV`, `*.dta` and the derived `*.pkl` / `*.npz` files,
which contain individual-level records and predictions.
## Results
### Sample
| Step | n |
|---|---|
| Birth records in the Births Recode file | 23,007 |
| Most recent births (BIDX = 1) | 5,846 |
| Valid newborn PNC outcome, PSU, stratum and weight | 3,970 |
| Complete predictors (modelling sample) | 3,892 |
Of the 3,970 newborns, 3,438 missed PNC and 532 received it.
### Prevalence
Weighted prevalence of missed newborn PNC was **86.3% (95% CI 84.0–88.7)**;
unweighted, 86.6%.
### Timing, provider and place among newborns who received PNC
| Timing of first check | Weighted % |
|---|---|
| Weeks after birth | 42.8 |
| 3–6 days | 14.8 |
| Immediately or within the first hour | 12.0 |
| 7 or more days | 11.6 |
| 1–23 hours | 10.6 |
| 1–2 days | 8.2 |
**30.7%** were checked within 48 hours. Checks were overwhelmingly performed by
a skilled or medical provider (96.3%), most often at a public health facility
(61.2%) or at home (35.0%).
### Model performance
Five-times repeated five-fold cross-validation, training partition only:
| Model | AUC | 95% …