Table 4. The relative risk of missing data and mean change in concordance score from the baseline to the follow-up RDQAs.
| Baseline N = 2411 |
Follow-up N = 2381 |
Unadjusted model | Adjusted model1 | |||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Relative risk of missing data | n | (%)2 | n | (%)2 | RR3 | (95% CI) | P-value | RR3 | (95% CI) | P-value |
| Baseline to follow-up, paper forms | 735 | (30) | 760 | (32) | 1.04 | (0.79–1.38) | 0.785 | 1.09 | (0.84–1.41) | 0.522 |
| Baseline to follow-up, KenyaEMR | 747 | (31) | 320 | (13) | 0.43 | (0.32–0.58) | <0.001 | 0.43 | (0.32–0.58) | <0.001 |
| Change in concordance score | Mean | (SD)4 | Mean | (SD)4 | β5 | (95% CI) | P-value | β5 | (95% CI) | P-value |
| Baseline to follow-up | 11.9 | (4.0) | 13.6 | (4.2) | 1.79 | (0.25–3.33) | 0.023 | 1.79 | (0.25–3.33) | 0.023 |
1 Adjusting for facility type, facility ownership, months of EMR implementation, and facility patient load.
2 The number (n) and percent of records with at least one missing value among nine required data elements.
3 GEE models were used to determine if RDQA round was associated with the proportion of records that had any missing values among nine required data elements. GEE models used a log link, binomial distribution, exchangeable correlation matrix, and robust standard errors that allow for clustering by facility.
4 20 data elements were incorporated into the concordance score, with one point awarded for each of the 20 elements that had matching values recorded on paper records and KenyaEMR (0 indicates no concordant elements, 20 indicates complete concordance).
5 GEE models were used to determine if the mean concordance score changed from the baseline to follow-up RDQAs. GEE models used an identity link, normal distribution, exchangeable correlation matrix, and robust standard errors that allow for clustering by facility.