Table 2. Overall 28-Day Mortality After Intensive Care Unit (ICU) vs General Ward Care for the Matched Sample.
| Estimator | Sample Sizea | ICU Deaths, No. (%)b | General Ward Deaths, No. (%)b | Risk Difference, % (95% CI)c |
|---|---|---|---|---|
| IV (PeT, Probit) | 9015 | 2090 (23.2) | 2534 (28.1) | −4.9 (−26.4 to 16.6) |
| IV (PeT, Logit) | 9015 | 2096 (23.2) | 2539 (28.2) | −4.9 (−24.4 to 16.6) |
| Regression | 9192 | 2594 (28.2) | 1914 (20.8) | 7.4 (5.0 to 9.8) |
| Unadjusted | 9192 | 2915 (31.7) | 1715 (18.7) | 13.1 (11.2 to 14.9) |
Abbreviations: IV, instrumental variable; PeT, person-centered treatment.
a
For each method, the maximum sample size was 9192. Observations were excluded if there is not mass at any value (rounded to 0.01) of the propensity score for both levels of exposure as recommended by Basu.18
b
The number of predicted deaths is rounded to the nearest whole number.
c
Normal-based 95% CI with standard error is calculated with the nonparametric bootstrap, allowing for clustering by hospital. Difference in percentage of deaths is from the PeT instrumental variable analysis.