Table 5. Observed outcomes by treatment arm and by the ACCORD-BP trial population’s predicted benefit/harm (validation cohort).
| Benefit/harm subgroup | Number of patients | Number of events (%) | Expected absolute risk difference (95% CI) | Observed absolute risk difference (95% CI), P value | Expected minus observed absolute risk difference (95% CI) | |||
|---|---|---|---|---|---|---|---|---|
| Intensive treatment | Standard treatment | All patients (N = 4,498) | Intensive treatment (N = 2,243) | Standard treatment (N = 2,255) | ||||
| CVD events/deaths | ||||||||
| 1 (lowest predicted benefit) | 795 | 842 | 293 (17.9) | 151 (19.0) | 142 (16.9) | 0.018 (−0.009, 0.098) | 0.021 (−0.016, 0.058), P = 0.261 | −0.003 (−0.059, 0.106) |
| 2 (middle predicted benefit) | 647 | 620 | 135 (10.7) | 73 (11.3) | 62 (10.0) | −0.020 (−0.011, −0.030) | 0.013 (−0.021, 0.047), P = 0.459 | −0.033 (−0.070, 0.002) |
| 3 (highest predicted benefit) | 801 | 793 | 235 (14.7) | 86 (10.7) | 149 (18.8) | −0.056 (−0.136, −0.030) | −0.081 (−0.115, −0.046), P < 0.001 | 0.025 (−0.082, 0.078) |
| Serious adverse events | ||||||||
| 1 (lowest predicted harm) | 118 | 114 | 19 (8.2) | 11 (9.3) | 8 (7.0) | −0.001 (−0.040, 0.005) | 0.023 (−0.047, 0.093), P = 0.522 | −0.024 (−0.088, 0.011) |
| 2 (middle predicted harm) | 956 | 1,013 | 192 (9.8) | 116 (12.1) | 76 (7.5) | 0.025 (0.007, 0.039) | 0.046 (0.020, 0.073), P = 0.001 | −0.021 (−0.034, −0.019) |
| 3 (highest predicted harm) | 1,169 | 1,128 | 274 (11.9) | 195 (16.7) | 79 (7.0) | 0.073 (0.041, 0.160) | 0.097 (0.071, 0.123), P < 0.001 | −0.024 (−0.040, 0.042) |
The lowest predicted benefit subgroup had a <1-percentage-point predicted absolute risk reduction in CVD, while the highest predicted benefit subgroup had a >3-percentage-point predicted absolute risk reduction. The lowest predicted harm subgroup had a <0.5-percentage-point predicted absolute risk increase in serious adverse events, while the highest predicted harm subgroup had a >4-percentage-point predicted absolute risk increase. Cut points were chosen to correspond to the tertiles of the distribution of predicted benefit and harm for the combined data from SPRINT and ACCORD-BP. The ACCORD-BP sample included N = 4,498 participants (95.0% of all ACCORD-BP participants) with sufficient data to calculate the risk scores; the other 235 participants were omitted due to missing predictor variables.
CVD, cardiovascular disease.