Table 3. Sensitivity (95% CIs), Overall and Within Racial/Ethnic Subgroups, for Suicide Prediction Modelsa.
| Percentile | All (N = 1543)b | White (n = 1092) | Hispanic (n = 144) | Black (n = 30) | Asian (n = 85) | American Indian/Alaskan Native (n = 15) | Not recorded (n = 167) |
|---|---|---|---|---|---|---|---|
| Logistic regression with LASSO: global thresholdsc | |||||||
| ≥75% | 75.0 (72.9-77.1) | 78.8 (76.5-81.2) | 73.6 (66.6-80.6) | 63.3 (46.7-80.0) | 80.0 (70.6-88.2) | 26.7 (6.7-53.3) | 56.9 (49.1-64.1) |
| ≥90% | 54.9 (52.4-57.3) | 62.2 (59.2-65.0) | 54.9 (47.2-63.2) | 10.0 (0-23.3) | 44.7 (34.1-56.5) | 6.7 (0-20.0) | 27.5 (21.0-34.7) |
| ≥95% | 41.1 (38.6-43.5) | 46.8 (43.7-49.7) | 36.8 (29.9-45.1) | 6.7 (0-16.7) | 31.8 (22.4-42.4) | 6.7 (0-20.0) | 23.4 (16.8-29.9) |
| ≥99% | 15.4 (13.6-17.2) | 18.4 (16.2-20.5) | 5.6 (2.1-9.7) | 3.3 (0-10.0) | 8.2 (2.4-14.1) | 0 (0-0) | 12.0 (7.2-17.4) |
| Random forest: global thresholds | |||||||
| ≥75% | 71.4 (69.3-73.6) | 73.5 (70.9-76.4) | 66.0 (58.3-73.6) | 56.7 (40.0-73.3) | 90.6 (84.7-96.5) | 33.3 (13.2-53.3) | 59.3 (52.1-66.5) |
| ≥90% | 48.6 (46.2-51.2) | 52.8 (50.0-55.8) | 47.2 (39.6-55.6) | 6.7 (0-16.7) | 61.2 (50.6-71.8) | 6.7 (0-20.0) | 29.3 (22.8-36.5) |
| ≥95% | 38.0 (35.6-40.2) | 40.6 (37.5-43.6) | 38.2 (31.2-46.5) | 3.3 (0-10.0) | 60.0 (49.4-70.6) | 6.7 (0-20.0) | 20.4 (14.4-26.3) |
| ≥99% | 12.6 (10.9-14.2) | 15.0 (12.9-17.3) | 5.6 (2.1-9.7) | 3.3 (0-10.0) | 4.7 (1.2-9.4) | 6.7 (0-20.0) | 9.0 (4.8-13.2) |
| Logistic regression with LASSO: race/ethnicity-specific thresholdsd | |||||||
| ≥75% | NA | 76.4 (73.6-78.9) | 80.6 (73.6-86.8) | 70.0 (53.3-86.7) | 83.5 (75.3-90.6) | 26.7 (6.7-53.3) | 35.3 (28.1-43.1) |
| ≥90% | NA | 59.6 (56.9-62.5) | 65.3 (57.6-72.9) | 56.7 (40.0-73.3) | 44.7 (34.1-55.3) | 6.7 (0-20.0) | 25.1 (18.6-32.3) |
| ≥95% | NA | 45.5 (42.6-48.5) | 47.2 (38.9-54.9) | 6.7 (0-16.7) | 32.9 (22.4-43.5) | 6.7 (0-20.0) | 22.2 (15.6-29.3) |
| ≥99% | NA | 17.1 (14.9-19.3) | 9.7 (4.9-14.6) | 3.3 (0-10.0) | 8.2 (3.5-14.1) | 0 (0-0) | 9.0 (4.8-13.8) |
| Random forest: race/ethnicity-specific thresholds | |||||||
| ≥75% | NA | 72.3 (69.9-74.7) | 77.1 (70.1-83.3) | 90.6 (84.7-96.5) | 76.7 (60.0-90.0) | 13.3 (0-33.3) | 49.7 (41.9-57.5) |
| ≥90% | NA | 52.1 (49.1-54.8) | 48.6 (41.0-56.9) | 61.2 (50.6-71.8) | 36.7 (20.0-53.3) | 6.7 (0-20.0) | 26.3 (20.3-32.9) |
| ≥95% | NA | 37.8 (34.8-40.8) | 43.1 (34.7-51.4) | 52.9 (42.3-62.4) | 3.3 (0-10.0) | 6.7 (0-20.0) | 18.6 (12.6-24.6) |
| ≥99% | NA | 13.8 (11.8-15.9) | 6.9 (2.8-11.1) | 4.7 (1.2-9.4) | 3.3 (0-10.0) | 6.7 (0-20.0) | 9.0 (4.8-13.8) |
Abbreviation: NA, not applicable.
Sensitivities are given as percentages. The 95% CIs were obtained from 1000 visit–level bootstrap samples. Prediction model performance is not reported for visits made by patients who were Native Hawaiian/Pacific Islander, multiracial, or other race because there were fewer than 10 visits with a suicide death within 90 days in the validation data set (eTable 10 in the Supplement).
Number of visits with suicide death within 90 days in validation data.
Global thresholds defined at given percentiles of the distribution of all risk predictions in the training data set.
Race/ethnicity-specific thresholds defined at given percentiles of the distribution of risk predictions among all visits from patients of the specified race/ethnicity within the training data set.