Table 3.
Multivariable Cox regression for OS in 2284 ER+ breast cancer patients
| Characteristic | Events | Adjusted HR (95% CI) | P a |
|---|---|---|---|
| AI–endocrine treatment ratiob | |||
| AI < 25% | 127 | 1.00 | |
| AI 25%-75% | 62 | 0.32 (0.21 to 0.49) | <.001 |
| AI 25%-75% * (follow-up time − 5)c | 1.42 (1.12 to 1.80) | .003 | |
| AI > 75% | 47 | 0.50 (0.34 to 0.74) | <.001 |
| Age | 1.05 (0.97 to 1.13) | .19 | |
| Trastuzumab | |||
| No | 233 | 1.00 | |
| Yes | 3 | 2.46 (0.72 to 8.40) | .15 |
| Grade | |||
| I | 43 | 0.33 (0.17 to 0.61) | <.001 |
| II | 12 | 0.55 (0.40 to 0.75) | <.001 |
| II * (follow-up time − 5)c | 1.22 (1.05 to 1.43) | .009 | |
| III | 82 | 1.00 | |
| Unknown | 99 | 1.41 (0.97 to 2.07) | .07 |
| Positive lymph nodes | |||
| 0 | 43 | 1.00 | |
| 1-3 | 93 | 1.34 (0.93 to 1.94) | .11 |
| 4-9 | 56 | 2.31 (1.54 to 3.47) | <.001 |
| >10 | 44 | 4.76 (3.06 to 7.38) | <.001 |
| pT stage | |||
| 1 | 78 | 1.00 | |
| 2 | 120 | 1.08 (0.81 to 1.45) | .57 |
| 3 | 25 | 1.27 (0.79 to 2.04) | .31 |
| 4 | 6 | 1.76 (0.73 to 4.20) | .20 |
| Unknown | 7 | 1.07 (0.48 to 2.41) | .85 |
| PR status | |||
| Negative | 54 | 1.00 | |
| Positive | 170 | 0.46 (0.34 to 0.64) | <.001 |
| Unknown | 12 | 0.59 (0.31 to 1.12) | .11 |
| HER2 status | |||
| Negative | 180 | 1.00 | |
| Positive | 33 | 0.99 (0.63 to 1.54) | .98 |
| Unknown | 23 | 0.95 (0.61 to 1.49) | .85 |
| Ovarian ablation | |||
| No | 196 | 1.00 | |
| Yes | 40 | 1.12 (0.79 to 1.59) | .50 |
P values are based on a 2-sided Wald test. AI = aromatase inhibitor; CI = confidence interval; ER+ = estrogen receptor positive; HR = hazard ratio; OS = overall survival; PR = progesterone receptor.
The AI–endocrine treatment ratio, included in the model as a time-dependent variable, is defined as the percentage of total endocrine treatment duration (AI+tamoxifen) spent on AI treatment.
Interaction between the covariates and follow-up time centered at 5 years was included to accommodate nonproportional hazards. At 5 years of follow-up, patients with an AI 25%-75% ratio had a smaller chance of an OS event then patients with a AI less than 25% ratio (adjusted HR = 0.32). The hazard ratio increases by 42% for each additional year of follow-up, so at 6 years of follow-up the adjusted hazard ratio for AI 25%-75% ratio vs AI less than 25% ratio = exp{ln(0.32) + (follow-up time-5) * ln(1.42)} = 0.45. At 5 years of follow-up, patients with grade II tumors had a smaller chance of an OS event then patients with a grade III tumor (adjusted HR = 0.55). The hazard ratio increases by 22% for each additional year of follow-up, so at 6 years of follow-up the adjusted hazard ratio for grade II tumors vs grade III tumors = exp{ln(0.55) + (follow-up time − 5) * ln(1.22)} = 0.67.