Table 3.
Generalized ordinal logistic regression: predictors of outcome in SWIFT + STAR
| Model 1 | Good outcome (mRS 0–2) |
Good to intermediate outcome (mRS 0–4) |
||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| OR | 95% CI | Coefficient | SE | P value | OR | 95% CI | Coefficient | SE | P value | |
| THRIVE score | 0·70 | 0·61–0·81 | −0·35 | 0·07 | <0·001 | 0·70* | 0·61–0·81 | −0·35 | 0·07 | <0·001 |
| Time to treatment (hours) | 0·67 | 0·56–0·80 | −0·40 | 0·09 | <0·001 | 0·87 | 0·71–1·06 | −0·14 | 0·10 | 0·18 |
| Treatment with Solitaire device |
1·4 | 0·65–3·15 | 0·36 | 0·40 | 0·37 | 4·81 | 2·22–10·4 | 1·57 | 0·39 | <0·001 |
| sICH | 0·07 | 0·01–0·43 | −2·60 | 0·89 | 0·004 | 0·07* | 0·01–0·43 | −2·60 | 0·89 | 0·004 |
| General anesthesia | 0·85 | 0·52–1·38 | −0·16 | 0·25 | 0·51 | 0·85* | 0·52–1·38 | −0·16 | 0·25 | 0·51 |
| Model 2 | Good outcome (mRS 0–2) |
Good to intermediate outcome (mRS 0–4) |
||||||||
| OR | 95% CI | Coefficient | SE | P value | OR | 95% CI | Coefficient | SE | P value | |
| THRIVE score | 0·69 | 0·60–0·80 | −0·37 | 0·07 | <0·001 | 0·69* | 0·60–0·80 | −0·37 | 0·07 | <0·001 |
| Time to treatment (hours) | 0·71 | 0·60–0·83 | −0·34 | 0·08 | <0·001 | 0·71* | 0·60–0·83 | −0·34 | 0·08 | <0·001 |
| Recanalization (TICI 2b/3) | 2·5 | 1·49–4·43 | 0·94 | 0·28 | <0·001 | 2·5* | 1·49–4·43 | 0·94 | 0·28 | <0·001 |
| sICH | 0·05 | 0·01–0·29 | −2·99 | 0·89 | <0·001 | 0·05* | 0·01–0·29 | −2·99 | 0·89 | <0·001 |
| General anesthesia | 0·80 | 0·49–1·31 | −0·22 | 0·25 | 0·38 | 0·80* | 0·49–1·31 | −0·22 | 0·25 | 0·38 |
Generalized ordinal logistic regression models showing the relative impact of predictors on outcome across the range of the mRS, trichotomized as good outcome (0–2) vs. intermediate outcome (3–4) vs. poor outcome (5–6). Odds ratios above 1 (with positive coefficients) represent a predictor that is associated with improved odds of a better outcome, and odds ratios below 1 (with negative coefficients) represent a predictor that is associated with reduced odds of a better outcome.
In each model, odds ratios are shown for good outcome (mRS 0–2, on the left) and for good to intermediate outcome (mRS 0–4, on the right). For predictors meeting the proportional odds assumption for ordinal logistic regression, the same odds ratio value is presented for both outcome levels, indicated by an asterisk. For predictors failing to meet the proportional odds assumption, two different odds ratio values are presented for the two outcome levels.
In Model 1, predictors are THRIVE score, time to treatment (onset to groin puncture in hours), treatment with the Solitaire device (vs. Merci device), occurrence of symptomatic intracerebral haemorrhage (sICH), and use of general anesthesia (vs. monitored anesthesia care without endotracheal intubation).
In Model 2, predictors are THRIVE score, time to treatment (onset to groin puncture in hours), successful recanalization (TICI 2b/3), occurrence of sICH, and use of general anesthesia (vs. monitored anesthesia care without endotracheal intubation).
Two separate models are shown because of significant interaction between use of the Solitaire device and the rate of successful recanalization. Patients treated with the Solitaire device were significantly more likely to have successful recanalization [(221/271 (81·5%)] than patients treated with the Merci device [22/53 (41·5%)] (P < 0·001).
CI, confidence interval; mRS, modified Rankin Scale; OR, odds ratio; SE, standard error; THRIVE, Totaled Health Risks in Vascular Events; TICI, Thrombolysis in Cerebral Ischemia.