Characteristics of the multiple linear regression model for predicting limb strength from clinical and MR imaging data
| Ankle Dorsiflexion (lb) | Hip Flexion (lb) | |||
|---|---|---|---|---|
| Medulla Included | Medulla Excluded | Medulla Included | Medulla Excluded | |
| Number of ankles* | 72 | 88 | 68 | 83 |
| Model coefficients† | ||||
| Midbrain, median MTR | −276 ± 86 (P = .002) | −224 ± 78 (p = .005) | −183 ± 68 (P = .01) | −158 ± 71 (P = .03) |
| Pons, median MTR | 300 ± 99 (P = .003) | 254 ± 83 (P = .003) | 175 ± 88 (P = .03) | 176 ± 75 (P = .004) |
| Medulla, median MTR | 153 ± 57 (P = .009) | 134 ± 48 (P = .006) | ||
| MS clinical subtype‡ | −21.7 ± 3.9 (P < .001) | −19.5 ± 3.7 (P < .001) | −22.9 ± 3.2 (P < .001) | −20.4 ± 3.3 (P < .001) |
| Constant | −42 ± 59 (P = .5) | −26 ± 48 (P = .6) | −21 ± 47 (P = .7) | −27 ± 43 (P = .5) |
| Partial correlation coefficients§ | ||||
| Midbrain, median MTR | −0.37 (P = .002) | −0.30 (P = .005) | −0.32 (P = .01) | −0.24 (P = .03) |
| Pons, median MTR | 0.35 (P = .003) | 0.32 (P = .003) | 0.27 (P = .03) | 0.26 (P = .02) |
| Medulla, median MTR | 0.31 (P = .009) | 0.34 (P = .006) | ||
| MS clinical subtype | −0.56 (P < .001) | −0.50 (P < .001) | −0.67 (P < .001) | −0.57 (P < .001) |
| Model performance | ||||
| Adjusted r2 | 0.36 (P < .0001) | 0.30 (P < .0001) | 0.45 (P < .0001) | 0.33 (P < .0001) |
| Predictions within 10% of actual strength | 28% | 21% | 5% | 30% |
| Median difference of prediction from actual strength | 13% | 21% | 50% | 17% |
| Model performance vs prior imaging and strength testing | ||||
| r2 | 0.34 (P < .0001) | 0.17 (P = .0004) | 0.38 (P < .0001) | 0.28 (P < .0001) |
| Predictions within 10% of actual strength | 24% | 15% | 25% | 22% |
| Median difference of prediction from actual strength | 21% | 23% | 32% | 29% |
* Each subject contributes up to 2 ankles (right and left). The number of hips and ankles listed here is lower than the overall total because MTR was not obtained for all subjects. The model does not account for correlations between ankles in the same subject. Pearson correlation coefficients are used.
† Coefficient errors are ± 1 standard deviation.
‡ MS clinical subtype is modeled as 1 for secondary progressive MS, 0 otherwise.
§ Pearson correlation coefficients for model parameter, holding the others constant.