Yoshitaka Wada, MD, Shigeru Sonoda, MD, PhD, Sayaka Okamoto, MD, PhD, Makoto Watanabe, OTR, MS, Hideto Okazaki, MD, PhD, Yuko Okuyama, RPT
Jpn J Compr Rehabil Sci 10: 71-76, 2019
Objective: We compared the accuracy of formulas for
predicting ADL outcome constructed by multiple
regression analysis in post-stroke patients admitted to
a Kaifukuki rehabilitation ward.
Methods: We divided 1,502 post-stroke patients into a
construction group used to generate prediction
formulas, and a validation group used to confirm
the prediction accuracy. Prediction formula S was
constructed by conventional multiple regression
analysis using Functional Independence Measure-motor score (mFIM) at discharge as the dependent
variable. Prediction formula R was constructed by
reciprocal multiple regression analysis. Prediction
equation E was constructed by calculating mFIM at
discharge via mFIM effectiveness. In the validation
group, predicted mFIM at discharge was calculated,
and intraclass correlation coefficient and absolute
value of residual were compared.
Results: Intraclass correlation coefficients were 0.86
using prediction formula S, 0.90 using prediction
formula R, and 0.89 using prediction formula E.
Absolute values of residual were 9.38+-6.62 using
prediction formula S, 7.30+-6.56 using prediction
formula R, and 7.56+-6.45 using prediction formula E.
The Steel-Dwass test detected a significant difference
between prediction formulas S and R, and between
prediction formulas S and E (both p<0.05).
Conclusion: The prediction accuracy of formulas for
predicting ADL outcome constructed by multiple
regression analysis is improved by adding a
transformation that brings the model toward linearity.
Key words: stroke, rehabilitation, multiple regression analysis, outcome prediction, Functional Independence Measure (FIM)