Assessing individual-level measurement precision of the Short Physical Performance Battery using the test information function

The Short Physical Performance Battery (SPPB) is a commonly used tool to assess lower extremity function, composed of three assessments (standing balance, gait speed, and chair stand). While its validity for group-level decisions has been positively demonstrated, the measurement precision at the individual level needs more clarification. We aimed to examine the SPPB's psychometric characteristics including its conditional measurement precision with Rasch methods in a sample of elderly patients admitted to cardiac rehabilitation. We analyzed prospectively collected SPPB data from 637 patients aged >= 75 years, admitted to our cardiac rehabilitation department (January 2018-July 2019). After classical Rasch analysis, we calculated the test information function to examine the measurement precision of the SPPB along the score continuum. The mean SPPB score was 6.3 points (SD 3.4) (potential range 0-12). Our results confirmed the SPPB unidimensionality (variance attributable to the main factor: 84.8%), appropriate functioning of rating scale categories, internal construct validity (infit and outfit mean-square statistics: 0.90-1.09), and no item local dependence (residual correlations <0.2). However, the measurement precision at the individual level was quite limited (SE > 0.94 logits): the confidence intervals for true scores were, at best, about one-third of the score range (four points). This limited measurement precision increases the risk of inappropriate clinical decisions about individuals in diagnostic classification or change assessment. Therefore, further research of high methodological quality is warranted on this point. Moreover, a clearer distinction between group-level vs. individual-level indicators of change is necessary.

Automated summary

The Short Physical Performance Battery (SPPB) is a commonly used tool for assessing lower extremity function. While its validity for group-level decisions has been demonstrated, its measurement precision at the individual level needs more clarification.