Letter to Editor Regarding “Cross‐Sectional Associations of Integrated Lifestyle‐Related Factors With Intrinsic Capacity in Community‐Dwelling Older Adults: The Kashiwa Cohort Study”

We read with interest Lyu et al.'s article entitled “Cross‐Sectional Associations of Integrated Lifestyle‐Related Factors With Intrinsic Capacity in Community‐Dwelling Older Adults: The Kashiwa Cohort Study.” The study tackles an important question and explores multidomain lifestyle‐related factors and intrinsic capacity [1]. But there are some statistical points to be made.

First, the composite IC score, the construction of which is required, requires further clarification. The five IC domains were combined by using logarithmic transformation, standardization, and sex‐specific confirmatory factor analysis (CFA) loadings. But comprehensive CFA fit indices such as comparative fit index, Tucker–Lewis index, root mean square error of approximation, standardized root mean square residual, and chi‐square statistic must be provided to show that the fit of the proposed latent structure is satisfactory. This is especially significant because IC is a multidimensional construct and may be measured through various measurement models which can result in different estimates of the underlying construct [2]. Using sex‐specific CFA loadings also has some concerns about the comparability of the resulting IC scores. Assessment of measurement invariance, such as configural, metric, and scalar, is essential prior to assuming the equivalence of meaning of the construct at the level of the construct across sexes [3, 4].

Secondly, it seems that there is a lot of overlap in the conceptual nature of the variables relating to exposure and the variables relating to the outcome. The physical factor was regular exercise, prolonged walking, and perceived walking speed, while the IC was measured by physical performance including the Short Physical Performance Battery. The same was true for the nutrition‐related factor (ability to eat hard foods) and nutritional status (vitality domain). There may also be a close relationship between social participation/social support and psychological capacity and depressive symptoms. Therefore, the observed associations might be partially explained by construct overlap, and independent associations between different lifestyle exposures and IC might still be observed. This concern can be especially important if a composite outcome consists of domains that are conceptually similar to the predictors.

Third, it is assumed that the cumulative exposure variable is a uniform unit, both nutritionally and physically and socially. Different rules were applied in this process to the domains, however: the nutrition‐related factor required both oral and dietary criteria, the physical factor required two of three criteria, and the social factor required all three criteria. However, all three domains were evenly contributing to a 0–3 count. Therefore, the extent of exposure may not be the same for one favorable factor by domain. A simple total count may, therefore, mask the actual difference in magnitude and clinical significance of the individual domains.

Fourth, a formal test of trend would add to the interpretation of a graded association. Rising regression coefficients for each exposure category alone do not provide statistical evidence of a dose–response relationship. Moreover, the potential for synergy is not exhibited by the sum of the factors; interaction terms and formal tests of the effect modification would be needed to determine statistical interaction. Formal interaction tests should also be used to study differences in sex‐specific estimates, not by comparing p values within the sex‐specific subgroups [5, 6].

Last, more detailed regression diagnostics and sensitivity analyses would be beneficial when reporting the results. The reporting of the variance inflation factors does not imply that the assumptions of linear regression (linearity, homoscedasticity, normally distributed residuals, and lack of influential observations) have been met. In addition, the multiple‐imputation procedure requires more transparency in terms of the amount and pattern of missing data, the reasons for the number of imputations created, and the sensitivity of the results to complete‐case analysis [7, 8]. Additionally, because of the cross‐sectional nature of the study, reverse causation is also of major concern because those with higher IC may be more able to exercise, have dietary diversity, and be more socially involved.

Overall, this study provides valuable hypothesis‐generating evidence regarding multidomain lifestyle characteristics and IC. However, the results should be considered associations, not as evidence of independent causal effects. Having more information on the validation of the CFA‐derived outcome and formal exploratory trend and interaction testing would further enhance the statistical interpretation of this important work. Formal testing for missing data sensitivity would benefit the statistical interpretation of this important work. The statistical interpretation of this important work would benefit from greater transparency about validation of the CFA‐derived outcome, measurement invariance, predictor–outcome construct overlap, and formal test for missing data sensitivity.

Author Contributions

Deep Shikha: conceptualization, data curation, writing – original draft, writing – review and editing. Sunita Sharma: writing – review and editing.

Funding

The authors have nothing to report.

Disclosure

The authors have nothing to report.

Ethics Statement

The authors have nothing to report.

Data Availability Statement

Data sharing not applicable to this article as no datasets were generated or analysed during the current study.