Fine motor performance and global cognitive function with incomplete data in the population-based CHRIS study

R. Melotti${}^a$, M. Banerjee${}^{a,b}$, A. J. Kaat${}^c$, M. Gögele${}^a$, F. Del Greco M.${}^a$, P. P. Pramstaller${}^a$, R. Gershon${}^c$, C. Pattaro${}^a$ and S. Wang${}^{a,c}$

${}^a$Institute for Biomedicine, Eurac Research, Italy, ${}^b$Department of Biostatistics, University of Michigan School of Public Health, MI, USA, ${}^c$Department of Medical Social Science, Feinberg School of Medicine, Northwestern University, IL, USA

Fine motor performance is a potential marker of cognitive impairment. Global cognitive function and its relationship with spiral drawing features were assessed under self-selection and missingness in the Cooperative Health Research in South Tyrol (CHRIS) study.

There were 8,230 out of 13,367 (61.6%) CHRIS participants without tremor related conditions and 6 replicates of the digital spiral drawing test.[1] Among these, 7,401 had full 30-item response, 216 had only partial item response, and 613 had full missing items to the Mini-Mental State Examination (MMSE). Mean drawing error from the smoothed spiral trace and median speed were averaged features of fine motor performance. A multiple imputation algorithm was developed, tested for trace convergence and applied with sampling weights to rescue missing data. Proportions of MMSE score categories [30, 29-27, 26-24, <24], as well as associations between the MMSE score and spiral features in adjusted generalized linear models were estimated either including or excluding imputed missing data.

Estimated proportions of MMSE score lowest two categories [26-24, <24] were 9.4$\%$ and 1.0$\%$, 12.0$\%$ and 1.7$\%$, and 12.3$\%$ and 2.2$\%$ in non-imputed, non-imputed weighted, and imputed weighted data, respectively. MMSE score had an inverse linear association with both mean drawing error and median speed in all models. However, the size of these associations was consistent between complete and imputed data for mean drawing error, not for median speed.

The proportion of low cognitive function was underestimated by self-selection and missingness in the CHRIS data. However, MMSE associations with different features of fine motor performance were not equally affected.

Keywords: CHRIS, Multiple Imputation, Selection Bias