Sometimes, the estimate for the variance of a random factor turns out to be negative. The problem of negative variance estimates in mixed and random effects models has plagued statisticians since their introduction (since the introduction of the models, that is, not since the introduction of statisticians :).
We'll present Smith&Murray's/Hocking's (very neat) solution to the
problem, raise some questions on the applicability and interpretation of
their solution, and touch on some further work that remains to be done
in this area.
Methods: A cohort of 2578 elders cognitively intact at baseline, aged 65 years and older was sampled from the Group Health Cooperative members in Seattle area from 1994 to 1996, then followed biennially. A multidimensional questionnaire assessing demographic characteristics, medical history, cognitive function measured by Cognitive Abilities Screening Instrument (CASI), behavior factors, psycho-social factors, and memory & functioning, as well as performance-based physical function tests were administrated at baseline and follow-up visits. Three outcome measures: activities of daily living (ADL), instrumental activities of daily living (IADL) and performance-based physical function (PPF) were constructed and evaluated simultaneously. Multiple linear regressions with Generalized Estimating Equation (GEE) were used to identify significant predictors of functional outcomes.
Results: Medical conditions including diabetes, arthritis, osteoporosis, hypertension, coronary artery disease and cerebrovascular disease, as well as low cognitive functioning, depression and smoking were associated with increased limitations in doing ADL and IADL activities and poorer performance on PPF. Regular exercise and moderate alcohol intake were associated with fewer limitations in doing ADL and IADL activities and better performance on PPF. Over the follow-up period, coronary disease and depression were associated with increased rates of functional decline, and moderate alcohol consumption associated with decreased rates of functional decline. We identified significant interactions such as exercise with coronary disease, alcohol use with cerebrovascular disease, alcohol use with arthritis and cognitive function with cerebrovascular disease.
Conclusions: Medical conditions such as diabetes, osteoporosis, arthritis,
hypertension, coronary artery disease and cerebrovascular disease, as well
as depression contribute to development of functional limitations in older
persons. Favourable lifestyles such as exercising regularly, drinking moderate
amount of alcohol and not smoking enhance maintaining good functioning
of elders. Exercising offers extra benefit to coronary disease patients,
and moderate alcohol use is beneficial to cerebrovascular disease and arthritis
patients. Our findings suggest possible interventions for delaying the
functional decline and maintaining functional independence of older adults,
since medical conditions, depression, as well as behavioral factors are
amenable to change.
Last Modification: 08 December 2000