Although age is the strongest known risk factor, not all people who reach old age develop dementia before they die. Recommendations on potentially modifiable risk factors for the prevention of dementia are based on evidence that is, at best, moderate in strength. There are major calls to strengthen the evidence on potentially modifiable risk factors of dementia. The European Prospective Investigation into Cancer in Norfolk (EPIC-Norfolk) is a prospective population study of 25 639 men and women aged 40–79 years first recruited in 1993-1997, who attended a health examination. Subsequent follow-ups have involved self-report of health and lifestyle and further health examinations. Cognitive measures (7 tests assessing a range of domains) were introduced as part of a third health examination between 2006 and 2011 (including data from a pilot phase 2004–2006) and are available on 8585 individuals. Almost complete follow-up for disease outcomes, including dementia and mortality, has been established via linkage to health records. Education was strongly associated with cognitive function for all abilities tested. Cross-sectional and prospective analyses showed those who were physically inactive during work, were less likely to have poor cognition (bottom tenth percentile of a composite cognition score); Odds Ratio (OR) = 0.68 (95% Confidence Interval or CI 0.54, 0.86 P=0.001). In contrast, inactivity during leisure time was associated with increased risk of poor performance in the cross-sectional analyses, although this association was not observed in the prospective analyses. Poor cognition was independently associated with higher risk of all-cause mortality and predictive of incident dementia. Associations were observed for the composite score (global cognition) as well as specific cognitive abilities. Poor cognition in four or more tests was associated with ten-fold increased risk of developing dementia compared with those who did not perform poorly in any test OR=10.82 (95% CI 6.85, 17.10 P<0.001). Addition of each cognitive measure strengthened prediction models of dementia further, Area under the curve (AUC) = 0.85 (95% CI 0.82, 0.87 P<0.001), with the single test for episodic memory having the strongest influence.

Routinely collected health records are increasingly encouraged and used for epidemiological research for dementia outcome ascertainment. The linkage of the cohort to diverse routine records enabled comparison of these data sources. I provide evidence for the need of a more consensus-based approach to the methods of data collection, coding and interpretation of health data across all sources examined (hospital inpatient, mortality and mental health services datasets). In summary, the findings from this dissertation suggest the relationships between lifestyle factors, poor cognition and dementia are complex. For stronger evidence, future studies need to account for characteristics of the sample population and for the test used to measure cognition. Furthermore, there is a need for a more nuanced approach to the way the exposure of interest as well as dementia outcomes are measured and to adequately address the issue of potential confounding.