Long‐term dementia risk prediction by the LIBRA score: A 30‐year follow‐up of the CAIDE study

Objective As no causal treatment for dementia is available yet, the focus of dementia research is slowly shifting towards prevention strategies. Therefore, this study aimed to examine the predictive accuracy of the “LIfestyle for BRAin Health” (LIBRA) score, a weighted compound score of 12 modifiable risk and protective factors, for dementia and mild cognitive impairment (MCI) in midlife and late‐life, and in individuals with high or low genetic risk based on presence of the apolipoprotein (APOE) ε4 allele. Methods The LIBRA score was calculated for participants from the Finnish Cardiovascular Risk Factors, Aging and Dementia (CAIDE) population‐based study examined in midlife (n = 1024) and twice in late‐life (n = 604) up to 30 years later. Diagnoses of MCI and dementia were made according to established criteria. Cox proportional hazards models were used to assess the association between LIBRA and risk of dementia and MCI in models adjusted for sex and education (age as timescale). Results Higher midlife LIBRA scores were related to higher risk of dementia (hazard ratio [HR] = 1.27; 95% confidence interval [CI], 1.13‐1.43) and MCI (unadjusted model: HR = 1.12; 95% CI, 1.03‐1.22) up to 30 years later. Higher late‐life LIBRA scores were related to higher risk of MCI (HR = 1.11; 95% CI, 1.00‐1.25), but not dementia (HR = 1.02; 95% CI, 0.84‐1.24). Higher late‐life LIBRA scores were related to higher dementia risk among apolipoprotein E (APOE) ε4 non‐carriers. Conclusions Findings emphasize the importance of modifiable risk and protective factors for dementia prevention.

Dementia is one of the core challenges facing our aging society. 1 Prevention is crucial given that there are no treatments available to stop or reverse dementia. 2 For this, evidence-based preventive strategies are needed, focusing on modifiable risk factors. [3][4][5] Such strategies should include the early identification of persons at risk for dementia, eg, in midlife, in order to target risk factors before irreparable brain damage and cognitive symptoms occur. The "LIfestyle for BRAin Health" (LIBRA) score was developed based on a systematic literature review and expert consensus study. 6 It consists of modifiable risk and protective factors that are promising targets for preventive strategies and thus reflects an individual's prevention potential for dementia. So far, LIBRA has been shown to explain variance in cognitive functioning and dementia risk in various population-and patient-based prospective cohort studies. [7][8][9][10][11] Yet, more research is needed into the predic- Therefore, the overall aim of the present study was to investigate the predictive validity of the LIBRA score in the longitudinal population-based Cardiovascular Risk Factors, Aging and Dementia (CAIDE) study. 12 The first aim was to investigate the performance of the LIBRA score, measured in the same individuals, midlife (40)(41)(42)(43)(44)(45)(46)(47)(48)(49)(50) years) and late-life (65-79 years) for predicting the risk of incident dementia and MCI up to 30 years later. The second aim was to investigate potential differences between persons with high and low genetic risk for dementia (APOE ε4 carriers versus non-carriers) regarding the relations of the LIBRA score with dementia and MCI risk.

| The CAIDE study
CAIDE participants were randomly selected from four independent population-based samples of the North Karelia Project and the FINMONICA study. Participants were examined in midlife 13-15 in 1972, 1977, 1982, or 1987. In 1998, a random sample of 2000 individuals aged 65 to 79 years from the towns and surroundings of Kuopio and Joensuu in Eastern Finland was invited for the first reexamination. 12 Of these, 1449 persons (72.5%) participated. A second re-examination took place between 2005 and 2008. Of the initial 2000 individuals, 1426 were still alive and living in the target areas in 2005, and 909 (63.7%) attended the re-examination ( Figure S1). In total, 1511 participants attended at least one re-examination, and 750 attended both re-examinations, with completed cognitive assessments. Mean follow-up time (SD) from midlife was 20.9 (4.9) years until the first re-examination, and 28.9 (5.0) years until the second re-examination. The study was approved by the local ethics committee of Kuopio University and Kuopio University Hospital, and written informed consent was obtained from all participants.

| Assessment of MCI and dementia
In both re-examinations, cognitive status was assessed with a threestep protocol (screening, clinical, and differential diagnostic phase). Individuals scoring less than or equal to 24 on the Mini-Mental State Examination (MMSE) 16 at screening were referred for additional clinical assessments. In 2005 to 2008, individuals with less than or equal to 24 points or a decline greater than or equal to three points on MMSE, less than 70% delayed recall in the CERAD word list, 17 or an informant expressing concerns about the participant's cognition were referred for more assessments. Both re-examinations had a clinical phase with detailed neuropsychological and medical assessments and a differential diagnostic phase. A review board consisting of a senior neurologist, senior neuropsychologist, study physician, and study neuropsychologist ascertained the final diagnosis based on all available information. In both re-examinations, diagnosis of MCI and dementia were made according to established criteria. [18][19][20]

| Design of the present study
For midlife LIBRA score analyses, the study population included 1024 CAIDE participants with available data for the midlife LIBRA score and who attended at least one re-examination with completed cognitive assessments ( Figure 1). Outcomes were incident dementia (n = 84) or MCI (n = 151) as diagnosed at the CAIDE re-examination visits.

| LIBRA score
The LIBRA score was developed after triangulation of results from a systematic literature review on risk and protective factors for dementia and an expert consensus study, 6 as part of the European (FP7) INnovative, Midlife INtervention for Dementia Deterrence (In-MINDD) project. 21 It consists of 12 modifiable risk and protective factors that can be targeted by tailored lifestyle interventions and primary prevention. Risk factors are coronary heart disease, diabetes, hypercholesterolemia, hypertension, depression, obesity, smoking, physical inactivity, and renal disease. Protective factors are low-to-moderate alcohol use, high cognitive activity, and healthy diet. A weight is assigned to each factor, based on the factor's relative risk (Table S1). 6 Weights are then standardized and summed up to yield the final LIBRA score (range from nurse verified the answers, and measured height, weight and blood pressure. A venous blood sample was taken to determine serum total cholesterol. APOE genotype was determined from blood leucocytes using polymerase chain reaction and HhaI digestion. 26 In CAIDE, data were available for all LIBRA factors, except for cognitive activity. Factors were dichotomized based on previously used cut-offs (Table S1). Hypertension was defined as a systolic blood pressure ≥ 140 mmHg or diastolic blood pressure ≥ 90 mmHg. Participants were classified as obese if their body mass index (BMI) exceeded 30 kg/m 2 . The cut-off point for hypercholesterolemia was greater than or equal to 6.5 mmol/L. Diabetes and coronary heart disease were based on self-reports of diagnoses made by a physician and diagnoses from the Finnish Hospital Discharge Register. Renal disease was based solely on diagnoses from the Finnish Hospital Discharge Register. A cut-off point for depressive symptoms was created based on the sum scores of two questions related to feelings of hopelessness. 27 Persons who engaged in physical activity at least twice a week, lasting at least 20 to 30 minutes each occasion, and causing sweating and breathlessness, were regarded as physically active. Low-to-moderate alcohol consumption was based on categorized frequency of alcohol use. For smoking, participants were divided into ever smokers and never smokers. Since information on diet was available for only a small group of approximately 240 participants in midlife, analyses including diet were conducted separately. Adherence to a healthy diet was based on previously used cut-offs of the CAIDE Healthy Diet Index. 28 Observed LIBRA scores ranged from −2.7 to +12.7.

| Statistical analysis
To examine differences in risk factors and demographic variables between participants with incident dementia, MCI and controls, oneway analysis of variance (ANOVA) and χ 2 tests were used. Cox proportional hazard regression models were used to test associations between the LIBRA score and dementia or MCI risk. Harrell concordance rate (C statistic) for censored data was calculated to examine predictive accuracy. The C-statistic indicates the probability that a randomly selected participant who developed the outcome (dementia or MCI) had a higher risk score than a participant who did not develop the outcome. It is equal to the area under the Receiver Operating Characteristic curve and ranges from 0.5 to 1. Model 1 was unadjusted (except for age as the time scale, see below), and model 2 was adjusted for the covariates sex and education. In addition, we tested for a multiplicative interaction between LIBRA score and APOE genotype (ε4 carriers versus non-carriers). All analyses were done in Stata 14 (StataCorp LP, TX), and the level of statistical significance was P < .05.
For midlife LIBRA analyses, stcox was used with age as time scale and age at first assessment (midlife) as origin. Right censoring was defined as the age at the first dementia diagnosis in CAIDE re-examinations or end of study (date of last available CAIDE re-examination).
The same approach was used in analyses with MCI as outcome (ie, considering date of first MCI diagnosis). The proportional hazard assumption was assessed based on the Schoenfeld residuals.
For late-life LIBRA analyses, stcox was used, with age as time scale and age during the later assessment wave in 1998 as origin. Censoring age at follow-up was defined analog to the midlife analysis at the 2005 to 2008 CAIDE re-examination.

| Population characteristics
Population characteristics are shown in Table 1. As expected, individuals with incident dementia were older, had a lower education level, and were more often APOE ε4 carriers (all P < .01). Sex distribution was not significantly different between diagnostic groups.
Midlife LIBRA scores were higher in participants who subsequently developed dementia (total population: n = 1024; mean LIBRA score

| Midlife LIBRA and incident dementia and MCI
Performance of the LIBRA score in predicting dementia or MCI is shown in Table 2 (continuous LIBRA score) and Figure 2  In the small group of participants with available midlife diet data, the LIBRA score including diet was not significantly related to dementia or MCI risk, although C-statistic values increased (Table 2). No significant interactions between midlife LIBRA score and APOE ε4 carrier status were found in any of the models (results not shown).

| Late-life LIBRA and incident dementia and MCI
Late-life LIBRA scores were not significantly related to dementia.

Significant interactions between late-life LIBRA scores and APOE
ε4 carrier status were found in relation to dementia risk (P = .011).

| DISCUSSION
In a general Finnish population, higher midlife LIBRA scores were related to higher risk of developing dementia or MCI up to 30 years   Score versus 0.65 for midlife LIBRA score in the present study). 31 The CAIDE risk score is based on a data-driven approach within the CAIDE cohort study and is developed to maximize prediction of persons atrisk of dementia. 32 The C-statistic improved after adding diet to the LIBRA score, although the small number of participants with midlife diet data, and lack of late-life diet data limited these analyses. The lack of data on cognitive activity in the present study may also have affected the LIBRA score predictive performance, given the growing evidence that high engagement in cognitive activities is associated with lower risk for cognitive impairment or dementia. 6 Higher midlife LIBRA scores were significantly related to higher dementia risk even after taking age, education, and sex into account, genetic and/or non-genetic risk and protective factors may also be important.
The major strengths of the present study include the populationbased design, the long follow-up period starting already in midlife, and detailed assessments at several time points during the second half of the life-course. For these reasons, CAIDE is a highly suitable cohort study for external validation of LIBRA. The risk prediction in this study applies only to individuals who actually survive to older ages, when they are more likely to develop dementia. As an inevitable part of all studies with a long-follow up period, mortality, and non-participation are often linked to poorer health, ie, people who are more likely to develop dementia, or die before dementia onset.
This might have led to selection of a healthier sample and therefore may result in an underestimation of the "true" association. While data for most LIBRA factors were available in the CAIDE study, reliance on register diagnoses for some chronic conditions (ie, only conditions severe enough to require hospitalization) may have affected the predictive performance of the LIBRA score. Also, complete data on pharmacological treatment for the included risk factors/ conditions were not available. In addition, interactions between risk factors were not taken into account in the design of the LIBRA score. Knowledge on possible interactions between risk factors is still incomplete in the available literature, and therefore more research on this matter is needed.
Findings from the present study emphasize the role of modifiable risk and protective factors in the development of MCI and dementia.
The LIBRA score may be useful for educational and motivational purposes in public health initiatives by emphasizing areas amenable to preventive lifestyle measures and for identifying at-risk individuals who may benefit from lifestyle interventions.