Specific samples are summarized in Figure 1. We performed sensitivity analyses in which we included Kibei and Issei participants.
Analysis 1 evaluated prevalent dementia outcomes at the time of examination 4. In Analysis 2, participants were considered at risk if they did not have dementia at examination 4. We then assessed risk of incident dementia outcomes at examination 5, 6, or 7. In Analysis 3, participants were considered at risk if they did not have dementia at examination 5, and subsequent risk of incident dementia diagnoses at examination 6 or 7 was assessed (Fig. 1). Participants were censored at the time of any dementia diagnosis, drop out, or death in Analyses 2 and 3.
We used logistic regression for Analysis 1. Since stratification was used to determine who was further evaluated to identify prevalent cases of dementia, we weighted these analyses based on the probability of being selected. (Results without weighting were similar.)
We used Cox proportional hazards regression for Analyses 2 and 3. We estimated time of dementia diagnosis halfway between the last visit free of dementia and the visit at which dementia was diagnosed. Age served as the time scale for all Cox models.24
For all analyses, we used separate models for each outcome (any dementia, Alzheimer disease, and vascular dementia). We initially fit unadjusted models, and then adjusted for age and education. We screened for additional confounding by adding ApoE ϵ4, income, and health variables one at a time to the age- and education-adjusted models. A potential confounder was retained in the model if it met 2 criteria. First, the difference in the value of the β coefficient corresponding to language group membership for models including and excluding the potential confounder had to differ by at least 15%. Second, if the change was greater than 15%, the absolute value of the β coefficient in the model including the potential confounder had to be at least 0.1 (so adjusted odds or hazard ratios had to be less than 0.90 or greater than 1.11). The rationale for this compound criterion for confounding was that even large percentage differences in β coefficients close to zero could nevertheless still result in very small (and clinically insignificant) overall effects. We attempted a model including all possible confounders but this proved unstable because of small cell sizes. We did not look for interactions between possible confounding variables; we only examined their main effects. We tested the assumption of proportional hazards for the final adjusted models using Schoenfeld residuals; these assumptions were tenable for all models. All statistical analyses were carried out using Stata software, version 9.2.25
We used R26 to construct a plot of the point estimates and 95% confidence intervals (CIs) for the primary exposure of interest, “both spoke and read Japanese” versus “spoke but did not read Japanese” for Analyses 1, 2, and 3a.
Of the 3139 participants in Analysis 1, 561 (18%) neither spoke nor read Japanese, 1847 (59%) spoke but did not read Japanese, and 731 (23%) both spoke and read Japanese. Demographic characteristics, health risk factors, and health conditions are shown in Table 2. On average, those who both spoke and read Japanese were older and had lower household incomes; were more likely to have spoken Japanese as their first language and to have had trouble with English during the examination; were less likely to have ever smoked; and were shorter, with smaller head circumference. BMI categories, ApoE ϵ4 allele frequencies, and medical conditions were similar across groups. Rates of medical conditions were similar across language exposure groups even after adjusting for education and age (eTable 1, http://links.lww.com/A1451). The distribution of education across exposure groups is shown in eTable 2 (http://links.lww.com/A1451), interviewer descriptions of language use across exposure groups are shown in eTable 3 (http://links.lww.com/A1451), and the distribution of baseline cognitive test scores across exposure groups is shown in the eFigure (http://links.lww.com/A1451).
The distributions of baseline cognitive test scores were similar across the 3 groups, though mean baseline scores were lower in those who both spoke and read Japanese (Table 2). There were 154 prevalent cases of dementia, 74 of Alzheimer disease, and 43 of vascular dementia for Analysis 1. There were 236 incident cases of dementia, 138 of Alzheimer disease, and 45 of vascular dementia over 13,838 person-years of follow-up for Analysis 2. For Analysis 3, there were 125 incident cases of dementia, 80 of Alzheimer disease, and 20 of vascular dementia over 7178 person-years of follow-up. The number of dementia evaluations, dementia outcome prevalence rates, and weighted adjusted odds ratios (ORs) of prevalent dementia outcomes at the time of examination 4 are shown in Table 3. The point estimates for odds ratios for dementia outcomes associated with reading Japanese were all greater than 1.0; all confidence intervals included 1.0. The number of dementia evaluations, rates and adjusted hazard ratios (HRs) of incident dementia outcomes at examinations 5–7 (Analysis 2) are shown in the middle section of Table 3. The point estimate for Alzheimer disease was slightly less than 1.0 (adjusted HR 0.84 [95% CI = 0.56–1.28]). The reduced hazard for Alzheimer disease for the “both spoke and read Japanese” group was of similar magnitude as that for the “neither spoke nor read Japanese” group (0.79 [0.48–1.69]).
We compared self-reports of Japanese language proficiency at Honolulu Asia Aging Study examination 5 with those of the same participants at Honolulu Heart Program examination 1 (eTable 4, http://links.lww.com/A1451). Overall there was consistency in self-reporting of language proficiency across the life span, though a sizable proportion of participants had different assessments in late life than they had in midlife of their ability to speak and read Japanese.
Rates and adjusted HRs of incident dementia outcomes at examinations 6–7 associated with self-reported language use in midlife (Analysis 3) are shown in the bottom section of Table 3. As in Analysis 2, the point estimate for the adjusted HR for Alzheimer disease for the group that was highly literate in Japanese was less than 1.0 (0.86 [0.38–1.93]), of similar magnitude as for the group that neither spoke nor read Japanese (0.73 [0.43–1.26]). Less than high degrees of literacy in English were associated with increased risk of dementia outcomes, though confidence intervals included 1.0. When combined in the same analysis, no consistent pattern emerged for literacy in Japanese and English. As indicated in Table 3, only a single covariate—height—met our criteria for confounding, and in only a few analyses. Models that included and excluded height were very similar. Models including and excluding income were also very similar (eTable 5, http://links.lww.com/A1451).
There were 335 Kibei participants, most of whom read and wrote Japanese. Odds and hazard ratios were essentially unchanged when we included the Kibei.
Figure 2 provides a graphical display of our primary results. All 95% CI regions include the null. With 2 exceptions, the point estimates are at least 1.0, contrary to the relationship expected based on the cognitive reserve hypothesis.
Self-reported use of written Japanese did not appear to be protective for dementia, Alzheimer disease, or vascular dementia. Arguably, in 2 analyses, use of written Japanese was associated with a barely reduced risk for Alzheimer disease (adjusted HR <1.0 in Analyses 2 and 3a; both 95% CI's include 1.0). However, in those same analyses, use of neither spoken nor written Japanese was associated with similar protection (Table 3). This pattern of findings is more consistent with random variation than with neuroprotection garnered from richly developed neural circuitry required to process the numerous pictograms and ideographs involved with Kanji script.
There are many potential explanations for a negative result. These include a true negative finding and several different types of error. In general, the point estimates associated with use of written Japanese suggest harm rather than protection, which does not seem biologically plausible (Fig. 2). Recall bias is always a concern in studies of dementia. Here, exposure data were collected in midlife for Analyses 1 and 2, decades before dementia onset. In Analysis 3 we excluded individuals found to be demented at the time of exposure ascertainment. Some of these individuals may have had milder forms of cognitive deficits that were not diagnosed. Recall bias that was nondifferential between exposure groups would produce bias towards the null, decreasing our ability to find an effect. Furthermore, findings from Analysis 2—not subject to recall bias—were similar to those from Analysis 3.
Confounding is an important issue in observational studies. We adjusted models for age and education, and considered confounding related to ApoE genotype, self-reported income, and numerous health factors. In this population, Japanese speakers and those with proficiency with written Japanese were older and had lower incomes in late life than those who did not have proficiency with written Japanese (Table 2). This fact permits us to disentangle proficiency with written Japanese from socio-economic status (SES) in this study; similar studies in Japan would not be able to separate the effects of these factors. Height was the only covariate that met our criteria for confounding, for some analyses, despite our use of fairly liberal criteria for confounding. Models that included and excluded height produced essentially the same results. We specifically assessed head circumference as a potential confounder and did not find confounding related to this variable. Nevertheless, it is theoretically possible that some additional unmeasured confounder could exist, and that adjustment for that confounder would show protection from dementia outcomes associated with spoken or written Japanese.
We limited these analyses to second generation sons of Japanese immigrants (Nisei). We obtained similar results when we included first generation immigrants (Issei). Our rationale for excluding the Issei was that their life experiences were very different from the Nisei, and factors relating to proficiency with Japanese written language could be very different for the 2 generations of participants. Similarly, in our primary analyses, we excluded Kibei participants who received some of their education in Japan. Our rationale was that we would be unable to exclude written Japanese use (expected to be much more common among participants with some education in Japan) from other exposures in Japan. A secondary rationale was that an earlier study found an inverse relationship between childhood years spent in Japan and cross-sectional cognitive performance in the Honolulu Asia Aging Study.12 If this finding represented higher proportions with current or developing dementia, including these participants in our primary analyses would be expected to show less protection from dementia outcomes associated with written Japanese use. In any case, results when including and excluding the Kibei were essentially unchanged. Among the Nisei, we found an inverse relationship between years of (Western) formal schooling and written Japanese language proficiency. This may be due to different priorities for families who chose to enroll their sons in special Japanese language schools. These data provide us with the opportunity to determine whether the choice to study Japanese language among children of Japanese immigrants might have a relationship with risk of dementia decades later. We found no evidence that this choice had any impact on dementia outcomes. There are many reasons one might choose to learn the language of one's ancestors, or any second language. These data suggest that prevention of dementia should not be part of the rationale for encouraging second language acquisition.
Several limitations should be considered for these analyses. Proficiency with spoken and written languages was assessed only through self-report. We found some consistency across decades in self-reporting of use of written Japanese, but less consistency for proficiency with spoken Japanese (eTable 4, http://links.lww.com/A1451). We excluded people with prevalent dementia at the time of Honolulu Asia Aging Study examination 5 from Analysis 3, but the accuracy of self-reported exposures could have been affected by undetected mental problems. The power to detect a beneficial effect of proficiency with written Japanese for vascular dementia was somewhat limited. Cross-sectional studies of dementia can be biased if the level of participation varies across exposure groups, because lack of participation is strongly associated with the presence of dementia. While in theory this could lead to bias in the prevalent dementia cases analyzed in Analysis 1, it should not interfere with the incident cases analyzed in Analyses 2 and 3, where we found similar results. Test bias related to language may lead to biased results from the neuropsychological battery used to evaluate cognitive ability in participants with low cognitive test scores. We did not have additional data on acculturation, vocabulary, or reading level.27 We limited our analyses to dementia, though we found similar results when we analyzed rates of change in cognition. The cognitive test has insufficient coverage to address changes in specific cognitive domains. While we used 2 complementary exposure definitions, a more robust measure of exposure—such as measured written and spoken Japanese skill in midlife—would have strengthened our confidence in our findings. We do not have measures of frequency of exposure to written Japanese. These analyses should thus be considered exploratory.
In summary, we did not find evidence to support the cognitive reserve hypothesis. Proficiency with written Japanese language was not associated with reduced risk for dementia outcomes. Our data provide no support for the hypothesis that acquisition of a second language might help to prevent dementia.
1. Le Carret N, Auriacombe S, Letenneur L, Bergua V, Dartigues JF, Fabrigoule C. Influence of education on the pattern of cognitive deterioration in AD patients: the cognitive reserve hypothesis. Brain Cogn
2. Whalley LJ, Deary IJ, Appleton CL, Starr JM. Cognitive reserve and the neurobiology of cognitive aging. Ageing Res Rev
3. Stern Y. What is cognitive reserve? Theory and research application of the reserve concept. J Int Neuropsychol Soc
4. Stern Y. The concept of cognitive reserve: a catalyst for research. J Clin Exp Neuropsychol
5. Scarmeas N, Stern Y. Cognitive reserve: implications for diagnosis and prevention of Alzheimer's disease. Curr Neurol Neurosci Rep
6. Graves AB, Larson EB, Edland SD, et al. Prevalence of dementia and its subtypes in the Japanese American population of King County, Washington state. The Kame Project. Am J Epidemiol
7. White L, Petrovitch H, Ross GW, et al. Prevalence of dementia in older Japanese-American men in Hawaii: the Honolulu-Asia Aging Study. JAMA
8. Graves AB, Rajaram L, Bowen JD, McCormick WC, McCurry SM, Larson EB. Cognitive decline and Japanese culture in a cohort of older Japanese Americans in King County, WA:
the Kame Project. J Gerontol B Psychol Sci Soc Sci
9. Marian V, Spivey M, Hirsch J. Shared and separate systems in bilingual language processing: converging evidence from eyetracking and brain imaging. Brain Lang
10. Bialystok E, Craik FI, Freedman M. Bilingualism as a protection against the onset of symptoms of dementia. Neuropsychologia
11. Matsuoka K, Uno M, Kasai K, Koyama K, Kim Y. Estimation of premorbid IQ in individuals with Alzheimer's disease using Japanese ideographic script (Kanji) compound words: Japanese version of National Adult Reading Test. Psychiatry Clin Neurosci
12. Yano K, Grove JS, Masaki KH, et al. The effects of childhood residence in Japan and testing language on cognitive performance in late life among Japanese American men in Hawaii. J Am Geriatr Soc
13. Borenstein Graves A, Mortimer JA, Bowen JD, et al. Head circumference and incident Alzheimer's disease: modification by apolipoprotein E. Neurology
14. Syme SL, Marmot MG, Kagan A, Kato H, Rhoads G. Epidemiologic studies of coronary heart disease and stroke in Japanese men living in Japan, Hawaii and California: introduction. Am J Epidemiol
15. Havlik RJ, Izmirlian G, Petrovitch H, et al. APOE-epsilon4 predicts incident AD in Japanese-American men: the Honolulu-Asia Aging Study. Neurology
16. Teng EL, Hasegawa K, Homma A, et al. The Cognitive Abilities Screening Instrument (CASI): a practical test for cross-cultural epidemiological studies of dementia. Int Psychogeriatr
. 1994;6:45–58; discussion 62.
17. Jorm AF, Jacomb PA. The Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE): socio-demographic correlates, reliability, validity, and some norms. Psychol Med
18. Crane PK, van Belle G, Larson EB. Test bias in a cognitive test: differential item functioning in the CASI. Stat Med
19. Gibbons LE, McCurry S, Rhoads K, et al. Japanese-English language equivalence of the Cognitive Abilities Screening Instrument among Japanese-Americans. Int Psychogeriatr
20. Morris JC, Heyman A, Mohs RC, et al. The Consortium to Establish a Registry for Alzheimer's Disease (CERAD). Part I: clinical and neuropsychological assessment of Alzheimer's disease. Neurology
21. American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders
. 3rd ed. Washington, DC: American Psychiatric Association; 1987.
22. McKhann G, Drachman D, Folstein M, Katzman R, Price D, Stadlan EM. Clinical diagnosis of Alzheimer's disease: report of the NINCDS-ADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimer's Disease. Neurology
23. Chui HC, Victoroff JI, Margolin D, Jagust W, Shankle R, Katzman R. Criteria for the diagnosis of ischemic vascular dementia proposed by the State of California Alzheimer's Disease Diagnostic and Treatment Centers. Neurology
. 1992;42(3 pt 1):473–480.
24. Korn EL, Graubard BI, Midthune D. Time-to-event analysis of longitudinal follow-up of a survey: choice of the time-scale. Am J Epidemiol
25. Stata Corporation. Stata Statistical Software: Release 8
. College Station, TX: Stata Corporation; 2003.
26. R Development Core Team. R: A Language and Environment for Statistical Computing
. Vienna, Austria: R Foundation for Statistical Computing; 2005.
27. Manly JJ, Byrd DA, Touradji P, Stern Y. Acculturation, reading level, and neuropsychological test performance among African American elders. Appl Neuropsychol
Supplemental Digital Content
© 2009 Lippincott Williams & Wilkins, Inc.