final hum

article_2_melisa.docx

Home >Psychology homework help >final hum

Abstract

Three hundred and four participants in the Lothian Birth Cohort 1936 study took a validated IQ-type test at age 11 years and a battery of cognitive tests at age 70 years. Three tests of health literacy were completed at age 72 years; the Rapid Estimate of Adult Literacy in Medicine (REALM), the Test of Functional Health Literacy in Adults (S-TOFHLA), and the Newest Vital Sign (NVS). Participants who had a lower childhood IQ exhibited poorer performance on all three tests of health literacy taken in older adulthood. Relative cognitive change from age 11 to 70 and education were also important factors influencing performance on health literacy tasks, independent of childhood cognitive ability. It is important to understand the determinants of low health literacy in order to support individuals in managing their own health.

Research highlights

► Childhood IQ, relative cognitive change (age 11 to 70) & education were measured. ► Participants with lower childhood IQ had poorer health literacy scores at 72 years. ► Cognitive change & education influenced health literacy independent of childhood IQ. ► Understanding predictors of health literacy, could improve self-management of health.

Keywords

Health literacy;

IQ;

Cognition;

Health outcomes;

Cognitive change;

Old age

1. Introduction

Increasingly, individuals are expected to manage their health and partake in behaviors to either prevent or self-manage disease. For those who lack the cognitive or financial resources, health knowledge, or access to healthcare services, meeting these expectations can be difficult. Health literacy, “the capacity to obtain, process and understand basic health information and services needed to make appropriate health decisions” (Ratzan & Parker, 2000) has been pinpointed as one key resource that individuals use to promote, protect, and manage their health. Recently, several brief assessment tools to measure health literacy skills have been developed and considered for clinical use. There is a well-established association between lower scores on these measures and an increased likelihood of adverse health outcomes (DeWalt, Berkman, Sheridan, Lohr, & Pignone, 2004), such as higher risk of hospital admission (Baker et al., 2002 and Baker et al., 1998), poor self reported health (Baker et al., 1997 and Wolf et al., 2005), pursuing poor health behaviors (Von Wagner, Knight, Steptoe, & Wardle, 2007), lack of knowledge about preventive healthcare (Scott, Gazmararian, Williams, & Baker, 2002), and higher mortality (Baker et al., 2007 and Sudore et al., 2006). Individuals with poor health literacy may not fully understand or appropriately apply health-related knowledge to manage their health. Therefore, health service providers are obliged to consider health literacy an important issue.

General cognitive ability (intelligence) – the capacity to learn and reason well, spot and solve problems, and use abstract ideas (Gottfredson, 2008) – has also been associated with health outcomes (Gottfredson, 2004). Gottfredson and Deary (2004) suggest that intelligence is a significant contributor to good health because it aids the management of chronic illnesses and the avoidance of risk factors associated with acquiring an illness (Gottfredson, 2004).

Some researchers have reported associations between scores on health literacy measures and cognitive function. In a review of health literacy research, Von Wagner, Steptoe, Wolf, and Wardle (2009) presented possible causal pathways involving health literacy and health actions. In their conceptualization of health literacy, environment circumstances, formal education and experiential learning, as well as cognitive ability, age-related cognitive decline, and pre-existing knowledge were highlighted as important contributors to health literacy (Von Wagner et al., 2009). Baker, Gazmararian, Sudano, et al. (2002) found a linear relationship between health literacy and Mini Mental State Examination (MMSE) scores. Despite the MMSE being a crude measure of cognitive ability, the findings provided basis for future investigation. In a sample of patients with chronic heart failure, cognitive ability measures explained 13% of variance in health literacy scores, independent of demographic and education characteristics (Marrow et al., 2006). In a further study of patients with hypertension, cognitive ability (processing speed and working memory) accounted for 24% of the variance in health literacy scores (Levinthal, Morrow, Tu, Wu, & Murray, 2008). Both studies (Levinthal et al., 2008 and Marrow et al., 2006) found cognitive ability to be a possible explanation for the previously observed decline in health literacy scores with age (Baker, Gazmararian, Sudano et al., 2000), suggesting that the age-related decline in health literacy scores may be attributed to age-related cognitive decline. Beier and Ackerman (2003) assessed cognitive ability, age, gender, personality and interest, as determinants of health knowledge. Unlike other studies, they used measures of both crystallized and fluid intelligence and tested health knowledge using a health knowledge battery as opposed to established health literacy tests. When other factors were equal, participants' knowledge of health issues was correlated with intelligence. Both crystallized and fluid intelligence were predictors of health knowledge, with crystallized intelligence showing the stronger association.

It is important to explore further which psychological and other constructs are actually being measured by health literacy tests. Individual differences in cognitive abilities show high stability of individual differences over the lifespan (Deary, Whalley, Lemmon, Crawford, & Starr, 2000). In older and medically unwell individuals, cognitive test scores might reflect cognitive deterioration from pre-morbid levels in fluid cognitive abilities such as processing speed and reasoning (Hedden & Gabrieli, 2004). However, knowledge is generally more stable in older age than more fluid cognitive abilities (Hedden & Gabrieli, 2004). If health literacy test performance mostly reflects knowledge and habits, health literacy might be stable across age. If it reflects the ability to process new information related to health, it may mirror fluid cognitive abilities and therefore be impaired in the same way as fluid cognitive functions with age. Evidence for the latter was found when health literacy scores decreased with age (Baker, Gazmararian, Sudano, & Patterson, 2000).

A full account of the association between cognitive ability and health literacy must therefore ask whether health literacy levels are associated with prior cognitive ability and, in addition, with any change in cognitive ability between the prior assessment of cognitive ability and the time of assessment of health literacy. An association between prior cognitive ability and health literacy measured several decades later would suggest that health literacy to some extent reflects the life-long trait of intelligence. That is, in epidemiological parlance, health literacy would be confounded by (explained by, and a reflection of) people's relatively stable level of intelligence, even that which is assessed prior to educational differences. A more exclusive association with later, contemporaneously-assessed cognitive ability would suggest that health literacy reflects primarily current cognitive capability. Thus, health literacy might reflect more fluid-type cognitive skills, which are decoupled from prior intelligence to the extent that there has been relative decline from a prior level of intelligence. Both associations might co-exist. The importance of investigating these associations is supported by the well-established associations between childhood cognitive ability, mortality and morbidity (Batty et al., 2007 and Gottfredson, 2008). And it is important, practically as well as theoretically, to know whether older people's health literacy can be predicted from their prior and/or current intelligence levels; if the former is correct to an important extent, then even people with some cognitive decline might have effective health literacy levels. It is just as important to know whether cognitive decline affects health literacy scores, as then it would give health systems a signal to identify those who may be at risk. If this former is true (childhood intelligence predicts later health literacy skills as effectively as current intelligence), then what is most interesting is that the construct of health literacy, or in the way some tests measure health literacy, are probably not picking up fluid learning skills but, rather, crystallized health knowledge.

In the present report, we examined the extent to which IQ at age 11, and relative cognitive change between age 11 and age 70, contributed to health literacy test performance in a relatively healthy sample of older adults at about age 72. Other possible contributors to health literacy (gender, personality traits, education and social status) were controlled. By having both age 11 and age 70 intelligence measures we could effectively form a variable of lifetime cognitive change which was orthogonal to the prior intelligence level at age 11. Therefore, we could ask whether early life intelligence and the unrelated amount of lifetime cognitive change were both contributors to health literacy differences. This capability is almost unique to the Lothian Birth Cohort 1936 that was tested in the present study.

2. Method

2.1. Participants

This study used a sub-sample of participants from the Lothian Birth Cohort 1936 (LBC1936). The LBC1936 cohort consists of 1091 relatively healthy participants, born in 1936 and residing in Scotland. A large amount of cognitive and health information is held on these participants, including a measure of childhood cognitive ability. More comprehensive information on this cohort is described elsewhere (Deary et al., 2007). Three hundred and four (165 males and 139 females) of the LBC1936 participants were given three health literacy tests. These were the first 304 consecutive participants who attended for a MRI brain imaging session. All had Mini-Mental State Examination (Folstein, Folstein, & McHugh, 1975) scores of 23 or higher.

2.2. Cognitive measures

2.2.1. Cognitive ability in childhood (age-11 MHT IQ)

Childhood cognitive ability (childhood IQ) was measured using a version of the Moray House Test no. 12 (MHT), which participants took in the Scottish Mental Survey 1947 (SMS1947) at age 11 years (Scottish Council for Research in Education, Mental Survey Committee, Thomson, & Scottish Council for Research in Education, 1949). The MHT is a group-administered test with a 45-minute time limit and has a maximum score of 76. The test was validated against the Terman-Merrill revision of the Binet–Simon intelligence test (Deary, Whalley, & Starr, 2009). Individual differences in MHT scores are highly stable across the life-course from childhood to old age (Gow et al., 2010). The LBC1936 participants had higher than average childhood MHT scores (N = 293, mean = 49.92, SD = 11.97) by comparison with the Scottish population born in 1936 (N = 70,805; mean = 36.74, SD = 16.10) (Scottish Council for Research in Education et al., 1949). However, we have shown elsewhere that this attenuates the correlations by only a little, typically much less than .1 (Johnson, Gow, Corley, Starr, & Deary, 2010). MHT scores for the LBC1936 were corrected for age in days at time of testing and converted into IQ-type scores (mean = 101.12, SD = 15.34 for this subsample, standardized to this scale within the sample).

2.2.2. Cognitive ability in adulthood

When participants were about 70 years old, four measures of adult cognitive ability were obtained (Deary et al., 2007).

2.2.2.1. Age-70 MHT IQ

Participants re-sat the MHT under the same conditions applied at age 11. Scores were corrected for age in days at time of testing and converted to IQ scale.

2.2.2.2. General cognitive (g) factor

Participants completed six Wechsler Adult Intelligence Scale IIIUK (WAIS-IIIUK) (Wechsler, 1998) subtests: Matrix Reasoning, Block Design, Letter-Number Sequencing, Symbol Search, Digit Span Backwards, and Digit Symbol. Scores from the first unrotated principal component were used as measure of general cognitive ability (g) at age 70 years.

2.2.2.3. General cognitive speed (g speed)

Participants completed five tests of speed of information processing. Symbol Search and Digit Symbol were from the WAIS-IIIUK. Simple Reaction Time (RT) mean and 4-Choice RT mean were obtained using a self-contained box that has been used in large UK population studies (Deary, Der, & Ford, 2001). The reaction time tasks are highly reliable over a 1-day test interval (Deary & Der, 2005). There were 8 practice trials in each test and 20 and 40 test trials for the simple and 4 choice RT tests, respectively. The final test was Inspection Time; a visual processing task with a two-alternative forced choice design in which participants indicate which of two vertical lines of markedly different lengths was longer (Deary et al., 2004). Responses were not speeded; only the correctness of each response was recorded. After practice items, there were ten repeats of each of 15 stimulus durations ranging from 6 ms to 200 ms; stimulus durations appeared at random. The first unrotated principal component was used to generate individuals' scores on a general cognitive speed factor. All tests and the derivation of the components and scores for g and g speed are more fully described elsewhere ( Luciano et al., 2009).

2.2.2.4. Mini mental state examination (Folstein et al., 1975)

The MMSE is commonly used as a brief clinical screening test for possible dementia. Scores less than 27 are sometimes considered indicative of impending dementia. We ran all analyses described here both including and excluding the 17 participants with MMSE scores below 27 (range = 23 to 26). The results (not shown, but available from the authors) were very similar, so we report the analyses based on data from all participants.

2.2.3. Measures of relative cognitive change

Three measures of relative cognitive change were calculated. IQ change (MHT 11–70 years) was computed using standardized residuals from a linear regression of age-70 IQ on age-11 IQ. ‘Residualized g’ and ‘residualized g speed’ were computed as standardized residuals from linear regressions of age-70 g and g speed, respectively, on age-11 IQ. These variables reflected relative rather than absolute changes. Note that these measures of cognitive change were orthogonal to age 11 MHT IQ test scores.

2.3. Demographic measures

At age 70 years, participants provided general demographical information (Deary et al., 2007). Social class numbers were assigned based on the Classification of Occupations 1980 (Office of Population Censuses and Surveys, 1980) as applied to participants' highest status occupation; I (professional) to V (unskilled). Women were assigned their husband's class when it was higher. Class III (skilled) was divided into IIIN (non-manual) and IIIM (manual). Education was measured by the number of years participants spent in full-time education. Personality traits were measured using the NEO-Five Factor Inventory (Costa & McCrae, 1992), a 60-item inventory consisting of twelve items for each of the five personality traits: Openness (O), Conscientiousness (C), Extraversion (E), Agreeableness (A) and Neuroticism (N).

2.4. Measures of health literacy

At a mean age of 71.92 years, participants took three health literacy tests. This was a mean of about two years after the cognitive and other data described above.

2.4.1. Rapid Estimate of Adult Literacy in Medicine (REALM)

This is a word recognition task assessing participants' ability to pronounce common medical words, including terms for body parts and illness (Davis et al., 1993). The rationale is as follows: if participants encounter difficulty reading a word, they are unlikely to comprehend it (Davis, Michieulutte, Askov, Williams, & Weiss, 1998). Participants are asked to read aloud three lists of 22 words. List one begins with the most straightforward words (fat, flu, pill). The words increase in difficulty, finishing with obesity, osteoporosis, and impetigo. One mark is given for each word pronounced correctly. The REALM has been used by researchers in American (Davis et al., 1993) and British (Ibrahim et al., 2008) populations, has high validity and reliability, and correlates strongly with other measures of reading fluency (Parker, Baker, Williams, & Nurss, 1995) and tests of health literacy (Davis et al., 1993 and Parker et al., 1995).

2.4.2. Shortened Test of Functional Health Literacy in Adults (S-TOFHLA)

The S-TOFHLA is a shortened version of the TOFHLA (Baker et al., 1999 and Parker et al., 1995). This assesses both numeracy and reading comprehension. The numeracy section has four items. Participants are provided with medical directions (such as directions for taking medication) and asked questions that assess their understanding of the information. In the 36-item reading comprehension section, participants are required to select the missing word from four possible options. The S-TOFHLA was developed for an American sample and contains a section which refers to Medicaid applications. Thus, like other authors (Von Wagner et al., 2007), we substituted Passage B of the S-TOFHLA with Passage B of Von Wagner et al.'s British Version of the TOFHLA (Von Wagner et al., 2007). The S-TOFHLA is a reliable and valid measure of health literacy (Davis et al., 1998).

2.4.3. Newest Vital Sign (NVS)

For the NVS, participants are presented with a U.S. nutrition label from a pot of ice cream. They are required to answer six questions about it (Weiss et al., 2005). In addition to demonstrating reading, comprehension and numeracy abilities, participants must process and extract the relevant information while ignoring the distracters. Weiss et al. (2005) found NVS to be reliable with good internal consistency (Cronbach α = .76) and good criterion validity, correlating moderately with TOFHLA test scores (r = .59, p < .001).

2.5. Statistical analysis

SPSS Version 14.0 (2005) and Mplus statistical packages were used. Although age had a significant effect only on S-TOFHLA test scores (Adjusted R squared = .018; F (1, 257) = 5.78, p = .017), new health literacy measures were computed as standardized residuals from linear regressions of each test on age in days when the health literacy test was taken. Bivariate Spearman's rank correlations were used to describe the associations among the three measures of health literacy (REALM, S-TOFHLA and NVS) and their associations with other variables.

NVS scores were basically normally distributed so we used linear regression to assess the contributions of the predictor variables (sex, age-11 MHT IQ, years of full-time education, personality traits, social economic status, IQ change [MHT 11 versus 70 years], residualized g, and residualized g speed) to NVS test scores. There were pronounced ceiling effects in REALM and S-TOFHLA test scores. Of the 304 participants who completed the REALM, 221 achieved the maximum score of 66. Of the 259 participants who completed the S-TOFHLA, 109 achieved the maximum score of 100. In such situations, the power of ordinary least squares linear regression to detect true effects is markedly reduced ( Wilcox, 2005). Moreover, variable transformation cannot spread out the values clustered at the test ceiling. To overcome these difficulties, we used zero-inflated Poisson regression as implemented in Mplus. Zero-inflated Poisson regression relies on a mixture model ( Muthen & Shedden, 1999) in which the complete distribution of the outcome is approximated by mixing a logistic regression model for an underlying ‘zero-not zero’ process and a Poisson regression model for an underlying process generating deviations from 0 ( Lambert, 1992). Of course, our data were stacked not at 0 but at the test ceiling; we therefore reverse-scored the tests so that perfect scores were 0. Conceptually, the mixing procedure can be thought of as modeling one psychological process that produced mastery of the health literacy concepts tested, and another psychological process that produced varying degrees of deviation from mastery. Implementation of this model allowed the possibility that variables contributing to the two processes could be different.

3. Results

3.1. Descriptive statistics

Table 1 presents descriptive information about the study sample for demographic, cognitive and health literacy measures. Of the 304 participants, we collected complete data from 259 participants. During the early stage of testing, only two health literacy tests (REALM and NVS) were administered. Further into testing, the original S−TOFHLA was replaced with the more appropriate revised version, as described above. This accounts for the majority of the missing data. Mean age of participants when taking the health literacy tests was 71.92 years (range 71.00−72.82 years).

Table 1.

Descriptive statistics and correlations with standardized health literacy test scores.

	n	M (SD)	Correlation coefficients⁎
			NVS	S-TOFHLA	REALM
NVS	303	−.00 (.99)	1.00	.46	.28
S-TOFHLA	259	.02 (1.00)	.46	1.00	.46
REALM	304	.00 (1.00)	.28	.46	1.00
Gender (coding: males + 1, females + 2)	304	–	.03	.05	.17
Years of Education	304	10.86 (1.14)	.39	.38	.23
SES (Social Class as a Number)	300	2.32 (.93)	−.32	−.32	−.17
Age-11 MHT IQ	293	101.12 (15.34)	.50	.37	.29
Age-70 MHT IQ	301	100.63 (14.03)	.50	.54	.31
IQ Change (MHT 11–70 years)	290	−.03 (.98)	.13	.37	.08
g	297	.33 (.97)	.50	.53	.35
g speed	291	.24 (.98)	.39	.47	.33
Residualized g	286	.00 (1.00)	.28	.42	.12
Residualized g speed	281	.00 (1.00)	.17	.25	.07
Agreeableness	277	45.86 (5.39)	.11	.11	.06
Conscientiousness	277	47.48 (5.96)	−.03	.04	.08
Extraversion	274	39.48 (5.88)	−.09	−.08	.03
Neuroticism	281	27.95 (7.57)	−.13	−.18	−.11
Openness	277	38.03 (5.88)	.18	.09	.12

⁎

Correlations > +/− .17 all significant at .001, correlations ranging from +/−.12 to .13 significant at .05, correlations <+/−.12 are none significant; N ranges from 231 to 300; g, g speed, A, C, E, N and O measured at aged 70 years; NVS, S-TOFHLA and REALM measured at aged 72 years.

3.2. Associations among health literacy test scores

Spearman's rank correlations showed all three health literacy measures to be positively and moderately correlated: NVS and REALM rrho = .28, p < .01; REALM and S-TOFHLA rrho = .46 p < .01; NVS and S-TOFHLA rrho = .46, p < .01.

3.3. Relationship of independent variables to health literacy scores

3.3.1. Bivariate correlations

Table 1 illustrates that childhood MHT IQ had moderate positive correlations with all three measures of health literacy taken over 60 years later: REALM, rrho = .29; S-TOFHLA, rrho = .37; and NVS, rrho = .50. All were significant at p < .01. All three health literacy test scores had strong significant positive Spearman's rank correlations with measures of cognitive ability at age 70; g factor, g-speed factor, and age-70 MHT IQ ( Table 1). Participants who had higher g, g-speed, and age-70 MHT IQ scores, tended to score better on all measures of health literacy. Of these measures of adulthood cognition, age-70 MHT IQ was the most highly correlated with health literacy: NVS, rrho = .50; S-TOFHLA, rrho = .54; and REALM, rrho = .31. All were significant at p < .01. Social class and education had small to moderate significant Spearman's rank correlations with health literacy scores ( Table 1). Table 1 shows that NVS and S-TOFHLA scores had significant small negative relationships with neuroticism, and small positive correlations with openness. There was no significant association between the other personality traits (agreeableness, conscientiousness, and extraversion) and health literacy scores.

3.3.2. Regression analysis

3.3.2.1. NVS

A series of planned linear regression analyses using SPSS was conducted with NVS scores as the dependent variables (Table 2). Predictor variables were grouped in three models. There were three versions of the third model; they differed in the measure of relative cognitive change (IQ change 11–70 years, residualized g cognition, and residualized g speed) entered. The ‘enter’ method was used in each model.

Table 2.

Linear regression analysis predicting associations with NVS scores.

Variable	Model 1 β	Model 2 β	Model 3a β	Model 3b β	Model 3c β
Gender (coding: males + 1, females + 2)	.01	.01	.00	.01	.03
Age-11 MHT IQ	.45⁎⁎⁎	.33⁎⁎⁎	.36⁎⁎⁎	.35⁎⁎⁎	.34⁎⁎⁎
Years of education		.19⁎⁎	.18⁎⁎	.19⁎⁎	.22⁎⁎
SES		−.11	−.07	−.03	−.03
Agreeableness		.08	.08	.09	.09
Conscientiousness		−.02	−.02	−.04	−.03
Extraversion		−.07	−.06	−.06	−.09
Neuroticism		−.05	−.03	−.02	−.05
Openness		.03	.02	.02	.04
Relative cognitive change			.16⁎⁎	.24⁎⁎⁎	.19⁎⁎
Model fit	31.92⁎⁎⁎	10.43⁎⁎⁎	10.11⁎⁎⁎	11.26⁎⁎⁎	10.47⁎⁎⁎
Adjusted R2	.20	.25	.27	.30	.28
ΔR2	.20⁎⁎	.07⁎⁎	.02⁎⁎	.05⁎⁎	.03⁎⁎

Note. A, C, E, N and O measured at aged 70 years. Model 3a = IQ Change (MHT 11–70 years); Model 3b = residualized g; Model 3c = residualized g speed.

⁎⁎

p < .01.

⁎⁎⁎

p < .001.

Sex and age-11 MHT IQ were entered in the first model accounting for 20% of the variance in NVS scores (Table 2). Age-11 MHT IQ was the only significant contributor (sex β = .01, ns; age-11 IQ β = .45, p < .01). Education, personality traits (O, C, E, A and N) and social class, which were added in model two, accounted for a further 7% of the variance in NVS scores. Only age-11 MHT IQ and education were significant predictors (age-11 IQ β = .33, p < .01; education β = .19, p < .01). In model three, when we added IQ Change (MHT 11–70 years) as a measure of relative cognitive change, an additional 2% of variance in NVS scores was accounted for. In addition, age-11 MHT IQ and education remained significant predictors (age-11 IQ β = .36, p < .01; education β = .18, p < .05, IQ change β = .16, p < .05). When we substituted residualized g cognition into model three as a measure of relative cognitive change, an additional 5% of variance in NVS scores was accounted for. Age-11 MHT IQ and education remained significant (residualized g cognition β = .24, p < .01; age-11 IQ β = .35, p < .01; education β = .19, p < .01). Finally, when residualized g speed was used as the measure of relative cognitive change in model three, an additional 3% of variance in NVS scores was accounted for. Age-11 MHT IQ and education remained significant predictors (residualized g speed β = .19, p < .01; age-11 IQ β = .34, p < .01; education β = .22, p < .05). For the models, the VIF values were below 10 and tolerance statistics were all below .2. Thus, there was no cause for concern regarding collinearity within the data. The Durbin–Watson statistics confirmed that the errors in the three regressions were independent (Durbin–Watson statistic = 2.09, 2.18 and 2.13 respectively).

3.3.2.2. S-TOFHLA

Zero-inflated Poisson regressions were carried out using Mplus, with S-TOFHLA scores as the dependent variable (Table 3). These regressions yielded estimated independent variable contributions to S-TOFHLA scores for a hypothetical class of participants whose literacy was less than perfect (‘less than mastery’) and a second, not necessarily consistent, set of estimated independent variable contributions to likelihood of participant membership in a hypothetical class that had uniformly perfect S-TOFHLA scores (‘mastery’).

Table 3.

Poisson regression analysis predicting mastery on S-TOFHLA and REALM Tests.

	S-TOFHLA	REALM
	Model 1	Model 2	Model 3a	Model 3b	Model 3c	Model 1	Model 2	Model 3a	Model 3b	Model 3c
Less than mastery
Gender (coding: males + 1, females + 2)	.23⁎⁎⁎	.06	.12	.06	.01	.60⁎⁎⁎	.46⁎⁎⁎	.44⁎⁎	.50⁎⁎⁎	.15
Age-11 MHT IQ	.95⁎⁎⁎	.66⁎⁎⁎	.64⁎⁎⁎	.56⁎⁎⁎	.61⁎⁎⁎	.73⁎⁎⁎	.63⁎⁎⁎	.61⁎⁎⁎	.53⁎⁎⁎	.40⁎⁎
Years of education		.17	.08	.10	.21		.28	.31⁎	.35⁎	.28
SES		−.27	−.03	−.10	−.02		−.13	−.12	−.13	−.07
Agreeableness		.21	.21	.24	.26		.00	.01	.03	.22
Conscientiousness		.15	.19	.06	−.05		.41⁎⁎⁎	.42⁎⁎⁎	.39⁎⁎	.06
Extraversion		−.20	−.16	−.26	−.32		.14	.15	.13	−.45⁎
Neuroticism		−.18	−.11	−.03	−.10		.28	.28	.30	.33
Openness		−.04	−.20	−.21	.02		−.17	−.16	−.17	.35⁎
Relative cognitive change			.55⁎⁎⁎	.58⁎⁎⁎	.50⁎⁎⁎			.03	−.02	.21
Mastery
Gender	−.06	−.07	−.02	−.06	−.06	−.01	−.24	−.26	−.22	.12
Age-11 MHT IQ	.44⁎⁎⁎	.26⁎⁎	.42⁎⁎⁎	.29⁎⁎⁎	.25⁎⁎	.40⁎⁎⁎	.08	.11	.13	.28⁎
Years of education		.13	.05	.12	.14		−.15	−.32	−.21	−.17
SES		−.17⁎	−.09	−.07	-.10		−.11	.07	−.01	−.20
Agreeableness		.02	−.02	.01	.04		−.17	−.25	−.22	−.31⁎
Conscientiousness		.09	.09	.05	.06		−.10	−.24	−.14	.26
Extraversion		−.05	−.01	−.04	−.09		−.06	−.12	−.06	.53⁎⁎⁎
Neuroticism		−.07	−.03	−.03	−.07		−.35	−.39	−.31	−.26
Openness		−.02	−.01	−.01	.02		.37	.41⁎	.36	−.26
Relative cognitive change			.44⁎⁎⁎	.39⁎⁎⁎	.20⁎			.26	.27	.05
-2LL	−888.3	−679.79	−545.67	−589.54	−543.15	−287.55	−192.03	−186.75	−183.89	−168.22
Number of free parameters estimated	6	20	22	22	22	6	20	22	22	22

Note. To interpret these effects, the coefficients need to be exponentiated. A, C, E, N, and O measured at aged 70 years.

Model 3a = IQ Change (MHT 11–70 years); Model 3b = Residualized g; Model 3c = Residualized g speed.

⁎

p < .05.

⁎⁎

p < .01.

⁎⁎⁎

p < .001.

Sex and age-11 MHT IQ were entered in the first model. In both regressions, age-11 MHT IQ was a significant predictor. Sex was a significant predictor when assessing the process of producing varying degrees of deviation from mastery, but was not a significant predictor of producing mastery on the S-TOFHLA test. In both the less-than-mastery and mastery regressions, age-11 MHT IQ remained a significant predictor when education, personality traits (O, C, E, A and N) and social class were added in model two (Table 3). As with the linear regression above, there were three versions of the third model; each differed in the measure of relative cognitive change (IQ change [MHT 11–70 years], residualized g cognition, residualized g speed) entered. When we entered IQ change (MHT 11–70 years) as a measure of relative cognitive change, it significantly predicted both the less-than-mastery and mastery of the S-TOFHLA task. Age-11 MHT IQ remained significant. The same pattern was observed when we entered residualized g cognition and residualized g speed into model three as measures of relative cognitive change ( Table 3). As can be seen from the log likelihood scores (− 2LL), all three measures of relative cognitive change improved the model fit.

3.3.2.3. REALM

Zero-inflated Poisson regressions were also carried out using Mplus with REALM scores as the dependent variable (Table 3). Sex and age-11 MHT IQ were entered in the first model. Age-11 IQ was a significant predictor of both mastery and less than mastery of the REALM task. Sex was not a significant predictor of mastery on the REALM task but did significantly predict less than mastery performance. When education, personality traits (O, C, E, A and N) and social class were added in model two, sex and age-11 MHT IQ only remained significant predictors of less than mastery. With the exception of C, which was a significant predictor of less than mastery performance, education, personality traits (O, E, A and N) and social class were not significant contributors in model two. Sex and age-11 MHT IQ remained significant (Table 3). To maintain consistency with the previous analyses, three versions of the third model were computed; each differed in the measure of relative cognitive change (IQ change [MHT 11–70 years], residualized g cognition, residualized g speed) entered. When IQ change (MHT 11–70 years) was entered, we found it was a significant predictor of mastery but was not a significant predictor for the process of producing varying degrees of deviation from mastery on the REALM test. Neither residualized g cognition, nor residualized g speed were significant predictors of either less than mastery or mastery of the REALM task. In Table 3, it can be seen that age-11 MHT IQ was a significant predictor of less than mastery when all the measures of relative cognitive change were controlled.

4. Discussion

In this study we investigated the extent to which age-11 IQ, relative cognitive change, age-70 IQ, gender, personality traits, education and social status contributed to health literacy test performance in a single relatively healthy sample of older adults. Based on previous research linking cognitive abilities to health literacy test performance (Baker et al., 2008, Baker et al., 2002, Levinthal et al., 2008 and Marrow et al., 2006), we hypothesized that both childhood cognitive ability and the relative change in cognitive ability between age-11 and age-70 years would be associated with health literacy measured in later life.

4.1. Effects of childhood IQ (Age-11 MHT IQ)

Consistent with the hypothesis, childhood IQ (age-11 MHT IQ) was significantly related to performance on all three health literacy tasks measured almost 60 years later. In all of our analyses, age-11 IQ remained a significant influence, despite the addition of gender, currently estimated personality traits, education, social status, and measures of relative cognitive change since childhood. The mean age 11 MHT score of our sample was relatively high and its range somewhat restricted compared with the mean age 11 MHT score for Scotland in 1947; therefore caution must be taken when making generalizations to the rest of the population, and it is likely that our coefficients of association are underestimates. They may also be underestimates because they were adjusted for level of education, and education probably shared common causal variance with MHT scores. The findings suggest that some of the variance in health literacy scores can be explained by the life-long trait of intelligence and the part it plays in accruing health literacy skills. This supports the view that, to some extent, tests of health literacy and health knowledge reflect general abilities to learn without assistance and solve problems across one's lifespan (Gottfredson, 2004). These are general abilities which can be applied to manage one's own health and aid in the acquisition of specific health knowledge and skills of the sort an individual may require when diagnosed with an illness.

Lifetime cognitive ability may be associated with either stronger health constitution from birth or a lifetime accumulation of health knowledge and habits or both. With regard to the latter, health in old age may be better than it otherwise would have been even if cognitive decline has set in and the health literacy is no longer useful. Age-related cognitive decline undermines both the ability to use previously accumulated knowledge and the ability to acquire and use new knowledge, but particularly the latter (Hedden & Gabrieli, 2004). When clinicians are assessing whether individuals are able to manage their own health, it is one's current ability and health literacy test scores that are pertinent.

In addition to their usefulness in working with individuals in clinical situations, health literacy tests have also been used in epidemiological studies to help understand health outcomes. To develop successful interventions to improve health literacy, it is important firstly to identify the constructs that are being assessed by the health literacy tests and the pathways through which health literacy is associated with mortality and morbidity. Authors commonly infer that associations between health literacy and health outcomes indicate that health literacy may be a direct and distinct (from more general cognitive ability) predictor of health outcomes (DeWalt et al., 2004). The current study suggests this may not be the case. Although limited in that we did not measure health outcomes, the unique data we hold includes access to participants' childhood IQs. From our findings, we are able to suggest that the accumulation and maintenance of health literacy may be one of the avenues through which childhood intelligence influences adult health. If so, it is premature to conclude that health literacy is a direct and distinct (from general intelligence) predictor of health outcomes. Thus, it is possible that any apparent effect of health literacy on health outcomes is confounded by prior intelligence, or it might be that the effects of prior intelligence on health outcomes are mediated via the gaining of health literacy; these possibilities require testing in future studies with appropriate data. The results imply that, along with cognitive change and education, epidemiologists should be considering intelligence when implementing interventions to improve health literacy, mortality and morbidity.

4.2. Effects of relative cognitive change

Three measures of relative cognitive change were used to assess the change in cognitive ability between youth and time of assessment of health literacy (IQ change, residualized g cognition and residualized g speed). All three measures were significant predictors of health literacy when measured using the S-TOFHLA and NVS tests but not using the REALM. These S-TOFHLA and NVS associations remained significant while controlling for gender, age-11 IQ, personality traits, education and social status. As the measures of relative cognitive change were independent of childhood cognitive ability, the results suggest that, in addition to the influence of the life-long trait of intelligence, factors associated with specific and relatively recent circumstances that have affected cognitive function may contribute to individual differences in health literacy scores. We have assumed that cognitive change between age 11 and old age is primarily indexing cognitive aging. However, it should be recognized that the change in cognitive function between 11 and the 70 s reflected a variety of factors; i.e., individual differences in improvements in abilities from age 11 to mature adulthood, and individual differences in declines after peaks in adulthood. Given the age of our cohort we think it likely that in terms of relative importance, aging outweighed early life maturation factors in its impact on the individual differences in changes in cognitive function from 11 to the 70s.

4.3. Perspective on current cognitive function and relative cognitive change

We structured our regression analyses to focus on relative cognitive change between ages 11 and 70 years because we perceived that the degree to which later-life health literacy reflects the generally stable trait of lifelong cognitive ability has not generally been appreciated. Equivalently, of course, we could have structured our analyses to focus on the extent to which age 70 cognitive function mediated the association between age 11 IQ and age 72 health literacy. Had we done this, we would report that an important association between age 11 IQ and age 72 health literacy was fully mediated by age 70 cognitive function for the S-TOFHLA and NVS, and substantially mediated for the REALM. From this alternative perspective,1 our results indicate that later-life cognitive function is directly related to health literacy, and early life cognitive function is related basically only indirectly, through its rather substantial contribution to later-life cognitive function. This places the emphasis where it belongs for evaluating the likely success of interventions intended to improve health literacy; i.e., on the relevance of current cognitive function. It obscures, however, an important feature in understanding of the developmental mechanisms involved in levels of later-life health literacy. This feature is their association with more general, lifelong cognitive function, the causes of which have, by age 11 years, established a rank order among people that substantially carries forward to cognitive ability and health literacy in the eighth decade. It was this feature to which we were particularly interested in calling attention, especially because few if any other datasets exist which can test this.

4.4. Education

Years of education were significantly associated with NVS scores after controlling for age-11 IQ and gender. This could be because people who received more years of education learned and acquired more health literacy skills at school itself and/or because their general intelligence and education helped them acquire these skills post schooling. We suggest that the latter is more probable.

4.5. Health literacy tests

Although NVS scores were basically normally distributed, pronounced ceiling effects existed in REALM and S-TOFHLA scores, inhibiting discrimination among high scoring participants. Statistically significant associations are more likely to occur when data has a normal distribution. This may be the reason that significant associations were found between relative cognitive change and NVS and S-TOFHLA scores, but not REALM scores. This issue is not unique to the current study. Reading recognition tasks such as the REALM are suitable for highlighting individuals who have low reading ability but not necessarily for distinguishing among higher attaining individuals (Davis et al., 1998). The NVS is more sensitive and facilitates making distinctions among higher scoring participants (Weiss et al., 2005). Our participants had higher than average cognitive abilities, which probably produced greater ceiling effects than might be observed in other samples.

However, there exist differences in the specific skills assessed by the tasks which may also have contributed to some of the variation in results across the three health literacy tests. The NVS and S-TOFHLA both assess the ability to read and comprehend novel information, as well as numeracy skills. In addition, for mastery on the NVS, document literacy skills are required (Baker, 2006). The REALM is a word recognition task. There is a stronger association between MHT scores and NVS and S-TOFHLA scores, than between REALM and MHT scores (Table 1). This may be because they are assessing similar cognitive abilities. Like the MHT, the NVS and S-TOFHLA require mostly fluid intelligence whereas the REALM requires crystallized intelligence.

The health mastery field of research needs more consistency in the complexity of assessment tools used, especially when using the tests in clinical settings. Currently the same participant can be identified as having good health literacy when measured using the REALM and the S-TOFHLA, but relatively poor health literacy when using the NVS. This means thinking more clearly not just about consistency across measures, but about the rationale for measuring health literacy. The specialized statistical analyses we used (zero-inflated Poisson regressions) to avoid distortions in results that can occur in the presence of the kinds of ceiling effects present in the REALM and S-TOFHLA make clear the issue involved. The zero-inflated Poisson regressions we used rely on the assumption that two processes underlie the test scores: one that propels participants to produce perfect scores and one that propels participants to produce scores that deviate to varying degrees from perfection. Doing this implies that the test is presumed to measure some level of health literacy that can be considered adequate. It is interesting to consider whether this was what the developers of the REALM and S-TOFHLA intended.

4.6. Conclusion and implications

Our findings support the incorporation of external variables (environment influences, formal education and experiential learning) as well as individual differences (cognitive ability, age-related cognitive decline, and pre-existing knowledge) when considering health literacy (Von Wagner et al., 2009, Wolf et al., 2009 and Wolf et al., 2010). These factors need to be considered by both researchers and clinicians when conceptualizing, and developing policies to improve health literacy. When it comes to managing health, individuals with poor health literacy may not fully understand or efficiently apply required knowledge to enable the maintenance of health and prevention of illness.