Associations and Implications of Low Health Literacy in Parkinson's Disease

Abstract

Background

Low health literacy (HL) indicates a limited ability to understand and use basic information to make appropriate health care decisions. Whereas low HL is associated with higher morbidity, mortality, and health care costs in multiple chronic conditions, little is known about HL and its associations in Parkinson's disease (PD).

Methods

This was a cross‐sectional study of nondemented adults with PD participating in the National Parkinson Foundation Parkinson's Outcomes Project at the University of Pennsylvania. Subjects were administered two brief HL assessments—the Rapid Estimate of Adult Literacy in Medicine‐Short Form (REALM‐SF), a word‐recognition test, and the Newest Vital Sign (NVS), a test of literacy, numeracy, and understanding of health information—as well as demographic and clinical questionnaires. Adverse outcomes included falls in the 3 months preceding the study visit and hospital admissions, emergency room visits, infections, or injuries in the preceding year. Caregiver burden was measured using the Multidimensional Caregiver Strain Index.

Results

A total of 168 subjects completed both HL screens (mean, 65.8 years; 65.5% male; 65.2% H & Y stage 2). Using the REALM‐SF, 97.6% of subjects had adequate HL. Using the NVS, however, 29.8% had low HL, which was associated with older age, lower education, male sex, greater disease severity, and poorer cognition. Low HL was associated with hospital admission and increased caregiver burden.

Conclusions

Low HL is common and associated with greater caregiver burden and a higher likelihood of hospitalization in patients with PD. Because HL is associated with both disease severity and adverse outcomes, it may be an important, modifiable contributor to morbidity.

In 2013, the American Association of Medical Colleges issued a list of competencies integral to physician development.1 Included is the capacity to “communicate effectively with patients.” To do so, physicians must recognize low health literacy (HL) as a potential obstacle to clear communication. HL has been defined as the ability to obtain and understand health information in order to make informed decisions regarding health care2 and is a prerequisite for comprehension of any health condition. Estimates of low HL prevalence range from 35% to 80% in elderly and disease‐specific cohorts.3, 4, 5, 6 Low HL is associated with poorer self‐management and higher hospitalization rates, health care costs, and mortality among older adults.7, 8, 9, 10, 11 HL is often erroneously interpreted as a surrogate marker of education; however, many individuals have low HL despite advanced degrees.12, 13

Little is known about the prevalence and implications of low HL in Parkinson's disease (PD). In the first study, to our knowledge, employing a validated HL screening instrument in a nondemented, general neurology population, nearly 21% had low HL.14 This is despite neurologists’ tendency to manage patients with complicated, chronic conditions, requiring extensive communication between providers, patients, and families. In the case of PD, affecting over 1 million adults in the United States and often involving multiple pharmacological and nonpharmacological interventions,15 adequate HL is critical. Adherence relies, in part, on adequate HL, and increasing regimen complexity is associated with both a higher degree of HL required for comprehension and a higher risk of medication nonadherence.13 We therefore sought to establish the validity of two brief HL screens—the Rapid Estimate of Adult Literacy in Medicine‐Short Form (REALM‐SF)16 and the Newest Vital Sign (NVS)17, 18—in a nondemented PD population. We aimed to identify patient factors associated with low HL in this population, explore the relationship between low HL and caregiver burden, and examine the potential role of low HL in the causal pathway between PD and adverse outcomes.

Materials and Methods

Sample

The institutional review board of the University of Pennsylvania approved this study. Subjects included all individuals participating in the National Parkinson Foundation Parkinson's Outcomes Project (NPF POP) at the University of Pennsylvania's Parkinson's Disease and Movement Disorders Center, an NPF Center of Excellence. The study design, inclusion and exclusion criteria, and methods of the POP are detailed elsewhere.19 In short, the POP is a worldwide, multicenter, longitudinal study of care practices in the management of individuals with PD. Participants complete an annual evaluation, including demographic and clinical information, a review of medical comorbidities, and adverse events, including falls, injuries, infections, emergency room visits, and hospital admissions occurring over the past year, and a battery of mobility and cognitive tests validated for use in PD. Medications reviewed include dopaminergic agents, antidepressants, and acetylcholinesterase inhibitors. Quality of life is assessed using the Parkinson's Disease Questionnaire (PDQ‐39), covering eight dimensions of quality of life (QOL). Each dimension is scored from 0 to 100, with lower scores indicating better QOL, and an overall summary index score is calculated as the average of the eight subscores, ranging from 0 to 100.20, 21 The emotional well‐being subscore is used as a surrogate measure of depression in this analysis. In addition, care partners are asked to complete the Multidimensional Caregiver Strain Index (MCSI).22, 23 The MCSI contains 18 items for a range of 0 to 72, where higher scores indicate greater strain. For the purposes of this study, individuals with a Montreal Cognitive Assessment (MoCA) score <23 were excluded from analysis24 given the likelihood of significant cognitive impairment confounding any associations with HL.

Interviews

Trained study personnel administered the POP battery, and if the participant was amenable to additional testing, introduced the current study. Participants completed the REALM‐SF,16 a brief, validated seven‐item instrument requiring reading and pronunciation of health‐related words (e.g., “menopause” and “jaundice”). The REALM‐SF is dichotomized at the eighth‐grade reading level, with 0 to 6 points corresponding to low HL and 7 points corresponding to ≥9th‐grade reading level and adequate HL. The subject was then administered the NVS,17, 18 a validated six‐item instrument wherein the subject views a nutrition label from an ice cream container. She must answer questions based on the label that test reading, interpretation, and numeracy—critical components of HL. The NVS is scored as 0 to 3 points indicating low HL and 4 to 6 points indicating adequate HL. Both tests have been validated as rapid and reliable screens for low HL in populations of older adults and in cohorts with neurological disease25, 26, 27 and have shown good criterion validity with lengthier gold‐standard measures of HL.28 Combined, these screens take <5 minutes and were selected because they are neither timed nor require writing—issues that may confound interpretation of results in a PD population.

Statistical Analysis

Statistical analyses were performed using STATA software (version 12.1; StataCorp LP, College Station, TX). Descriptive statistics were calculated for each variable, including performance on both the REALM‐SF and NVS assessed as both continuous and categorical variables according to established cutoffs.16, 17, 18 Correlation between the two instruments was calculated using Spearman's rank correlation. Two‐tailed t tests, chi square (χ²), and Wilcoxon's rank‐sum tests were used for bivariate analyses, as appropriate, comparing the demographic and clinical characteristics of individuals with low versus adequate HL. We then constructed a multivariable model to predict NVS scores, with all covariates with a P value of <0.20 in bivariate analyses as candidates for inclusion in the predictive model. Next, we used t tests, χ² and Wilcoxon's rank‐sum tests to compare the frequency of adverse outcomes in individuals with low versus adequate HL. We built multiple logistic regression models to predict the occurrence of individual and any adverse outcomes in 1 year, incorporating demographic, clinical, HL, and PD‐specific factors. Finally, we conducted bivariate analyses as described above to identify factors associated with increased caregiver burden and built a multivariable regression model to predict burden based on HL, demographic, and clinical factors. For all models, manual backward step‐wise elimination was conducted with each step assessed for confounding. We confirmed the assumptions of linearity and normality of the error distribution for each model. Given the correlation between age and PD duration, each model included only one of these variables to avoid multicollinearity. Model fit was assessed by likelihood ratio testing, and a P value <0.05 was considered significant.

Results

Participant Characteristics

A total of 168 individuals participated having a mean age of 65.8 years (standard deviation [SD]: 8.4) and 65.5% were male. Subjects had a median 16 years of education (interquartile range [IQR]: 15.5–18.0) and 90.5% had a care partner. Table 1 displays the demographic and clinical characteristics of the cohort. Polypharmacy was common, with 75% of subjects taking at least two dopaminergic medications. Antidepressants were taken by 15.6% and acetylcholinesterase inhibitors by 5.4%. Falls occurred in 44.6% of subjects in the past 3 months. Twenty‐two percent of the cohort visited an emergency room and 23.2% required hospitalization in the past year for any cause.

Table 1.

Characteristics of participants (n = 168)

Male, %	65.5
Caucasian, %	99.4
Mean age, years (SD)	65.8 (8.4)
Median education, years (IQR)	16 (15.5–18.0)
Median duration of PD, years (IQR)	9 (6–14)
PD severity by H & Y stage, %
1	5.5
2	65.6
3	24.5
4 or 5	4.3
Caregiver present, %	90.5
≥3 medical comorbiditiesa, %	23.8
≥2 antiparkinsonian medications, %	75.0
Median MoCA score (IQR)	27 (26–29)
Quality of life
Median PDQ‐39 Summary Index (IQR)	15.6 (9.0–26.3)
Median PDQ‐39 Emotional Well‐being	16.7 (6.7–26.7)
Dimension Subscore (IQR)

MoCA: possible range of 0 to 30. PDQ‐39 contains eight dimensions of well‐being, each dimension with subscore 0 to 100 where lower score indicates better quality of life. PDQ‐39 Summary Index calculated as average of eight dimension subscores (range, 0–100). PDQ‐39 Emotional Well‐being subscore (range, 0–100) used as surrogate marker of depression.

^aSurveyed comorbidities include heart disease, respiratory problems, arthritis, diabetes mellitus, other neurologic comorbidities, and other medical comorbidities.

Comparison of Health Literacy Screens

The median score on the REALM‐SF was 7 of a possible 7 (range, 0–7), with 2.4% of subjects scored as low HL. However, the median score on the NVS was 5 of a possible 6 (IQR, 3–6; range, 0–6), with 29.8% of subjects scored as low HL (≤3) as shown in Figure 1. There was no significant correlation between REALM‐SF and NVS total scores (scored continuously: Spearman's rho = 0.05, P = 0.54; scored categorically: kappa, 0.03, P = 0.18).

Figure 1.

Distribution of Newest Vital Sign scores.

Factors Associated With Low Health Literacy

Based on the ceiling effect demonstrated by the REALM‐SF in this highly educated population, the weak correlation between the HL screens, and the more comprehensive assessment of the NVS instrument, all further HL analyses were based on NVS scores alone. In bivariate analyses, 35.45% of men versus 18.97% of women had low HL (P = 0.026). Older age was significantly associated with low HL, with a median age of 66 versus 68 years for adequate and low HL groups, respectively (P < 0.001). An educational level of high school or below was more common among individuals with low (12.00%) versus adequate HL (3.39%; P = 0.067). Subclinical cognitive impairment was more common among those with low HL, with a median of 4 versus 2 missed points on the MoCA for low and adequate HL, respectively (P < 0.001). There was a modest correlation between MoCA and NVS scores (Pearson, r = 0.28). Finally, there was a trend toward greater disease severity among those with low HL, with 38.00% versus 27.97% rated as H & Y stage 3 or higher (P = 0.110).29 After adjusting for depression, low HL remained significantly associated with male sex, high school education or lower, older age, PD severity, and cognitive impairment, as shown in Table 2.

Table 2.

Factors associated with total score on the Newest Vital Sign

Characteristic	Change in NVS Score (Points)	95% CI	P Value
Male sex	−1.14	−1.66, −0.61	<0.01
Every year of age >66 a	−0.08	−0.11, −0.05	<0.01
H & Y stage 3 or higher	−0.55	−1.11, 0.01	0.05
Every point missed on MoCA	−0.17	−0.29, −0.04	<0.01
High school education or less	−1.52	−2.57, −0.46	<0.01
Every point on PDQ‐39 emotional subscale	−0.01	−0.03, 0.00	0.14

Final model constructed using all demographic variables and all significant variables in bivariate analyses with backward elimination. Intercept in this model = 4.75; R ² = 0.30. Bolded characteristics indicate independently significant factors associated with low HL. NVS range: 0 to 6. MoCA: possible range of 0 to 30. PDQ‐39 Emotional subscale: range of 0 to 100 where lower score indicates better quality of life.

^aFor ease of interpretation, age was centered at the mean of 66 years.

Relationship of Low Health Literacy With Caregiver Burden and Adverse Outcomes

Of 134 subjects with complete data on the MCSI, low HL was associated with greater caregiver burden (median score of 4 and IQR 1–9 in individuals with adequate HL compared to median score of 10 and IQR 2.5–17.0 in those with low HL; P < 0.001). Each point missed on the NVS was associated with a 1.35‐point increase in the MCSI score (P = 0.02; 95% confidence interval [CI]: 0.25–2.44). After adjustment for PD severity, cognitive function, and multicomorbidity, however, the relationship between low HL and MCSI was attenuated (0.95‐point increase in the MCSI for every NVS point missed: P = 0.086; 95% CI: −0.14 to 2.04), as shown in Table 3.

Table 3.

Factors associated with total score on the Multidimensional Caregiver Strain Index

Characteristic	Change in MCSI Score (Points)	95% CI	P Value
Every point missed on NVS	0.95	−0.13, 2.04	0.09
H & Y stage 3 or higher	10.54	6.48, 14.59	<0.01
Every point missed on MoCA	−0.49	−1.45, 0.46	0.31
Three or more comorbidities	2.31	−2.10, 6.71	0.30

Final model constructed using all demographic variables and all significant variables in bivariate analyses with step‐wise, backward elimination performed manually by J.F. R‐squared for model = 0.21. NVS range: 0 to 6. MoCA: possible range of 0 to 30. Bolded characteristics indicate factors independently associated with MCSI.

There was a 2.6‐fold increased odds of hospitalization in individuals with low HL by NVS criteria (P = 0.01; 95% CI: 1.23–5.48). Adjusting for multiple comorbidities, social support, and duration of disease, low HL remained significantly associated with hospitalizations (odds ratio [OR]: 2.89; P < 0.01; 95% CI: 1.30–6.42). None of the other individual adverse outcomes surveyed—falls, emergency room visits, injuries, pneumonia, or infection in general—were associated with low HL. As shown in Table 4, after combining all adverse outcomes into a binary summary outcome measure, we found that greater PD severity (H & Y ≥ 3) and multicomorbidity were independently associated with adverse events after adjusting for demographics, PD duration, cognitive function, and depression. Low HL was not independently associated with the summary adverse outcomes measure (P = 0.83).

Table 4.

Factors associated with adverse outcomes in the past year^a

Characteristic	Hospitalization			Any Adverse Outcomea
	OR	95% CI	P Value	OR	95% CI	P value
Low HL (by NVS criteria)	2.89	1.30, 6.42	<0.01	1.09	0.51, 2.31	0.83
PD duration, years	1.04	0.97, 1.11	0.26	1.05	0.99, 1.12	0.14
H & Y stage 3 or higher	1.16	0.49, 2.76	0.74	2.69	1.17, 6.18	0.02
Three or more comorbidities	3.93	1.74, 8.88	<0.01	8.32	2.73, 25.32	<0.01

Final model constructed using all demographic variables and all significant variables in bivariate analyses with step‐wise, backward elimination performed manually by J.F. Pseudo‐R‐squared for model predicting hospitalizations: 0.11. Pseudo‐R‐squared for model predicting all adverse outcomes: 0.14. Age, sex, race, depression, and cognition were not significant. Bolded characteristics indicate independently significant factors associated with adverse outcomes.

^aAdverse outcomes assessed include falls (in the 3 months preceding study visit) and injuries, infections, pneumonia, emergency department visits, or hospital admissions in the year preceding study visit.

Discussion

In this cross‐sectional study of nondemented individuals with PD, nearly 30% of subjects met NVS criteria for low HL despite high levels of education. Factors independently associated with low HL included male sex, older age, greater disease severity, lower education, and cognitive impairment. Low HL was associated with higher caregiver burden; however, this was no longer significant after accounting for PD severity and cognitive impairment. Low HL was, however, independently associated with hospitalizations.

We tested two screening instruments for this study—the REALM‐SF and NVS—which have been validated in other disease populations. Our choice of instruments was based upon feedback from previous work using the Short‐Form Test of Functional Health Literacy in Adults (S‐TOFHLA).30 Although the S‐TOFHLA has been extensively validated and used in a previous study of HL in neurology,14 we opted for the REALM‐SF and NVS here for several reasons. First, subjects in the previous study complained of the time required to take the S‐TOFHLA. Furthermore, individuals with PD commented on the difficulty presented by the S‐TOFHLA for people with bradykinesia and micrographia. Both the REALM‐SF and NVS are shorter than the S‐TOFHLA and do not require writing. In addition, the S‐TOFHLA is timed and unanswered questions are incorrect. Bradykinesia is thus a potentially significant confounder of S‐TOFHLA performance. Whereas both instruments have been compared with and found to have moderate correlation with the S‐TOFHLA,28, 31 one recent study found that REALM‐SF and NVS scores were weakly correlated with each other and similarly noted a prominent ceiling affect with REALM‐SF scores.28 These results corroborate our finding that 2.4% of our population had low HL by REALM‐SF criteria, whereas 29.8% had low HL according to the NVS.

There are several possible explanations for these divergent findings. The REALM‐SF predominantly measures literacy and highly correlates with education.16 Our findings on the REALM‐SF, as compared to the NVS, may be explained by our highly educated sample, a pattern consistent with other studies that compare these instruments.17, 28 It is also interesting to note the higher prevalence of low HL in this sample, as compared to our previous finding of 20.5% prevalence in a diagnostically heterogeneous cohort of neurology outpatients.14 The two studies are not directly comparable because of the use of different HL screens. Language—the primary cognitive domain tested by the REALM‐SF—tends to be relatively spared in PD. However, the shared cognitive domains affected by PD and assessed by the NVS—namely, attention, thought processing, decision making, and working memory32—may explain the poorer NVS performance compared to the REALM‐SF. Indeed, the large Health Literacy and Cognition in Older Adults study demonstrated that the relationship between low HL and adverse outcomes was significantly attenuated after adjusting for cognitive function.25 The PD‐specific declines in particular cognitive domains may cause a rising prevalence of low HL in this population out of proportion to the prevalence attributable to older age. Because of the overlap in these cognitive domains, the NVS may be a rapid and practical screening instrument for subtle cognitive impairment in the PD population, although longitudinal studies to establish reliability and comparison to formal neuropsychological testing are necessary for validation.

Further validating the NVS as a measure of HL in PD were the associations noted with sex, education, age, and cognition. Studies in other conditions have shown similar relationships between low HL and male sex, education, age, and cognitive impairment.4, 6, 11 Here, the MoCA serves as a measure of global cognition because of its inclusion in the POP.19, 33 After excluding individuals with dementia, even subclinical cognitive impairment was significantly associated with low HL. Disease severity—H & Y ≥ 3—was also associated with low HL. Therefore, clinicians must have a low threshold to suspect poor understanding of complicated disease management plans. Although one may assume that patients coping with PD for several years have greater disease knowledge, the converse may be the case.14, 34 Advancing age and disease severity make clear, plain language communication critical. Additionally, apathy—a common PD symptom—must be evaluated in future studies to determine its role in the relationship between HL and patient outcomes.

The finding of increased caregiver burden among patients with low HL bears consideration. Although bivariate analyses supported this relationship, it was attenuated once known risk factors for caregiver burden, namely, motor disability and cognitive impairment, were included in the model.23 This is of particular interest given that our population was limited to individuals without severe cognitive impairment, suggesting that even subclinical cognitive decline remains a powerful influence on caregiver strain. Future work capturing apathy, depression, and other known predictors of high caregiver burden is indicated to better understand the complex relationship between HL and caregiver burden.35, 36, 37 Though some caregivers may be able to compensate for the low HL of their charge, data on HL of caregivers themselves are sparse and disheartening.3 In studies of patients with congestive heart failure (CHF) and their caregivers, 29% to 31% of caregivers had low HL.38, 39 To our knowledge, there are no similar studies involving caregivers of patients with neurological disease or PD specifically. Future work on the prevalence and implications of low HL among caregivers may reveal potential interventions to mitigate patient morbidity and mortality.

In another study of patients with CHF, for example, low HL was independently tied to a greater risk of hospitalization and mortality.40 The researchers hypothesized that low HL was an indicator of suboptimal self‐care behavior. This is particularly detrimental when a medical condition depends on adherence to medications, engaging in physical activity, and monitoring one's symptoms. Low HL was associated with PD severity, which predicts morbidity, and low HL was also independently associated with hospitalizations. PD severity was strongly associated with all adverse outcomes. Thus, HL may represent a critical and modifiable contributor in the causal pathway in which greater PD severity leads to adverse outcomes.

Self‐care in PD frequently involves a complex medication regimen marked by polypharmacy and frequent dosing. In a decade‐old study, most individuals with PD were taking ≥2 dopaminergic medications, often dosed ≥3 times daily.41 In the current study, that percentage is nearly 75% and does not account for any nondopaminergics or medications for other conditions. Low HL may account for poor understanding of the need for each prescription and dose, leading to nonadherence. In cardiovascular disease, for example, and in multiple cohorts of elderly patients, low HL has repeatedly been a risk factor for both nonadherence and excess morbidity and mortality.8, 42, 43, 44, 45 In two large, multicenter studies of individuals with PD, poor adherence independently predicted worsening disability.46, 47 Therefore, medication nonadherence may be the critical link in the pathway between HL and disease severity. Future studies are needed to address adherence as an actionable source of excess morbidity and mortality in PD.

Limitations of this study include its cross‐sectional design that only allows us to measure associations, and not causality, among the different variables, and the potential selection bias inherent in drawing participants from an academic referral center. Individuals observed at a specialty clinic may be skewed toward those with greater disease severity. Furthermore, uninsured or underinsured patients may have limited access to such specialty care and are often the most likely to have low HL.5 This limitation would bias our results toward the null, underestimating the prevalence and impact of low HL in PD.

Despite the limitations of this study, this remains, to our knowledge, the first investigation of HL in PD and provides evidence for the use of the NVS to measure HL in PD. Based on the NVS, nearly 30% of educated, nondemented individuals with PD failed to demonstrate the skills required for simple health care understanding. Low HL was independently associated with hospitalizations. Initiatives exist to begin addressing this situation. The Agency for Healthcare Research and Quality provides helpful resources for engaging in HL‐sensitive patient care (http://www.ahrq.gov/qual/literacy/). Also, the NPF has created a low literacy version of their Aware in Care kit, designed to empower patients and caregivers while hospitalized. Further effort is needed to raise awareness among neurologists of this barrier to care and to create successful, cost‐effective interventions to overcome low HL.

Author Roles

(1) Research Project: A. Conception, B. Organization, C. Execution; (2) Statistical Analysis: A. Design, B. Execution, C. Review and Critique; (3) Manuscript Preparation: A. Writing of the First Draft, B. Review and Critique.

J.F.: 1A, 1B, 1C, 2A, 2B, 2C, 3A, 3B

K.S.: 1C, 3B

W.F.: 1C, 3B

N.D.: 1A, 1B, 2A, 2C, 3B

Disclosures

Funding Sources and Conflicts of Interest: This study received support from the National Parkinson Foundation Parkinson Outcomes Project and the Parkinson Council. Dr. Fleisher received support from the National Institutes of Health (T32‐NS‐061779). Dr. Dahodwala is supported by the National Institute on Aging (K23 AG034236). The authors report no conflicts of interest.

Financial Disclosures for Previous 12 Months: Dr. Fleisher received research support from NIH and Parkinson Council. Dr. Dahodwala received research support from NIH, Parkinson Council, National Parkinson Foundation, and Teva.