Key Question 1. Relationship of Health Literacy to Various Outcomes and Disparities

We identified 86 good- or fair-quality articles reporting on 72 unique studies for this topic. Some studies report on more than one key question. These studies report results about the relationship between health literacy and use of health care services, health outcomes, and costs of health care and disparities between specific racial, ethnic, cultural, or age groups. Fourteen studies were of good quality and 72 of fair quality, according to the criteria described in Chapter 2. In addition, we identified 40 studies which were considered to be of poor quality and therefore not included in the analysis (poor-quality studies are listed in Appendix E; we do not discuss them further in this review.) In the text below, we identify only studies of good quality; all others for which quality is not specifically called out are fair quality. Most studies had a cross-sectional design (N = 64), but 22 were cohort designs (Table 5).

Table 5

Overview of health literacy studies.

Multiple studies reported results using the same data. For instance, eight articles reported results collected during the “Prudential study.” This study was conducted with 3,260 new members in a Prudential Medicare managed care plan of enrollees in Cleveland, Ohio, Houston, Texas, and Tampa and south Florida.^61-68 Other studies reported in multiple articles include four articles reporting on a sample of patients at Chicago, Illinois, and Shreveport, Louisiana, HIV clinics,^69-72 two articles reporting on pharmacy patients in Atlanta, Georgia,^73,74 and three articles reporting on patients in three primary care clinics in Chicago, Illinois; Shreveport, Louisiana; and Jackson, Michigan.^75-77

Studies examined a variety of outcome measures including use of health care services (hospitalization and emergency department visits and screening and immunizations), access to care, and health outcomes (adherence, self-efficacy, health behaviors, health-care-related skills, disease prevalence and severity, health status, and mortality). Studies also examined differences in costs and disparities related to health literacy level (Table 5).

Table 6 groups KQ 1 health literacy studies based on the health literacy measurement tool used in the analysis and, further, the skill-level groupings used to distinguish study participants. We found that health literacy was mostly measured with the Rapid Estimate of Adult Literacy in Medicine (REALM; 33 articles) or the Test of Functional Health Literacy in Adults (TOFHLA) or Short Test of Functional Health Literacy in Adults (S-TOFHLA; 42 articles). Three articles used the National Assessments of Adult Literacy (NAAL), and, unlike our earlier review, no article used the Wide Range Achievement Test (WRAT; a general literacy measure that was commonly used in studies included in our earlier review Literacy and Health Outcomes¹). Several other literacy measures (in contrast to health literacy measures intended to be used in a health care environment) were included in one study apiece: the Cape Area Panel Study Literacy and Numeracy Evaluation, a reading comprehension instrument in Nepalese, an instrument for the diagnosis of reading, and the Woodcock Language Proficiency Battery. Although the validity and reliability of the Woodcock battery⁴² is well known, information about these other literacy measures is quite limited. The health literacy levels used to compare study participants evaluated using the REALM, TOFHLA, or S-TOFHLA varied among studies, ranging from a continuous measure to two, three, or even more groups. In some studies, three groups were identified (i.e., inadequate, marginal, and adequate); in others, two of the three groups were combined in the statistical analysis. Studies varied concerning whether the two lower or the two higher groups were combined. Conceptually, an individual's health literacy level could change over time. However, the instruments included in the reviewed studies capture only static measures of health literacy or numeracy.

Table 6

Measurement tools and criteria used to measure health literacy or literacy in KQ 1 articles.

In contrast to our earlier review, studies reviewed in the update by and large include multivariate analyses (rather than just unadjusted bivariate analyses) (Table 5). However, the choice of variables controlled for in analyses varied greatly across studies. Potential confounders (related to health literacy and health outcomes) controlled for in many studies include education, age, race, gender, and income.

KQ 1a. Use of Health Care Services

We identified 24 articles reporting on 23 unique studies examining the relationship between health literacy skills and the use of health care services. Three studies were of good quality and 21 were of fair quality. Nine studies included cohort designs; the rest were cross-sectional. These studies focused on emergency department admissions or hospitalizations, general preventive screenings (mammogram, colon, Papanicolau [Pap], sexually transmitted infection testing, and influenza and pneumococcal vaccination), and access to office visits and insurance.

Hospitalization and emergency department rates. Six studies—one good-quality prospective cohort study (hereafter, the Prudential study),⁶⁸ two fair-quality prospective cohort study,^78,79 one retrospective cohort study,⁸⁰ and two cross-sectional studies^81,82—examined the risk of hospitalization by health literacy level (Table 7). All but one study showed a statistically significant association of increased hospitalization and use of inpatient services with lower health literacy level. Populations included the elderly,^68,81 patients with asthma,^79,80 and patients with congestive heart failure.⁷⁸ The one study that did not find an association with hospitalizations included a cross-sectional subpopulation of HIV-positive adolescents, which may be a healthier population compared to the other studies.⁸² One of the larger cohort studies, the Prudential study, examined the impact of low health literacy on medical care use among 3,260 Prudential Medicare managed care enrollees.⁶⁸ Patients with low health literacy had higher probabilities of using inpatient services than those with adequate health literacy (mean differences in probability of use, 0.05; 95% confidence interval [CI], 0.00-0.09). Enrollees with marginal and adequate health literacy did not differ in use of inpatient services. The strength of evidence is moderate (Table 8 and Appendix F). These findings are consistent with previous findings in our 2004 systematic review.¹

Table 7

Summary of studies of the relationship between health literacy and emergency department and hospitalization rates (KQ 1a).

Table 8

KQ 1a health literacy studies: strength of evidence grades by health care service outcomes.

Nine studies, including two good-quality prospective analyses from the Prudential study,^62,68 three other prospective cohorts,^78,79,83 one retrospective cohort,⁸⁰ and three cross-sectional studies,^81,82,84 examined emergency and urgent care visits by literacy level (Table 7). All but two studies^82,84 showed an association of greater emergency department use and low health literacy. The Prudential study⁶² examined the association of emergency department visits with health literacy level. After controlling for multiple confounders, both the inadequate health literacy and the marginal health literacy groups had a higher rate of two or more emergency department visits when compared with those with adequate health literacy (marginal literacy relative risk [RR], 1.44; 95% CI, 1.01-2.02; inadequate literacy RR, 1.34; 95% CI, 1.00-1.79).

The two studies that did not find an association with health literacy examined associations of parent health literacy and child asthma care among children with persistent asthma⁸⁴ and the HIV-positive adolescents described above.⁸² The other study, a cross sectional study of 499 children with persistent asthma, examined parental health literacy and multiple aspects of asthma care (preventive medicine use, acute care, unmet needs, parental worry, and parental quality of life). Parental health literacy was not associated with children's use of any urgent care. This particular outcome was limited because the outcome of urgent care visits was measured by parental self-report. The strength of evidence is moderate (Table 8 and Appendix F). No studies of emergency department use were reported in our earlier report.

General screening. We found one good⁸⁵ and seven fair studies^81,86-91 examining the association of health literacy with general screening services. These services included colon screening (Table 9), Pap testing (Table 10), mammography (Table 11), and testing for sexually transmitted diseases (Table 12).

Table 9

Summary of studies of the relationship between health literacy and colon cancer screening (KQ 1a).

Table 10

Summary of studies of the relationship between health literacy and Pap tests (KQ 1a).

Table 11

Summary of studies of the relationship between health literacy and mammography (KQ 1a).

Table 12

Summary of studies of the relationship between health literacy and sexually transmitted infections testing (KQ 1a).

Colon screening. Five cross-sectional studies found mixed results for the probability of having received colon screening by health literacy level (Table 9).^81,86-89 Of note, the two larger studies found a lower probability of colon screening in patients with lower health literacy.^81,86 The largest study⁸⁶ found a decreased probability of colon cancer screening among those 65 years of age and older with below-basic health literacy compared with those with proficient skills in a nationally representative US cross-sectional study of 18,100 individuals examining multiple self-reported preventive services (data not reported [NR]; P < 0.05). The three studies not finding an association with health literacy were smaller in size (samples of 50 to 136) and limited to one geographic area.^87-89 The strength of evidence is low (Table 8 and Appendix F). No studies of colon screening use were reported in the earlier 2004 report.¹

Pap tests. Three cross-sectional studies found that women with lower health literacy had a lower probability of ever having had a Pap test (Table 10).^81,86,91 However, this result was present only in certain age cohorts. In a nationally representative sample, researchers found that women less than 40 years of age with below-basic health literacy had a lower probability of having a Pap test than women in the same age group with proficient health literacy (NR; P < 0.05), but the probabilities did not differ by literacy level in women 40 to 64 years of age.⁸⁶ Results also seemed to differ by degree of lower health literacy (inadequate vs. marginal). One study examined Pap screening in 205 low-income Spanish-speaking Latinas in New York City.⁹¹ In adjusted analyses, controlling for age, years in the United States, education, and having a source of care and health insurance, these investigators found that women with inadequate health literacy were less likely to have ever had a Pap test than women with adequate literacy (odds ratio [OR], 0.06; 95% CI, 0.01-0.55). However, the marginal and adequate health literacy groups did not differ significantly (OR, 0.14; 95% CI, 0.01-1.41). This discrepancy in findings between inadequate and marginal groups is consistent with an earlier study⁹² in the 2004 report.¹ Thus, the overall strength of evidence is low (Table 8 and Appendix F).

Mammography. Four cross-sectional studies examined use of mammography by health literacy group (Table 11).^81,85,86,90 All studies found a lower use of mammography in the lower health literacy group compared with the adequate group. However, one study found a difference in receipt of mammograms among older women⁸⁶ and another found differences between groups by frequency of mammograms.⁹⁰ In the Prudential study, women ages 65 and older with low health literacy had a lower probability of having a mammogram than those with adequate health literacy (NR; P < 0.05); health literacy was not associated with the probability of having mammography among women ages 40 to 64.⁸⁶ Another study evaluated mammography rates in 97 women in three community health clinics in Philadelphia; inadequate health literacy was associated only with significantly lower odds of ever having a mammogram (OR, 0.88; 95% CI, 0.79-0.98), but not with having a mammogram in the past year, past 3 years, or as part of a check-up.⁹⁰ The strength of evidence is moderate (Table 8 and Appendix F). These results are consistent with the 2004 report.

Sexually transmitted infection testing. Researchers conducted a cross-sectional study (N = 372) of HIV test acceptors in an inner-city urgent care hospital (Table 12).⁹³ Subjects with inadequate health literacy had greater odds of accepting an HIV test result than those with adequate health literacy (OR, 2.02; 95% CI, 1.19-3.42). In the 2004 report, the one study about this type of service showed a lower probability of having received a gonorrhea test in the past year among those in the low-literacy group.⁹⁴ The strength of evidence is low (Table 8 and Appendix F).

Immunizations. One good cohort⁶³ and three cross-sectional studies^85,86,95 found inadequate health literacy associated with lower receipt of influenza vaccine (Table 13). In a Prudential study analysis, controlling for age, sex, race, ethnicity, education, income, site, morbidity, and smoking, researchers found lower odds of receiving an influenza vaccine in the inadequate health literacy group than in the adequate group (OR, 0.76; P = 0.020), but no significant differences in the marginal health literacy group compared with the adequate health literacy group.⁶³ These findings are similar to those in our 2004 report. Age also appears to be a factor in a study⁸⁶ that found a lower receipt of influenza vaccine by health literacy level among adults under 40 years of age and 65 or older (NR; P < 0.05), but no differences by health literacy level in adults 40 to 64 years of age (NR; P = nonsignificant [NS]). The strength of evidence is moderate (Table 8 and Appendix F).

Table 13

Summary of studies of the relationship between health literacy and immunizations (KQ 1a).

Pneumococcal vaccine did not follow a pattern similar to influenza vaccine (Table 13). In the two studies that examined pneumococcal vaccine,^63,86 no significant association between pneumococcal vaccine and health literacy level was found. The strength of evidence is insufficient (Table 8 and Appendix F).

Access to care. Four cohort^62,68,96,97 and five cross-sectional studies^{82,86,95,98-100} examined various measures of access to office visits and general care; these types of services included pharmacy visits, dental visits, and vision checkups as well as hospital choice and transplant waitlists (Table 14). Two good cohort analyses from the Prudential study did not find an association of inadequate health literacy level with number of physician visits⁶² or pharmacy services used.⁶⁸ These results are consistent with the one study¹⁰¹ described in the 2004 report. Similarly, one prospective cohort of 68 individuals did not find differences in time to follow up after an abnormal Pap test by health literacy level.⁹⁶ However, results were mixed for dental and vision visits in one Prudential study analysis.⁸⁶ Another large study (N =2,512) of Medicare recipients found less access to medical care by lower health literacy groups.⁹⁵

Table 14

Summary of studies of the relationship between health literacy and access to care and access to insurance (KQ 1a).

One interesting retrospective cohort study involved 62 patients in five outpatient dialysis units in San Francisco, California.⁹⁷ After controlling for multiple confounders, the investigators found a significantly longer time from start of dialysis to referral to a transplant list in patients with inadequate health literacy (hazard ratio [HR], 4.54; 95% CI, 1.67-12.5). However, they saw no subsequent differences in time from being on a transplant list to making the waitlist for transplant. The strength of evidence is insufficient given the variation among studies (Table 8 and Appendix F).

Access to insurance. One nationally representative cross-sectional study¹⁰² of 6,100 parents examined parental health literacy and their children's access to health insurance. After controlling for multiple confounders, the odds of having at least one child without health insurance in their household was higher among parents with below-basic literacy compared to parents with proficient health literacy (OR, 2.4; 95% CI, 1.1-4.9). The strength of evidence is low because there is only one study and there are biases associated with using self-reported measures as the outcome (Table 8 and Appendix F).

KQ 1b. Health Outcomes

We identified 72 articles reporting on 60 unique studies examining the relationship between literacy skills and health outcomes. Of these, 13 articles were of good quality and 59 were fair quality.

Adherence. Eleven studies, reported in 15 articles, evaluated the relationship between health literacy level and adherence in adjusted analyses (Table 15).^{61,69-74,81,82,103-108}

Table 15

Summary of studies of the relationship between health literacy and adherence (KQ 1b).

Five studies reported in 8 articles examined nonadherence in taking HIV medication and found mixed evidence of a direct relationship.^{69-72,82,103-105} Studies found no relationship examining 100 percent adherence to medications over 3 days among patients with a history of alcohol problems,¹⁰⁵ 90 percent adherence over the past 3 days among adolescents,⁸² and less than 95 percent adherence over the past 3 months among a small sample (N = 87) of clinic patients.¹⁰⁴ In the last study, the relationship between health literacy level and nonadherence was examined, comparing the unadjusted relationship with an adjusted model, controlling only for the potential mediation of a patient's norms about an acceptable level of adherence and no potential confounding variables. Norms were found to mediate the relationship.

In contrast, in study using self-reported pill counts and controlling for education and other variables, researchers found a positive relationship between lower health literacy level (measured as a TOFHLA score of less than 90 percent correct rather than more commonly used categories) and probability of nonadherence (OR, 3.77; 95% CI, 1.46-9.93).¹⁰³ Similarly, based on findings from a study of 204 patients in clinics in Shreveport, Louisiana, and Chicago, Illinois, researchers found a positive relationship: nonadherence to HIV regimen was higher among those with low health literacy than those with adequate health literacy (OR, 2.12; 95% CI, 1.93-2.32).^69,72 However, this study found no difference between the marginal and adequate groups. In subsequent analyses of this sample, the researchers conducted formal mediation analyses and found that the relationship between low health literacy and nonadherence to HIV medications was mediated by the combination of HIV treatment knowledge and medication self-efficacy in one analysis⁶⁹ and by stigma related to taking HIV medications in another.⁷¹

Medication-taking adherence, refill adherence, and adherence to procedural instructions were examined in various other patient populations with mixed results. Among 110 caregivers of infants in pediatric clinics, a combined group of those with low or marginal health literacy were significantly more likely to be adherent in providing vitamins to their infants than those with adequate health literacy (OR, 2.4; 95% CI, 1.37-4.2).¹⁰⁸ However, no significant differences by health literacy level emerged in other patient populations for medication-taking, refill adherence, or adherence to procedural instructions. Studies included patients at an anticoagulation clinic missing doses of warfarin,¹⁰⁶ seniors at two clinics filling any medication prescriptions on time,⁸¹ seniors refilling medications for cardiovascular disease,⁶¹ preoperative clinic patients following fasting and preoperative medication instructions,¹⁰⁷ and adults reporting adherence at hospital pharmacies in Atlanta, Georgia.⁷⁴ However, in the Atlanta study, researchers found that the relationship between health literacy and adherence was moderated by social support; at the highest levels of social support, patients with adequate health literacy reported better adherence, and, at the lowest levels of social support, patients with lower health literacy reported better adherence.⁷⁴

Three studies examining the relationship between health literacy level and adherence assessed outcome differences between individuals in the marginal- and adequate-health-literacy groups but found no significant difference.^61,69-72,105

Our research team found mixed evidence of a relationship between health literacy and health outcomes resulting in a strength of evidence grade of insufficient, which may be the result of differences in adherence measure, disease state, and adjustment for relevant confounders (Table 16 and Appendix F). Our earlier review also found mixed results across studies. One study reported a significant relationship between lower literacy and poorer self-reported adherence; three found no significant relationship.^109-112

Table 16

KQ 1b health literacy studies: strength of evidence grades by health outcomes.

Self-efficacy. Five studies examined the relationship between participant health literacy level and self-efficacy for a variety of behaviors^{70,82,87,113,114} (Table 17). One study found greater self-efficacy for taking HIV medications in the adequate-health-literacy group than in the low-health-literacy group, but no difference between the adequate and marginal groups.⁷⁰ A second study found greater self-efficacy for colorectal cancer screening among individuals with higher health literacy levels (measured by the UK TOFHLA).¹¹⁴ In contrast, another study found no difference between groups in relation to self-efficacy for taking medications or keeping appointments among adolescent HIV patients.⁸² Furthermore, self-efficacy for obtaining a fecal occult blood test or colonoscopy was not related to limited health literacy level (low and marginal groups combined) compared with a group with adequate literacy in a small, potentially underpowered adjusted analysis of 99 patients at one clinic.⁸⁷ Finally, although higher self-efficacy for taking hormone therapy among postmenopausal women was correlated with higher health literacy level, this was in an unadjusted analysis.¹¹³

Table 17

Summary of studies of the relationship between health literacy and self-efficacy (KQ 1b).

Based on the mixed results in these studies, our research team graded the strength of evidence as insufficient (Table 16 and Appendix F). Our earlier review included no self-efficacy studies.

Health Behaviors. We identified studies reporting on a variety of health behaviors including smoking, alcohol and drug use, healthy lifestyle, review of prescription information, HIV risk behaviors, and sexual activity.

Smoking. Two large studies evaluated the relationship between health literacy level and self-report of smoking in adjusted analyses (Table 18); results were statistically different even though odds ratios were fairly similar.^64,115 A study examining current smoking status in a national sample of British adults (N = 719) found that higher health literacy, measured as a continuous variable, was associated with a small increased likelihood of not smoking (OR, 1.02; 95% CI, 1.003-1.03).¹¹⁵ In contrast, among the Prudential sample of American seniors (N = 2,923), researchers found no relationship between health literacy level and participants' smoking status (never, former, or current).⁶⁴ Due to these mixed results, the strength of evidence was graded as insufficient (Table 16 and Appendix F). We reported mixed results in our earlier review through one adjusted analysis of adolescents (boys and girls reported separately) and two unadjusted analyses examining outcomes of smoking in adults; therefore, these studies do not modify our evaluation of the strength of evidence.^116-118

Table 18

Summary of studies of the relationship between health literacy and health behaviors (KQ 1b).

Alcohol and drug use. The Prudential study also examined the relationship between health literacy level and current alcohol consumption; they found no relationship.⁶⁴ Among adolescents with HIV, higher health literacy was associated with greater substance use.⁸² Neither study adjusted for comorbid depression. With only one study concerning alcohol consumption and one concerning substance use, strength of evidence was graded as insufficient (Table 16 and Appendix F). In our earlier review, we included one study of alcohol consumption among adolescents and no significant relationship with health literacy was found.¹¹⁸

Healthy lifestyle. Eight studies addressed the relationship between health literacy level and various measures of healthy lifestyle, including level of physical activity, eating habits, seat belt use, and weight^{9,10,64,65,81,95,115,119} (Table 18).

Two studies, discussed above for smoking outcomes, measured level of physical activity. Neither study found significant differences by health literacy level.^64,115

Healthy eating, overall healthy lifestyle, and seat belt use were examined in one study each. In a sample of British adults, higher health literacy level was associated with a small but significantly higher probability of eating five or more servings of fruits or vegetables per day (OR, 1.02; 95% CI, 1.003-1.03).¹¹⁵ Among 489 seniors receiving care at two clinics in Chicago, health literacy level did not have a direct effect on a composite measure, the Health-Promoting Lifestyle Profile, which assesses a combination of exercise, nutrition, and health responsibility.⁸¹ Only one unadjusted analysis examined the relationship between health literacy level and seat belt use. The researchers found no significant differences.⁶⁴

Among obese children, body mass index (BMI) was inversely related to the child's health literacy level, controlling for their parent's health literacy level and other confounders.¹¹⁹ Four additional studies examined differences in rates of obesity or BMI by health literacy level in unadjusted analyses.^9,10,65,95 Results were mixed.

The research team judged the strength of evidence as insufficient (Table 16 and Appendix F) for the relationship between health literacy and physical activity, eating habits, and seat belt use as a group based on mixed findings. The strength of evidence concerning weight or obesity was also insufficient (Table 16 and Appendix F). Our earlier review included no studies with any healthy lifestyle outcomes.

Review of prescription information. One adjusted analysis examined the relationship between health literacy and review of prescription information (Table 18). Clinic patients (N = 251) in Shreveport, Louisiana, were asked to report on whether they ever looked at the consumer information included with their prescriptions.¹²⁰ After controlling for potential confounders, including the number of prescriptions taken, those with low health literacy were less likely to look at the material than persons of adequate health literacy (OR, 2.5; 95% CI, 1.2-5.2). The marginal- and adequate-health-literacy groups did not differ. The strength of evidence was low (Table 16 and Appendix F).

HIV risk behaviors and sexual activity. Two adjusted analyses examined the relationship between health literacy and sexual behaviors (Table 18). One study of female inmates did not find a relationship between health literacy level and HIV risk behaviors (sex without a condom or sharing injecting equipment), controlling for age, race, and problem drinking.¹²¹ A large study of adolescents and young adults (N = 4,751) in Cape Town, South Africa, found that higher literacy level (measured using the Cape Area Panel Study Literacy and Numeracy Evaluation) was associated with a lower probability of sexual debut but not first pregnancy, controlling for socioeconomic variables.¹²² The research team judged the strength of evidence to be insufficient based on mixed findings (Table 16 and Appendix F). Our earlier review included no studies with these outcomes.

Health care-related skills. Eleven studies reported in 13 articles included outcomes concerning a variety of health care-related skills (Table 19). Among these were appropriate medication use;^47,123-127 interpreting prescription medication, nutritional labels, and health messages;^{9,75-77,102,128} and asthma self-care skills.⁷⁹

Table 19

Summary of studies of the relationship between health literacy and the outcome of health care related skills (KQ 1b).

Taking medications appropriately. Three studies directly observed whether participants could take prescription medications appropriately; their results generally found a relationship with health literacy level. In one study we rated good quality, researchers required 152 coronary heart disease patients to perform four tasks relating to their medication: identify the appropriate medication, open the container, select the correct dose, and report the appropriate timing of doses.¹²³ The researchers found no difference across health literacy levels in patients' scores from completing all four tasks in an unadjusted analysis. However, after controlling for age, education, and cognitive functioning, low health literacy (but not marginal health literacy) was associated with poorer performance on one of the tasks—being less likely to identify all of one's medications (OR, 12.00; 95% CI, 2.57-56.08). Using a similar approach, a second team of researchers conducted a mock exercise concerning successful medication management (Medication Management Test) among HIV-positive patients.⁴⁷ Patients with higher health literacy scored significantly higher in an adjusted analysis. Similarly, in a small sample of seniors in Texas (N = 57), researchers found that lower health literacy (measured continuously) was associated with poorer ability to open and take one's own medications, in adjusted analysis.¹²⁴

Three additional adjusted analyses examined other measures of whether patients take medications properly, the first through self-report, the second through direct observation, and the third through biologic test results, and found limited evidence of a relationship with health literacy level.^125-127 One study examined whether health literacy level was associated with parents' use of nonstandardized dosing instruments (such as kitchen spoons) when providing medications to their children; they found no relationship in an analysis adjusting for all identified potential confounding variables.¹²⁵ However, after removing from the adjusted analysis only the variables in the analysis that were confounded with health literacy level (caregiver's education, country of origin, language, and socio-economic status), participants with marginal/inadequate health literacy (combined into one group) were more likely to use nonstandardized instruments than those with adequate health literacy (OR, 1.9; 95% CI, 1.0-3.5). In a second study, researchers tested parents' health literacy level using the Newest Vital Sign and evaluated whether they made dosing errors using common dosing instruments (i.e., dosing cups, droppers, dosing spoons, and syringes).¹²⁷ Parents with a high likelihood of limited health literacy and those with possible limited health literacy were significantly more likely to make a dosing error (greater than 20 percent deviation) than parents with adequate health literacy, in adjusted analyses; parents with a high likelihood of limited health literacy were significantly more likely to make a large dosing error (greater than 40 percent deviation). One study examined warfarin control measured by international normalized ratio (INR) variability. Results did not differ by health literacy level, controlling only for age, in a population of adults 50 years of age and older.¹²⁶

Interpreting labels and health messages. Two studies examined participants' ability to interpret labels (prescription medications and nutrition); both found a positive relationship with health literacy level. One study among 395 adult patients in three primary care clinics in Shreveport, Louisiana, Jackson, Michigan, and Chicago, Illinois, examined interpretation of prescription medication labels.^75-77 Participants demonstrated their ability to understand prescription label instructions by describing to physicians how they would take five medications in adjusted analyses, those with inadequate health literacy (RR, 2.32; 95% CI, 1.26-4.28) as well as those with marginal health literacy (RR, 1.94; 95% CI, 1.14-3.27) had a greater probability of misunderstanding one or more label instructions than those with adequate health literacy.⁷⁵ A further (unadjusted) examination of participants' correct interpretation of each of the five primary labels found significant differences in interpretation of four of five primary medication labels. They also found differences in whether participants attended to auxiliary labels in two of five comparisons.⁷⁶ Lastly, researchers found in an adjusted analysis that those with lower health literacy (less than high school level) were less likely to understand nutrition labels.⁹

One study examined health literacy and the ability to give an organized oral health narrative. Among a community sample of mothers of young children in Nepal, higher literacy level was associated with greater ability to give an organized health narrative (a skill associated with higher oral health literacy) in an adjusted analysis.¹²⁸

Asthma self-care. One study examined self-care skills relating to asthma among hospitalized adults.⁷⁹ In adjusted analysis, those with inadequate health literacy, compared with those with adequate literacy, were less likely to have mastery of their dose inhaler (OR, 0.29; 95% CI, 0.08-1.00). We had found a similar result in our earlier review.¹²⁹

Health care-related skills strength of evidence. The research team separately determined that the strength of evidence concerning taking medications appropriately and interpreting labels and health messages was moderate and the strength of evidence concerning asthma self-care was low (Table 16 and Appendix F). Our earlier review included one health-care-related skills study concerning asthma self-care. ¹²⁹

Disease prevalence and severity. We found multiple studies examining the relationship between health literacy level and disease prevalence (specifically, mental health diagnoses and chronic conditions) or disease severity (specifically, HIV, asthma, diabetes, hypertension, and prostate cancer).

Mental health outcomes. Eight of ten studies evaluating the relationship between depression and health literacy level found that patients with lower health literacy were more likely to have symptoms of depression or to be considered depressed; however, the majority of studies controlled for a limited number or no potential confounders.^{68,95,103,130-135} One additional study examined the relationship between health literacy level and psychological distress⁸² (Table 20). In the most rigorous study of depression (a prospective cohort conducted among 390 patients receiving inpatient detoxification from alcohol and substance abuse), depression symptomatology did not differ between health literacy groups at baseline, but was higher among those with lower health literacy at 2-year followup, controlling for a number of potential confounders including sociodemographic characteristics, primary substance of choice, and mental state.¹³⁰ Other analyses were conducted among subpopulations with limited adjustments for potential confounders. One reported that depression was greater in the lower-health-literacy group among HIV-positive adults in five urban clinics, controlling for Hispanic nationality.¹³¹ A second reported that depression was also greater among pregnant patients with lower (but not marginal) health literacy, controlling for Mexican nativity and marijuana use.¹³² Finally, a third that depression scores were higher among recent Spanish-speaking immigrants in the low-health-literacy groups, controlling for a scale measuring the demands of immigration.¹³⁵ In unadjusted analyses, lower health literacy was also related to depression among rheumatology and diabetes patients^133,134 and among seniors in two community samples.^68,95 However, no difference by health literacy level was found among HIV-positive patients in Atlanta.¹⁰³ In relation to psychological distress, differences were not found by health literacy level among HIV-positive adolescents.⁸²

Table 20

Summary of studies of the relationship between health literacy and the outcome of prevalence of depression and other mental health outcomes (KQ 1b).

The research team judged the strength of evidence to be low because, although studies generally found consistent results, only one rigorously controlled for potential confounders (Table 16 and Appendix F). Results of studies evaluating differences in depression across different levels of health literacy in our earlier review were mixed, including among the two studies that controlled for potential confounders.^136-140

Chronic disease outcomes and prevalence. Three studies examined differences in rates of chronic disease (defined in a group as any long-term illnesses) by health literacy level (Table 21).^9,65,141Four additional studies examined differences in rates of specific diseases by health literacy level.^{66,68,95,142,143}

Table 21

Summary of studies of the relationship between health literacy and the outcome of prevalence of chronic diseases (KQ 1b).

Using the large, nationally representative NALS (N = 23,889), researchers found that lower health literacy was associated with higher odds of having a long-term illness (one lasting more than 6 months) and greater odds of having a condition that would keep the individual from working after controlling for various sociodemographic characteristics including education.¹⁴¹ In other studies with unadjusted analyses, the number of chronic conditions among seniors and the percentage with a chronic disease among adults in a clinic population did not differ by health literacy level.^9,65

Three studies, discussed in four articles, examined differences in rates of specific diseases by health literacy level; one used a well-designed adjusted analysis and the others used unadjusted analyses.^66,68,95,142 All analyses were limited to senior citizens. In adjusted good-quality analyses of the Prudential sample, inadequate compared with adequate health literacy was associated with significantly higher rates of diabetes and heart failure, but not with higher rates of hypertension, coronary heart disease, bronchitis, asthma, arthritis, or cancer.⁶⁶ In contrast, the investigators found no differences in rates of specific diseases between those with marginal and adequate health literacy. Potential limitations of this analysis are that respondents' outcomes are self-reported shortly after joining the health plan and differences in prior access to care may have resulted in differences in knowledge concerning their disease state. Also, by testing multiple outcomes, significant differences were more likely to be found in at least some of the comparisons. Two unadjusted analyses measured the probability of differences in prevalence of chronic disease across three health literacy levels; however, their design was insufficient to determine if differences existed between any two groups (inadequate compared with adequate or marginal compared with adequate).^68,95 A third unadjusted analysis among seniors in Korea found that health literacy was associated with significantly higher rates of arthritis and hypertension, but not sensory disease, diabetes, or pulmonary or heart disease.¹⁴²

Among individuals with diabetes, heart failure rates were higher in the limited health literacy group in one bivariate comparison.¹⁴³

Overall, the body of evidence found mixed results and was limited by differences in outcomes across studies with the majority of studies not controlling for potential confounders. Given these issues, the strength of evidence was graded insufficient (Table 16 and Appendix F). Our earlier review found one study of children with migraines and no relationship was found.¹⁴⁴

HIV infection severity and symptoms. Three adjusted and one unadjusted analyses of individuals with HIV did not find differences in severity of HIV (measured by viral load suppression, CD4 cell counts, and number of HIV symptoms) by health literacy level (Table 22).^{82,103,105,145} In contrast, higher health literacy was associated with greater symptom intensity in one study controlling only for Hispanic ethnicity.¹³¹ In this study, health literacy was measured as a continuous variable among a population with relatively high health literacy (REALM mean score = 59.1). Even though four of five studies found no relationship, the research team evaluated the strength of evidence as low because these studies included limited control for confounding and had small sample sizes (Table 16 and Appendix F). Our earlier review was limited to unadjusted analyses and found mixed results.^138,146,147

Table 22

Summary of studies of the relationship between health literacy and HIV patient symptoms (KQ 1b).

Asthma severity and control. The relationship between health literacy and asthma severity of children was examined in two studies reporting a mix of adjusted and unadjusted analyses (Table 23).^80,84 Both studies measured asthma severity by parent report. In one, an adjusted analysis concluded that lower-health-literacy parents of children with asthma were more likely to report that their children were in fair or poor health; however, in an unadjusted comparison, these same parents' reports of their children's asthma control did not differ by health literacy level.⁸⁴ In a different unadjusted analysis, parents with lower health literacy reported greater use of albuterol (a bronchodilator) by their children, indicating poorer asthma control.⁸⁰ Overall, the strength of evidence was insufficient (Table 16 and Appendix F).

Table 23

Summary of studies of the relationship between health literacy and asthma patient symptoms (KQ 1b).

Diabetes control, complications, and related outcomes. Five adjusted studies examined the relationship between glycosylated hemoglobin (HbA1c) level and health literacy level and found mixed results (Table 24).^134,148-151 One good-quality study measuring the HbA1c levels in 1,002 diabetic adults in Vermont found no relationship with health literacy level after measuring health literacy as a continuous variable using the TOFHLA and controlling for demographic characteristics and several factors related to successful diabetes control, such as duration, diabetes education, medication, and alcohol use.¹³⁴ Similarly, a second good-quality study conducted with diabetic patients in the Midwest also found no relationship between HbA1c and health literacy levels after controlling for different factors related to successful diabetes control including patient trust, depression, diabetes knowledge, and performance of self-care activities. The lack of a finding of association between health literacy and the outcome may be due to over-adjustment given that researchers controlled for potentially mediating variables in this analysis.¹⁵¹ In contrast, a very small study (N = 68) from one general internal medicine clinic found significant differences in HbA1c between the four health literacy levels; each increasingly higher level of health literacy, however, was not associated with better control.¹⁴⁹ In a good-quality study, using a path analysis statistical technique and controlling for potential confounders, researchers found that higher health literacy was related to better glycemic control and that health literacy mediated the direct relationship between education and HbA1c level.¹⁵⁰ Also, in a study conducted in Hong Kong, higher-health-literacy diabetic patients had better glycemic control.¹⁴⁸

Table 24

Summary of studies of the relationship between health literacy and diabetes control (KQ 1b).

The large study of diabetic patients in Vermont, did not find health literacy level to be related to blood pressure, cholesterol level, or the probability of having other potential side effects of poor diabetes control (retinopathy, nephropathy, foot or leg problems, gastroparesis, cerebrovascular disease, or coronary artery disease) after adjusting for confounders.¹³⁴

The strength of evidence relating to diabetes outcomes from this review was insufficient (Table 16 and Appendix F). In our earlier review, diabetes-related results were mixed.^129,152,153

Hypertension control. Two studies examined blood pressure control among patients diagnosed with hypertension; results were mixed (Table 25).^154,155 The larger study (N = 1,224), measuring health literacy using the REALM, did not find a significant main effect between systolic blood pressure and health literacy level (limited compared to adequate), controlling for education level, diabetes status, medication adherence, smoking, exercise, and participatory decisionmaking.¹⁵⁴ However, the interaction between health literacy and health care system was significant, indicating that the relationship between blood pressure and health literacy differed in the Veterans Administration vs. the private health care system. A second analysis (N = 330) measured health literacy using the S-TOFHLA subdivided into five categories and found that those in the lowest category were less likely than those in the highest category to have controlled blood pressure (less than 140 mmHg systolic and less than 90 mmHg diastolic [or less than 130 mm Hg systolic and less than 80 mm Hg diastolic among those with diabetes] RR, 2.68; 95% CI, 1.54-4.70) after controlling for sociodemographic characteristics, education level, insurance status, number of comorbid conditions, and years treated for hypertension.¹⁵⁵ In this study, the percentage of patients with controlled blood pressure was not consistently larger with every category of increasingly higher health literacy, and only some comparisons between various other health-literacy-level groups were significantly different. Based on mixed results, the research team judged the strength of evidence to be insufficient (Table 16 and Appendix F). Our earlier review did not find a relationship in hypertensive patients between blood pressure control and health literacy level in an adjusted analysis from the one study reviewed with this outcome.¹⁵⁶

Table 25

Summary of studies of the relationship between health literacy and hypertension control (KQ 1b).

Prostate cancer control. Prostate cancer patients with low health literacy (sixth grade or less) were more likely than those with adequate health literacy (ninth grade or higher) to have an elevated prostate-specific antigen (PSA) level in an adjusted good-quality study (OR, 2.5; 95% CI, 1.5-4.2) (Table 26).¹⁵⁷ In contrast, the marginal-health-literacy (seventh or eighth grade) group and the functional-health-literacy group did not differ. With only a single study, the strength of evidence was low (Table 16 and Appendix F). In our earlier review, stage of presentation of prostate cancer did not differ by health literacy level, in an adjusted analysis.¹⁵⁸

Table 26

Summary of studies of the relationship between health literacy and prostate cancer control (KQ 1b).

Global health status measures. Twelve studies reported in 14 articles examined health status differences by health literacy level among a variety of populations, including all adults, seniors, and adults with various specific disease states (Table 27).^{63,65,66,81,85,95,100,131,142,159-163} Health status was measured using an assortment of measures, including self-report of overall health status (excellent/very good/good/fair/poor) and physical and mental health subscales of the 12-Item Short Form Health Survey (SF-12) and SF-36, among others.

Table 27

Summary of studies of the relationship between health literacy and health status (KQ 1b).

Only one study measured self-reported health status among all adults (ages 18 to 85).¹⁵⁹ Limited to one clinic population in Canada, this work indicated that self-reported health status was not related to health literacy level after adjustment for confounders. With only a single study, the strength of evidence was low (Table 16 and Appendix F). Our earlier review found similar results in two adjusted analyses.^101,164

In studies limited to senior citizens, five studies, reported in six articles, all found differences in self-reported health status by health literacy level.^{63,81,85,95,142,160} Within a nationally representative sample (N = 2,668), one good-quality study reported that lower health literacy level measured through the NAAL was related to poorer self-reported health status, after adjusting for potential confounders.⁸⁵ Self-reported health status was also poorer in lower health literacy groups in three additional adjusted analyses: among Medicare patients in Chicago, Illinois,^81,160 in the Prudential study comparing differences between the low- and adequate-literacy groups (but not marginal- and adequate-literacy groups),^63,65 and among older Korean adults.¹⁴² The relationship was also found in one unadjusted analysis of 2,512 seniors in Pittsburgh, Pennsylvania, and Memphis, Tennessee.⁹⁵ The research team judged the strength of evidence to be moderate (Table 16 and Appendix F). In our earlier review, one unadjusted analysis from the Prudential study also found poorer overall health status among those with lower health literacy.¹⁶⁵

Three of the studies limited to seniors reported additional health status measures and results were mixed. In adjusted analyses, the Prudential study found lower health literacy to be associated with poorer physical- and mental-health-related quality of life and physical functioning in both the inadequate- and the marginal-literacy groups (SF-36) compared with the adequate group.^63,65,66 In contrast, a sample of Medicare beneficiaries in Chicago, Illinois, was not found to differ in physical or mental functioning by health literacy level.¹⁶⁰ One of these two studies, the Prudential study, also found that persons with inadequate health literacy had higher probabilities of having activity limitations, fewer accomplishments, and greater pain related to physical health than those with adequate health literacy.⁶⁶ Among Korean seniors, physical functioning (SF-12) did not differ by health literacy level in adjusted analyses, but significant differences were found in limitations in activities and pain that interfered with normal work.¹⁴² Given mixed results, the research team judged the strength of evidence to be insufficient (Table 16 and Appendix F).

Five studies examined differences in a variety of health status measures in adult populations with various diseases, including persons who were HIV-positive¹³¹ and patients with glaucoma,¹⁶¹ asthma,¹⁰⁰ spinal cord injuries,¹⁶² and cancer.¹⁶³ No more than one study examined each disease state, and results were mixed by disease state and outcome measure (e.g., general health, physical health, mental health, disease-specific quality of life). In HIV patients, better global physical health (using a scale developed by the researchers) was related to lower health literacy.¹³¹ In glaucoma patients, those with lower health literacy had poorer physical, but not vision or mental, quality of life based on quality-of-life scores.¹⁶¹ Among patients with spinal cord injuries, lower health literacy was associated with poorer physical morbidity, but not with mental health morbidity, physical health, or mental health status (SF-12).¹⁶² In cancer patients of all types, Functional Assessment of Cancer Therapy scores (related to physical and emotional functioning) and general health scores measured by the SF-36 showed no difference by health literacy level.¹⁶³ In asthma patients, lower health literacy was associated with poorer asthma quality of life (Asthma Quality of Life Quotient) and physical health status (SF-36), adjusting for asthma severity and asthma self-sufficiency.¹⁰⁰ However, the relationship with both outcomes was no longer significant after the investigators added age, education, depressive symptoms, and knowledge confounders to their analyses. Based on mixed results, the research team judged the strength of evidence as insufficient (Table 16 and Appendix F). In our earlier review of studies of global health measures, two unadjusted studies found no significant relationship.^139,166

Mortality. Differences in all-cause mortality rates of seniors were related to health literacy in adjusted analyses in two good-quality studies reported in three articles (Table 28).^65,67,167 The Prudential study reported higher mortality rates in the inadequate health literacy group than in the adequate health literacy group—first in an analysis controlling for cognitive functioning⁶⁷ and second in an analysis not controlling for cognitive functioning but instead controlling for baseline measures of disease, physical functioning, and healthy lifestyle.⁶⁵ Both analyses did not find significant differences between the marginal- and the adequate-health-literacy groups. In a population of seniors in Pittsburgh, Pennsylvania, and Memphis, Tennessee, those with limited health literacy had a higher all-cause mortality rate than those with adequate health literacy.¹⁶⁷ The Prudential study also reported, in adjusted analyses, higher cardiovascular-related mortality in the inadequate- and marginal-health-literacy groups than in the adequate group, but no differences in cancer-related mortality across health literacy levels.⁶⁵ The research team graded the strength of evidence as high (Table 16 and Appendix F). No studies examining the association between health literacy and mortality were included in our earlier review.

Table 28

Summary of studies on the relationship between health literacy and mortality (KQ 1b).

KQ 1c. Costs of Health Care

KQ 1c concerns differences in health literacy level and costs of health care (Table 29). The Prudential study of new Medicare managed care enrollees examined costs over a 1-year period. In adjusted analyses, inadequate- and marginal-health-literacy groups had higher emergency department costs; however, no other patterns of differences were uncovered in relation to overall, inpatient, outpatient, or pharmacy costs.⁶⁸ In contrast, total Medicaid costs were higher in the lower literacy group (less than third grade) among a small sample of beneficiaries in Arizona (N = 74).¹⁶⁸ Our earlier review found no relationship between literacy and Medicaid costs.¹⁶⁹

Table 29

Summary of studies of the relationship between health literacy and costs (KQ 1c).

In summary, the strength of evidence concerning differences by health literacy level in costs of health care (KQ 1c) was insufficient (Table 30 and Appendix F). The two relevant studies examined different payment sources (Medicaid and Medicare), found inconsistent results, and included different patient populations. No studies examined differences in costs among those with private health insurance coverage or no coverage.

Table 30

KQ 1c health literacy studies: strength of evidence grades by costs of health care.

KQ 1d. Disparities in Health Outcomes or Health Care Service Use

Eight studies examined whether health literacy mediates the relationship between race/ethnicity and health outcomes or use of health care services, and one study examined whether health literacy moderates the effect between race/ethnicity and health outcomes (Table 31). As described in more detail in Chapter 2, health literacy would be considered a mediator of racial differences in health outcomes, if differences in health literacy level between racial groups explain all or a portion of the outcome differences observed by race. Analytically, health literacy level is determined to be a mediator when health literacy is related to race or ethnicity and an outcome and when the coefficient for the race or ethnicity variable is smaller or becomes statistically insignificant after health literacy is added to the analytic model. Alternatively, the relationships can be observed through a path analysis.¹⁷⁰ Health literacy was found to mediate the effect of race on a variety of health outcomes in a variety of populations: on health conditions that keeps respondents from working and having a long-term illness in a nationally representative sample of adults included in the NALS,¹⁴¹ on self-reported health status and receipt of an influenza vaccine among seniors included in the nationally representative NAAL sample,⁸⁵ on physical and mental-health-related quality of life and self-reported health among seniors included in the Prudential study,⁶³ PSA levels among newly diagnosed prostate cancer patients in Chicago,¹⁵⁷ on nonadherence to HIV medications in a population of HIV patients,⁶⁹ on child health insurance among parents included in the NAAL sample,¹⁰² and misinterpretation of medication label instructions among adults.⁷⁷ The relationship was not found in relation to receipt of a mammogram or a dental checkup or parents' difficulty understanding over-the-counter medication labels in the NAAL study,^85,102 rate of receipt of vaccines in the Prudential study,⁶³ or glycemic control in diabetic adults.¹⁷¹

Table 31

Summary of studies of the relationship between health literacy and disparities (KQ 1d).

Only the NAAL study examined whether health literacy mediated the effect of ethnicity (Hispanic vs. white) on a health outcome, and this relationship was not found.⁸⁵ In contrast, only the study examining misinterpretation of medication label instructions in adults investigated whether health literacy was also a potential mediator of the relationship between gender and the outcome, as well as race; the relationship was found in this comparison as well.⁷⁷

Health literacy is determined to be a moderator of the relationship between race/ethnicity and health outcomes when the relationship is different in magnitude or direction between the two race/ethnicity groups. Only one study examined moderation and found no differences in the relationship between mortality and health literacy level in blacks and whites or males and females.¹⁶⁷

The strength of evidence was low in relation to health literacy level explaining racial differences in health outcomes based on findings of effect in some outcomes (Table 32 and Appendix F). The strength of evidence was low in relation to health literacy level explaining differences in health outcomes between Hispanics and whites and between males and females (Table 32 and Appendix F). Data were not available to examine disparities related to cultural or age group differences. In our earlier review, only one study was available to examine this issue, and it did not find that health literacy was a mediator of differences between black and white patients in late-stage prostate cancer diagnosis.¹⁵⁸

Table 32

KQ 1d health literacy studies: strength of evidence grades by disparities across health outcomes.

In summary, our research team found that health literacy mediates or partially explains disparities in health outcomes between white and black participants for a variety of outcomes; the strength of evidence for this conclusion is low because only one study examined each outcome (Table 32 and Appendix F). Health literacy was found to mediate outcome differences between blacks and whites in relation to the following outcomes: a health condition that keeps respondents from working or having a long-term illness, self-reported health status, receipt of an influenza vaccine, physical and mental-health-related quality of life, self-reported health among seniors, prostate-specific antigen levels among newly diagnosed prostate cancer patients, nonadherence to HIV medications, children's lack of health insurance, and misinterpretation of medication labels. We cannot know whether health literacy level would also mediate racial disparities for other health outcomes that have not been tested. Only one study examined whether health literacy level mediated the relationship between race and health outcomes for persons of Hispanic ethnicity and whites, and one study examined the relationship between males and females. The strength of evidence for these relationships was low. We found no studies that evaluated disparities related to differences in age, cultural group, or other sociodemographic characteristics.

Key Question 1. Relationship of Numeracy to Various Outcomes and Disparities

We identified 16 unique studies of the relationship between numeracy and outcomes of interest (Table 33). Nearly all studies examining the relationship of numeracy to health outcomes were cross-sectional in design.^{9,10,24,47,98,125,171-179} Four studies were randomized controlled trials (RCTs) that analyzed their data in a cross-sectional manner for this analysis,^{24,98,172,173} and one used a prospective cohort design.¹²⁶ Fifteen studies were of fair quality; only one was of good quality.¹⁷¹

Table 33

Overview of numeracy studies.

Studies employed a wide variety of numeracy measures. These included the WRAT-3, the Lipkus numeracy test, the Schwartz and Woloshin numeracy test (or adaptations thereof), the Diabetes Numeracy Test, the Black and Toteson numeracy test (or adaptations thereof), and the TOFHLA numeracy test. Using these measures, populations studied had a varying proportion of individuals with low numeracy (ranging from 5 percent to 74 percent).

Studies also examined a wide variety of outcome measures. Among them were the accuracy of the use of health care services, accuracy of risk perception, knowledge, self-efficacy, actual behaviors, skills, disease prevalence and severity, and disparities. No studies measured intent for behavior, adherence, quality of life, or costs.

Six studies measured both literacy and numeracy.^{9,47,98,125,126,171} This allowed assessment of whether these exposures affect health outcomes differently.

KQ 1a. Use of Health Care Services

One cross-sectional study¹⁷⁸ examined the effect of numeracy on use of health care services (Table 34). This study¹⁷⁸ focused on the effects of numeracy on use of screening services.

Table 34

The relationship between numeracy level and use of health care services (KQ 1a).

Screening services. In adjusted analyses, researchers reported no effect of numeracy level on up-to-date screening for either breast or colon cancer in women presenting for primary care.¹⁷⁸ However, the sample for colon cancer screening was small (N = 152; 58 percent of the total sample due to age ineligibility for screening for colon, but not breast cancer), and the authors provided no power calculations for either analysis.

Summary. In summary, only one study addressed the relationship between numeracy and use of health care services and reported no effect, possibly due to inadequate power. Based on this study, our research team judged the strength of the evidence for the relationship between numeracy and use of health care services to be low (Table 35 and Appendix F).

Table 35

KQ 1 numeracy studies: strength of evidence grades by use of health care services and health outcomes.

KQ 1b. Health Outcomes

Accuracy of risk perception. Five studies addressed the effects of numeracy level on accuracy of risk perception (i.e., whether individuals correctly perceived their health risks and treatment benefits) (Table 36). Three were RCTs^24,172,173 and two were cross-sectional studies,^173,176 although all analyzed their data in cross-sectional fashion to answer this question. Two examined the effects of numeracy on the accuracy of perceived risk^175,176 and four on the accuracy of perceived treatment benefit.^{24,172,173,176} All used the Schwarz and Woloshin 3-item numeracy test to assess numeracy level.

Table 36

The relationship between numeracy level and accuracy of risk perception (KQ 1b).

The two studies examining perceived risk found no effect of numeracy level on the accuracy of perceived risk of breast cancer or breast cancer survival over 5 years.^175,176 One study, however, reported that for every additional numeracy question answered incorrectly (scale range 0-3), participants' error in estimating lifetime risk increased by 18 percent (95% CI, 5-30%).¹⁷⁵

Four studies examined the effect of numeracy on the accuracy of perceived treatment benefit and found mixed results. Three studies reported lower accuracy of perceived treatment benefit at lower levels of numeracy (0-1 questions correct vs. 3 questions correct).^24,172,173 Notably, the size of the effect was smaller in the one study that adjusted for covariates including age, income, education, and the framing of information about treatment benefit (e.g., relative risk reduction or absolute risk reduction).²⁴ The fourth study, which also performed adjusted analysis, reported no significant difference between groups,¹⁷⁶ but the authors dichotomized their numeracy exposure variable differently (0-2 questions correct vs. 3 of 3 questions correct).

Interestingly, results varied across studies by how the investigators assessed accuracy. The differences in accuracy of perceived treatment benefit were greater between low- and high-numeracy participants who were asked to calculate an exact treatment benefit than between those who were asked merely to say which of two treatments provided more benefit.^172,173

Considering all of these studies in aggregate, our research team judged the overall strength of evidence about the relationship between numeracy and accuracy of risk perception to be insufficient due to mixed results by task and study (Table 35 and Appendix F).

Knowledge. We found four cross-sectional studies addressing the effect of numeracy level on knowledge (Table 37).^{125,174,177,178} These focused on different types of knowledge as well as different health topics and conditions, including diabetes,¹⁷⁴ general health and HIV,¹⁷⁷ breast and colorectal cancer screening guidelines,¹⁷⁸ and medication dosing.¹²⁵ Results were mixed.

Table 37

Relationship between numeracy level and knowledge (KQ 1b).

Three studies,^174,177,178 including two that adjusted for relevant covariates,^177,178 showed significantly lower knowledge about diabetes, HIV, and breast cancer screening with lower numeracy. These same studies, however, showed no effect of numeracy on general health knowledge or colorectal cancer screening, although nearly half of the sample queried about colorectal cancer screening included individuals who were too young to be eligible for screening. A fourth study showed lower numeracy to be related to lower knowledge about medication dosing in an analysis controlling for some confounders;¹²⁵ however, results became nonsignificant after additional adjustment for education, acculturation, and socioeconomic status.

Considering these studies in aggregate, our research team judged the overall strength of evidence regarding the relationship between numeracy and knowledge to be insufficient (Table 35 and Appendix F).

Self-efficacy. One cross-sectional study examined the effects of numeracy level on self-efficacy (Table 38).¹⁷⁴ In an unadjusted analysis, this study found significant reductions in self-efficacy (a 4-point reduction on the Perceived Diabetes Self-management scale ranging from 8 to 40) among those who scored in the lowest vs. the highest quartile of the Diabetes Numeracy Test. Based on this single unadjusted analysis, the overall strength of evidence about the relationship between numeracy and self-efficacy was insufficient (Table 35 and Appendix F).

Table 38

Relationship between numeracy and self-efficacy (KQ 1b).

Intent for behavior. We found no studies that examined the effect of numeracy on intent for behavior.

Behavior. One cross-sectional study examined the effects of numeracy level on behavior (Table 39).¹⁷⁴ In unadjusted analysis, this study found no significant differences in diabetes self-management behaviors in four of five domains of the Diabetes Self-Care Activities Scale, including general diet behavior, specific diet behavior, exercise behavior, or blood glucose testing. However, there were small increases in foot care behavior (+2.25 on a scale of 0-7; P < 0.001) among those in the lowest vs. highest quartile of numeracy; these unexpected results (as well as the negative results for analyses of other self-care behaviors) may be the result of confounding. Based on this single unadjusted analysis, our research team judged the overall strength of evidence about the relationship between numeracy and self-efficacy to be insufficient (Table 35 and Appendix F).

Table 39

Relationship between numeracy level and behavior (KQ 1b).

Health-related skills. Six studies examined the effects of numeracy level on health-related skills (Table 40). One was a cohort study,¹²⁶ four were cross-sectional studies,^9,47,125,179 and one was an RCT that analyzed data in cross-sectional fashion.⁹⁸ The skills included taking medication, reading nutrition labels, and assessing health plan materials.

Table 40

Relationship between numeracy level and skills (KQ 1b).

The four studies that focused on skills in taking medication found mixed results. In analyses adjusted for age, one found mixed effects of numeracy on two different but related variables denoting medication-taking skill: the proportion of INR tests within range (adjusted absolute difference, NR; P = 0.35) and INR variability (adjusted absolute difference, NR; P = 0.03).¹²⁶ Other studies measured medication-taking skill more directly and still found mixed effects. One study found a relationship between numeracy and HIV medication management capacity after adjusting for gender, education, health literacy, and time since HIV diagnosis (0.5-point increase in Medication Management skill [range 2-16] for every 1-point increase in the Applied Problems subtest of the Woodcock Johnson Test; P < 0.01).⁴⁷ Another study reported that, after adjustment for some confounders, poor caregiver numeracy resulted in use of nonstandardized dosing instruments for administering medications to children.¹²⁵ Additional adjustment for education, acculturation, and socioeconomic status, however, led to nonsignificant differences between groups, based on TOFHLA numeracy scores split at the median. Finally, a third study found that poor caregiver numeracy (second through eighth grade on the WRAT-math) was associated with (1) an increased likelihood of thinking a potentially harmful over-the-counter medication to be suitable (adjusted OR, 1.25; 95% CI, 0.99-1.58), although results were not statistically significant, and (2) increased intent to use potentially harmful over-the-counter cold medicines in a 13-month-old (adjusted OR for each decrease in numeracy skill level, 1.19; 95% CI, 1.01-1.41). This study also reported that, paradoxically, for caregivers with higher numeracy (9th-16th grade), each increase in numeracy grade level made them more likely to intend to use over-the-counter cold medicines (adjusted OR for each increase in numeracy skill level, 1.78; 95% CI, 1.07-2.96). Investigators attributed this finding to heavier reliance on independent judgment. Importantly, however, analyses were not adjusted for potentially relevant confounders, such as prior physician prescriptions for these medications. Based on these studies, our research team judged the overall strength of evidence regarding the relationship between numeracy and skills in taking medication to be insufficient (Table 35 and Appendix F).

The studies assessing other outcomes—skill at reading nutrition labels⁹ and at reviewing health plan materials⁹⁸—found lower comprehension of reviewed materials in participants with lower numeracy. However, only the nutrition label study adjusted for potential confounders. Additionally, the health plan study found fewer participants choosing a higher quality hospital among those with lower numeracy.⁹⁸ Interestingly, this result was moderated by patient activation; subjects who were more motivated to process information were also more likely to make higher quality choices, regardless of their numeracy level.

Based on these studies, our research team judged the overall strength of evidence regarding the relationship between numeracy and skill in interpreting health information as insufficient (Table 35 and Appendix F).

Disease prevalence and severity. Three cross-sectional studies examined the effect of numeracy level on disease prevalence and severity (Table 41).^9,10,174 These studies addressed the effects of numeracy on BMI,^9,10 HbA1c,¹⁷⁴ and illness requiring dietary restriction.⁹

Table 41

Relationship between numeracy level and disease prevalence and severity (KQ 1b).

The two studies addressing the effect of numeracy (measured by the WRAT-3 numeracy test) on BMI found mixed results in patients drawn from the same academic medicine practice. In one study, those scoring below the ninth-grade level on the WRAT-3 had higher mean BMIs (adjusted beta coefficient, 0.14; P = 0.01).¹⁰ By contrast, the other study reported no effect of differential WRAT-3 scores on obesity (BMI greater than 30) in unadjusted analysis.⁹ The differences in findings may be attributable to a combination of differences in recruiting (physician referral in the Huizinga study), handling of the outcome variable (continuous in the Huizinga study, categorical in the Rothman study), and adjustment in analysis (adjusted in the Huizinga study, unadjusted in the Rothman study).

Findings on other health outcomes were also mixed. One study reported modest effects of numeracy on HgbA1c (adjusted beta coefficient 0.09 for every 10-percentage-point decrease in the proportion of correct responses on the Diabetes Numeracy Test).¹⁷⁴ A second study, however, reported no effects of numeracy on the proportion of individuals with illness requiring diet restriction in unadjusted analysis.⁹

Given the mixed nature of results, our research team judged the overall strength of evidence regarding the relationship between numeracy and disease prevalence to be insufficient (Table 35 and Appendix F).

Summary. In summary, studies of the relationship between numeracy skill level and many health outcomes (including accuracy of risk perception, knowledge, skills taking medication, and disease prevalence and severity) found mixed results. Based on these findings, we judged overall strength of evidence for its relationship to these outcomes to be insufficient.

The relationship between numeracy skill level and other outcomes is also uncertain. One study suggests a possible relationship between numeracy skill level and label-reading skill. Additionally, only one study each addressed the relationships between numeracy and self-efficacy or behavior (both with unadjusted analyses), making conclusions impossible.

KQ 1c. Costs

We found no study that examined the effect of numeracy level on costs.

KQ 1d. Potential Mediator of Disparities

We found two studies that addressed the effects of numeracy as a potential mediator of disparities in health outcomes.^47,171 One examined numeracy as a potential mediator of the relationship between race and HgbA1c.¹⁷¹ The other examined numeracy as a potential mediator of the relationship between gender and HIV medication management capacity.⁴⁷ Both used formal mediational analyses.

Table 42

Relationship between numeracy level and disparities (KQ 1d).

In the study examining numeracy as a potential mediator of the relationship between race and HgbA1c, investigators used path analysis and structural equation models to examine the relationships between race, numeracy, and HgbA1c in a cross-sectional sample of 383 diabetic patients who received care at primary care and diabetes specialty clinics at three medical centers. Investigators demonstrated significant negative relationships between both African-American race and numeracy (standardized path coefficient, -0.46; P < 0.001) and numeracy and HgbA1c (standardized path coefficient, -0.15; P < 0.01). They additionally demonstrated that the relationship between African-American race and HgbA1c (standardized path coefficient, 0.12; P < 0.01) lessens and becomes nonsignificant with the addition of numeracy (standardized path coefficient, 0.10; P = NS), suggesting partial mediation of racial disparities by numeracy.

In the study examining numeracy as a potential mediator of the relationship between gender and HIV medication management capacity, investigators also used path analysis to examine the relationships between gender, numeracy, and HIV medication management capacity in a cross-sectional sample of 155 HIV-positive patients recruited from clinics or drug assistance programs in Miami, Florida. In this study, investigators demonstrated a significant negative relationship between female gender and numeracy (path coefficient, -0.428; P < 0.01) and a significant positive relationship between numeracy and medication management capacity (path coefficient, 0.644; P < 0.01). They additionally demonstrated that the correlation between female gender and medication management capacity (path coefficient = NR) lessened and became nonsignificant (path coefficient, 0.073; P = NS) with the addition of numeracy to the model. These findings suggest partial mediation of gender disparities in medication management capacity by numeracy. Our research team judged the overall strength of evidence to be low (Table 35 and Appendix F).