From the 4765 titles and abstracts screened, 114 articles were included in the full-text phase. 52 articles subsequently met the eligibility criteria and were included in the rapid review. Two further papers were added based on hand-search of reference lists of the included articles, bringing the total number of articles analysed to 54 (Fig. 1).
Main reasons for exclusion during the full-text screening were not reporting modifiable predictors (n = 29), the study population not matching inclusion criteria (n = 18) and not assessing HL as an outcome but as an exposure for another outcome (n = 8). Six further articles that did not provide the mean age of the study population were not included in the main analyses, but in sensitivity analyses (see Additional file 3).
The 54 studies included in the review were conducted in 26 different countries, with the majority being conducted in the USA (n = 11) (20–30), in Iran (n = 8) (31–38), in Turkey (n = 5) (39–43) and in Australia (n = 3) (44–46).
With one exception (cohort study), all included studies had a cross-sectional (n = 53) design. The sample sizes of the studies ranged from n = 75 to n = 8362 and the average sample size was n = 818. All studies were written in English and published between 2009 and 2021.
The study populations were heterogeneous and included university, nursing and college students (26, 33, 36, 41, 47–51), teachers (including preschool, primary and secondary class teachers) (43, 52), clinical populations/patients attending healthcare clinics (25, 29, 44, 45, 53–58), immigrant populations (20, 21, 28), general adult populations (according to the place where the study was conducted) (35, 37–39, 46, 59–64) and prisoners (65).
Nevertheless, the objectives were similar across included articles. Two main aims could be identified: (a) depicting HL levels of the population under investigation and (b) assessing determinants and associated factors of HL. Further aims included the validation of HL survey tools, oral HL tools and e-HL scales.
Health literacy outcomes and instruments
Studies assessed general HL (n = 34), e-health literacy (n = 9), oral health literacy (n = 8) and mental health literacy (n = 2). The instruments employed most often were the HLS-EU-Q47 (61, 65, 66), the HLS-EU-Q16 (40, 67), the Newest Vital Sign (NVS) (23, 28, 29, 43, 60) the HLS-SF12 (64, 68) and the S-TOFHLA (25, 30, 53). All tools are given in the data extraction table (see Additional file 1).
Risk of bias analysis
All included studies were peer-reviewed and we rated the quality of the articles as good or fair (via NIH Quality Assessment Tool). The quality of 22 articles was rated as good and 32 articles as fair (Fig. 2, for more details see Additional file 4). However, the majority of the included articles had a cross-sectional design, presenting a limitation in itself when associated factors or determinants are studied.
Modifiable predictors of HL
In n = 54 articles, we found more than 20 potentially modifiable predictors of HL. We included n = 41 articles to perform 9 meta-analyses on the associations between modifiable determinants and HL (language proficiency, frequency of internet use, internet as information source, watching health-related TV, smoking, alcohol consumption, physical activity, oral health behaviours and health status) (Fig. 3–11). The remaining studies either examined individual determinants exclusively, or the data reported could not be converted to the effect size r. These studies are described narratively (see Additional file 1).
Ten studies examined the relationship between language proficiency and HL (Fig. 3), and the z-standardized correlation coefficients ranged from − 0.52 to 0.78 with 9 studies showing a positive association of high language proficiency with adequate HL. The pooled correlation was significant and positive, r = 0.38 [95% CI: 0.17, 0.58], and the Q-test suggests significant heterogeneity between studies (Q(9) = 200.69 p < 0.0001, tau²=0.11, I²=98.81%). One study (Morris et al. 2020) had a studentized residual larger than ± 2.81 and may be a potential outlier and might be overly influential according to Cook's distances. There was no indication of funnel plot asymmetry (see Additional file 5).
Frequency of internet use and computer skills
Six studies and nine analyses examined the relationship between frequency of internet use/computer skills and HL (Fig. 4) and the z-standardized correlation coefficients ranged from 0.06 to 0.72. All studies showed a positive association between high frequency of internet use and adequate HL with a pooled correlation coefficient of r = 0.35 [95% CI: 0.21, 0.50]. There was significant heterogeneity between studies, (Q(8) = 114.98, p < 0.0001, tau²=0.04, I²=91.93%). No outliers or overly influential studies were identified. There was no indication of funnel plot asymmetry (see Additional file 5).
Using the Internet as a source of health information/ for health-related purposes
Seven studies examined the relationship between using the internet as health information source and HL (Fig. 5) with six studies showing a positive association between using internet as an information source and adequate HL. The z-standardized correlation coefficients ranged from − 0.12 to 0.52 and the pooled correlation coefficient was r = 0.24 [95% CI: 0.09, 0.39]. The Q-test indicated significant heterogeneity (Q(6) = 67.84, p < 0.0001, tau²=0.04, I²=95.07%). One study (Kahouei et al. 2015) had a studentized residual larger than ± 2.69 and may be a potential outlier, none of the studies was overly influential and no indication of funnel plot asymmetry was given (see Additional file 5).
Watching health-related TV
Four analyses of three studies examined the relationship between watching health-related TV or reading health news and HL (Fig. 6), all showing a positive association with HL. Z-standardized correlation coefficients ranged from 0.30 to 0.65, and a pooled r of 0.48 [95% CI: 0.31, 0.65] was found. The Q-test showed significant heterogeneity. No outliers or overly influential studies were identified. The regression test (p < 0.0001), but not the rank correlation test (p = 0.33) indicated funnel plot asymmetry (see Additional file 5).
A total of eight studies examined the association between not smoking and HL (Fig. 7). Results differed substantially across the studies with z-standardized correlation coefficients ranging from − 0.27 to 0.68, and a non-significant pooled correlation coefficient of r=-0.06 [95% CI: -0.13, 0.25]. The Q-test suggests significant heterogeneity (Q(7) = 205.38, p < 0.0001, tau²=0.07, I²=98.57%). One study (Panahi et al. 2019) had a residual larger than ± 2.73 and may be a potential outlier. No study was considered to be overly influential. There was no indication for funnel plot asymmetry.
Five studies examined the relationship between alcohol consumption and HL (Fig. 8), showing no statistically significant association between low alcohol consumption and HL. The z-standardized correlation coefficients ranged from − 0.41 to 0.05 with a pooled r of -0.08 [95% CI: -0.24, 0.09]. The Q-test suggests significant heterogeneity (Q(4) = 101.12, p < 0.0001, tau²=0.03, I²=96.73%). One study (Yigitalp et al. 2021) may be a potential outlier according to residually, and might be overly influential according to Cook's distances. There was no indication for funnel plot asymmetry.
All five studies examining the relationship between not exercising regularly and HL (Fig. 9) indicated a negative correlation, whereby lack of regular exercise was associated with lower HL with z-standardized correlation coefficients ranging from − 0.22 to -0.10 and a pooled correlation coefficient of r=-0.16 [95% CI: -0.21, -0.10]. According to the Q-test, correlations were heterogeneous (Q(4) = 13, p = 0.01, tau²=0.003, I²=68.04%). No outliers or overly influential studies were identified and there was no indication for funnel plot asymmetry (see Additional file 5).
Oral health behaviour (dental visits and tooth-brushing behaviour)
Nine analyses in seven studies examined the relationship between oral health behaviours (brushing frequently or dental visits) and oral health literacy (Fig. 10), showing a positive association between oral health behaviours and oral health literacy with z-standardized correlation coefficients ranging from − 0.13 to 0.63 and a pooled r of 0.28 [95% CI: 0.10, 0.46]. The Q-test indicated significant heterogeneity (Q(8) = 156.26, p < 0.0001, tau²=0.07, I²=94.18%). There was no indication of outliers or overly influential studies. Funnel plot asymmetry was suggested by the regression test (p = 0.02) but not the rank correlation test (p = 0.12) (see Additional file 5).
All eight studies on the relationship between health status and HL showed a positive association between ‘good ‘health and adequate HL (Fig. 11). The z-standardized correlation coefficients ranged from 0.07 to 1.07, and the pooled effect was calculated as r = 0.29 [95% CI: 0.06, 0.52]. The Q-test indicated significant heterogeneity (Q(7) = 168.19, p < 0.0001, tau²=0.07, I²=98.57%). One study (Noor et al. 2019) may be a potential outlier based on residuals and overly influential according to Cook´s distances. The regression test indicated funnel plot asymmetry (p = 0.08) but not the rank correlation test (p = 0.18) (see Additional file 5).
Further modifiable predictors: narrative review
Three studies reported multivariate analyses only and thus could not be considered for the meta-analyses. Michou et al. reported a significant association between not smoking, not drinking alcohol, being physical active and adequate HL (66). Kayupova et al. found an association of HL with low frequency of watching health-related TV in a multivariate linear regression model (61) and Shah et al. identified Body Mass Index (BMI) as being associated with HL using a logistic regression model (29). BMI was also identified being associated with HL in two further studies (40, 64).
A number of Additional potentially modifiable determinants were examined in one study only. All examinations converged in that higher levels of these determinants were associated with higher levels of HL. Some of these determinants were study time, (32), last cervical cancer pap test within 36 months (69), being member of a social organization (26), engaging in social groups (27), better information/access to books (60), adherence to Mediterranean diet (47), use of specific medical websites (33) and being a member of health club/welfare group in their community within last six months (52) (see Additional file 1 for summary of all study findings).
In six articles we could not identify if the study population matched our inclusion criteria. We looked at them separately and found similar results as in our main findings. These were: English proficiency, consuming all types of information on the internet and using more search strategies while looking for the information on the internet. Poor health behaviors as smoking, drinking alcohol, having a higher BMI and physical inactivity were associated with lower HL scores (see Additional file 3 for more details).