A Simple Scoring Algorithm for Health Literacy in Community-Dwelling Older Adults

Health literacy (HL) is the capacity to access, understand, appraise, and apply health information to make appropriate health decisions. This study aimed to establish a predictive algorithm for identifying community-dwelling older adults at a high risk of low HL. Methods: A total of 648 older adults were included and 85% was used to generate the prediction model for scoring algorithm while 15% was used to test the tness of the model. Pearson’s chi-squared test and multiple logistic regression were used to identify the factors associated with the HL level. An optimal cutoff point was identied based on the maximum sensitivity and specicity. 350 patients (54.6%) was classied as the low HL level. Twenty-four variables were identied for signicantly differentiating between high and low HL. Eight factors including socio-environmental determinant and health outcome related factors signicantly predicted low HL. The scoring algorithm yielded an area under the curve of 0.71 and optimal cutoff of 5 represented mediocre sensitivity (62.0%) and good specicity (76.2%). This simple scoring algorithm and effectively identify community-dwelling older low HL.

usage (14), and inadequate empowerment (15) and participation (16). Therefore, early and accurate prediction of the high risk of low HL among older adults has become essential worldwide in order to provide prompt and appropriate health care strategies.
Various HL measures been developed for older adults (17). However, most of them lack underlying theoretical basis and fail to su ciently cover comprehensive dimensions of HL across different clinical environments. Furthermore, to the best of our knowledge, no HL prediction model that can enable early and precise identi cation of HL levels in older adults has been developed. Therefore, developing a simple, cost-effective algorithm that can be applied in clinical settings to accurately identify older adults at a potential high risk of low HL is essential. For this purpose, this study conducted a survey among community-dwelling older adults to identify the factors in uencing low HL and constructed an optimal scoring algorithm for predicting HL.

Participants
In this cross-sectional study, by convenience sampling, we recruited eligible community participants who were 65 years or older from six senior service centers and three health check-up clinics in northern, central, and southern Taiwan between June and September 2018. Individuals with cognitive impairment based on screening by using the Mini-Cog instrument were excluded (18).

Procedure
The objective of the study was explained to respondents before they expressed their willingness to participate by trained interviewers. The survey was anonymous, and the respondents were allowed to suspend the interview at any time. This study was approved by Taipei Medical University-Joint Institutional Review Board (N201804046) and National Taiwan University Hospital (201804057RIND). After signing informed consent forms for participation in the study, the participants completed a self-administered questionnaire of 52 potential predictors, including personal, situational, and socio-environmental determinants and factors related to health service use, health costs, health behavior, health outcomes, participation, and empowerment, based on the theoretical model of the European Health Literacy Survey Consortium (19).

Outcome Measures
The 47-item European Health Literacy Survey Questionnaire (HLS-EU-Q), developed by the European Health Literacy Consortium, was used to assess the HL of the study participants. The HLS-EU-Q measures four HL competencies (access, understand, appraise, and apply health information) required under three health domains: HC (16 items), DP (15 items), and HP (16 items). Each item assesses the self-perceived di culty in performing selected healthrelated tasks on a 4-point scale ranging from "very easy" (4) to "very di cult" (1). Higher scores indicate higher HL. For ease of comparison, each domain (i.e., HC, DP, and HP) score was linearly transformed to a score between 0 and 50 by using a scale validated with satisfactory psychometric properties used in the European Health Literacy Survey (20). Based on the scores, HL was divided into four categories as following: inadequate (0-25), problematic (26)(27)(28)(29)(30)(31)(32)(33), su cient (34-42), and excellent (43-50) (21,22). We dichotomized the HL into "high" and " low" based on the cutoff value of 34, as de ned by the European Health Literacy Survey (21).

Statistical analysis
The dichotomized outcome is de ned using the HL level as follows: To develop a scoring algorithm for predicting low HL, the core data set was divided using strati ed random sampling without the replacement method as follows: 85% of the core data set was categorized into the training data set that was used for training the prediction model to create the scoring rule, and 15% of the data set was categorized as the validation test data set that was used for validating the scoring algorithm (23). The prediction model was generated using the training data set as follows: (1) Pearson's chi-squared test was used to assess the association of the HL level with each of the 52 self-administrated HL predictors. To select the most relevant predictors, variables with a p value of < 0.1 were included in the multiple logistic regressions. (2) Multiple logistic regressions with forward selection were used to examine relationships between low HL and the potential predictors classi ed into domains of personal determinants, situational determinants, socio-environmental determinants, health service use, health costs, health behavior, health outcomes, participation, and empowerment. The potential predictors with a p value of < 0.05 were further identi ed from the multiple logistic regression models (24,25). The multiple regression equation is as follows: where p denotes the probability of low HL in older adults, α is the intercept of the multiple regression, and β i is the slope of the main predictor (i = 1, 2, …, n). Odds ratio (OR) was estimated using exp(β i ). The measured β i, exp(β i ) or p is usually applied to calculate the clinical score for predicting health risk (26,27). A total of eight signi cant predictors were identi ed from multiple logistic regressions. (3) A simple algorithm was created based on the signi cant predictors identi ed from the multiple logistic regressions. Signi cant predictors that were positively associated with low HL were assigned a value of + 1, whereas those that were negatively associated with low HL were assigned a value of − 1.
A separate 15% of the participants were used to validate the proposed scoring algorithm. Based on the algorithm obtained from the training data set, the total score for each older adult in the test data set ranged from 0 to 8. Overall accuracies of low HL were classi ed with sensitivity, speci city, positive predictive value (PPV), and negative predictive value (NPV) (28). The model t was assessed on the basis of McFadden's pseudo R-square (measuring the reduction in maximized log-likelihood from the intercept only model) and c-statistic [area under receiver operating characteristic (ROC) curve, AUC] values (29,30). A two-sided 95% con dence interval (CI) for AUC was used to denote the uncertainty (31), and a p value of > 0.05 in the Hosmer-Lemeshow tting test was used to indicate the algorithm performance. The optimal cutoff score denoting the optimal classi cation threshold was the maximum value of sensitivity + speci city. All statistical analyses were computed using the SAS 9.4 software (SAS Institute, NC, USA).

Results
A total of 648 older adults were recruited. Figure 1 presents the age-speci c HL levels, which indicate that nearly half (41.6-46.1%) of the participants had problematic HL. A large portion of participants (72.6%) aged ≥ 81 years old had low HL. The sociodemographic characteristics of participants in the training and test data sets are presented in Table 1. Sex, age, education level, marital status, occupation, and monthly income were similarly distributed between the training and test data sets.

NTD: New Taiwan Dollar
In the training data set (n = 552), of the 52 variables of the original self-administrative questionnaire (Appendix Table 1), 24 factors (i.e., 5 personal determinants, 2 situational determinants, 2 socio-environmental determinants, 1 factor related to health service use, 1 factor related to health costs, 3 factors related to health behavior, 6 factors related to health outcomes, 1 factor related to participation, and 3 factors related to empowerment of HL) associated with the HL level (p < 0.1) were identi ed using Pearson's chi-squared tests ( Table 2).   Table 4. The optimal cutoff point was considered to be 5, yielding a sensitivity and speci city of 62.0% and 76.2%, respectively. By using a score of 5 out of 8 to predict the low HL level, the obtained PPV and NPV were 75.6% and 62.7%, respectively. Figure 2 presents the predictive ability of the scoring algorithm among older adults in the test data set. The indicators of model performance revealed a reasonably satisfactory performance with an AUC of 0.71 (95% CI: 0.61-0.81).

Discussion
To the best or our knowledge, this is the rst study to develop a model for predicting the HL of community-dwelling older adults. This algorithm-based model was well calibrated by integrating HL-related factors in the model of the European Health Literacy Survey Consortium and is useful in HL risk prediction among older adults. In addition, it has a modest ability to discriminate between older adults with high HL and low HL.
In this study, we integrated variables associated with both medical and public health perspectives in the aforementioned HL model of the European Health Literacy Survey Consortium and proposed a simple scoring algorithm. The scoring system dichotomizes older adults into high-risk (cutoff ≥ 5) and low-risk (cutoff < 5) populations to maximize the sensitivity and speci city of low HL prediction. Based on the proposed cutoff points, among the 92 older adults in the test data set, 63 (68.5%) with a cutoff ≥ 5 were recommended to undergo further HL intervention, although only 31 (62.0%) actually had low HL, resulting in a positive predictive value of 75.6%. Given the importance of early identi cation and strategy provision for community-dwelling older adults at high risk of low HL, the proposed scoring algorithm proposed can be considered useful in community practice.
This conceptual framework integrating medical and public health perspectives developed by the European Health Literacy Survey Consortium is suitable for exploring the most relevant determinants of HL levels in older adults. Eight predictors were identi ed to be signi cantly associated with HL levels: one socio-environmental determinant (i.e., dominant spoken dialect) and seven HL-related factors including health services (i.e., having a family doctor), health cost (i.e., self-paid pneumonia vaccination), health behaviors (i.e., searching online health information), health outcomes (i.e., assistance while visiting a doctor and activities of daily living), participation (i.e., attending health classes), and empowerment (i.e., self-management during illness). The results for seven identi ed predictors of HL-related factors were consistent with those of previous studies, for example, having a family doctor (7), costs for self-paid vaccination (32), searching online health information (9), functional status such as di culty in daily activities and assistance while visiting doctors (32, 33), participation in health classes (34), and self-e cacy in disease management (35). However, our study found that personal and situational factors did not affect the HL among older adults. Previous studies have documented that personal determinants of age, education level, and working status as well as situational and environmental determinants including marriage and residential area were signi cantly associated with HL levels (14,36). This difference might be because personal and situational determinants were proximal factors of HL, which are in uenced and displaced by a more distal and upstream factor (societal and environmental determinants) (37).
Our risk prediction tool provides primary public health workers with an easy-to-use scoring system that examines relevant variables. Users can rapidly predict low HL and thus identify community-dwelling older adults who may require further health assistance by evaluating their HL-related personal, situational, and environmental factors as well as the health behavior and outcomes. Hospitalization and mortality due to poor HL in older adults can be avoided through early identi cation and intervention. Therefore, this assessment tool should be promptly extended to broader communities.
Our study had some limitations. First, this was a cross-sectional study by convenience sampling from northern, central, and southern Taiwan. Therefore, potential selection bias might also exist. Second, this study relied on the 47-item HLS-EU-Q self-reported questionnaire for the criteria for HL. Further more objective HL assessments might be required to recognize the functional HL in order to avoid the potential for outcome misclassi cation bias. Third, the high prevalence rate of low HL (54.9%) among our sample may in uence the capacity of prediction (i.e., PPV) of this algorithm when applied in other populations. Therefore, when it applies to a population with a lower prevalence of low HL, the older adults with positive results of low HL may in fact have higher HL. Additionally, we excluded older adults who could not pass the Mini-Cog screening or follow instructions to complete the assessment. Our model may, therefore, not be generalizable to the entire population of older adults. Thus, this model is not recommended to be used in individuals with cognitive impairments or dementia who may have di culty understanding the instructions. Larger population studies with prospective longer term outcome measures are necessary to validate our study.

Conclusion
We proposed a simple clinical scoring algorithm with substantial sensitivity and satisfactory speci city to assess the risk of low HL among community-dwelling older adults.

Practice implications
This scoring algorithm not only helps clinicians to assess and identify the HL level in older adults but also assists researchers to establish intervention strategies for predictors of low HL. However, for further population-based application for early detection of older adults at a high risk of low HL, prospective trials should study the implementation and utility of this algorithm in the community.

Consent for publication
Not applicable.

Availability of data and materials
The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Competing interests
All of authors declare that have no competing interests Funding: This study was supported by research grants from the Taiwan National Health Research Institutes (MOHW107-TDU-M-212-133001-107-FR-04). The funder was not involved in study design, and will not be involved in the collection, analysis or interpretation of data.
WHH is expected to have made substantial contributions to the conception, design of the work, the acquisition, analysis, interpretation of data and have drafted the work or substantively revised it; YMC is expected to have made substantial contributions to the conception; MJC is expected to have made substantial contributions to the acquisition, analysis, interpretation of data; HWT is expected to have made substantial contributions to design of the work and analysis, interpretation of data; CTS is expected to have made substantial contributions to the conception, the acquisition of data and have drafted the work or substantively revised it; DSH is expected to have made substantial contributions to the conception, the acquisition of data and have drafted the work or substantively revised it; DCC is expected to have made substantial contributions to the conception and have drafted the work or substantively revised it; KNK is expected to have made substantial contributions to the conception and have drafted the work or substantively revised it; CYL is expected to have made substantial contributions to have drafted the work or substantively revised it. The author (s) read and approved the nal manuscript.  Figure 1 Age-speci c health literacy levels ROC curve and c-statistics of the tting test in the test data set. The AUC was 0.71 (95% CI: 0.61-0.81), indicating acceptable discrimination