Table 3. Risk of Bias.

Study	Loss to follow-up	Statistical adjustment	Limitations
Bhat et al.(2008) [12]	17% (n = 121) did not do the REALM-not reported why, these might possibly be themost illiterate people	Age, gender, race, BMI, educational, marital and work status.	Data from two RCTS was combined but one was of 8, the other of 20 weeks duration, one was a behavior modification the other an exercise intervention; 6-months follow up data was only collected in the intervention group not in the control group.
Buchbinderet al. (2006)[13]	3 of 83 refused to participate, (not able toread well = 2, reason not given = 1). Of the 80 participants,1 did not attempt the TOFHLA and 15 did not do the TORCH-reasons not reported.	Not reported	Limited generalisability as sample was regular attendees at private clinic. Two patients unable to participate due to inability to read, while one patient did not complete TOFHLA.
Hirsh et al.(2010) [15], Hirshet al. (2011)[16]	118 recruited but 8 withdrew (due to illiteracy = 2, reasons not given = 6). Of the remaining 110, 2 did not complete patient assessment.	Multivariate analysis adjustedfor use of biologic agents,education, sex, age	About 70% of eligible patients agreed to participate, but 8/118 then withdrew, 2 due to illiteracy, and 1 due to poor command of English. 2/110 did not complete patient assessment.
Gordon et al. (2002) [14]	4 of 127 refused to take part (3 of thosesaid they were unable to read).	Not reported	Limited generalizability as sample was from tertiary care centre. Four subjects refused to participate (three of whom could not read)
Kim et al.(2009)[17]	7 of 110 subjects agreeing to participate(survey response rate 65%) in the studywere excluded because of vision problemsand incomplete questionnaire.	General linear model comparing literacy groups adjusting for age, education and monthly income	Cross-sectional nature makes it difficult to draw causal pathway between health literacy and rates of arthritis.Seven subjects were excluded because of their vision problems and incomplete questionnaire.
Rudd et al.(2009)[19]	134 consented and 127 of those were randomized.	Multivariate models were runwith and without adjustmentfor covariates that differedat baseline between the groups	There may have been a ceiling effect since participants had higher baseline education, literacy skills, satisfaction, adherence and attendance. This also limits generalizability. The average disease duration in this population was 17years thus they might have been very experienced with arthritis and arthritis care already.
Quinlan et al. (2010) [18]	27 of 157 (17%) did not participate dueto various reasons, it is possible that somerefused due to poor health literacy/inabilityto fill out the instruments	Stepwise regression with race, education, neighbourhood income, duration of disease,study location, age, healthknowledge, medications.	Use of self-modified version of TOFHLA, unclear if any validation
Swearingenet al. (2010)[20]	No attrition or refusals	Not reported	Socioeconomic status not described. Participants in this study were used to filling out questionnaires at each visit therefore their health literacy/skills completing questionnaires might be better than other population

BMI = body mass index.