S. We used the t-test to compare groups on continuous outcomes and the chi-squared test for binary outcomes. All statistical tests were performed using twosided tests at the 0.05 level of significance. The Bonferroni method was used to adjust the level of significance for testing of secondary outcomes. For group comparisons, the results are expressed as mean difference 
(MD) and risk ratio (RR) for binary outcomes, corresponding two-sided 95 confidence intervals (95 CIs) and associated p-values. Adjusted analyses using baseline covariates (age, gender, educational level, duration on ART, HIV staging,Ancillary analysesRegression. We performed regression analyses 12926553 to determine the impact of baseline covariates on the primary outcomes. Higher levels of education (OR = 5.32, 95 CI 2.51, 11.30; p,0.001) and being on a second line regimen (OR = 11.06, 95 CI 3.75, 32.65, p,0.001) were statistically significant predictors of adherence .95 . The Hosmer and Lemeshow goodness-of-fit test was as follows: chi-squared 56.7, degrees of freedom = 8, p,0.001. We also added an interaction term between covariates and the intervention variable. The interaction terms were not statistically significant for age group (p = 0.633) and gender (p = 0.268), butText Messages for Adherence in HIVFigure 1. CONSORT flow diagram for CAMPS trial. doi:10.1371/journal.pone.0046909.gstatistically significant for level of education (p,0.001) and regimen (p,0.001). Adverse events. One female in the intervention arm requested to withdraw from the study because she felt it had compromised her undisclosed status. No other undesirable effects were reported. Satisfaction. inhibitor Satisfaction with the text message was measured using four questions (Table 4): rating of the SMS; if it helped improve adherence; if they wanted it to continue and if they would recommend it to a friend. Moderate levels of satisfaction (65 reported that the messages were good, very good or excellent) were reported by the participants who received the text messages.DiscussionOur study did not find a significant effect of motivational SMS texts on improving adherence to ART over a 3 to 6 month period. This trial was unique in that it was the first to report the effect of an SMS intervention on ART adherence among treatment experienced patients in Africa. Two recent trials in Kenya demonstrated improved ART adherence among patients initiating ART, but used different intervention protocols [8,9]. Our study had some important inhibitor limitations. Firstly, our primary measures of adherence (by interviews) might have resulted in overestimates of the true adherence rate [23,24] and the adherence reported for the last week may not adequately reflect adherence behaviors over longer periods because patients maybecome more adherent in the few days preceding their appointment [25]. However, pharmacy refill data showed similar findings to self-reports. The trial design did not interfere with patient care by providing medication or lab tests so there are large amounts of missing data for CD4-positive-T-lymphocyte count and viral load. Drug stock-outs were also frequent in the last two months of the trial. This may explain why some participants missed their scheduled appointments. Importantly, participants in the control arm were not prohibited from using other reminder methods, so additional benefits may have been difficult to detect. Previous studies conducted in the same parent population suggest that up to 25 have systematic r.S. We used the t-test to compare groups on continuous outcomes and the chi-squared test for binary outcomes. All statistical tests were performed using twosided tests at the 0.05 level of significance. The Bonferroni method was used to adjust the level of significance for testing of secondary outcomes. For group comparisons, the results are expressed as mean difference (MD) and risk ratio (RR) for binary outcomes, corresponding two-sided 95 confidence intervals (95 CIs) and associated p-values. Adjusted analyses using baseline covariates (age, gender, educational level, duration on ART, HIV staging,Ancillary analysesRegression. We performed regression analyses 12926553 to determine the impact of baseline covariates on the primary outcomes. Higher levels of education (OR = 5.32, 95 CI 2.51, 11.30; p,0.001) and being on a second line regimen (OR = 11.06, 95 CI 3.75, 32.65, p,0.001) were statistically significant predictors of adherence .95 . The Hosmer and Lemeshow goodness-of-fit test was as follows: chi-squared 56.7, degrees of freedom = 8, p,0.001. We also added an interaction term between covariates and the intervention variable. The interaction terms were not statistically significant for age group (p = 0.633) and gender (p = 0.268), butText Messages for Adherence in HIVFigure 1. CONSORT flow diagram for CAMPS trial. doi:10.1371/journal.pone.0046909.gstatistically significant for level of education (p,0.001) and regimen (p,0.001). Adverse events. One female in the intervention arm requested to withdraw from the study because she felt it had compromised her undisclosed status. No other undesirable effects were reported. Satisfaction. Satisfaction with the text message was measured using four questions (Table 4): rating of the SMS; if it helped improve adherence; if they wanted it to continue and if they would recommend it to a friend. Moderate levels of satisfaction (65 reported that the messages were good, very good or excellent) were reported by the participants who received the text messages.DiscussionOur study did not find a significant effect of motivational SMS texts on improving adherence to ART over a 3 to 6 month period. This trial was unique in that it was the first to report the effect of an SMS intervention on ART adherence among treatment experienced patients in Africa. Two recent trials in Kenya demonstrated improved ART adherence among patients initiating ART, but used different intervention protocols [8,9]. Our study had some important limitations. Firstly, our primary measures of adherence (by interviews) might have resulted in overestimates of the true adherence rate [23,24] and the adherence reported for the last week 
may not adequately reflect adherence behaviors over longer periods because patients maybecome more adherent in the few days preceding their appointment [25]. However, pharmacy refill data showed similar findings to self-reports. The trial design did not interfere with patient care by providing medication or lab tests so there are large amounts of missing data for CD4-positive-T-lymphocyte count and viral load. Drug stock-outs were also frequent in the last two months of the trial. This may explain why some participants missed their scheduled appointments. Importantly, participants in the control arm were not prohibited from using other reminder methods, so additional benefits may have been difficult to detect. Previous studies conducted in the same parent population suggest that up to 25 have systematic r.