Selection of Studies
Articles were retrieved from databases including PubMed (552), EMBASE (344), ScienceDirect (168), Cochrane (92) and grey literature from Google Scholar (78) and reference lists (3), as shown in the flow diagram for selection processes (Figure 1)
Figure 1: PRISMA flow diagram for inclusion of articles
From the total of 1237 articles retrieved, 152 articles excluded due to duplication; 1039 excluded with title review and 24 articles excluded due to ineligibility. Finally, 23 articles were fully read and 9 studies were excluded from the review with different reasons, as shown in Table 1. Some articles were excluded because of using video-based directly observed therapy [33–37] the intervention was different from phone text, audio, graphic and video messages. SMS intervention has an unrelated purpose [38] and study participants were not matching with this review [5, 39, 40].
Table 1: Characteristics of excluded studies
Study
|
Reasons for exclusion
|
Garfein, et’al, 2015
|
Different intervention (Video Based Directly Observed Therapy)
|
Holzschuh, et’al, 2017
|
Video Based Directly Observed Therapy
|
Buchman, et’al, 2017
|
Different intervention (Skype Observed Therapy)
|
Hoffman, et’al, 2010
|
Remote (video) mobile Direct Observation of Treatment
|
DeMaio, et’al, 2001
|
Focus on videophone technology
|
Lorent, et’al, 2014
|
SMS was used for issuing of test results not for treatment reminder
|
Bassett, et’al, 2016
|
Study subjects were HIV/TB co-infected patients
|
Chaiyachati, et’al, 2013
|
Intervention for Health care workers
|
Howard, et’al, 2016
|
Participants were HIV/TB co-infected
|
Characteristics of selected Studies
After exclusion of ineligible studies, fourteen studies were selected and reviewed for the qualitative synthesis of evidence. Among included studies, four in Africa, seven in Asia, one in South in one in North America, and one study enrolled patients from North America, Europe, Asia, and Africa. See details in Table 2 in the Appendix.
All studies implemented mobile phone messaging interventions on top of the standard TB treatment and care. The various types of phone messaging interventions were applied. Nine of fourteen studies used text-based phone messaging, two studies used only phone call and two studies applied a text and/or voice messages. One study implemented text and graphic messages for TB patients to adhere to their treatment. Among all studies, the messaging intervention was interactive (two-way) in seven studies, one-way in five and two studies didn’t report on the messaging model. The ultimate purpose of mobile messaging in 11 studies were to remind patients towards their anti-TB medication. In 3 studies the messaging intervention aimed to identify patients’ concerns, educate and motivate patients to engage on their own medication.
A study in Thailand [41] reported a significant effect of SMS messaging on TB treatment success rate. Whereas, other randomized control study in Argentina [42], in Pakistan [43, 44], in China [45] found no significant difference between the SMS or control groups for treatment success.
Two studies didn’t calculate significance test on the effect of SMS messaging on TSR of TB due to inappropriate or lack of comparator group [46, 47]. One study reported only about the level of patient adherence on TB treatment but not on treatment outcome [48]. The characteristics of each selected studies have explained in Table 2. Two randomized controlled studies in USA, Spain, Hong Kong and South Africa [49], and in Canada [50], reported that phone messaging did not significantly improve LTBI completion rates compared to standard care. Successful treatment of Latent Tuberculosis Infection (LTBI) also plays key role in eliminating TB [51]; however, difficult to ascertain treatment success in LTBI cases.
Quantitative synthesis of evidence
Only eight studies [41–45, 52–54] were found to be fitted for the pooled estimation of the effect of mobile phone messaging on successful TB treatment outcome. Six studies were excluded with the following reasons. One study reported treatment completion and unable to calculate the treatment success rate from the reported data [55]. Two studies focused on LTBI treatment completion [49, 50] couldn’t be pooled to determining treatment success. One study used a different tool to measure the outcome variable [48] and studies excluded due to lack of comparison groups [46, 47].
Risk of bias assessment for the included studies
Among the selected studies for meta-analysis, performance bias was the major challenge in the studies, because of the nature of the intervention, participants could not be blinded. Blinding of outcome assessors was also major gap that could lead to detection bias. All studies were jugged free of selection and reporting bias. Two studies have attrition bias (Figure 2). Based on Cohen’s Kappa level of agreement, the two reviewers (KDG and ZAM) have shown 83.3% agreement with k = 0.686, p<0.0001 for the included papers.
Figure 2: Risk of bias summary and graph: authors’ judgments for each included study
Pooled estimation on the effect of phone messaging on TB treatment success
Overall, 5680 TB patients, 2733 in intervention and 2947 in control groups were involved in the pooled analysis. The overall Anti-TB treatment success was 2390/2733 (87.4%) in intervention and 2490/2947 (84.5%) in control groups with heterogeneity level (I2 = 7%, p<0.0002). Fixed-effect model has shown that phone messaging group had higher treatment success rate compared to standard care (RR 1.04, 95% CI 1.02 to 1.06), see the Forest plot in Figure 3.
Figure 3: Forest Plot of the effect of mobile-phone messaging on Anti-TB treatment success.
Sensitivity analysis was carried out to see the effect size of each model by excluding every one of the studies. All studies were taken out of the analysis one by one and the output indicated that no single study separately influenced the pooled effect of phone messaging on TB treatment success. The Forest plot for sensitivity analysis for the two big weighted studies (Mohammed et al and Liu et al) presented on Additional file 3.
Sub-group analysis
Sub-group analysis was carried out to identify the effect of interventions by the level of national TB burden (high Vs low) and model of phone messaging (one-way Vs two-way) applied by individual studies. The finding indicated that phone messaging has a modest effect on TSR in high TB burden (RR = 1.04; 95% CI 1.00–1.07) and in low TB burden countries (RR = 1.06; 95% CI 1.01–1.11) with moderate (I2 = 53%) and no (I2 = 0%) heterogeneity of studies in both groups respectively see Additional file 4. Similarly, a sub-group analysis by model of phone messaging (one-way Vs two-way) shown that interactive phone messaging has modest effect on TSR of TB (Additional file 5). Further sub-group analyses (purpose of the messaging, frequency of messaging, type of messaging and national income) were not conducted due to unable to get sufficient studies in the pre-determined sub-groups.
Publication bias
A subjective visual inspection of funnel plot of the included studies pointed out symmetrical observation that shows less likely occurrence of publication bias. We used Egger’s test to objectively measured publication bias. The finding revealed that there was no evidence of publication bias (P = 0.753), (Figure 6).
Figure 6: Funnel Plot of studies comparing phone messaging and standard of care on TSR of TB
Rating quality of evidence
Based on GRADE quality of evidence assessment approach, the overall quality of evidence of this review was rated low, mainly due to limitations of performance, detection and selection biases for more details see Table 3.
Table 3: GRADE rating of the quality of evidence
Outcomes
|
№ of participants (studies) Follow up
|
Certainty of the evidence (GRADE)
|
Relative effect (95% CI)
|
Anticipated absolute effects* (95% CI)
|
TB treatment success assessed with: 6-9 months
|
5680 (8 Studies)
|
⊕⊕⊝⊝ LOW 1 2
|
RR 1.04 (1.02 to 1.06)
|
Assumed risk
|
Usual care group
|
Risk difference with phone Messaging
|
901 per 1,000
|
36 more per 1,000 (18 more to 54 more)
|
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; RR: Risk ratio
1High performance bias, detection and selection biases
2Some studies used self-assessment tools that subjectively measure treatment completeness