Author(s): Bobby Tang
Purpose: Non-attendance for healthcare appointments can adversely affect patients’ clinical outcomes as well as being costly to healthcare systems. The aim of this study is to assess the role of mobile health (mHealth) interventions in tackling this problem.
Methods: A systematic review was conducted on several databases including the Cochrane Central Register of Controlled Trials (CENTRAL), MEDLINE, Embase and PubMed. Overall 12 articles were selected for the review. A random effects model was used to estimate an overall effect size.
Results: SMS reminders significantly decreased the rate of non-attendance compared to no reminders (Risk ratio [RR] 0.77; 95% Confidence Interval [CI] 0.71, 0.84) with moderate heterogeneity (I2=32%) between studies. A funnel plot indicated no evidence of reporting bias.
Conclusions: SMS reminders significantly improve healthcare attendance rates across a wide range of clinical and socioeconomic settings. Utilization of SMS reminders are a cheap and effective method of improving patient attendance rates
Low attendance rates for healthcare appointments pose significant problems, both in developed and developing countries. In 2003, the World Health Organization (WHO) stated that interventions aiming to improve adherence may have a far greater impact on the health of the population than any improvement in specific medical treatments and that patients should be supported in order to improve compliance [1]. One setting where the importance of attendance has been well established is in cancer screening. For example, epidemiological studies on cervical cancer survival have estimated that screening prevents 66-74% of cervical cancer deaths [2-4]. Despite this, cervical cancer screening in the UK only covered 73.5% of the eligible population in 2015, a figure that has fallen steadily since 2005 and it has been estimated that if everyone attended screening regularly in England, an additional 13% of cervical cancer deaths would be prevented each year [4,5].
Adverse effects of poor attendance have been shown for other non-communicable diseases (NCDs). For example, patients with diabetes mellitus who miss a significant proportion of their appointments (>30% of scheduled appointments) have significantly poorer glycemic control (HbA1c 0.70 - 0.79% higher) than those who attend regularly (after adjusting for demographic factors, clinical status, and health care utilization) [6]. As well as this, non-attendance has economic implications, not only with higher treatment costs associated with delayed diagnoses but also due to lost productivity. In 2012-2013, the cost of missed appointments in the National Health Service (NHS) was estimated at around £225 million [7].
One potential solution to improving attendance rates is the use of SMS reminders. Globally, there are over 5 billion wireless subscribers and commercial wireless signals cover over 85% of the world?s population, thus making mobile health interventions very promising due to their wide penetration and low cost [8]. A recent review on mHealth by Gurol-Urganci et al. in 2013 compiled evidence from seven studies to conclude that text messaging reminders increase attendance at healthcare appointments compared to no reminders (RR 1.14; 95% CI 1.03, 1.26). Text messaging reminders were also shown to have the same impact as phone call reminders while being a less expensive alternative (RR 0.99; 95% CI 0.95, 1.02) [9]. However, there was a substantial degree of study heterogeneity in the pooled effects model and the quality of evidence was deemed low to moderate. The authors recommended further high-quality trials in mHealth before conclusive policy decisions can be made.
This systematic review includes RCTs which compare the effect of SMS reminders to no reminders on patients? attendance to healthcare appointments. Our objective is to update the existing data in order to inform further research.
Types of studies
Only randomised controlled trials were included for this review.
Potential studies could be published or unpublished in any language with no restrictions placed on publication date.
We included all patients attending a healthcare appointment irrespective of their gender, age and ethnicity. No restrictions were placed on disease type or the appointment setting whether it was from primary care, secondary care or community services.
We included studies where SMS reminders were used in isolation. This meant that studies using SMS reminders as part of a larger multifactorial intervention (e.g. text messaging and a phone call reminder) were excluded. In addition, the initial intervention had to be sent to the patient directly and not to a relative or carer. We only included studies which compared the intervention against no reminders as this is often the standard of care in low and middleincome countries.
The primary outcome was the proportion in each group who did not attend their next healthcare appointment, either scheduled or unscheduled.
Search methods for identification of studies
A systematic search was conducted on the following databases: Cochrane Central Register of Controlled Trials (CENTRAL), MEDLINE, Embase and PubMed. In addition to the listed databases an ongoing trial register was screened (WHO Clinical Trial Search Portal) and a manual search of references from included studies was also conducted. The following search terms were used: SMS, MMS, text messaging, attendance, adherence, randomised or randomized.
Study selection
Study selection was conducted by the researcher who performed
the literature search and another independent researcher. Both
retrieved full text articles for potential inclusion after reviewing
titles and abstracts. Any disagreements on study inclusion was
resolved by discussion between the researchers.
The following data was extracted from included studies: general information (title, authors, source, publication details), setting (geographical and type of healthcare setting), study methods (eligibility criteria for participants, time of interventions), outcomes (proportion attending next appointment, cost analysis, adverse events, timing for measurements).
This was assessed by the primary researcher using the Cochrane
Risk of Bias Tool for Randomised Controlled Trials, which grades
each criterion as ?low?, ?high? or ?unclear? risk and the overall
quality was rated according to thresholds specified by the Agency
for Healthcare Research and Quality (AHRQ) [10]:
• Good quality: low risk of bias for all domains.
• Fair quality: One high risk criterion or two unclear criteria,
and the assessment that this was unlikely to have biased the
outcome.
• Poor quality: One high risk criterion or two unclear criteria,
and the assessment that this was likely to have biased the
outcome
• Poor quality: Two or more criteria listed as high or unclear
risk of bias
Although part of the Cochrane Risk of Bias tool, masking was removed as one of the assessment criteria due to the nature of notification interventions and our judgment that the outcome is unlikely to be influenced by the lack of masking in these studies.
Heterogeneity is the variation in study outcomes which cannot be explained by sampling error. This was examined by calculation of the I2 statistic. Cochrane Collaboration offers the following interpretation of I2: 0-30% may represent little or no heterogeneity, 30-60% represents moderate heterogeneity, 50-90% represents substantial heterogeneity and 75-100% represents considerable heterogeneity [10]. With low to moderate heterogeneity, a metaanalysis will be conducted using a fixed or random effects model as appropriate and with high heterogeneity, individual study characteristics will be assessed for potential reasons.
As our outcome is a binary variable, we used risk ratio for our effect measures. A random or fixed effects model will be used in our pooled analysis depending on the heterogeneity between studies. With moderate or high heterogeneity heterogeneity (I2>30%) we will use a random effects model to obtain a more conservative estimate of the confidence interval.
Due to the high degree of between-study heterogeneity identified by Gurol-Urganci et al. (2013) [9], we plan to conduct subgroup analyses for the following subgroups: studies from upper-middle and lower-middle income countries, studies from secondary care facilities, studies where the intervention consisted of a single reminder sent within 72 hours of appointment and studies of multiple reminders.
Assessment of reporting bias
We assessed reporting bias using funnel plots.
All statistical analysis and generation of tables and figures were conducted on Review Manager (RevMan) Version 5.3. Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014.
No ethics approval was required as all included articles are published in the public domain.
The search strategy identified 489 studies. After initial screening of the titles for basic relevance we retained 120 studies. This was reduced to 16 after review of abstracts to remove any studies that did not meet eligibility criteria and removal of duplicates. These were retained for full text review, although 3 of these were conference abstracts and full text articles had yet to be published and 1 was excluded as it did not meet eligibility. In total, 12 studies with 9,524 participants were selected for inclusion in this review (Figure 1) [11-23]. From the included studies we extracted information on the number and gender distribution of participants, the setting and type of appointment, disease type, timing of the intervention and whether the follow-up appointment was scheduled or unscheduled.
Figure 1: Flowchart of study selection
Our included studies were wide-ranging geographically with three from the UK two from Malaysia [18,19] and one each from the USA, China, Switzerland, Kenya, Nigeria, Australia and Saudi Arabia [12,17-19,22]. Five of the studies were from low-middle income countries. Seven of the studies were from hospital clinics or secondary care facilities, four were from primary care or general practice and one was from a health promotion center [11-22]. With the exception of Van Ryswyk et al. in which patients with gestational diabetes were sent text reminders to attend for a follow up appointment within 6 months, all other studies had scheduled appointments. (Appendix 1-2) [11].
In eight studies, a single reminder was sent within 72 hours of the appointment [12,14-15,17-19,21-22]. One study sent reminders at 6 weeks, 3 months and 6 months post-partum [11]. Another sent reminders at 7, 3 and 1 day(s) before the follow-up at an emergency department [13]. Another sent daily informational messages to post-circumcision patients for 7 days with reminders to re-attend on the day before and day of the appointment and another sent reminders to patients attending a psychiatric clinic at 5 and 3 days before the appointment [16, 20].
The mean age of study participants ranged from under 20 to over 50. Most of the interventions were appointment reminders only, however, the study by Odeny et al. included daily messages on post-operative care and the study by Narring et al. included a text back option to cancel or reschedule the upcoming appointment [15,16]. The outcome of this study was the proportion of unexplained missed appointments without prior notification, although, the number of participants who cancelled their appointment was small (n=6, 0.61%) and therefore, this is unlikely to affect the overall results.
Figure 2 illustrates the results of our pooled analysis. Events are defined as non-attendance to the scheduled appointment or in the case of the study by Van Ryswyk et al. non-attendance within 6 months post-partum as recommended by the physician [11]. Overall, SMS reminders decreased the rate of non-attendance compared to no reminders (RR 0.77, 95% CI 0.71 to 0.84). There was moderate heterogeneity (I2=32%) between the studies and therefore a random effects model was used in the meta-analysis.
Figure 2: Forest plot with all 12 included studies
In the prespecified subgroup analyses, studies from upper-middle and lower middle-income countries (appendix 3) showed a significant reduction in non-attendance rates for the SMS group compared to no intervention (RR 0.78, 95% CI 0.70 to 0.86) with a low level of between study heterogeneity (I2 = 13%). Similarly, studies from secondary care facilities (appendix 4), showed a significant reduction in non-attendance rates for the SMS group (RR 0.79, 95% CI 0.72 to 0.87) with no study heterogeneity (I2 = 0%). Studies which involved a single SMS reminder sent within 72 hours of the appointment (appendix 5) had a greater reduction in non-attendance rates for the SMS group compared to no SMS (RR 0.73, 95% CI 0.65 to 0.83) and studies which used multiple reminders (appendix 6) had a similar effect although the reduction in the SMS group was not as great (RR 0.83, 95% CI 0.74 to 0.93).
Using AHRQ thresholds, one study was rated as good quality, eight were deemed to be of fair quality and three were poor quality (Appendix 7) [11-22].
The shape of the funnel plot in Figure 3 does not indicate any evidence of reporting bias; however, the relatively small number of studies and the paucity of low-powered studies makes it harder to interpret. The high-powered studies are over-represented in the top half of the graph, likely due to the ease in recruiting large numbers of participants to these trials.
Figure 3: Funnel plot with all 12 included studies
This review includes 12 studies from a wide range of countries and various healthcare settings. It provides strong evidence that SMS interventions can be used to improve healthcare attendance rates where no reminders are the standard of care. Our pooled estimate had considerably less study heterogeneity (I2=32%) compared to previous reviews [9,23].
This may be due to our eligibility criteria which specified that studies had to be randomized controlled trials and that the SMS intervention was compared to no intervention. Furthermore, as we only included studies where the SMS reminder was sent to the patient directly, we excluded studies of pediatric populations from our analyses. Subgroup analyses show that text messaging interventions are also effective in middle income countries and in secondary care settings, although the use of multiple reminders has not been shown to be more effective in reducing non-attendance rates.
A strength of this study is the strict eligibility criteria which enabled the composition of a large homogenous data set. Overall, most studies were assessed to be of fair or good quality, although three had a high risk of bias. In the study by Leong et al. the outcome was defined as attending on the day of the appointment. In Malaysia the concept of ?walk-in? clinics is common and 48% of patients who attended on alternative days were classified as non-attenders although their overall healthcare outcomes were unlikely to be affected. In this case, classification bias might have caused the effect of the intervention to be underestimated [19].
Van Ryswyk et al. assessed the effect of SMS reminders on attendance to a post-partum diabetes clinic in women with gestational diabetes. This was limited by the high proportion of participants who also received postal reminders through a national reminder scheme (>83%). In addition, many of the patients? GPs were sent letters recommending further assessments (81%). Therefore, any difference in follow up between the intervention and control group is likely to have been minimized [11].
In the study by Fairhurst et al. the clustering effect of repeat appointments for the same patient was not accounted for. Overall, 415 appointments with 172 different patients were included in the study. This may have skewed the overall effect in our meta-analysis although this study had the smallest weighting of 2.7% and is unlikely to have had a large impact in the pooled analysis.
This systematic review provides evidence that SMS interventions improve attendance rates to healthcare appointments by 20-25%, both in high income countries and middle to low-income countries. The results may be used to inform further research, for example, in providing an estimate for sample size calculations and in highlighting a need for research on how to optimize text messaging reminders.
Funding: The author declares that no funds, grants, or other support were received during the preparation of this manuscript
Competing interests: The author has no relevant financial or non-financial interests to declare
Appendix| Included study | Country | Healthcare setting | Disease | Intervention timing | Appointment Type |
|---|---|---|---|---|---|
| Arora 2015 | USA | Hospital/ secondary care | Various | 7, 3 and 1 day beforehand | Scheduled appointment |
| Chen 2008 | China | Health promotion centre | Various | 72hrs prior to appointment | Scheduled appointment |
| Fairhurst 2008 | UK | Primary care | Various | Morning of appointment or afternoon before | Scheduled appointment |
| Kerrison 2015 | UK | Breast cancer screening | Various | 7, 3 and 1 day beforehand | Scheduled appointment |
| Leong 2006 | Malaysia | Primary care | Acute, chronic or preventative care | 24-48 hours beforehand | Scheduled appointment |
| Liew 2009 | Malaysia | Primary care | Chronic diseases | 24-48 hours beforehand | Scheduled appointment |
| Narring 2013 | Switzerland | Hospital/ secondary care | Various | 1 day beforehand | Scheduled appointment |
| Odeny 2012 | Kenya | Hospital/ secondary care | Post-circumcision | Daily post-op text messages for 7 days | Scheduled appointment (post op day 7) |
| Taylor 2012 | UK | Hospital/ secondary care | Physical therapy | 1 or 2 days beforehand | Scheduled appointment |
| Thomas 2017 | Nigeria | Hospital/ secondary care | Psychosis | 5 and 3 days beforehand | Scheduled appointment |
| Van Ryswyk 2015 | Australia | Hospital/ secondary care | Gestational diabetes | 6 weeks / 3 months / 6 months post-partum | Unscheduled - within 6 months |
| Youssef 2014 | Saudi Arabia | Hospital/ secondary care | Various | 48 hours beforehand | Scheduled appointment |
| Study | No. of patients (M/F) | Mean age (SMS/control) | SMS group (DNA/total) (%) | Control group (DNA/total) (%) | Intervention characteristics |
|---|---|---|---|---|---|
| Arora 2015 | 156/172 | 44.9/46.1 | 40/146 (27.4) | 69/182 (38.0) | Reminder only |
| Chen 2008 | 716/518 | 50.0/51.1 | 77/615 (12.5) | 121/619 (19.5) | Reminder only |
| Fairhurst 2008 | 156/259 | 33.1/33.1 | 22/226 (9.7) | 39/189 (20.6) | Reminder only |
| Kerrison 2015 | 0/2240 | N/A | 400/1122 (35.7) | 457/1118 (40.9) | Reminder only |
| Leong 2006 | 229/435 | 38.4/37.8 | 135/329 (41.0) | 174/335 (51.9) | Reminder only |
| Liew 2009 | 276/341 | 58.2/60.8 | 48/308 (15.6) | 71/309 (23.0) | Reminder only |
| Narring 2013 | 241/750 | 17.7/17.7 | 76/462 (16.5) | 106/529 (20.0) | Reminder + option to cancel/ reschedule |
| Odeny 2012 | 1188/0 | N/A | 205/592 (34.6) | 240/596 (40.3) | Reminder + post-op instructions |
| Taylor 2012 | 263/416 | 37.5/36.9 | 37/342 (10.8) | 55/337 (16.3) | Reminder only |
| Thomas 2017 | 88/104 | 33.5/33.9 | 45/95 (47.4) | 60/97 (61.9) | Reminder only |
| Van Ryswyk 2015 | 0/168 | 32.1/32.8 | 30/134 (22.4) | 31/134 (23.1) | Reminder only |
| Youssef 2014 | 213/289 | 52.0/53.0 | 66/251 (26.3) | 100/251 (39.8) | Reminder only |
| Random Sequence Generation | Allocation concealmen | Selective reporting | Other bias | Blinding of Outcome Assessment | Incomplete Outcome Data | Overall Quality | |
| Arora 2015 | Fair | ||||||
| Chen 2008 | Poor | ||||||
| Fairhurst 2008 | Fair | ||||||
| Kerrison 2015 | Fair | ||||||
| Leong 2008 | Poor | ||||||
| Liew 2009 | Fair | ||||||
| Narring 2013 | Fair | ||||||
| Odeny 2012 | Fair | ||||||
| Taylor 2012 | Good | ||||||
| Thomas 2017 | Fair | ||||||
| Van Ryswyk 2015 | Poor | ||||||
| Youssef 2014 | Fair |
| Key: | |
|---|---|
| Low | |
| Unclear | |
| High |
