1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Limitations
  7. Conclusions
  8. Acknowledgments
  9. References

ACADEMIC EMERGENCY MEDICINE 2012; 19:949–958 © 2012 by the Society for Academic Emergency Medicine


Objectives:  Nonadherence to prescribed medications impairs therapeutic benefits. The authors measured the ability of an automated text messaging (short message service [SMS]) system to improve adherence to postdischarge antibiotic prescriptions.

Methods:  This was a randomized controlled trial in an urban emergency department (ED) with an annual census of 65,000. A convenience sample of adult patients being discharged with a prescription for oral antibiotics was enrolled. Participants received either a daily SMS query about prescription pickup, and then dosage taken, with educational feedback based on their responses (intervention), or the usual printed discharge instructions (control). A standardized phone follow-up interview was used on the day after the intended completion date to determine antibiotic adherence: 1) the participant filled prescription within 24 hours of discharge and 2) no antibiotic pills were left on the day after intended completion of prescription.

Results:  Of the 200 patients who agreed to participate, follow-up was completed in 144 (72%). From the 144, 26% (95% confidence interval [CI] = 19% to 34%) failed to fill their discharge prescriptions during the first 24 hours, and 37% (95% CI = 29% to 45%) had pills left over, resulting in 49% (95% CI = 40% to 57%) nonadherent patients. There were no differences in adherence between intervention participants and controls (57% vs. 45%; p = 0.1). African American race, greater than twice-daily dosing, and self-identifying as expecting to have difficulty filling or taking antibiotics at baseline were associated with nonadherence.

Conclusions:  Almost one-half (49%) of our patients do not adhere to antibiotic prescriptions after ED discharge. Future work should improve the design and deployment of SMS interventions to optimize their effect on improving adherence to medication after ED discharge.

Transitions of care from the emergency department (ED) to home provide an opportunity to influence patient health-related behavior.1,2 For example, patients often fail to take prescribed medicines, which are a source of poor postdischarge outcomes3,4 and preventable treatment failures.5 Which ED patients are at high risk for nonadherence is unknown;6 also, effective tools to measure patient adherence after discharge are expensive or impractical for short-course treatments7 and interventions to improve adherence have mixed effectiveness.8

Electronic communication technologies such as mobile phone text messaging (short message service [SMS]) are a potentially effective way to communicate with patients after ED discharge.9 An estimated 83% of U.S. adults operate mobile phones, and three-quarters of mobile phone owners regularly use SMS.10 SMS programs are well suited to assist self-monitoring and self-disclosure of nonadherence with minimal inconvenience to the patient.11 Additionally, SMS programs can provide real-time feedback to potentially alter medication-taking behavior.12–14 Finally, SMS programs can utilize health information technology to provide automated dialog, allowing for low-cost, large-scale implementation.

For our primary outcome, we sought to assess the utility of using an automated SMS program to improve oral antibiotic adherence in patients discharged from the ED. We chose to study dosing of antibiotics because it is a common therapy.15 We hypothesized that an SMS query about medication pickup and daily queries about dosage taken for the length of the prescription would improve adherence by increasing the proportion of patients who had filled their prescription within 24 hours of discharge and had no pills left on the day following intended completion of prescription, when compared to no exposure (control group). Secondary outcomes included patient and medication characteristics associated with nonadherence, to aid ED clinicians in optimal deployment of interventions.


  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Limitations
  7. Conclusions
  8. Acknowledgments
  9. References

Study Design

The Improving Prescription Adherence through Computerized Text-messaging (IMPACT) study was a randomized controlled clinical trial among patients discharged from the ED with prescriptions for oral antibiotics. The institutional review board at the University of Pittsburgh approved the protocol. All enrolled participants provided written informed consent, and we registered our trial before recruitment (NCT01388465). Three participants who completed the final assessments were randomly chosen to receive $100 gift cards.

Study Setting and Population

We recruited subjects in a single, urban ED in western Pennsylvania with an annual census of 65,000 visits per year. Recruitment occurred on approximately 60% evening shifts (4 p.m. to 12 a.m.) and 40% day shifts (8 a.m. to 4 p.m.) when the research associate (RA) was available in the ED. We recruited adults from June to September 2011.

The RA identified patients who were being discharged from the ED through the triage report viewed from the ED tracking software. The RA then asked the clinician (physicians, nurses, or physician extenders) caring for a particular patient if he or she was being discharged on an antibiotic and if that patient would be interested in hearing about the study. After ED providers recognized a potentially eligible subject, the RA confirmed eligibility and obtained written informed consent. Those eligible were at least 18 years old, able to speak English, owned a personal cellular phone with text messaging, and were prescribed oral antibiotics at ED discharge. We excluded those who did not administer their own medications. Additionally, we excluded those who were prescribed once daily dosing because those patients have a low baseline rate of nonadherence and are unlikely to benefit from an intervention.16 The RA recorded age, chief complaint category, and acuity in those not eligible or interested. All participants were told that “In this research study, we will compare the ability of patients to comply with antibiotic prescriptions and the effect of text-message support on antibiotic use.” As well, they were told “We will randomly select you (that is, selected by chance) to either answer daily text-message questions about your antibiotic use OR not.”

Study Protocol

Participants completed a self-administered instrument that collected age, sex, race, ethnicity, highest education level, marital status, current employment, and medical insurance. Participants also provided information on frequency of visits to their primary care providers, routine daily medications, and number of antibiotic prescriptions in the past 12 months. We assessed intrinsic motivation by asking “How important do you feel that antibiotics are in making you better?” asking to choose from: not at all, somewhat, and very important. We assessed barriers by asking “Do you think you will have difficulty: 1) filling your antibiotic prescription, or 2) taking the antibiotic as prescribed?” We assessed the baseline use of cell phone text messaging by asking “What is the average number of daily text-messages that you send per day?” We collected patient contact information, including cell phone number and e-mail account for follow-up. Additionally, we asked the emergency physician (EP) prescribing the antibiotic his or her opinion about perceived barriers to the patient filling or taking the antibiotic, stated as: “Do you think your patient will have difficulty: 1) filling the antibiotic prescription, or 2) taking the antibiotic as prescribed?”

After baseline assessments, an RA opened a consecutively numbered, sealed opaque envelope containing assignment information prepared using a computer-generated set of random numbers in blocks of 20. Participants in both groups were told they would be contacted by phone on the day after the intended completion of their antibiotics. At discharge, the RA obtained the antibiotic name, dose, frequency, and duration from the patient medical record. The RA then entered the information into a database through a Web-based case report form. Participants in the control group received no study-related SMS during their prescribed duration of antibiotics.

At exactly 1 hour after entry, the participants assigned to the intervention group received the following text message: “Welcome to the IMPACT antibiotic study. Text back ‘yes’ when you have picked up your prescription for [Antibiotic],” where [Antibiotic] was populated with the drug name. If participants did not reply within 6 hours of first text, a second text was sent out, stating, “We’re concerned you have not picked up your prescription for [Antibiotic]. Text us “yes” when you do.” If the participant did not respond to four total texts (sent every 6 hours), we sent no further SMS because we assumed they either did not wish to participate further or did not have an active SMS account. When the participant responded, he or she received: “Thanks for letting us know you picked up your [Antibiotic]. Keep in mind that they work best if taken exactly as prescribed.” Exactly 24 hours following their response, participants received the text: “IMPACT antibiotic study: How many doses of [Antibiotic] did you take between [0:00 PM] yesterday and [0:00 PM] today?” where [0:00] was populated with the time stamp of the text message when the participant responded “yes” that they had picked up their prescription. If the participant picked up his or her prescription after 9 p.m. and before 9 a.m., the computer system defaulted to 9 a.m. to avoid sending messages to participants in the middle of the night. If a participant replied with a number outside a predetermined range (0 to 10), or other response outside the range of expected responses, he or she received the following: “Please respond with a number between 0 and 10.” If the participant responded with a correct dose amount, he or she received: “Thank you for your response. You are taking your [Antibiotic] as prescribed. Please continue to do so.” If the participant took too many doses, he or she received: “We are concerned you have taken too many doses of [Antibiotic], which may be dangerous to your health.” immediately followed by “Remember to take only [X] doses per 24 hour period and to separate doses as recommended” where [X] is populated with information from the computer server. If the participant took too few doses, he or she received: “We are concerned you have taken too few doses of [Antibiotic] to be effective in fighting your infection” followed by “Remember to take only [X] doses per 24 hour period and to separate doses as recommended.”

The design of our SMS intervention was based on prior literature supporting self-monitoring as an effective tool for behavior change,17 previous research showing that self-monitoring improves medication adherence,18 and modeled on prior effective SMS interventions for medication adherence.12–14 All intervention dialog was automated and conducted through a text messaging platform we had previously built and tested.19 Dialog was developed by content experts in emergency medicine and computer programming. A computer server within hospital firewalls logged time and content of all SMS sent and received. Investigators periodically reviewed the log to assess for performance and aberrant responses.


Primary outcomes were based on patients’ self-report questionnaires, which have been shown to effectively measure adherence,20,21 and show moderate to high correlation with medication monitoring devices.22 Each participant received a phone call on the day following the intended completion of their antibiotic. We made up to three calls over 3 days before we designated the participant lost to follow-up. The follow-up assessments were interview-based questionnaires with the RA unaware of treatment allocation. We included the following questions, adapted from a prior study:23“How many days did it take you to pick up your prescription?” and “How many pills do you have in your bottle?” (If the participant was reached after the day following intended completion, we asked “How many pills did you have [X] days ago, where [X] was calculated from the day following intended completion). We also collected self-reports of difficulty filling antibiotic prescription, difficulty taking antibiotic as prescribed, and any reported side effects. Participants in the intervention group were asked additional questions about their opinions of the SMS program, including: “How useful was our SMS system to remind you to pick up your antibiotics?,”“How useful was our SMS system to remind you to take your antibiotics?,” and “How likely would you be to use a SMS service next time you are prescribed an antibiotic?” where they answered “not at all,”“somewhat,” or “very.” To determine follow-up patterns, we asked participants: “Have you seen or spoken with your primary care doctor since your emergency visit?” To determine reported difficulty with taking antibiotics as prescribed, we asked a single question taken from the four-item Morisky scale:24“Did you have any problem remembering to take your antibiotic?” We defined adherence from telephone follow-up self-report as a patient who both 1) had filled his or her prescription within 24 hours of discharge and 2) had no antibiotic pills left on the day following intended completion of prescription.

Data Analysis

A single RA entered data into a database and they were verified by a second RA. We used STATA 10.0 (StataCorp, Inc., College Station, TX) to analyze data. We described study participants by baseline demographic and health utilization variables, prestudy motivation, self-efficacy, and SMS usage. We summarized these variables by calculating frequencies with percentages for categorical data and means and standard deviations (SDs) for count data. We used plots and examination of skewness and kurtosis to identify evidence of normality for continuous variables. The effectiveness of randomization assessment used Pearson’s chi-square test or Fisher’s exact test when the number of values in any cell was ≤10 (for categorical variables) and Student’s t-test (for continuous variables) with the comparison alpha error set at 0.05. We examined the degree of agreement between patient baseline report of expecting difficulty filling or taking his or her antibiotic as prescribed and EP report using Cohen’s kappa.

We assessed utility of SMS to measure daily medication use after ED discharge by examining the proportion of intervention participants who responded to daily SMS queries, from pickup through the entire length of the prescription. We present the proportion of participants completing SMS queries with 95% confidence intervals (CIs). For our primary outcome of interest, we examined whether treatment group was related to adherence by comparing proportions between treatment groups using chi-square. We thought that a minimum of 25% increase in the proportion adherent, from 50% to 75%, would be needed to be clinically relevant.25 To reach an 80% power at an alpha of 0.05 (using a two-sided test), we estimated that we would need 66 patients per group (132 total). We expected that up to 30% of patients would be lost to follow-up, and therefore we enrolled 200 patients.

We compared the proportion of intervention participants (prescribed at least 7 days of treatment) who had taken the correct dose on day 1 compared with day 7 using chi-square analysis. We also explored patient and medication factors associated with non-adherence using univariate logistical regression. With an estimated 50% of participants with nonadherence, 100 “nonadherence events” would allow us to examine up to 10 covariates.26 Those factors with a univariable association p < 0.20 contributed to a forward stepwise multivariable logistic regression model, presented as adjusted odds ratios with 95% CIs. We developed a score where the presence of each factor was allocated a point given the similar regression coefficients in the multivariable model. Discrimination of the final model assessment used the area under the receiver operator curves (AUC) and calibration testing by the Hosmer-Lemeshow statistic.


  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Limitations
  7. Conclusions
  8. Acknowledgments
  9. References


We approached 275 patients over 64 unique days. We screened 230 subjects and enrolled 200 (72%, Figure 1). Screened subjects who declined to enroll were not different in age, chief complaint category, or acuity when compared to those enrolled. Baseline characteristics of the sample are shown in Table 1 for the 200 patients enrolled in the study. Around 10% thought they would have either difficulty filling or taking their antibiotics as prescribed. Among those who reported expecting to have difficulty filling their prescriptions, common reasons included cost (32%) and lack of transportation (32%). Among those who reported expecting to have difficulty taking antibiotics as prescribed, common reasons included trouble remembering (50%) and trouble swallowing pills (50%). The EP prescribing the antibiotic thought that 5% of participants would have difficulty filling their prescription and 9% would have difficulty taking the antibiotic as prescribed, with no statistically significant differences between treatment groups. There was only slight agreement between patient self-report and EP report of expecting difficulty filling prescription (κ = 0.13; p = 0.02) and difficulty taking antibiotic (κ = 0.02; p = 0.4).


Figure 1.  Recruitment and randomization of study participants.

Download figure to PowerPoint

Table 1.    Characteristics of Participants at Baseline
VariableTotal (N = 200)Intervention (n = 100)Control (n = 100)p-value
  1. All values are n (%) unless specified otherwise.

  2. No p-values were < 0.05.

  3. GED = general equivalency diploma; PCP = primary care physician.

Age (years), mean (SD)33 (12)34 (13)31 (11)0.1
Male sex, n (%)63 (31)29 (29)34 (34)0.5
African American race99 (50)49 (49)50 (50)0.5
Hispanic ethnicity7 (4)5 (5)2 (2)0.3
Highest level of education
 <High school42 (21)23 (23)19 (19)0.9
 High school graduate or GED69 (35)33 (33)36 (36)
 Some college63 (31)33 (33)30 (30)
 College graduate21 (11)9 (9)12 (12)
 Postgraduate5 (3)2 (2)3 (3)
Single marital status138 (69)72 (72)66 (66)0.6
 Unemployed88 (44)42 (42)46 (46)0.8
 Part-time employment39 (20)22 (22)17 (17)
 Full-time employment58 (29)29 (29)29 (29)
 Student/nonemployed15 (8)7 (7)8 (8)
Medical insurance
 None37 (19)19 (19)18 (18)0.7
 Public/Medicaid91 (46)45 (45)46 (46)
 Private45 (23)22 (22)23 (23)
 Other/don’t know27 (14)14 (14)13 (13)
Last time seen PCP
 No PCP/never18 (9)7 (7)11 (11)0.1
 0–3 months ago76 (38)34 (34)42 (42)
 3–6 months ago47 (24)30 (30)17 (17)
 >6 months59 (30)29 (29)30 (30)
Routine daily medications
 0112 (56)59 (59)53 (53)0.4
 133 (17)16 (16)17 (17)
 2 to 327 (14)15 (15)12 (12)
 >328 (14)10 (10)18 (18)
Antibiotic prescriptions in past 12 months
 Never60 (30)35 (35)25 (25)0.3
 One72 (36)31 (31)41 (41)
 2 to 349 (25)26 (26)23 (23)
 >318 (9)8 (8)19 (10)
How important are antibiotics
 Somewhat important31 (16)15 (15)15 (15)0.9
 Very important169 (85)85 (85)84 (84)
Difficulty filling antibiotic in next 24 hours22 (11)10 (10)8 (8)0.8
 Cost7 (5)5 (7)2 (3)0.3
 Lack of transportation7 (4)3 (3)4 (4)0.9
 Too busy000
Difficulty taking antibiotic in next 24 hours18 (9)10 (10)8 (8)0.8
 Dislike taking medication1 (0.5)01 
 Too busy4 (2)2 (2)2 (2) 
 Trouble swallowing9 (5)6 (6)3 (3)0.5
 Trouble remembering9 (5)6 (6)3 (3) 
Mean (SD) number of daily text messages sent
 011 (6)3 (3)8 (8)0.1
 1 to 529 (15)15 (15)14 (14)
 6 to 1531 (16)8 (8)23 (23)
 16–5067 (34)39 (39)28 (28)
>5062 (31)35 (35)27 (27)
Satisfaction with ED care
 Somewhat satisfied20 (10)10 (10)10 (10)0.9
 Very satisfied177 (90)88 (90)89 (90)

Antibiotic and infection types across treatment groups are shown in Table 2. A total of 11 different antibiotic prescriptions were administered to participants, the most common being trimethoprim/sulfamethoxazole (24%), cephalexin (16%), and penicillin (15%). Antibiotic dosing was predominantly twice daily (53%), with more twice-daily dosing assigned to the intervention group (61%) than the control group (42%; p = 0.03). There was a significantly lower proportion of dental infections in the intervention group than control group (16% vs. 24%; p = 0.04) and a higher proportion of urinary infections in the intervention group (27% vs. 11%; p = 0.04). Most treatment was prescribed for 7 (38%) or 10 days (47%), with no statistically significant differences between treatment assignments.

Table 2.    Comparison of Participants Antibiotic and Infection Category by Treatment Condition
VariableTotal (N = 200)Intervention (n = 100)Control (n = 100)p-value
  1. Differences between treatment groups were calculated using Pearson chi-square test or Fisher’s exact test for categorical variables.

  2. DS = double strength; SMX = sulfamethoxazole; TMP = trimethoprim.

Antibiotic name/type
 TMP/SMX DS48 (24)27 (27)21 (21)0.3
 Cephalexin32 (16)13 (13)19 (19)
 Penicillin31 (15)14 (14)17 (17)
 Amoxicillin/Augmentin23 (12)14 (14)9 (9)
 Flagyl18 (9)8 (8)10 (10)
 Clindamycin16 (8)5 (5)11 (11)
 Doxycycline10 (5)6 (6)4 (4)
 Ciprofloxacin9 (5)7 (7)2 (2)
 Other13 (6)8 (8)5 (5)
 Twice daily105 (53)61 (61)42 (42)0.03
 Three times daily25 (13)9 (9)16 (16)
 Four times daily70 (34)29 (29)41 (41)
Length of treatment, days
 310 (5)8 (8)2 (2)0.06
 515 (8)6 (6)9 (9)
 775 (38)41 (41)34 (34)
 1095 (47)44 (44)51 (51)
 145 (2)1 (1)4 (4)
Category of infection
 Dental/oral40 (20)16 (16)24 (24)0.04
 Urinary/kidney38 (19)27 (27)11 (11)
 Genital/pelvic17 (9)9 (9)8 (8)
 Skin and soft tissue78 (39)35 (35)43 (43)
 Other27 (13)13 (13)14 (14)

SMS Interaction

In the intervention group, 67% (95% CI = 57% to 76%) responded to the antibiotic pickup question. Among those who completed the pickup question, the proportion who did not respond to SMS increased from 19% (95% CI = 10% to 29%) on day 1 to 34% (95% CI = 22% to 46%) on day 7 (p = 0.05), as shown in Figure 2. There was a higher proportion of those with less than high school education in those who did not interact with the SMS program (33%) than those who did (18%; p = 0.02). The proportion taking the correct dose increased from 37% (95% CI = 25% to 49%) on day 1 to 61% (95% CI = 49% to 73%) on day 7 (p = 0.01), with the greatest change exhibited between Day 1 and Day 2. This is reflected in the reduction of those who took both too many (21% to 0%) and too few pills (23% to 5%) from day 1 to day 7.


Figure 2.  Daily text message responses of antibiotic dose over first 7 days of treatment.

Download figure to PowerPoint

Phone Follow-up

Among the 200 participants, 144 (72%, 95% CI = 67% to 77%) completed the phone follow-up with no differences between treatment groups. Most phone follow-ups were completed on the day after intended prescription completion (87%), with only 5% at completion day +3. There were no differences in baseline patient demographics; chief complaint category; or antibiotic type, dosing, or length between those who were lost and those who completed follow-up. Among those with whom we were able to follow up, there were no differences in the percentage of those who had seen or spoken with their primary care doctor since ED discharge (39% in intervention vs. 34% in control; p = 0.6).


As shown in Table 3, the proportion of participants adherent in the intervention group was 57% (95% CI = 44% to 67%) versus 45% (95% CI = 37% to 59%) in the control group (p = 0.16). The proportion of intervention participants who had filled their prescriptions in the first 24 hours following discharge was 78% (95% CI = 66% to 87%) versus 69% (95% CI = 57% to 80%) in control participants (p = 0.26). The proportion of intervention participants who had no pills left was 68% (95% CI = 55% to 78%), versus 59% (95% CI = 47% to 71%) in control participants (p = 0.30). Changing the definition of adherence to >80% of pills taken at follow-up in sensitivity analysis did not reveal any difference in distribution of results. The proportion of participants with any reported problem remembering to take their antibiotics in the intervention group was 19% (95% CI = 17% to 38%) versus 26% (95% CI = 11% to 30%) in the control group (p = 0.32). Using this question alone at follow-up would have a sensitivity of 44% (95% CI = 32% to 57%) and a specificity of 99% (95% CI = 93% to 100%) to identify participants with nonadherence.

Table 3.    Comparison of Participant Outcomes at Follow-up by Treatment Condition
VariableTotal (N = 144) Intervention (n = 72)Control (n = 72)p-value
  1. All values are n (%). Differences between treatment groups were calculated using Pearson chi-square test or Fisher’s exact test for categorical variables.

Days to fill
 0106 (74)56 (78)50 (69)0.2
 128 (19)12 (17)16 (22)
 ≥210 (7)4 (5)6 (8)
Pills left in bottle
 090 (63)48 (68)42 (59)0.5
 1 to 525 (18)12 (17)13 (18)
 >527 (19)11 (15)16 (23)
 Unknown2 (2)11
Adherent to antibiotic prescription73 (51)41 (57)32 (45)0.16
Report of difficulty filling antibiotic11560.4
 Cost4 (3)3 (4)1 (1) 
 Too busy5 (4)1 (1)4 (5) 
Report of difficulty taking antibiotic33 (23)14 (20)19 (26)0.4
 Dislike taking meds000 
 Too busy4 (3)3 (4)1 (1) 
 Trouble swallowing101 
 Trouble remembering14 (10)5 (7)9 (12) 
 Other17 (12)8 (11)9 (12) 
Side effect of antibiotic43 (30)22 (31)21 (29)0.9

Factors Associated With Nonadherence

As shown in Table 4, there was a higher proportion of African American participants, those who thought they either would have trouble filling their antibiotic or taking their antibiotic as prescribed, and those who were prescribed greater than twice-daily dosing who were nonadherent than adherent. In a multivariable model, African American race had an adjusted OR of 1.8 (95% CI = 0.9 to 3.5), trouble filling or taking antibiotic had an adjusted OR of 3.3 (95% CI = 1.2 to 9.2), and greater than twice-daily dosing had an OR of 1.8 (95% CI = 0.9 to 3.5). a Hosmer-Lemeshow chi-square of 1.18 (p = 0.88) showed good calibration of the model. The presence of any one of the above attributes identified at baseline would have resulted in a sensitivity of 84% (95% CI = 74% to 92%) and a specificity of 32% (95% CI = 21% to 43%) to identify participants with nonadherence, whereas the presence if at least two factors resulted in a sensitivity of 44% (95% CI = 32% to 56%) and a specificity of 81% (95% CI = 69% to 89%). The model AUC was 0.7.

Table 4.    Comparison of Baseline Variables by Adherence Outcome
VariableAdherent (n = 73)Nonadherent (n = 70)p-value
  1. All values are n (%) unless specified otherwise. Differences between treatment groups were calculated using Pearson chi-square test or Fisher’s exact test for categorical variables. Difficulty filling and taking antibiotics were assessed at baseline with the questions: “Do you think you will have difficulty filling your antibiotic prescription” and “Do you think you will have difficulty taking the antibiotic as prescribed?”

Age (years), mean (SD)33 (12)33 (12)0.9
Male sex27 (30)17 (33)0.9
African American race32 (44)41 (56)0.01
Hispanic ethnicity1 (1)4 (6)0.2
Highest level of education, <high school16 (22)12 (17)0.6
Single marital status51 (70)51 (73)0.1
Unemployed33 (45)29 (41)0.2
Medical insurance, none17 (23)8 (11)0.06
Difficulty filling antibiotics in next 24 hours4 (6)11 (16)0.05
Difficulty taking antibiotics as prescribed2 (3)9 (13)0.03
Either6 (8)18 (25)0.005
 Twice daily45 (63)32 (46)0.04
 Three times daily4 (6)12 (17)
 Four times daily23 (31)26 (37)
Length of treatment in days, median (IQR)7 (7–10)8.5 (7–10)0.7

Acceptance of SMS Program

Among the 71 participants in the intervention group who completed follow-up, 91% (95% CI = 83% to 97%) thought that the SMS program was at least somewhat useful to remind them to pick up their antibiotics, with 52% (95% CI = 40% to 64%) who thought it was very useful. Also, 97% (95% CI = 90% to 100%) thought the SMS program was at least somewhat useful to remind them to take their antibiotics, with 61% (95% CI = 48% to 72%) who thought it was very useful. Finally, 94% (95% CI = 86% to 98%) thought they would be at least somewhat likely to use an SMS service the next time they are prescribed an antibiotic, with 58% (95% CI = 45% to 69%) who thought they would be very likely. There were no differences in adherence rates across the different perceptions of usefulness.


  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Limitations
  7. Conclusions
  8. Acknowledgments
  9. References

In this preliminary study, which to our knowledge is the first published report testing an SMS-based intervention for antibiotic adherence after ED discharge, we found no statistically significant effect of our intervention. We did, however, demonstrate that SMS may be a useful way to track medication use after ED discharge. Adherence to prescriptions administered at ED discharge is a key component of translational care, but nonadherence largely goes unnoticed and unmeasured in routine ED clinical practice and clinical trials. In clinical practice, for example, failure of an infection to improve could be misattributed to failure of the antibiotic, as opposed to poor circulating concentrations due to nonadherence. In a clinical trial, failure of an antibiotic to improve length of symptoms may result in Type II errors, where efficacy may be masked by poor drug concentrations due to nonadherence.

Prior studies have shown that between 7% and 35% of ED patients fail to fill discharge prescriptions, and 6% to 31% report being nonadherent with daily dosing of prescriptions.3,4 These large variations are likely due to the differences in the definition and measurement of adherence across prior studies.27 Our findings fall somewhere near the high end of these ranges, suggesting that our definition of adherence is conservative and/or we have an especially high-risk group of patients. When we changed the definition of adherence to >80% of doses, there was no significant change in the proportion of patients labeled nonadherent, suggesting that there were more than sporadic omissions as the cause of nonadherence.

Nonadherence can take several forms, including delayed or noninitiation, overdosing, omission of doses, and/or early discontinuation. In our cohort, delayed or noninitiation occurred in a quarter of patients, overdosing and omission of doses each occurred in up to 20% of patients in the first day but rapidly decreased, and early discontinuation did not systematically occur. Taken together, these suggest that delayed pickup of prescriptions and misdosing in the first day are critical in nonadherence and that once a patient has picked up the prescription and gotten into a routine, he or she is likely to continue. Therefore, future studies may wish to test initiatives that incorporate ED medication dispensing28 or more intensive initial education or counseling sessions25,29 prior to ED discharge.

Our preliminary findings show promise that an automated two-way SMS dialog could result in a small increase (12%) in the proportion of participants adherent to short courses of antibiotics when compared to a control. In a recent systematic review of 24 articles examining use of text messaging for health-related behavior, seven addressed medication adherence specifically.30 Among those seven, three used simple one-way reminders of medication dose,31–33 two others utilized more sophisticated one-way communication,34,35 and two used a two-way system that asked patients to confirm medication use.36,37 Among the four that were randomized controlled trials, only one showed significant improvement with intervention.34 This study by Strandbygaard et al.34 randomized 26 adults with asthma to receive daily SMS reminders to take their inhaled corticosteroid for 8 weeks, or no SMS, and found an absolute difference in mean adherence rate of 17.8% between the two groups after 12 weeks. It remains unknown what components of SMS dialog, including use of personalized messages, frequency of interaction, and type of feedback, are needed to optimize effect size; however, this will be limited by what can be conveyed in 160 characters and what will be tolerated by the patient.

In exploratory analysis, several factors were identified that could be easily elicited from a history and treatment plan to identify patients at risk for nonadherence. We found that African American race, those expecting to have trouble filling or taking the antibiotic, or greater than twice-daily dosing of antibiotic would identify over 80% of patients who become nonadherent. Our observations fit with others,38,39 suggesting that unknown factors make this racial group especially vulnerable. The current finding that those who expected to have difficulty either filling or taking their prescriptions were more nonadherent is consistent with the Theory of Planned Behavior, where intention is influenced by perceived self-efficacy and barriers.40 The observation that those with greater than twice-daily dosing were more likely to be nonadherent is consistent with other studies examining the burden of dosing schedules on patient adherence.41


  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Limitations
  7. Conclusions
  8. Acknowledgments
  9. References

First, our results are based on self-report, which can either over- or underrepresent the true adherence rates,20 but in general have been found to be valid.21,22 We cannot rule out bias related to social desirability, which could have underestimated the amount of nonadherence. We do not believe that this was a significant factor given the similar rates on nonadherence reported through SMS and phone follow-up. It is also possible that SMS may have improved recall of nonadherence at follow-up, potentially minimizing the measured difference between groups. Second, we experienced a high attrition rate, which may have led to either a under- or overestimation of adherence rates. Although attrition is not unusual in ED-based trials, 28% missing at less than 14 days is a potential threat to validity. Third, there may be patient factors associated with nonadherence that we did not measure, such as cognitive impairment,42 depression,43 or health literacy.44 Fourth, we did not inform our design by qualitative research prior to system design, which could have resulted in a suboptimal deployment. We believe that this was not a major factor given the high level of perceived usefulness reported among intervention participants. Fifth, we had group imbalances that occurred despite randomization. A higher proportion of intervention participants prescribed twice-daily dosing would favor higher adherence. A higher rate of urinary tract infections and lower rate of dental infections in intervention participants would have an unknown effect on adherence. Finally, we did not directly assess whether patient nonadherence was due to medication changes made by the primary care doctor. We believe that this is unlikely given that there were no differences between groups in the proportion who had communicated with their physicians, and local practice is for patients to continue antibiotic course through completion.


  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Limitations
  7. Conclusions
  8. Acknowledgments
  9. References

In this preliminary study, self-monitoring and basic feedback on prescription pickup and dosage taken using short message service alone did not alter adherence, although it improved contact with patients after ED discharge. Certain baseline ED factors are more associated with nonadherence. Future work should improve the design and deployment of SMS interventions to optimize its effect on improving adherence to medication after ED discharge.


  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Limitations
  7. Conclusions
  8. Acknowledgments
  9. References

The authors acknowledge Jack Doman in the Office of Academic Computing at the Western Psychiatric Institute at the University of Pittsburgh, for all computer programming support.


  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Limitations
  7. Conclusions
  8. Acknowledgments
  9. References
  • 1
    Engel KG, Heisler M, Smith DM, Robinson CH, Forman JH, Ubel PA. Patient comprehension of emergency department care and instructions: are patients aware of when they do not understand? Ann Emerg Med. 2009; 53:45461.
  • 2
    Forster AJ, Rose NG, van Walraven C, Stiell I. Adverse events following an emergency department visit. Qual Saf Health Care. 2007; 16:1722.
  • 3
    Thomas EJ, Burstin H, O’Neil A, et al. Patient noncompliance with medical advice after the emergency department visit. Ann Emerg Med. 1996; 27:4955.
  • 4
    Lam F, Stevenson FA, Britten N, et al. Adherence to antibiotics prescribed in an accident and emergency department: the influence of consultation factors. Eur J Emerg Med. 2001; 8:1818.
  • 5
    Horwitz RI, Horwitz SM. Adherence to treatment and health outcomes. Arch Intern Med. 1993; 153:18638.
  • 6
    Hohl CM, Abu-Laban RB, Brubacher JR, et al. Adherence to emergency department discharge prescriptions. CJEM. 2009; 11:1318.
  • 7
    Blaschke TF, Osterberg L, Vrijens B, Urquhart J. Adherence to medications: insights arising from studies on the unreliable link between prescribed and actual drug dosing histories. Ann Rev Pharmacol Toxicol. 2012; 52:275301.
  • 8
    Haynes RB, Ackloo E, Sahota N, McDonald HP, Yao X. Interventions for enhancing medication adherence. Cochrane Database Syst Rev. 2008; Apr 16(2):CD000011.
  • 9
    Cole-Lewis H, Kershaw T. Text messaging as a tool for behavior change in disease prevention and management. Epidemiol Rev. 2010; 32:5669.
  • 10
    Pew Research Center. Americans and Text Messaging. Available at: Accessed May 21, 2012.
  • 11
    Heron KE, Smyth JM. Ecological momentary interventions: Incorporating mobile technology into psychosocial and health behaviour treatments. Br J Health Psychol. 2010; 15(Pt 1):139.
  • 12
    Petrie KJ, Perry K, Broadbent E, Weinman J. A text message programme designed to modify patients’ illness and treatment beliefs improves self-reported adherence to asthma preventer medication. Br J Health Psychol. 2012; 17:7484.
  • 13
    Pop-Eleches C, Thirumurthy H, Habyarimana JP, et al. Mobile phone technologies improve adherence to antiretroviral treatment in a resource-limited setting: a randomized controlled trial of text message reminders. AIDS. 2011; 25:82534.
  • 14
    Dick JJ, Nundy S, Solomon MC, Bishop KN, Chin MH, Peek ME. Feasibility and usability of a text message-based program for diabetes self-management in an urban African-American population. J Diabetes Sci Technol. 2011; 5:124654.
  • 15
    Raofi S, Schappert SM. Medication therapy in ambulatory medical care: United States, 2003-04. Vital Health Stat 13. 2006; (163):140.
  • 16
    Claxton AJ, Cramer J, Pierce C. A systematic review of the associations between dose regimens and medication compliance. Clin Ther. 2001; 23:1296310.
  • 17
    Glanz K, Rimer BR, Lewis FM (eds). Health Behavior and Health Education: Theory, Research, and Practice. San Francisco, CA: Jossey-Bass, 2008.
  • 18
    Feldman R, Bacher M, Campbell N, Drover A, Chockalingam A. Adherence to pharmacologic management of hypertension. Can J Public Health. 1998; 89:I168.
  • 19
    Suffoletto B, Callaway C, Kristan J, Kraemer K, Clark DB. Text-message-based drinking assessments and brief interventions for young adults discharged from the emergency department. Alcohol Clin Exp Res. 2012; 36:55260.
  • 20
    Haynes RB, Taylor DW, Sackett DL, Gibson ES, Bernholz CD, Mukherjee J. Can simple clinical measurements detect patient noncompliance? Hypertension. 1980; 2:75764.
  • 21
    Walsh JC, Mandalia S, Gazzard BG. Responses to a 1 month self-report on adherence to antiretroviral therapy are consistent with electronic data and virological treatment outcome. AIDS. 2002; 16:26977.
  • 22
    Shi L, Liu J, Koleva Y, Fonseca V, Kalsekar A, Pawaskar M. Concordance of adherence measurement using self-reported adherence questionnaires and medication monitoring devices. Pharmacoeconomics. 2010; 28:1097107.
  • 23
    Park LC, Lipman RS. A comparison of patient dosage deviation reports with pill counts. Psychopharmacologia. 1964; 6:299302.
  • 24
    Morisky DE, Green LW, Levine DM. Concurrent and predictive validity of a self-reported measurement of medication adherence and long-term predictive validity of blood pressure control. Med Care. 1986; 24:6774.
  • 25
    Colcher IS, Bass JW. Penicillin treatment of streptococcal pharyngitis: a comparison of schedules and the role of specific counseling. JAMA. 1972; 222:65759.
  • 26
    Peduzzi P, Concato J, Kemper E, Holford TR, Feinstein AR. A simulation study of the number of events per variable in logistic regression analysis. J Clin Epidemiol. 1996; 49:13739.
  • 27
    Farmer KC. Methods for measuring and monitoring medication regimen adherence in clinical trials and clinical practice. Clin Ther. 1999; 21:107490.
  • 28
    Ginde AA, Von Harz BC, Turnbow D, Lewis LM. The effect of ED prescription dispensing on patient compliance. Am J Emerg Med. 2003; 21:3135.
  • 29
    Al-Eidan FA, McElnay JC, Scott mg, McConnell JB. Management of Helicobacter pylori eradication--the influence of structured counseling and follow-up. Br J Clin Pharmacol. 2002; 53:16371.
  • 30
    Wei J, Hollin I, Kachnowski S. A review of the use of mobile phone text messaging in clinical and healthy behaviour interventions. J Telemed Telecare. 2011; 17:418.
  • 31
    Mao Y, Zhang Y, Zhai S. Mobile phone text messaging for pharmaceutical care in a hospital in China. J Telemed Telecare. 2008; 14:4104.
  • 32
    Ollivier L, Romand O, Marimoutou C. Use of short message service (SMS) to improve malaria chemoprophylaxis compliance after returning from a malaria endemic area. Malar J. 2009; 8:e236.
  • 33
    Dunbar PJ, Madigan D, Grohskopf LA. A two-way messaging system to enhance antiretroviral adherence. J Am Med Inform Assoc. 2003; 10:115.
  • 34
    Strandbygaard U, Thomsen SF, Backer V. A daily SMS reminder increases adherence to asthma treatment: a three-month follow-up study. Respir Med. 2010; 104:16671.
  • 35
    Márquez Contreras E, de la Figuera vonWichmann M, Gil Guille′n V. Effectiveness of an intervention to provide information to patients with hypertension as short text messages and reminders sent to their mobile phone (HTA-Alert). Aten Primaria. 2004; 34:399405.
  • 36
    Miloh T, Annunziato R, Arnon R. Improved adherence and outcomes for pediatric liver transplant recipients by using text messaging. Pediatrics. 2009; 124:e84450.
  • 37
    Cocosila M, Archer N, Haynes RB, Yuan Y. Can wireless text messaging improve adherence to preventive activities? Results of a randomized controlled trial. Int J Med Inform. 2009; 78:2308.
  • 38
    Gerber BS, Cho YI, Arozullah AM, Lee SY. Racial differences in medication adherence: a cross-sectional study of Medicare enrollees. Am J Geriatr Pharmacother. 2010; 8:13645.
  • 39
    Gellad WF, Haas JS, Safran DG. Race/ethnicity and nonadherence to prescription medications among seniors: results of a national study. J Gen Intern Med. 2007; 22:15728.
  • 40
    Ajzen I. The theory of planned behavior. Org Behav Hum Decis Process. 1991; 50:179211.
  • 41
    Wagner JH, Justice AC, Chesney M, et al. Patient- and provider-reported adherence: toward a clinically useful approach to measuring antiretroviral adherence. J Clin Epidemiol. 2001; 54(Suppl 1):S918.
  • 42
    van Servellen G, Chang B, Garcia L, Lombardi E. Individual and system level factors associated with treatment nonadherence in human immunodeficiency virus infected men and women. AIDS Patient Care STDS. 2002; 16:26981.
  • 43
    Stilley CS, Sereika S, Muldoon MF, Ryan CM, Dunbar-Jacob J. Psychological and cognitive function: predictors of adherence with cholesterol lowering treatment. Ann Behav Med. 2004; 27:11724.
  • 44
    Giaj Levra N, Cuniberti FA, Rava A, Vietti G, Sciascia S. Health literacy and discharge instruction adherence [letter]. J Gen Intern Med. 2012; 27:273.