Electronically Monitored Adherence to Short-Term Antibiotic Therapy in Children

A prospective, blinded, electronically monitored, observational study was performed at Shamir Medical Center, a tertiary medical institution in central Israel. The study was approved by the Shamir Medical Center Institutional Review Board (0274-18-ASF).

Patients between the ages of 2 months and 5 years who were discharged from the department of pediatrics or the pediatric emergency unit after being diagnosed with an acute bacterial infection requiring short-term (5–10 days) oral antibiotic monotherapy with either amoxicillin or cephalexin were recruited between January 2018 and October 2021. The following diagnoses, for which these medications are standard of care, were included: acute otitis media (AOM), pneumonia, cellulitis, pharyngitis, urinary tract infection (UTI), and bacteremia. We excluded children with any concurrent or recent (past month) use of other medications on the basis of their medical history, and patients whose guardians refused antimicrobial treatment. A study investigator reviewed the electronic patient census daily for patients meeting inclusion criteria. Importantly, any treatment decisions were made by the duty physician unrelated to the study. Patients meeting our a priori definitions were provided with a bottle containing the prescribed antibiotic suspension, closed with a Medication Event Monitoring System (MEMS). Administration instructions were written on the prescription bottles. A guardian of the patient was asked to participate in the evaluation of a new type of child-proof medication cap. After agreeing and signing a consent form, the participants were instructed to dose directly from the bottle, use no other container and keep the bottle closed between doses. Unknown to the guardian, the prescribing physician was asked to predict the level of adherence of the patients, and information about physician seniority was noted. In addition, contact details and prescribed treatment duration were logged in an investigator file.

On the day of the last planned dose, an investigator contacted the guardians and arranged an in-person visit. First, information about the antibiotic treatment, as remembered by the parent, was collected, including the name of the antibiotic, treatment duration, and the number of daily doses. The guardian was then asked whether they had administered all prescribed doses and, if not, the reason for nonadherence. Then the true purpose of the research was explained, and permission to use the data recorded in the cap and in the previous short interview was sought. In case of refusal to participate, the adherence data were discarded, and no further information was collected. If the guardians agreed to participate, they were reconsented by using a relevant consent form, the caps were collected, and questionnaires were completed to collect information on select patient characteristics and family demographics. Questions included patient ethnicity, sex, age, previous illnesses, and previous treatments. Parental age, marital status, education, self-defined level of income, occupation and work schedule, number of children, and chronic treatments were also collected.

The MEMS (Aardex, Switzerland) is a cap with an integrated microcircuit recording the time and date whenever the patient opens the bottle. The system has sufficient memory to accumulate information on 4000 actuations, has a failure rate of ∼2% and is accurate to within 30 seconds per month. After collection of the caps, the dosing history data are stored on centralized, secured servers (Supplemental Table 2). To account for redundant cap manipulation, we did a temporal analysis of all registered medication events, and whenever 2 or more events were noted within minutes of each other, they were combined into 1 event.

For the purpose of our study, we defined 2 main measures of adherence: (1) overall adherence, defined as the percentage of doses taken relative to the overall number of doses prescribed by the treating physician, and (2) timing adherence, defined as the proportion of prescribed doses taken within ±20% of the prescribed interval. In our study, all treatment regimens were 2 or 3 times daily (ie, administration every 12 or 8 hours), meaning the prescribed intervals should be 9.5 to 14.5 hours or 6.5 to 9.5 hours, respectively. For both outcomes, we defined good adherence as >75% of doses by consensus agreement of study investigators.

The primary outcomes included overall and timing adherence rates of pediatric patients treated for acute infections with antibiotic suspensions. Secondary outcomes included factors influencing adherence rates, evaluating whether the prescribing physician can accurately predict the adherence rates of their patients, agreement between parental perception of the recommended antibiotic dose and the prescription and reported reasons for nonadherence. For the primary outcome, we hypothesized that only one-half of the pediatric patients treated with antibiotic suspensions for acute bacterial infections take 75% or more of the prescribed doses.

Continuous variables were summarized with mean and standard deviation (SD) or median and interquartile range and compared between groups by using the Mann-Whitney test. Categorical variables were summarized by using frequency and percentages, and the Fisher exact test or χ² test were used to compare differences between groups. A logistic regression model was applied to identify factors related to adherence, as measured by the MEMS. The following variables were included in the model: age, diagnosis, antibiotic type, parental marital status, work hours, partner work hours, income, and parental chronic treatments, with a threshold of P <.1 for inclusion. Age, diagnosis, and antibiotic type were forced in the model. No imputations for missing data were applied and P <.05 was considered statistically significant. Univariate analyses were conducted by using the R Core Team software (2021, Vienna, Austria). Multivariate analyses were conducted by using SPSS-27 software (IBM, Armonk, NY, USA).

One hundred and twenty-three patients meeting all inclusion/exclusion criteria were approached at the time of discharge. Nineteen caregivers (15.4%) refused to participate in the evaluation of the “child-proof” cap. Of the remaining 104, 3 participants discarded the caps before the second enrollment and 1 parent refused to participate in the adherence study and the collected data were erased, resulting in a final sample size of 100 participants (Fig 1).

Of the 100 participants, 49 were male, mean age was 1.87 years (range 0.2–5.1), and 93 were of Jewish ethnicity. The most common diagnosis was AOM (18 participants), followed by cellulitis and UTI (16 and 14, respectively). Sixty-six participants were prescribed amoxicillin, and 34 cephalexin, for an overall mean (SD) of 15.22 (5.87) antibiotic doses to be administered at home (Supplemental Table 2). The recruited guardian was the mother in 83 cases, with 90 being married and 10 being single parents. The mean (range) number of children in the family was 2.55 (1–8) and 46 were from below-average income families. See Table 1 for complete participant and treatment characteristics.

Overall adherence (administration of >75% of prescribed doses) was 62%, whereas timing adherence (administration of >75% of prescribed doses at the recommended time interval) was only 21%. The mean number of administered doses was 11.81 (SD 5.84), whereas the mean number of on-time administered doses was 5.69 (SD 3.98). Only 11 participants received all the recommended doses, whereas 4 children were “over adherent,” receiving more doses than recommended by the prescribing physician.

In the univariate analysis, 3 characteristics significantly differed between overall adherent and nonadherent participants: patient age (mean [SD] age 1.69 years [1.37] vs 2.16 years [1.25]; P = .032], enrolled parents’ work schedule (unemployed/day job/shifts 25.8%/53.2%/21.0% vs 25.0%/72.2%/2.8%; P = .036) and partner work schedule (unemployed/day job/shifts 21.0%/71.0%/8.1% vs; 18.9%/48.6%/32.4%; P = .046) (Table 1). Parental status was of borderline significance (95.2% vs 81.6% married; P = .060). After applying a logistic regression model, the only factor remaining significant was parental status, with single parents having increased odds for nonadherence (odds ratio = 5.7; 95% confidence interval [1.07–30.3]) (Supplemental Tables 3 and 4). For timing adherence, only frequency of administration was significant (timing adherent 20/21 prescribed twice-daily administration vs 43/79 in the nonadherent group, P <.001) (Supplemental Table 5).

Thirty-four participants reported different dosing or duration than recorded in the prescription, with 18 reporting being instructed to give fewer doses than recorded and 16 reporting being asked to give more doses than recorded. When asked about actual medication administration parents reported higher adherence rates than those recorded by the MEMS, with 82 guardians reporting full adherence and only 18 admitting to giving less than 75% of requested doses (P = .001 vs overall adherence, even when calculating overall adherence relative to the treatment regimen perceived by the parent). Of the 18 guardians reporting low adherence, the main reasons stated were stopping treatment early because the child was feeling well (5 cases, 27.7%), forgetting (4 cases, 22.2%), child opposition to treatment (3 cases, 16.3%) and treatment cessation because of side effects of the medication (2 cases, 11.1%). No participant sought additional medical care because of treatment failure or recurrence of symptoms by the day of data collection.

Sixty-three adherence predictions were collected at the time of discharge from the treating physician. Prescribers tended to overestimate adherence, defining 49 (77.7%) participants as likely adherent (P = .039 for the difference vs true overall adherence). A physician prediction of adherence occurred in a similar proportion of actual adherent and nonadherent patient/parent pairs. (31/47 adherent participants predicted to be adherent vs 6/16 nonadherent participants predicted to be adherent, P = .77). Physician seniority was not associated with prediction accuracy.

In this study, we found an overall adherence rate of 62% for short-term antimicrobial therapy in previously healthy children. Previous studies have revealed rates ranging between 25% and 80%, depending on the methodology used, with most using parental surveys or prescription filling data, methods that are highly prone to bias.^14,16–18  To the best of our knowledge, this is the first study using objective electronic monitoring to assess short-term treatment adherence in this patient population.

Although adherence research has largely been focused on chronic conditions and treatments due to the difficulties associated with obtaining prolonged adherence, it is increasingly recognized that nonadherence to acute, short-term interventions can result in serious consequences.^3,4,6  Although we defined good compliance liberally, as administration of 75% of prescribed doses, overall compliance was suboptimal at 62% despite the fact that we recruited patients who were discharged from the pediatric ward or emergency department, meaning the parents were worried enough to present to the hospital. In the past years, as the consequences of superfluous antimicrobial administration are becoming evident, there is a trend to reduce the duration of treatment. Accordingly, an increasing number of clinical studies have revealed the noninferiority of shorter treatment regimens in several acute pediatric infectious conditions. For example, a 5-day treatment course was sufficient for children with uncomplicated pneumonia in 2 recent publications,^24,25  with a single study in a developing country even suggesting a 3-day course.²⁶  Treatment adherence during clinical trials is superior to real-life experience. As treatment regimens are shortened, the potential consequences of even partial nonadherence might be more pronounced. For example, 30% nonadherence in a 5-day, twice-daily dosing regimen will result in the child receiving only 7 doses, with unknown consequences on treatment efficacy. Future adherence studies should also focus on short-term antimicrobial treatments in the pediatric population.

In addition to overall adherence, we report the timing adherence, the concept of adhering to the dosing interval prescribed. Only 48% of the doses administered were given within the recommended time interval, with only 21% of participants administering >75% of doses on time. If timing is taken into account, these results suggest that compliance is much lower than previously reported. The impact of overstepping the recommended time interval depends on the pharmacokinetics and pharmacodynamics of the prescribed antimicrobial. For most β-lactams, the main pharmacodynamic parameter found to be important in predicting efficacy is “time above minimal inhibitory concentration” (MIC), and, thus, less influenced by strict adherence to time intervals. However, many other antimicrobial agents, like fluoroquinolones, macrolides, and glycopeptides, are dependent on “area under the curve” to MIC, in which prolonging the dosing interval may significantly impact efficacy.^27,28  In addition, recent studies have shown that, for some antibiotics, the “post antibiotic recovery period” is critical for bacterial persistence and the development of resistant strains; in other words, prolonging the dosing interval might result in treatment failure and even encourage resistance development.²⁹  On the basis of our results, additional studies about timing adherence are needed to better understand this component of general adherence. In the meantime, we suggest emphasizing the importance of proper dosing intervals when giving medication instructions to patients.

Another interesting finding of our study is the fact that the treating physician not only overestimated adherence, as revealed in previous studies¹¹  but was completely unable to differentiate between adherers and nonadherers. These data imply that any interventional measures to increase adherence in this patient population should be uniformly implemented, with little consideration given to provider input.

Our study has several strengths, most importantly the use of electronic monitoring without the participants’ knowledge, ensuring unbiased medication behavior. Another strength is the extremely low dropout rate, with only 1 participant withdrawing from the study after learning the true purpose of the cap. Finally, because of the personal interaction with participants, we had 100% completion of study questionnaires. Nevertheless, the study has some limitations. First, because of the complexity of our study flow, we limited the sample size to 100 participants. A caveat of electronic medication monitoring is the fact that the cap will register any opening of the package as medication administration. If this was a frequent occurrence, it could potentially result in “over adherence” and is not supported by our results. In addition, to account for redundant cap manipulations, we did a temporal analysis of all registered medication events, and whenever 2 or more events were noted within minutes of each other, they were combined into 1 event. Others have suggested that MEMS are associated with underestimation of adherence.²³  We did not measure residual medication volume to corroborate our electronic data. Third, because we provided patients with the prescribed antibiotic, we limited inclusion to only 2 commonly used medications. Although we found no differences in adherence rates between these medications, we were limited in evaluating the influence of factors like differences in palatability on adherence. Finally, the fact that we provided patients with the prescribed medication eliminated the need for patients to fill the prescription at a pharmacy and removed the step of “primary adherence.” In Israel, all citizens belong to an HMO, pharmacies are easily accessible, and the cost of medication is negligible, thus we do not expect this to have a major impact on our results. In addition, if anything, this would result in an overestimation of adherence rates, whereas we describe low overall adherence in the population.

In the past decade, microelectronic devices have become the gold standard in adherence research. However, studies have suggested that informing the patients about the purpose of the device inherently enhances adherence, resulting in an overestimation of adherence. This significant effect has even been used for the purpose of improving adherence, as an adjunct to direct observed therapy.^24,25  Therefore, we did not inform the patients, at first, of the purpose of the MEMS, and even went so far as to misinform them about the cap. Although this process may be considered by some to be against common Good Clinical Practice, we felt that it was justifiable to obtain valid results. In case of refusal to participate after a full explanation was provided, all collected data were discarded, and no further information was collected. Only 1 participant refused enrollment at this stage of reconsent. Importantly, the study had no influence whatsoever on the workup or treatment of the patients and was found to be ethical by our local review board.

Adherence of children to the short-term antimicrobial treatment of an acute infection is suboptimal, with only 62% showing good adherence and 21% with good adherence at proper timing intervals. Providers were unable to predict the adherence of their patients. These data are important when considering recommended treatment durations and developing interventional programs to increase adherence.

AOM

acute otitis media

MEMS

medication event monitoring system

SD

standard deviation

UTI

urinary tract infection