Mobile Text Messaging and Adherence of Patients to Medication Prescriptions: A txt a dA keeps da doctR awA?

Abstract

Low patientmedication adherence is one of the 2 largest unsolved gaps in health care, with the other being suboptimal therapyprescription. For example, oral anticoagulation (OAC) is firmlyrecommended inpatientswithatrial fibrillationathigh risk for stroke, but typically about half of the eligible patients donot receiveOAC.1Of those patients who receive OAC, approximately half will not begin therapy or will stop therapy within a year,2 and of those patients persisting with an OAC regimen, only half take their pills consistently as prescribed.3As a consequence,most patientswill not reap the full health benefits fromproven therapies such asOAC. There is rising interest inmedication adherence research,4 but these well-meaning efforts have not yet led to substantial and reliable adherence improvements when looking at the entire adherence research field. Are we perhaps missing the signal of specific interventions? In this issueof JAMAInternalMedicine, Thakkar et al5 provideafocusedandthoroughmeta-analysisof randomizedclinical trials that aimed to improve medication adherence in patients with chronic disease by the specific means of mobile telephone textmessaging (TM), amongwhom they observed adoubling of the relative odds of improving adherence. Technology-mediated interventions are logical candidates for innovations inadherenceresearch,especially if theyareaswidely used as mobile telephones. Research testing new technologies naturally lags behind the introduction and uptake of the technology,6 which is why most of the included randomized clinical trialswerepublished in thepast 5years, renderingmobile telephone TM the “new kid on the block.” The kid has shown potential, but—as Thakkar et al5 indicate—caution is warrantedwhen interpreting the results.Most studieshad fundamental methodological weaknesses that are common in medication adherence research to date,7 and several of these shortcomings are lethal to the case of whether TM interventions areworth the effort and resources required to create and conduct them. First, did the pooled studies truly show an improvement in adherence for chronic disease? Most studies used selfreport to measure adherence, but self-report typically overestimatesadherence,especially inunmaskedtrialsas theserandomized clinical trialswere.Unobtrusivemeasurements such as pharmacy claims data would be more objective and accurate. Furthermore, lasting effects are crucial for chronic disease, but these studies had amedian interventionduration of only 12 weeks, and no study reported on adherence beyond the active intervention period. Second, and even more important, measuring a clinical outcome could assess whether the adherence improvement translated into benefits important to the patient, but the randomized clinical trials in the meta-analysis by Thakkar et al5 did not measure such outcomes, nor would it have been feasible to do so in these small and short-term studies. Therefore, TM has not been subjected to the acid test of whether it makes a clinically useful difference. Third, because the meta-analysis focused on one type of intervention, the authors chose to conveniently pool the results using standardized effect sizes, but it is hard to ignore the many differences among the studies. The substantial statistical heterogeneity (I2 = 62%) across the included clinical trials reflects obvious differences in the following characteristics: (1) The clinical conditions varied, with some patients requiring chronic daily lifelong treatment (eg, for human immunodeficiency virus) vs intermittent or seasonal treatment (eg, for allergic rhinitis) vs treatments that are typically required in the long term but may be discontinued after clinical remission (eg, in the case of epilepsy). (2) The medication regimens used differed, including the timing and frequency of intake, absence of noticeable benefits (eg, blood pressure lowering), severity of adverse effects, and route of administration. (3) Different adherence outcomes were used, including self-report, pill count, medication event monitoring systems, and pharmacy dispensing data, reported as continuous measures or dichotomous measures, with cutoffs for good adherence ranging from 80% to 95%. Adding to the adherence outcome heterogeneity is the large variation in follow-up time, and the absence of (or ambiguity about using) intent-to-treat analyses in half of the studies. (4) There was variation in baseline adherence, with only 2 studies using low baseline adherence as an eligibility criterion. Failure to focus on those patients who can benefit from interventions minimizes the effect sizes that can be observed, limits the power of studies to detect differences, and wastes resources. (5) The study settings differed, with a wide variety represented—from hospitals in Africa (9 of the studies were set in developing countries) to the general population in Denmark—indicating substantive differences in cultures, languages, resources, health services, priorities, and personal finances. (6) The intervention design varied, such that different TM approaches were used, with various message frequencies, timing, content, framing, degree of personalization, ability for patients to respond, and degree of automation. These issues of heterogeneity might be mitigated by distinguishing studies based on methodological rigor and relRelated article page 340 Research Original Investigation Text Messaging for Medication Adherence in Chronic Disease