Background
Poor adherence to treatment is an important problem in the management of chronic diseases [
1]. Non-adherence is widely prevalent, yet, frequently under-recognized and is associated with higher mortality and morbidity, as well as increased treatment costs [
2]. Non-adherence is multi-dimensional and determined by five major interacting domains; namely socio-economic, health care system related, therapy related, disease related and patient related factors [
1]. Poor socio-economic status, complexity of treatment regimen, poor health literacy and comorbidities (such as depression and cognitive impairment) predispose patients with end stage kidney disease (ESKD) on dialysis to become non-adherent with their medications [
3,
4]. The problem can be particularly challenging, when poor adherence doesn’t have any noticeable short-term effect on symptoms [
5]. Poor treatment adherence is predictive of increased mortality among dialysis patients [
4], but the reporting of adherence in clinical trials involving dialysis patients is inconsistent.
Cardiovascular mortality is 10–20 fold greater in dialysis patients, compared to age and sex-matched controls without chronic kidney disease (CKD) [
6]. Traditional risk factors account for up to 50% of cardiovascular disease in CKD [
7], while non-traditional factors unique to renal disease, like anemia, disordered bone mineral metabolism and oxidative stress, also contribute to poor cardiovascular outcomes. Trials evaluating cholesterol lowering medications like HMG-CoA reductase inhibitors, which have proven efficacy in reducing cardiovascular outcomes in the general population, have shown no significant benefits in patient on dialysis [
8‐
11]. Modification of the risk factors like correction of anaemia [
12], homocystine lowering therapies [
13], treatment with omega-3 fatty acids [
14], control of hyperphosphataemia [
15,
16], and treatment of secondary hyperparathyroidism [
17] have also shown no consistent benefit in improving cardiovascular mortality or significant clinical events in dialysis patients. To explain this lack of efficacy of cardio-protective pharmacological interventions, it has been suggested that the pathogenesis of cardiovascular disease in ESKD, might be different from that in the general population, making it less amenable to interventions [
6]. Whether poor medication adherence contributes to the lack of efficacy of these pharmacological interventions is unknown. In clinical trial settings, a high frequency of non-adherence (i.e. failure to adhere to prescribed treatments) can result in failure to detect a true difference, due to the loss of statistical power [
18]. In addition, a high frequency of study drug discontinuation, which can be due to poor treatment adherence as well as several factors, such as adverse events, drop-out from the study or withdrawal due to protocol specified events like kidney transplantation, can also lead to a false negative study outcome due to loss of statistical power. Consistent reporting of the causes of drug discontinuation is needed to compare studies with respect to the contribution of non-adherence to discontinuation and evaluate their impact on clinical outcomes.
In this review, we sought to examine whether the important issue of adherence to prescribed treatment and study drug discontinuation were adequately and consistently assessed, reported and appropriately addressed in the randomized clinical trials (RCTs) evaluating self-administered cardioprotective medications compared to controls (placebo, another active medication or usual care) in improving cardiovascular or mortality outcomes in patients undergoing dialysis.
Methods
We included all RCTs published as full-text journal articles, over a ten-year period (2005–2015) in this systematic review. The time period was chosen because of the improved awareness of the need to monitor medication adherence in clinical outcomes of intervention trials in recent years. The studies that investigated the effect of any self-administered pharmacological treatment in ESKD patients undergoing dialysis, and reported clinical cardiovascular events or mortality, as the pre-specified primary or secondary outcomes were included.
a. Search strategy
Electronic database searches were performed in MEDLINE, EMBASE and Cochrane CENTRAL register of controlled trials for articles published in English, from 2005 onwards using standard search strategies. Medical subject headings included: ‘clinical trial’, ‘trial’, ‘randomized trial’, ‘single blind’ or ‘double blind’; ‘cardiovascular disease’, ‘cardiovascular outcome’, or ‘mortality’; and ‘dialysis’, ‘renal dialysis’ or ‘peritoneal dialysis’, with the search limited to between 1st January 2005-31st December 2015. Search results in the form of titles and abstracts were analyzed by two authors (KM, JC) to ensure inter-rater agreement, regarding which studies to include in the final review, based on inclusion criteria outlined below. Any disagreement was resolved by discussion among all authors. References within included articles and other important reviews regarding the topic were hand-searched to identify reports that might have been missed in the previous search.
b. Study selection criteria and characteristics
Studies published as full-text articles that included ESKD patients undergoing haemodialysis or peritoneal dialysis alone were considered eligible. Trials, that recruited both dialysis and non-dialysis patients were included, only if the article provided information on the mortality or cardiovascular outcomes for the sub-group of participants on dialysis. This review included only trials comparing a self-administered pharmacological intervention to a control therapy (placebo, another active therapy or usual care). The pre-specified primary or secondary outcomes had to report at least one clinical cardiovascular outcome, which could include fatal or non-fatal cardiovascular events, a composite or death due to any cause. Studies reporting surrogate cardiovascular endpoints, including radiological (e.g. vascular calcification) or biochemical markers (e.g. troponin levels) of cardiovascular disease, as the only primary or secondary outcome were excluded.
c. Data abstraction and synthesis
A standard check-list, for specific data as described below, was used to abstract information from the included studies. Two authors (KM, JC) abstracted data to generate independent datasheets, comprising of quantitative and qualitative information, which were compared to verify inter-rater agreement.
The abstracted data from each study included the: year of publication; journal; first author’s surname; funding source; study acronym; study period; number of participants in the intervention and control arms; study population; inclusion and exclusion criteria; the trialed intervention and control treatments; primary and secondary outcomes; randomization method; information on allocation concealment; blinding of participant, investigator and/or outcome assessment; analysis type (e.g. intention to treat); completeness of outcome data; likelihood of selective reporting; follow-up duration; drop-out rate; whether the study was positive or negative for the outcome of interest; significant secondary outcomes; and reported death from all causes.
We specifically examined whether medication adherence was evaluated, reported and addressed in the included trials. If reported, the method of measuring adherence and its prevalence were assessed. Any method of measurement of individual patient’s adherence was considered acceptable. We recorded the number of subjects discontinuing study medication during the course of the study, reasons for study medication withdrawal and whether non-adherence was identified as a contributor. Where medication adherence was not reported in an article, we contacted the authors to understand whether it was evaluated in the trial. All authors were asked the same question: whether medication adherence was assessed, if yes, what method was used and what was the reported level of adherence.
d. Statistical methods
Inter-rater reliability was assessed using Cohen’s kappa statistics. The average medication adherence and study drug discontinuation from the included trials were reported as a mean percentage. Proportions were expressed as percentages and Fisher’s exact test was used to compare proportions. The analyses were conducted using Stata® version 12.1.
Discussion
In this systematic review, we sought to examine, how the issue of medication adherence was assessed, reported and addressed in dialysis patient trials evaluating cardiovascular or mortality outcomes. Non-adherence to therapy is important in the treatment of dialysis patients, because, at the individual patient level, it can lead to poor clinical outcomes [
4] and in a clinical trial setting, a high degree of non-adherence can lead to failure to detect a true treatment effect [
18]. To our knowledge this is the first systematic review, exploring the problem of non-adherence in dialysis patient trials and we have noted striking inconsistencies and inadequacies in the way in which medication adherence was reported, assessed and addressed in the eligible trials.
We noted that only 27% of the included trials have measured medication adherence to any extent and 23% reported the results of adherence based on medication possession. The low prevalence of individual patient level adherence reporting, probably reflects a failure to recognize the importance of treatment adherence as a major factor influencing clinical outcomes. It is also possible that in many of the trials cited in this review, medication adherence was actually measured, but not reported, as part of the findings. The adherence reporting, we observed in dialysis patient trials is somewhat consistent with the systematic review findings of Gossec et al. [
18], evaluating the treatment adherence in six chronic diseases; namely HIV, Diabetes, Rheumatoid arthritis, Asthma, Hypertension and Osteoporosis. They found that medication adherence was assessed in 33% of the included trials, while only 25% of the trials provided results of adherence.
Consolidated Standards of Reporting Trials (CONSORT) was developed in 1996 and updated in 2001 and 2010 to improve the quality of reporting of RCTs [
34]. The consensus statement has highlighted the importance of distinguishing attrition, as a result of loss to follow up, which is often unavoidable, from exclusions due to other reasons such as withdrawal from treatment and poor adherence to trial intervention [
34]. The CONSORT flow diagram illustrating patient flow through the trial including components of attrition is frequently presented in publications, but the information given is often not detailed enough to ascertain the true extent and nature of non-adherence [
35]. The flow diagram was provided in most of the included trials but the information was highly inconsistent making it difficult to compare between studies.
In our review, we observed that discontinuation of study medication was common but the reporting of reasons for discontinuation were not consistent between studies. The cited reasons like ‘patient choice’, ‘administrative reasons’ and ‘other reasons’ reported in the included trials appear to classify elements of non-adherence under different categories. This makes it difficult to get an accurate estimate of this problem in any given study and to compare these estimates between studies. Our review suggests that trials with a study-drug discontinuation of over 20% are more likely to yield negative study outcomes. The loss of statistical power due to high drop-out or drug discontinuation can lead to false negative outcome results [
18]. In this review, since there was no consistent reporting of the causes of drug discontinuation between studies and only a small proportion of studies reported measuring adherence, we were unable to estimate the true prevalence of non-adherence and its contribution to discontinuation between studies. This made the assessment of their impact on study outcomes virtually impossible.
Some degree of non-adherence is inevitable during the conduct of any intervention trial. Addressing non-adherence can be considered in the design, conduct or analysis phase of the trial. Excluding patients who are likely to be non-adherent, is the most efficient strategy in the design phase and this was utilized by four [
11,
14,
21,
30] trials included in this review. Though patients who participate in the trial may be more motivated to adhere to the prescribed treatment than those in the general population, the intensity of the trial protocol may precipitate non-adherence [
35]. Liaising with the patient’s caregivers to elicit a history of poor treatment compliance has been used as a screening strategy, but the inherent difficulty in recognizing adherence in routine clinical practice may reduce its reliability. Screening during a run-in phase before randomization to unmask non-adherent behavior to exclude non-compliers, is another approach and was reported to have been used in one (8) of the studies included in this review. However, these methods are not foolproof and there is no guarantee that patients selected in this manner will remain adherent to medications throughout the RCT study period.
Efforts to increase the medication compliance during the conduct of the study in dialysis patients pose several challenges. Dialysis patients are frequently frail and chronically ill with several comorbidities and a heavy pill burden, which predispose to drop out due to trial fatigue [
17]. Increasing complexity of treatment is an important factor that precipitates non-adherence [
1]. These factors are highly relevant to the participants in the current systematic review.
Methods to address the effect of non-compliance in the analysis phase of the trial are prone to bias. In our review, 91% of the included trials were analyzed as “intention to treat”. When the level of non-adherence is high, the principle of assigning success or failure to an intervention, which was never received by the subject has some limitations. However, analysis by actual treatment received (TR) invalidates the assumptions underlying randomization and thereby the probabilistic meaning of reported p-values [
36]. Despite this serious limitation, analysis by TR has been tried in several forms in trials where non-adherence is an issue: a) non-compliers can be counted by the treatment they actually received (‘naïve’ TR); b) non-compliers can be excluded; or c) non-compliers can be treated as censored at the time or shortly after they have stopped the treatment being tested [
36]. One of the studies [
17] included in our review has reported analysis with lag censoring, where data was censored six months after participants discontinued the study drug, and showed significant improvements in hazards of the primary composite outcome for the active treatment, while the ‘intention to treat’ analysis was negative. Estimators of the effect size in analysis with lag censoring, may however be biased, as analysis by TR is constrained by the same limitations as in observational epidemiology, such as confounding [
36]. Nevertheless, in the setting of high trial drug discontinuation, especially for non-protocol specified reasons, such pragmatic approaches should be considered in context.
It is important to understand the difference between “efficacy”, which implies whether a specific intervention works under ideal circumstances and “effectiveness” which denotes its effect in the ‘real-world’. It could be argued that non-adherence is a "real-world" issue and in order to understand how drugs perform in the real world, it may be necessary to allow for non-adherence to occur in a clinical trial, as it occurs in usual clinical practice. However, failure to recognize and account for non-adherence in a clinical trial setting, especially when it is frequent, can mask the efficacy of the intervention being investigated. Furthermore, if the level of non-adherence is recognized to be higher than originally thought during the conduct of the clinical trial, false negative outcomes could potentially be avoided by increasing the sample size, if feasible or extending duration of study follow up.
If a specific drug is less acceptable to the patient and promotes non-adherence for this reason, its effectiveness in the ‘real-world’ is going to be lower than the “efficacy” demonstrated in a clinical trial setting.
Our study has both strengths and limitations. One of the major strengths of this review is that it is the first analysis of adherence reporting in randomized control trials evaluating cardiovascular or mortality outcomes in dialysis patients. Another strength is that our review examines the means to address the vexing problem of non-adherence in the setting of dialysis patient trials. From a limitations perspective, the number of eligible studies included in our review was small and the overall reporting of adherence was even smaller. The inconsistency in reporting of adherence and causes of trial drug discontinuation made it difficult to compare studies and combined with the varied nature of the pharmacological interventions made it meaningless to derive a pooled estimate. We would recommend the adoption of a more comprehensive and uniform approach to evaluating and reporting non-adherence in future clinical trials to assess its impact on outcomes. This should include the development of a broadly acceptable definition of non-adherence, consistent methodologies (like pill count) to measure the problem and routine reporting of measured adherence similar to other standard items reported as per CONSORT guidelines. Defining medication adherence as the intake of more than 80% of the prescribed treatment, as done by Baigent et al. [
9] may be acceptable for most situations, but a blanket approach is not appropriate – for instance, drugs like anti-HIV medications and immunosuppressant drugs would warrant more stringent criteria. We also recommend a standardized approach to reporting causes of trial drug discontinuation, which will help us to compare the impact of different causes of therapy discontinuation on outcomes between different trials. Adverse events, which may or may not be related to the medication, are important causes of non-adherence and consistent reporting of the causes of non-adherence is the only way to evaluate their contribution to this problem. Considering the heterogeneous nature of the problem of adherence and treatment discontinuation, these strategies pose difficult challenges, but are nevertheless possible to achieve.