Scoping Review of Randomized Trials With Discontinuation of Medicines in Older Adults

We systematically searched PubMed (coverage 1966 to present), Embase (coverage 1947 to present), and the Cochrane Library (coverage 1992 to present) for publications on randomized discontinuation trials in older patients defined as patients aged ≥65 years. The search was last updated on the 11th of June 2021. The search strings are available in Supplementary Material 2.

All references were imported into Covidence, where 2 authors (J.K. and C.B./A.M.S.S.) independently screened abstracts and full texts and included studies according to the eligibility criteria. Consensus was reached through discussion or through judgment of a third author (M.B.C.).

The eligibility criteria were as follows:

• •
  Randomized controlled trials investigating the discontinuation of a single medicine or a single medicine class, where medicine class was defined as medicines having a similar pharmacologic effect as judged by the authors
• •
  Mean age ≥ 65 years in total or in 1 or more arms

The exclusion criteria were as follows:

• •
  No randomized discontinuation intervention
• •
  Full text not in English
• •
  More than 1 medicine or medicine class discontinued
• •
  Discontinued medicines targeting cancer, transplant complications, or infections, because these treatments are highly specialized or of limited duration
• •
  Not published as full-length, peer-reviewed manuscript (eg, poster publications or clinical trial registrations)
• •
  Temporary discontinuation, that is, pausing of medicines

We limited the studies to discontinuation of a single medicine or a single medicine class with similar pharmacologic effect for ease of interpretability. As an example, antihypertensives cover a diverse range of pharmacologic effects and the choice of antihypertensive medicine for older patients may be influenced by the patient's comorbidities, other concurrent medication, frailty and fall risk, which makes generalizing the feasibility of discontinuation of such a medicine class difficult.

Data were extracted to a customized Google Sheet, and aggregated with the tidyverse package in R via RStudio. Figures were produced using ggplot2 and plotly in R.

The extracted data items included author, publication year, number of patients, discontinued medicine or medicine class, population, setting and treatment indication, study design including blinding, “feasibility of discontinuation,” other relevant results such as particular treatment-related efficacy/harms, and comments (eg, limitations to the study).

To describe the “feasibility of discontinuation,” we extracted 2 measures: (1) the “dropout” rate defined as the proportion of patients who did not complete the entire study per protocol in the discontinuation and continuation groups; and (2) the “disease recurrence” rate defined as the proportion of patients with disease appearance for preventive treatments (eg, fractures after discontinuation of bisphosphonate treatment) or clinical worsening of the treated disease only for primarily symptomatic treatments (eg, pain after opioid treatment withdrawal) in the discontinuation and continuation group. Although adverse drug withdrawal events (ADWEs) can be due to either disease recurrence or physiological reactions to the discontinuation of medicines, this measure focused on disease recurrence as the dropout rate should capture serious ADWEs due to physiological reactions. Thus, if beta-blockers are used to treat congestive heart failure and a patient experiences worsening in heart failure, then it would be considered disease recurrence. In contrast, if the patient experiences tachycardia then it is not, but it may be captured in the dropout rate if the patient exits the study because of the ADWE. To identify disease recurrence, we categorized treatment as primarily “symptomatic” if used to alleviate symptoms present (eg, fentanyl for pain), “preventive” if primarily used to prevent future events in a condition that is asymptomatic (eg, statins for prevention of cardiovascular events), or both “symptomatic and preventive” (eg, selective serotonin reuptake inhibitors for depression) as judged by the authors. We defined a difference as less than 10% in both dropout and disease recurrence rate as “small.” For trials with more treatment groups than just randomized discontinuation and continuation, we only extracted and present data from the randomized discontinuation and continuation groups.

Existing risk of bias evaluation tools were not suitable in their entirety. We used the first domain from Cochrane Risk of Bias tool 2 to assess the risk of bias arising from the randomization process. In addition, we assessed the comparator bias that could lead to a differential dropout rate by answering the following 4 questions: (1) Were the discontinued medicine(s) primarily symptomatic, preventive or both? Our hypothesis was that symptomatic treatments are more prone to differential dropout rates because of poor blinding. (2) Was tapering needed and if so, was the tapering strategy adequate to avoid withdrawal symptoms? (3) Was the study adequately blinded? and (4) Any other design choices which may result in differential dropout rates? Two authors (J.K. and A.M.S.S.) independently answered these questions for each trial and assessed the risk of bias as “low risk,” “some concern,” or “high risk.” In general, nonblinded studies were classified as high risk for symptomatic medicines and some concern for preventive medicines, studies with inadequate tapering were high risk, and studies where tapering was needed and mentioned, but not described in detail, were classified as some concern. Consensus was reached through discussion or via the judgment of a third author (M.B.C.).

To identify disease areas and medicines where discontinuation trials exist, we aggregated the data on medical specialty, indication for the treatment, and the medicine(s) discontinued. To provide an overview of the feasibility of discontinuation of the individual medicines, we plotted the dropout rates and disease recurrence rates in the discontinuation and continuation group, the size of the studies, and the risk of bias. To identify treatment areas where discontinuation may not be feasible, we compared the difference between the discontinuation and the continuation group for both the dropout and disease recurrence rate. A priori, the dropout and disease recurrence rates were expected to be greater in the discontinuation group compared with the continuation group. Therefore, the differences in these rates were compared using a 1-sided Fisher exact test in R version 3.6.3, and a 1-sided P < .05 was considered statistically significant. To identify treatment areas where discontinuation may be feasible, we extracted the trials where the above-mentioned differences were both statistically insignificant and where the absolute difference between groups for both rates was less than or equal to 10%, corresponding to a number to discontinue of 10 for 1 additional dropout and/or 1 additional disease recurrence. Other relevant outcomes could not be readily synthesized and are presented narratively or in tables.

In total, we screened 3304 records and included 41 publications comprising 40 unique trials.^{,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,  ,}  See Figure 1 for study selection.

An overview of the included trials is presented in Figure 2, where the medical specialty, indication for the treatment, and the discontinued medicine(s) are visualized. Additional information on the included trials is available in Supplementary Table 1. Of the included trials, 65% (26/40) of the discontinued medicines were primarily symptomatic, 15% (6/40) were primarily preventive, and 20% (8/40) were both preventive and symptomatic. The setting of the patient recruitment was outpatient clinics (20/40 = 50%); nursing homes (5/40 = 12.5%); the community, for example, using advertisements (5/40 = 12.5%); general practitioners (4/10 = 10%); unknown, that is, multicenter studies with no description of the setting (4/40 = 10%); hospital admissions (1/40 = 2.5%); and palliative care (1/40 = 2.5%).

The designs of the 40 included trials are presented in Figure 3 (designs A-E) along with the designs of the 64 excluded trials (designs F-J). Most of the included trials (20/40 = 50%) randomized patients already on the study medicine (design A), 10% (4/40) were an extension to a parent trial with randomized discontinuation for patients on the study medicine (design B), 15% (6/40) were randomized discontinuation of 1 medicine after combination therapy with 2 medicines (design C), 23% (9/40) were randomized discontinuation of the study medicine for patients with initial effect of the study medicine during a run-in phase (design D), and 1 trial (3%) was a crossover trial for patients already on the study medicine (design E). Of the included trials, 25% (10/40) were nonblinded and 75% (30/40) were placebo-controlled.

The bias assessments are presented in Supplementary Table 2. In summary, 5% of the trials (2/40) had high risk of bias regarding randomization whereas 45% (18/35) had “some concern” (most of them due to an inadequate description of the randomization process). Regarding differential dropout rates, 48% (19/40) of the trials had bias disfavoring the discontinuation group. Of these 19 trials, 53% (10/19) had high risk of bias, while 47% (9/19) had “some concern.” There was low risk of bias across both domains in 25% of the trials (10/40).

The disease recurrence rate was reported in 80% (32/40) of the included trials, and disease recurrence in the discontinuation group compared with the continuation group is presented in Figure 5. The disease recurrence rate varied from 0% in both groups to a maximum of 80% in the discontinuation group and 70% in the continuation group. The difference in disease recurrence rates was statistically significantly greater in the discontinuation group compared with the continuation group in 28% of the trials (9/32). Of these 9 trials, 5 trials also had greater dropout rates in the discontinuation groups and were described above, that is, amantadine, risperidone, fentanyl patch, citalopram, and fluticasone propionate. The remaining 4 trials not previously described investigated discontinuation of tamsulosin for lower urinary tract syndrome, citalopram or sertraline for patients living in nursing homes without a history of depression, SSRIs for neuropsychiatric symptoms due to dementia, and nitrates for angina pectoris. Of the 9 trials, 3 trials had a high risk of bias because of inadequate tapering of SSRIs,^, amantadine, or nonblinded discontinuation of nitrates.

In total, 31 trials reported both dropout rates and disease recurrence rates. In 13 of these 31 trials, the difference in both dropout rate and disease recurrence rate were statistically insignificant and less than 10% for the discontinuation group compared with the continuation group. These 13 trials investigated discontinuation of glucosamine for osteoarthritis, donepezil for mild to moderate dementia due to Alzheimer's disease and with uncertain clinical benefit of the treatment, lithium as adjunct to treatment for unipolar depression, alendronate for osteoporosis,^{,  ,}  statins as primary or secondary prevention of cardiovascular disease in patients with limited life expectancy, droxidopa for neurogenic orthostatic hypotension, tamsulosin for lower urinary tract symptoms, donepezil for dementia and neuropsychiatric symptoms due to Alzheimer's disease, sertraline for patients with major depressive episode, cholinesterase inhibitors for dementia due to Alzheimer's disease in patients living in nursing homes, and fentanyl patch for low back pain or pain due to osteoarthritis.

We identified 40 unique discontinuation trials with inclusion of older patients. Of these trials, 12 trials had dropout rates and/or disease recurrence rates that were statistically significantly higher in the discontinuation group compared with the continuation group, suggesting that discontinuation may not be feasible for these medicines in these populations. Further, we identified 13 trials where the difference in both dropout rates and disease recurrence rates were small (≤10%) and statistically insignificant, suggesting that discontinuation may be feasible. Interestingly, one study investigating discontinuation of nitrates for angina pectoris in stable patients resulted in a statistically significantly higher risk of disease recurrence (8/80 = 10% vs 0/40 = 0%), but with an absolute difference in disease recurrence of 10%. This highlights the complexity of the interpretation of these studies where statistical significance does not always equate clinically relevant differences and underlines why the results presented in this review must be interpreted in the context of the diseases studied. In addition, this example shows that for some of the therapeutic areas where there was a statistically significant difference in dropout rate or disease recurrence, discontinuation attempts may still be feasible in clinical practice, that is, in this study 90% of the attempts were successful and the few disease recurrences were successfully treated with reinstated therapy.

Another complexity highlighted in the present review is the relatively high proportion of patients with disease recurrence in certain therapeutic areas, also in continuation groups (Figure 5); for example, 41% (13/32) of the trials reported disease recurrence rates ≥20% in both groups. A high baseline recurrence risk complicates discontinuation in clinical practice because patients may experience disease recurrence during discontinuation attempts that may in fact be unrelated to the discontinuation attempt, but this is in practice impossible to know and may lead to reinitiation of treatment. This underlines the importance of randomized discontinuation trials as the baseline risk of disease recurrence cannot be estimated from single-arm discontinuation trials.

The identified discontinuation trials covered many of the most prevalent diseases in older patients. However, of the most commonly used medicines in older adults, discontinuation trials were lacking for proton pump inhibitors and platelet inhibitors. Trials on discontinuation of treatment for common diseases such as type 2 diabetes and cardiovascular diseases such as atrial fibrillation were also lacking. In general, trials with discontinuation of purely preventive treatments were not very common (6/40 = 15%). We identified 5 trials with discontinuation of bisphosphonates for the treatment of osteoporosis (5/6 = 83%)^{,  ,  ,  ,}  and only 1 trial investigating the effect of discontinuing statins for patients at the end of life. This is probably due to the difficulty of conducting these trials as they require a long follow-up period. Trials with discontinuation of purely symptomatic medicines are simpler to conduct. Here, any recurrence of symptoms should be reversible with reinstated treatment of the discontinued medicines. It is therefore remarkable that we only identified 2 studies^, with discontinuation trials of medicines for the treatment of pain because analgesics are among the most commonly used medicines for older citizens. Further studies regarding discontinuation of analgesics in older patients seem warranted.

As expected, the included trials were clinically and methodologically diverse with regard to therapeutic area, setting, study design, sample size, reporting of dropouts, harms, and risk of bias. Recently, the need to standardize and optimize the reporting of results from discontinuation trials was highlighted.

Based on this scoping review, identified trials can be divided into 10 different trial designs with an element of discontinuation (Figure 3). The numerous different designs of the included trials are a challenge for the external validity and renders it almost impossible to summarize and meta-analyze the findings. The design that most resembles the clinical situation is where patients are already receiving a given medicine (design A without a run-in period; Figure 3) or a combination of medicines (design C without a run-in period; Figure 3), but changes in prescription patterns over time (eg, new guidelines) or differences in prescription patterns across geographical regions (eg, countries) can limit the external validity of the findings, because the success of discontinuation will depend on how good those who have prescribed the medicine have been to prescribe to the right patients. Compared to these designs, trials where the medicine is started as part of the discontinuation trial (design A and C with a run-in period or design B; Figure 3) may have a more precisely defined population, but also a higher concentration of responders and are thereby more prone to showing lower feasibility of discontinuation. This feature is exacerbated in trials that only include patients who, in the short term, both tolerate the medicine and has apparent effect of the medicine (design D; Figure 3). Lastly, crossover trials with discontinuation (design E; Figure 3) may have problems with carryover effect and the high dropout rate in many discontinuation trials is problematic when analyzing data from these trials.

Our risk of bias assessment revealed that adequate tapering, blinding, and randomization are key when designing future discontinuation studies, and that high dropout rates seem unavoidable even in studies with low risk of bias; for example, overall 45% (17/38) had dropout rates ≥20% in both groups (Figure 4). Therefore, study outcomes that are less affected by high dropout rates should be considered such as time-to-event analyses as seen in Arai et al.

Limitations to this study include the exclusion of trials with patients with a mean age <65 years and the exclusion of trials investigating discontinuation of more than 1 medicine. Another limitation is that the proxies for feasibility of discontinuation we used, that is, disease recurrence rate and dropout rate, only imperfectly describe the feasibility of discontinuation. Disease recurrence, although easy to interpret, is not reported by all trials, and will generally favor continuation because no treatment-related harms (eg, adverse effects) are captured. In contrast, the dropout rate is more difficult to interpret, but this measure is reported univariately and should reflect a conservative estimate of the proportion of patients, where discontinuation is not feasible in clinical practice. As an example of the sometimes conflicting and complementary information these measures convey, a trial with discontinuation of beta-blockers for heart failure in patients with normalized ejection fraction following cardiac resynchronization therapy reported a similar low disease recurrence rate in the continuation and discontinuation groups, but a much higher dropout rate in the discontinuation group (because of withdrawal tachycardia). Thus, although a combination of the 2 measures provide some reassurance for the feasibility of discontinuation, a good measure of the feasibility of discontinuation is lacking.